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.pntd.0002773 | T-Cell Regulation in Lepromatous Leprosy | Regulatory T (Treg) cells are known for their role in maintaining self-tolerance and balancing immune reactions in autoimmune diseases and chronic infections. However, regulatory mechanisms can also lead to prolonged survival of pathogens in chronic infections like leprosy and tuberculosis (TB). Despite high humoral responses against Mycobacterium leprae (M. leprae), lepromatous leprosy (LL) patients have the characteristic inability to generate T helper 1 (Th1) responses against the bacterium. In this study, we investigated the unresponsiveness to M. leprae in peripheral blood mononuclear cells (PBMC) of LL patients by analysis of IFN-γ responses to M. leprae before and after depletion of CD25+ cells, by cell subsets analysis of PBMC and by immunohistochemistry of patients' skin lesions. Depletion of CD25+ cells from total PBMC identified two groups of LL patients: 7/18 (38.8%) gained in vitro responsiveness towards M. leprae after depletion of CD25+ cells, which was reversed to M. leprae-specific T-cell unresponsiveness by addition of autologous CD25+ cells. In contrast, 11/18 (61.1%) remained anergic in the absence of CD25+ T-cells. For both groups mitogen-induced IFN-γ was, however, not affected by depletion of CD25+ cells. In M. leprae responding healthy controls, treated lepromatous leprosy (LL) and borderline tuberculoid leprosy (BT) patients, depletion of CD25+ cells only slightly increased the IFN-γ response. Furthermore, cell subset analysis showed significantly higher (p = 0.02) numbers of FoxP3+ CD8+CD25+ T-cells in LL compared to BT patients, whereas confocal microscopy of skin biopsies revealed increased numbers of CD68+CD163+ as well as FoxP3+ cells in lesions of LL compared to tuberculoid and borderline tuberculoid leprosy (TT/BT) lesions. Thus, these data show that CD25+ Treg cells play a role in M. leprae-Th1 unresponsiveness in LL.
| Leprosy is a curable infectious disease caused by Mycobacterium leprae (M. leprae) that affects the skin and peripheral nerves. It is manifested in different forms ranging from self-healing, tuberculoid leprosy (TT) with low bacillary load and high cellular immunity against M. leprae, to lepromatous leprosy (LL) with high bacillary load and high antibody titers to M. leprae antigens. However, LL patients have poor cell mediated response against M. leprae leading to delayed clearance of the bacilli. A possible explanation for this bacterial persistence could lie in the presence of more regulatory cells at infection sites and in peripheral blood. This study shows the recovery of the cell mediated response by depletion of CD25+ cells in a subset of LL patients, while another patient subset was not affected similarly. Moreover, an increased frequency of FoxP3+ T cells together with anti-inflammatory macrophages was observed in LL patients' skin biopsies. Thus, these data show that CD25+ Treg cells play a role in M. leprae-unresponsiveness in leprosy patients.
| The human immune system strives to maintain the delicate balance between preventing host susceptibility to various pathogens and limiting immunopathology due to an exacerbated immune response to infections. Sub-populations of T-cells previously identified as suppressor T-cells and later as Treg cells are the major players in the regulatory network of the immune system [1], [2]. Although the idea of suppressor T-cells was a key topic of research already in the 70's and 80's it was not successfully established because of poor cellular characterization, and it took until mid-1990's before Treg cells were recognized as a different lineage [1]. More recently, studies clearly demonstrated the suppressive ability of this sub-population contributing to the re-acceptance of suppressor T-cell as a different T-cell lineage [3], [4].
Characterization of this T-cell sub-population has continued and currently the thymus-derived Treg cells (tTreg cells) and peripherally derived Treg cells (pTreg cells) [5] are the two widely accepted categories of Treg cells [1], [6], [7]. Both T-cell subtypes play a role in limiting immune reactions in autoimmune diseases and chronic infections [8]–[11]. In addition, CD39+ Treg cells have also been reported as a subset of the CD4+ CD25highFoxP3+ Treg cells in association with chronic infections like tuberculosis (TB) [12], hepatitis B (HBV) and in graft rejections [13], [14] and the ability of CD8+ CD39+ Treg cells to suppress antigen specific CD4+ proliferation clearly demonstrated the importance of this sub-population [15].
Leprosy is a chronic infectious disease leading to more than 200,000 new cases every year [16]. The remarkable inter-individual variability in clinical manifestations of leprosy closely parallels the hosts' abilities to mount effective immune responses to M. leprae. This is clear from the well-known immunological and clinical spectrum in those who progress to disease ranging from polar T helper 1 (Th1) to Th2 responses. TT and BT show more dominant Th1 responses which limit M. leprae growth resulting in clinical paucibacillary (PB) leprosy whereas, BL/LL patients demonstrate dominant Th2 responses as well as more permissive growth of M. leprae resulting in clinical multibacillary (MB) leprosy. TT/BT patients in general show high cellular responses and low antibody titers to M. leprae antigens, and develop localized granuloma with often no detectable bacilli in their lesions. The LL/BL patients at the opposite pole are incapable to generate M. leprae specific Th1 cell responses, show high antibody titers to M. leprae antigens, and poor granuloma formation with numerous bacilli in their lesions. The borderline states of leprosy are immunologically unstable. The different outcomes of infection in leprosy are most likely caused by host defense mechanisms [17]. However, the mechanism underlying the M. leprae-specific T-cell anergy in LL patients is still not completely understood.
In chronic bacterial or viral infections, evidence exists that Treg cells suppress effector T-cells (Teff cells) in order to limit damage to the host caused by the immune responses against pathogens [18]. In this situation, the regulatory activity of Treg cells may lead to prolonged survival of pathogens in the host [9], [19]. As evidenced in a previous study, higher levels of CD4+CD25+FoxP3+ Treg cells were observed in active TB patients in the periphery compared to latently infected individuals and healthy controls [20], [21]. Also, an increased number of Treg cells expressing FoxP3, cytotoxic T-lymphocyte antigen 4 (CTLA-4) and glucocorticoid-induced tumour-necrosis-factor-receptor-related protein (GITR) were reported in lymphnodes from children with tuberculosis lymphadenitis [22]. Similarly, in leprosy, higher numbers of Treg cells in PBMC from BL and LL patients stimulated with M. leprae cell wall antigen (MLCWA) were observed compared to TT/BT forms, indicating the possibility that Treg cells may have a role in persistence of M. leprae bacteria as well as unresponsiveness of Th1 cells in BL/LL patients [23]. Recently, the mechanism of action of FoxP3 in CD4+CD25+ T cells derived from BL/LL leprosy patients was shown to result from increased molecular interactions of FoxP3 with Histone deacetylases (HDAC7/9) in the nucleus of CD4+CD25+ T cells derived from BL/LL patients [24].
In the presence of pathogens, Treg cells can also be induced by certain macrophages as evidenced by the anti-inflammatory, CD163+ macrophages, known as type 2 macrophages (mφ2), that exert a suppressive effect on Th1 responses [25], [26]. On the other hand, IL-10 induced phagocytosis of M. leprae by mφ2 without induction of microbicidal activity in LL lesions has been described [27] indicating the role of IL-10 producing Treg cells in the persistence of the pathogen within the host. Similarly, the presence of higher IL-10 expression correlated with increased CD163 and indoleamine 2,3-dioxygenase (IDO) proteins in tissues and sera of LL patients further evidenced their potential [28].
In this study, we have investigated the functional role of CD25+ Treg cells in M. leprae unresponsiveness of LL patients as well as the frequency of CD25+ and FoxP3+ cells in the PBMC of leprosy patients. Additionally, lesions of LL and TT/BT patients were assessed for the presence of FoxP3+ cells and CD163+ macrophages (mφ2).
Ethical approval of the study protocol was obtained from the National Health Research Ethical Review committee, Ethiopia (NERC # RDHE/127-83/08) and the Nepal Health Research Council (NHRC #751). Participants were informed about the study objectives, the required amount and kind of samples and their right to refuse to take part or withdraw from the study at anytime without consequences for their treatment. Written and Informed consent was obtained from study participants before enrollment.
The following HIV-negative individuals were recruited on a voluntary basis: newly diagnosed, non reactional leprosy patients from Ethiopia (ALERT hospital, Addis Ababa, Ethiopia) classified as LL (n = 40) and TT/BT (n = 16) and healthy endemic controls from health centers in Addis Ababa (EC; n = 5); Treated, non reactional LL (n = 6) and TT/BT (n = 9) patients and EC (n = 10) from Anandaban Hospital, (Kathmandu, Nepal); and non-endemic Dutch healthy controls (NEC; n = 13). Leprosy was diagnosed based on clinical, bacteriological and histological observations and classified by a skin biopsy evaluated according to the Ridley and Jopling classification [17] by qualified microbiologists and pathologists. All patients were enrolled before treatment was initiated. EC were assessed for the absence of clinical signs and symptoms of tuberculosis and leprosy. Individuals working in health facilities were excluded as EC.
PBMC were isolated by Ficoll-Hypaque density gradient method, cells were washed and suspended in 20% fetal calf serum (FCS) in AIM-V (Invitrogen, Carlsbad, CA) and kept cool on ice, counted and frozen using a cold freshly prepared freezing medium composed of 20% FCS, 20% dimethyl sulphoxide (DMSO) in AIM-V. Cells were kept at −80°C for 2–3 days and transferred to liquid nitrogen until use. During thawing, cells were transported in liquid nitrogen to a water bath (37°C) for 30 to 40 seconds until thawed half way and resuspended in 10% FCS in AIM-V (37°C) containing 1/10,000 benzonase until completely thawed, washed 2 times (5–7 minutes each) and counted. The percentage viability obtained was >75% and cells were incubated with anti-CD25 magnetic beads or used for FACS analysis.
Frozen PBMC were thawed, washed and incubated with 20 µl of the CD25 micro beads II, human (Miteny Biotec, Bergisch Gladbach, Germany) in 80 µl MACS buffer (Phosphate-buffered saline (PBS) with 0.5% Bovine serum albumin (BSA) and 2 mM EDTA) for 20 minutes at 4°C. Cells were washed and added to MS column attached to Magnetic Cell Sorter (MACS) (Milteny Biotec) where CD25− cells were collected as flow through and the CD25+ population was collected by detaching the column from the magnetic cell sorter. Cells were washed with MACS buffer and resuspended in AIM-V medium. The purity of the CD25− and CD25+cell populations was >80% (supplementary figure S2A and S2B).
Total PBMC (150,000 cells/well), CD25− cells (150,000 cells/well) or CD25− cells with proportionally added CD25+ cells (10,000 and/or 25,000) were added in triplicate into 96 well U bottom tissue culture plates and cultured with M. leprae whole cell sonicate (WCS; 10 µg/ml), phytohaemagglutinin (PHA; 1 µg/ml) or AIM-V medium at 37°C with 5% CO2 and 70% humidity. After 6 days, supernatants were collected and kept frozen until used in ELISA.
Irradiated armadillo-derived M. leprae whole cells were probe sonicated with a Sanyo sonicator to >95% breakage. This material was kindly provided by Dr. J.S. Spencer through the NIH/NIAID “Leprosy Research Support” Contract N01 AI-25469 from Colorado State University (now available through the Biodefense and Emerging Infections Research Resources Repository listed at (http://www.beiresources.org/TBVTRMResearch Materials/tabid/1431/Default.aspx).
IFN-γ levels were determined by ELISA (U-CyTech, Utrecht, The Netherlands) [29]. The cut-off value to define positive responses was set beforehand at100 pg/ml. The assay sensitivity level was 40 pg/ml. Values for unstimulated cell cultures were typically <40 pg/ml.
After depletion, the total PBMC, CD25− or CD25+ populations (25,000 to 200,000 cells) were stained for CD3 (clone SK7, PerCP; Becton, Dickinson and Company, New Jersey, USA), CD4 (clone SK3, FITC; BD) and CD25 (PE; MACS) to check the purity.
Frozen PBMC of patients and healthy controls (2×106 cells/ml) were thawed, washed and treated with benzonase (10 U/ml, Novagen, Merck4Biosciences, Merck KGaA, Darmstadt, Germany) for 2 hours prior to in vitro stimulation with PMA (20 ng/ml)/ionomycine (500 ng/ml) in the presence of 1 µg/ml anti CD28 (Sanquin, the Netherlands) and 1 µg/ml anti CD49d (BD Biosciences, Eerbodegem, Belgium). After 4 hours, Brefeldin A (Sigma Aldrich) was added at 3 µg/ml and cells were left for an additional 16 hours in the incubator at 37°C with 5% CO2 and 70% humidity. After live/dead staining with Vivid (Invitrogen, Life technologies, Merelbeke, Belgium), surface staining was performed for 30 minutes at 4°C with the labeled antibodies directed against: CD14- and CD19-Pacific Blue, CD3-PE-TexasRed (all Invitrogen, Life technologies), CD8-Horizon V500, CD4-Pe-Cy7, CD25-APC-H7 (all BD Biosciences), and CD39-PE (Biolegend, ITK Diagnostics, Uithoorn, The Netherlands). Samples were washed, fixed and intracellular staining was performed using the intrastain kit (Dako Diagnostics, Glostrup, Denmark) with IFN-γ -Alexa700 (BD Biosciences), IL-10 APC (Miltenyi Biotec GmbH, Bergisch Gladbach, Germany), and FoxP3 PE-Cy5 (eBioscience, Hatfield, UK) labeled antibodies. Cells were acquired on a FACS LSR Fortessa with Diva software (BD Biosciences, The Netherlands) and analyzed with FlowJo version 9.4.1 (Tree Star, Ashland, OR, USA). The full gating strategy for live CD4+ CD3+ cells or CD8+ CD3+ cells (supplementary Figure S1A and S1B) was performed in compliance with the most recent MIATA [30] guidelines according to the following procedure: events were first gated using a forward scatter area (FSC-A) versus height (FSC-H) plot to remove doublets. Subsequently, the events were subjected to a lymphocyte gate using a side scatter (SSC) followed by a live/dead gating. Then, live CD3+ cells were gated and CD14+ and CD19+ events were excluded from analysis using a dump channel. Finally, CD3 live cells were separated in to CD4+ and CD8+. After the gates for each function were created, we used the Boolean gate platform to identify all functions within each cell preparation using the full array of possible combinations.
Skin biopsies taken from leprosy lesions of LL (n = 10) and TT/BT (n = 4) patients were fixed in formalin and embedded in paraffin. Tissue sections with 4 µm thickness were prepared using a microtome (LEICA RM 2165). The prepared tissues sections were stained for hematoxylin and eosine (H & E; images are shown in supplementary figure S3) and also used as previously described [31] for immunofluorescence staining. Tissue sections were deparaffinised and rehydrated using graded concentrations of ethanol to distilled water. Antigen retrieval was performed in boiling Tris-EDTA buffer (10 mM Tris Base, 1 mM EDTA Solution, 0.05% Tween 20, pH 9.0) for 12 minutes. After two hours of cooling at room temperature in antigen retrieval buffer, slides were washed twice in distilled water and twice in PBS, blocked for 15 min with 5% goat serum in PBS, washed again with PBS and stained with primary antibodies for FoxP3 (1∶100, mouse anti-human IgG1 Abcam; Cambridge, UK), CD8 (1∶100 mouse anti-human IgG2b, Abcam), CD68 (mouse anti-human IgG2a AbD serotec/Bio-Rad; Veenendaal, The Netherlands), CD163 (1∶400, mouse anti-human IgG1, Leica; Rijswijk, The Netherlands) and CD39 (1∶100, mouse anti-human IgG2a, Abcam). Two antibodies were used per tissue section: FoxP3 with CD68, CD163, CD39 or CD8; CD68 with CD163 and CD39 with CD163. After overnight incubation at room temperature in the dark, sections were washed and incubated for 1 hour in the dark with secondary antibodies; goat-anti-mouse IgG1 coupled with Alexa 488 (1∶200) (Invitrogen,Bleiswijk The Netherlands), goat-anti-mouse IgG2a or goat-anti-mouse IgG2b with Alexa 546 (1∶200) (Invitrogen). Tissue sections were then washed three times with PBS and mounted with Vectashield (DAPI, 4′, 6-diamidino-2-phenylindole; Vector Laboratories, Brussels, Belgium). Immunofluorescence of skin sections was examined and images were taken from 5 different fields per section using a Leica-TCS-SP5 confocal laser scanning microscope (Leica Microsystems, Mannheim, Germany). Nucleated cells that positively stained for the specific marker were counted from five different fields per section by two laboratory persons independently. Average counts for each marker per section were compared for all samples.
Differences in cytokine concentrations were analyzed with the two-tailed Mann-Whitney U test or Wilcoxon signed rank test for non-parametric distribution using GraphPad Prism version 5.01 for Windows (GraphPad Software, San Diego California USA; www.graphpad.com) P-values were corrected for multiple comparisons. The statistical significance level used was p<0.05.
To analyse the role of CD25+ cells in the production of IFN-γ, PBMC from Ethiopian LL patients (n = 17) and Dutch healthy controls (n = 12) were depleted of CD25+ cells and cell subsets with and without re-added CD25+ cells were stimulated with M. leprae WCS in 6 days culture.
PBMC from treated Nepali LL (n = 6), BT (n = 9) patients and EC (n = 10) were depleted for CD25+ cells but only the total PBMC and CD25−cell subset were stimulated with M. leprae WCS. When compared according to clinical classification, there was a trend of higher IFN-γ production in PB compared to MB samples. IFN-γ production of total PBMC (undepleted fraction) from LL patients in response to M. leprae (WCS) was significantly lower (p = 0.001) compared to responses by PBMC from TT/BT patients, whereas IFN-γ responses to PHA were high in both groups (Fig. 1). These data further confirm the M. leprae-specific lack of cell mediated immunity (CMI) in LL patients.
Analysis of IFN-γ production in response to M. leprae (WCS) by CD25− cells alone or CD25− cells (150,000 cells per well) supplemented with the CD25+ fraction (10,000 or 25,000 cells/well) discriminated two groups of LL patients: those that produced IFN-γ in response to M. leprae after CD25+ cell depletion and those that did not (Fig. 2A, 2B and 2E). Among the 18 LL Ethiopian patients, 7 (38%) responded to M. leprae WCS after depletion of CD25+ cells whereas they lacked any response in total PBMC. IFN-γ production in response to PHA in both groups was not affected by the depletion of or enrichment with CD25+ cells.
In the LL patient group, in which recovery of IFN-γ responses was observed to M. leprae WCS after depletion of CD25+ cells, this could be reversed proportionally by the addition of CD25+ cells (Fig. 2A). In the patient group in which CD25+ cell depletion did not reverse anergy to M. leprae, there was no effect observed by addition of CD25+ cells to the depleted fraction (Fig. 2B).
In similar analysis of treated leprosy patients (LL and BT) and endemic controls from a Nepali population, PBMC responded to M. leprae WCS in the presence of CD25+ cells and a slight increase in IFN- γ levels after CD25+ cell depletion was also observed (Fig. 2C). Similarly, healthy Dutch controls (n = 8) responding to M. leprae WCS before depletion of CD25+ cell showed a slight increase after depletion (Fig. 2D left panel) as well, while other NEC (n = 5) remained unresponsive after CD25+ cell depletion (Fig. 2D right panel).
For cell subset analysis, PBMC from Ethiopian LL (n = 13), TT/BT (n = 5) and EC (n = 7) and Dutch healthy controls (NEC; n = 4) were stained for surface and intra-cellular markers. The frequency of FoxP3+ CD8+CD25+ cells was significantly higher in PBMC of LL patients compared to TT/BT patients (p = 0.02) (Fig. 3). Although not statistically significant (p = 0.05), we also observed a higher frequencies of FoxP3+ CD4+ CD25+ T-cell in the LL group compared to the TT/BT patients (Fig. 3). In contrast, analysis of the frequency of IL-10 producing CD4+ CD25+ or CD8+CD25+ T-cell showed no significant differences between patients and healthy controls. The frequency of IL-10 production in CD4+ CD25+ or CD8+CD25+ T-cell in general was very low in all groups.
Confocal analysis of two-colour immunofluorescence was used to localize specific cell markers in skin biopsies of Ethiopian LL (n = 10) and TT/BT (n = 4) leprosy patients. Higher number of CD68+ cells in LL lesions (p = 0.02) (Fig. 4A, 5A and B) indicated the presence of more infiltrating macrophages compared to TT/BT (Fig. 5C and D). In addition, CD68+ CD163+ cells (mφ2) and FoxP3+ cells were present to a larger extent in LL patients' lesions (p = 0.02) compared to TT/BT (Fig. 4B, 4C, 5C and 5D). With respect to the numbers of CD68+ CD163+ cells (mφ2) and FoxP3+ cells, differences were observed among the LL patients which could be explained by variations in the time elapsed since skin lesions were noticeable or by influence of other host factors. Although we found significantly higher frequency of CD8+FoxP3+ in PBMC, we could not clearly detect CD8+FoxP3+ in the skin lesions indicating CD4+FoxP3+ cells could play a regulatory role in these tissues. In addition, skin lesions were stained with CD39 combined with FoxP3 to localize CD39+FoxP3+ regulatory T-cells. However, in most skin tissues, CD39+ cells were not detected except for two LL skin tissues in which CD39 and FoxP3 positivity was observed simultaneously in macrophage-like shaped cells (Fig. 4E). Thus, these results indicate the induction of more FoxP3+ but not CD39+ Treg cells in LL patients' skin lesions probably by the presence of type 2 macrophages.
Decreased M. leprae-specific T-cell mediated immunity is the hall mark of lepromatous multibacillary leprosy and can be assessed by in vitro unresponsiveness to M. leprae (antigens) or clonal anergy [2], [23], [32]. In this study, we confirm the M. leprae-specific unresponsiveness by the absence of IFN-γ responses to M. leprae WCS.
Several studies have investigated the possible causes leading to hyporesponsiveness in LL patients such as formation of foamy macrophages in presence of IL-10 [27], cholesterol dependent dismantling of HLA-DR raft in macrophages of BL/LL [33] and other factors, including Treg cells. Some of these studies on Treg cells have shown their presence and role either in the periphery or in skin lesions through measuring Treg associated markers, mainly CD25, TGF-β, CTLA4, IL-10, and FoxP3 [23], [24], [34], [35]. Recently, Teles et al. showed higher expression of IFN-γ and the downstream vitamin D-dependent antimicrobial pathway related genes including CYP27B1 and VDR (Vitamin D receptor) in TT/BT as well as an increased IL-10 expression induced by IFN-β in LL lesions [36]. Some reports have revealed the limitations of the available Treg markers due to their lack of specificity [37]–[39]: CD25, for example, is expressed on activated T and B cells and is not exclusively found on Treg cells. However, noting that CD25 is still a crucial marker for Treg cells in the unstimulated situation, we performed depletion of CD25+ cells from unstimulated PBMC to isolate the Treg cells and demonstrated their involvement in M. leprae-specific unresponsiveness in LL patients.
The BL/LL patients are known for their poor CMI and this is commonly assessed by measuring IFN-γ responses to M. leprae WCS. The total PBMC of the LL patients were analysed along with the CD25+ depleted and enriched fraction for their IFN-γ responses to M. leprae WCS and was negative. However, the depletion of CD25+ cells from total PBMC of LL patients showed an enhanced pro-inflammatory response as measured by the level of IFN-γ in response to M. leprae WCS in some but not all patients. Two distinct groups of LL patients were identified after depletion of CD25+ cells; 38% (7/18) of the LL patients showed enhanced IFN-γ responses in the CD25− population while the remaining 62% of the LL patients did not respond to M. leprae WCS at all. The recovered IFN-γ production in the first group was reversed by addition of CD25+ cells, clearly indicating that this CD25+ cell population conferred the unresponsiveness in these LL patients. However, we did not stain the CD25+ cell populations with FoxP3 which could have allowed more detailed characterization as CD25high FoxP3 or CD25low FoxP3 sub-populations which might have explained differences between the responders and non-responders. Nonetheless, the presence of non-responding LL patients after depletion of CD25+ cells indicates that CD25+ Treg cells do not represent the sole factor responsible for T-cell anergy in LL leprosy. As the Th1 arm is responsible for killing and clearing bacilli, there could have been enormous damage to tissues in BL/LL patients where high load of bacilli and antigens are available. However, the presence of Treg in these patients represents one important factor that can avoid tissue damage but, on the other hand, creates a convenient environment for bacilli to survive through suppression of Th1 response. In addition, the significant IFN-γ production observed in treated LL patients in our study before depletion of CD25+ T cells showed how treatment and thereby the level of bacillary load can influence the Th1 response and Treg. Similar findings were reported for TB patients with recovered IFN-γ production and reduced number of Treg cells after treatment [21], [40]. The slight increases observed in IFN-γ production after depletion of CD25+ T cells in treated LL and BT patients and in EC tested in the depletion experiments could also indicate the regular presence of Treg cells to maintain homeostasis in the host. However, the overall ratio of CD25+ Treg cells to effector T cells will be crucial in determining the outcome of M. leprae infection in the host.
Previous studies which aimed at identifying potential factors for M. leprae-specific unresponsiveness in LL used the addition of IL-2 [2], [41]–[43] or anti-DQ monoclonal antibodies [44] or offered isolated antigenic fractions of M. leprae. Interestingly, each of the studies similarly identified two groups of LL patients, in one of which M. leprae unresponsiveness could be reversed. This indicated that the unresponsive phenotype in LL patients is likely mediated through the collective effects of various molecules. The more recent observation of cholesterol-dependent dismantling of HLA-DR raft and an increased membrane fluidity in BL/LL patients which causes a major defect in antigen presentation provides additional evidence for the presence of multiple different factors leading to T-cell anergy [33]. Thus, M. leprae specific unresponsiveness/anergy in LL patients very likely is a complex phenomenon mediated by multiple host and pathogen associated factors, one of which is represented by Treg cells.
Several studies have reported on the ex vivo frequency of Treg cells in peripheral blood of LL and TT/BT patients in unstimulated or M. leprae antigens stimulated PBMC [23], [35]. Attia et al. showed, elevated frequencies of circulating Treg cells (CD4+CD25highFoxP3+) in TT patients [35] whereas Palermo et al., showed that PBMC stimulated with M. leprae antigen for 6 days in culture had significantly higher number of Treg cells (CD4+ CD25+FoxP3+) in LL patients [23]. Recently, Saini et al., further confirmed the importance of Tregs in LL non-responsiveness by measuring TGF-β producing CD4+ CD25+FoxP3+ cells in stimulated PBMC culture [45]. In this study, we analysed the frequency of Treg cells in PBMC briefly activated with PMA/ionomycin. The frequency of CD4+ CD25+FoxP3+ cells was higher in LL compared to BT but not statistically significant (Fig. 3). However, with the visible difference observed between LL and BT and with the evidences from previous studies, their presence and role in BL/LL patients cannot be denied. For example, the recent molecular analysis of FoxP3 in CD4+CD25+ T cells nuclei has revealed that the FoxP3 interaction with histone deacetylases drives the immune suppression by CD4+ CD25+ Tregs in BL/LL unlike in other forms of leprosy [24].
On the other hand, the frequency of CD8+ CD25+FoxP3+ cells found in this study was significantly higher in LL (Fig. 3). This suggests that FoxP3+ CD8+ CD25+ Treg cells may also play a role in unresponsiveness in LL although not specifically analyzed for their functional role in our depletion experiments. Although lower in frequency compared to the CD4+ CD25+FoxP3+, Saini et al., also reported higher numbers of CD8+ CD25+FoxP3+ in LL compared to BT but without induction of TGF-β [45]. Most studies focused on CD4+ CD25+FoxP3+ in leprosy [23], [35]. In contrast one study on LL lesions showed the presence of increased numbers of CD8+ T cells with suppressive type in LL indicating the importance of CD8+ Treg cells in leprosy [46]. In addition few other studies identified CD8+ Treg as a potential suppressive sub-population [47], [48]. Recent evidence from an in vitro study also revealed CD8+ Treg cells (CD8+ LAG-3+ FoxP3+CTLA-4+) induced by matured plasmacytoid dendritic cells (pDC) with suppression activity on allo-reactive T memory cells [49]. In our opinion, the CD8+ Treg population is not sufficiently studied in leprosy and we believe further analysis of this population in all forms of leprosy in periphery and lesionary tissues will be vital.
The low IL-10 frequency measured by FACS analysis in all groups did not allow detection of significant differences among groups as expected in view of the crucial role of IL-10 as an anti-inflammatory cytokine in the unresponsiveness in LL patients [27], [36]. This could be due to the short PMA/ionomycin stimulation inherent to the procedure for ex vivo determination of the frequency of CD25+ cells. However, 6 days stimulation of PBMC from BL patients with M. leprae induced high levels of IL-10 [50].
Although, it will not be easy to generalize or conclude on frequencies and numbers of CD4+ CD25+FoxP3+ Treg cells in different forms of leprosy since the experimental procedures used in each study vary, most of the studies including ours, point to the presence of increased numbers of Treg cells in LL patients either in periphery as well as lesions. Detailed characterization of Treg cell subsets in large cohorts of leprosy patients as well as the ratio to effector T cells may provide additional insights in this area.
The dominant presence of CD163+ macrophages in LL lesions [27], [28] and the significantly higher expression of IL-10 and CTLA4 in LL tissues have been reported previously [25]. The role of Treg cells (FoxP3+ GITR+ CD25+) and their induction by CD163+ anti-inflammatory human macrophages was demonstrated in vitro since CD4+ T-cells gained a potent regulatory/suppressor phenotype and functions after activation by mφ2 [25]. In the current study, we show the presence of significantly higher number of CD68+ CD163+cells (mφ2) in the vicinity of FoxP3+ cells in LL lesions compared to TT/BT lesions. These findings support the involvement of both cell types in the induction and/or maintenance of M. leprae directed Treg cells in LL lesions.
Since a suppressive effect of CD4+CD39+FoxP3+ Treg cells was described in TB patients [12], we also analysed the frequency of CD39+FoxP3+ cells in PBMC but observed no differences between LL and TT/BT patients except for few LL skin lesions, in which macrophage-shaped CD39+ cells were observed. A recent study has shown that CD39 expression on macrophages has an important role in self-regulation mechanism during inflammation [51]. These cells may also play a similar role in LL patients but this has to be further analysed.
In summary, this study clearly show that CD25+ Treg cells play a role in unresponsiveness in LL, and that there are two subtypes of M. leprae unresponsive LL patients. Furthermore, the co-existence of Treg cells with mφ2 in LL lesions further supports the potential role of these regulatory cell subsets at the site of infection.
|
10.1371/journal.pcbi.1002472 | Optimizing Provider Recruitment for Influenza Surveillance Networks | The increasingly complex and rapid transmission dynamics of many infectious diseases necessitates the use of new, more advanced methods for surveillance, early detection, and decision-making. Here, we demonstrate that a new method for optimizing surveillance networks can improve the quality of epidemiological information produced by typical provider-based networks. Using past surveillance and Internet search data, it determines the precise locations where providers should be enrolled. When applied to redesigning the provider-based, influenza-like-illness surveillance network (ILINet) for the state of Texas, the method identifies networks that are expected to significantly outperform the existing network with far fewer providers. This optimized network avoids informational redundancies and is thereby more effective than networks designed by conventional methods and a recently published algorithm based on maximizing population coverage. We show further that Google Flu Trends data, when incorporated into a network as a virtual provider, can enhance but not replace traditional surveillance methods.
| Public health agencies use surveillance systems to detect and monitor chronic and infectious diseases. These systems often rely on data sources that are chosen based on loose guidelines or out of convenience. In this paper, we introduce a new, data-driven method for designing and improving surveillance systems. Our approach is a geographic optimization of data sources designed to achieve specific surveillance goals. We tested our method by re-designing Texas' provider-based influenza surveillance system (ILINet). The resulting networks better predicted influenza associated hospitalizations and contained fewer providers than the existing ILINet. Furthermore, our study demonstrates that the integration of Internet source data, like Google Flu Trends, into surveillance systems can enhance traditional, provider-based networks.
| Since the Spanish Flu Pandemic of , the global public health community has made great strides towards the effective surveillance of infectious diseases. However, modern travel patterns, heterogeneity in human population densities, proximity to wildlife populations, and variable immunity interact to drive increasingly complex patterns of disease transmission and emergence. As a result, there is an increasing need for effective, evidence-based surveillance, early detection, and decision-making methods [1]–[3]. This need was clearly articulated in by a directive from the Department of Homeland Security and the Centers for Disease Control and Prevention to develop a nationwide, real-time public health surveillance network [4], [5].
The U.S. Outpatient Influenza-Like Illness Surveillance Network (ILINet) gathers data from thousands of healthcare providers across all fifty states. Throughout influenza season (CDC mandating reporting during weeks , which is approximately October through mid-May), participating providers are asked to report weekly the number of cases of influenza-like illness treated and total number of patients seen, by age group. Cases qualify as ILI if they manifest fever in excess of F along with a cough and/or a sore throat, without another known cause. Although the CDC receives reports of approximately million patient visits per year, many of the reports may use a loose application of the ILI case definition and/or may simply be inaccurate. The data are used in conjunction with other sources of laboratory, hospitalization and mortality data to monitor regional and national influenza activity and associated mortality. Similar national surveillance networks are in place in EU countries and elsewhere around the globe [6]–[9].
Each US state is responsible for recruiting and managing ILINet providers. The CDC advises states to recruit one regularly reporting sentinel provider per residents, with a state-wide minimum of sentinel providers. Since , the Texas Department of State Health Services (DSHS) has enrolled a total of volunteer providers. Participating providers regularly drop out of the network; Texas DSHS aims to maintain approximately active participants through year-round recruitment of providers in heavily populated areas (cities with populations of at least ). DSHS also permits other (non-targeted) providers of family medicine, internal medicine, pediatrics, university student health services, emergency medicine, infectious disease, OB/GYN and urgent care to participate in the network. During the influenza season, the Texas ILINet included providers with approximately reporting most weeks of the influenza season.
A number of statistical studies have demonstrated that ILI surveillance data is adequate for characterizing past influenza epidemics, monitoring populations for abnormal influenza activity, and forecasting the onsets and peaks of local influenza epidemics [10]–[16]. However, the surveillance networks are often limited by non-representative samples [17], inaccurate and variable reporting [12]–[14], and low reporting rates [6]. Some of these studies have yielded specific recommendations for improving the performance of the surveillance network, for example, inclusion of particular categories of hospitals in China [12], preference for general practitioners over pediatricians in Paris, France [14], and a general guideline to target practices with high reporting rates and high numbers of patient visits (per capita) [6]. Polgreen et al. recently described a computational method for selecting ILINet providers so as to maximize coverage, that is, the number of people living within a specified distance of a provider [17]. They applied the approach to optimizing the placement of the providers in the Iowa ILINet. While their algorithm ensures maximum coverage, it is not clear that maximum coverage is, in general, the most appropriate criterion for building a statistically informative ILINet.
In , Google.org launched Google Flu Trends, a website that translates the daily number of Googles search terms associated with signs, symptoms, and treatment for acute respiratory infections into an estimate of the number of ILI patients per people. It was shown that Google Flu Trends reliably estimates national influenza activity in the US [18], the state of Utah [18], and in some European countries [19], but it provided imperfect data regarding the H1N1 pandemic in New Zealand [20]. We assessed the correlation between Google Flu Trends for Texas and Texas' ILINet data and found a correlation of , similar to those presented in Ginsberg et al. 2009 [18] (See Text S1). The Google Flu Trends website includes ILI-related search activity down to the level of cities (in beta version as of November ). Thus, Google Flu Trends may serve as a valuable resource for influenza detection and forecasting if effectively integrated with public health data such as those coming from state ILINets.
Here, we present an evaluation of the Texas Influenza-Like-Illness Surveillance Network (ILINet), in terms of its ability to forecast statewide hospitalizations due to influenza (ICD9 and ) and unspecified pneumonia (ICD9 ). Although we henceforth refer to this subset of hospitalizations as influenza-like hospitalizations, we emphasize that these data do not perfectly reflect influenza-related hospitalizations: some unrelated pneumonias may be classified under ICD9 , and some influenza cases may not be correctly diagnosed and/or recorded as influenza. Nonetheless, this subset of hospitalizations likely includes a large fraction of hospitalized influenza cases and exhibits strong seasonal dynamics that mirror ILINet trends. The inclusion of all three ICD9 codes was suggested by health officials at Texas DSHS who seek to use ILINet to ascertain seasonal influenza-related hospitalization rates throughout the state (Texas DSHS contract numbers and ). Hospitalizations associated with these three codes in Texas accounted for between and of all hospitalizations due to infections and roughly billion dollars of hospitalization payments in (See Text S1).
Using almost a decade of state-level ILINet and hospitalization data, we find that the existing network performs reasonably well in its ability to predict influenza-like hospitalizations. However, smaller, more carefully chosen sets of providers should yield higher quality surveillance data, which can be further enhanced with the integration of state-level Google Flu Trends data. For this analysis, we adapted a new, computationally tractable, multilinear regression approach to solving complex subset selection problems. The details of this method are presented below and can be tailored to meet a broad range of surveillance objectives.
Using a submodular ILINet optimization algorithm, we investigate two scenarios for improving the Texas ILINet: designing a network from scratch and augmenting the existing network. We then evaluate the utility of incorporating Google Flu Trends as a virtual provider into an existing ILINet.
To construct new sentinel surveillance networks, we choose individual providers sequentially from a pool of approximately mock providers, one for each zip code in Texas, until we reach total providers. At each step, the provider that most improves the quality of the epidemiological information produced by the network is added to the network. We optimize and evaluate the networks in terms of the time-lagged statistical correlation between aggregated ILINet provider reports (simulated by the model) and actual statewide influenza-like hospitalizations. Specifically, for each candidate network, we perform a least squares multilinear regression from the simulated ILINet time series to the actual Texas hospitalization time series, and use the coefficient of determination, , as the indicator of ILINet performance. Henceforth, we will refer to these models as ILINet regression models.
We compare the networks generated by this method to networks generated by two naive models and a published computational method [17] (Figure 1). Random selection models an open call for providers and entails selecting providers randomly with probabilities proportional to their zip code's population; Greedy selection prioritizes providers strictly by the population density of their zip code. Submodular optimization significantly outperforms these naive methods, particularly for small networks, with Random selection producing slightly more informative networks than Greedy selection. The Geographic optimization method of Polgreen et al. [17] selects providers to maximize the number of people that live within a specified “coverage distance” of a provider. Submodular optimization consistently produces more informative networks than this method at a mile coverage distance (Figure 1) (, , and mile coverage distances perform worse, not shown). To visualize the relative performance of several of these networks, we compared their estimates of influenza-like hospitalizations (by applying each ILINet regression model to simulated ILINet report data) to the true state-wide hospitalization data (Figure 2). The time series estimated by a network designed using submodular optimization more closely and smoothly matches true hospitalizations than both the actual Texas ILINet and a network designed using geographic optimization (each with providers).
The submodular optimization algorithm is not guaranteed to find the highest performing provider network, and an exhaustive search for the optimal provider network from the pool of providers is computationally intractable. However, the submodular property of the objective function allows us to compute an upper bound on the performance of the optimal network, without knowing its actual composition (Figure 1). The performance gap between the theoretical upper bound and the optimized networks may indicate that the upper bound is loose (higher than the performance of the true optimal network) and/or the existence of better networks that might be found using more powerful optimization methods.
The networks selected by submodular optimization reveal some unexpected design principles. Most of the Texas population resides in Houston and the “I-35 corridor” – a North-South transportation corridor spanning San Antonio, Austin, and Dallas (Figure 3a). The first ten provider locations selected by submodular optimization are spread throughout the eastern half of the state (Figure 4a, pink circles). While most of the providers are concentrated closer to Texas' population belt, only two are actually located within Texas' major population centers (in this case, College Station).
The submodular networks are qualitatively different from the networks created by the other algorithms considered, which focus providers within the major population centers (Figure 4b). The higher performance of the submodular ILINets suggest that over-concentration of providers in major population centers is unnecessary. Influenza levels in the major population centers are strongly correlated (Figure 3b). Thus, ILINet information from San Antonio, for example, will also be indicative of influenza levels in Austin and Dallas. This synchrony probably arises, in part, from extensive travel between the major Texas population centers.
Using submodular optimization, we augment the 2008 Texas ILINet by first subsampling from the enrolled providers and then adding up to new providers. When subsampling, performance does not reach a maximum until all providers are included in the network (Figure 5), indicating that each provider adds predictive value to the network. However, the theoretical upper bound plateaus around providers, suggesting that smaller (more optimally chosen) networks of equal predictive value may exist. During the second stage, additional providers improve the objective by . Most of these providers are located in relatively remote areas of the state.
We also considered inclusion of Internet trend data sources as virtual providers, specifically, the freely available Google Flu Trends data for the state of Texas [21]. Google Flu Trends alone is able to explain about of the variation in state-wide hospitalizations; it outperforms the 2008 Texas ILINet and matches the performance of a network with traditional providers constructed from scratch using submodular optimization (Figure 6). However, the best networks include both traditional providers and Google Flu Trends. For example, by adding providers to Google Flu Trends using submodular optimization, we improve the objective by a third and halve the optimality gap (from a trivial upper bound of one). The additional providers are located in non-urban areas (Figure 4a, green circles) distinct from those selected when Google Flu Trends is not allowed as a provider.
To further validate our methodology, we simulated the real-world scenario in which historical data are used to design an ILINet and build forecasting models, and then current ILINet reports are used to make forecasts. Specifically, we used data to design ILINets and estimate multilinear regression models relating influenza-like hospitalizations to mock provider reports, and then used data to test the models' ability to forecast influenza-like hospitalizations. For networks with fewer than providers, the ILINets designed using submodular optimization consistently outperform ILINets designed using the other three strategies (Figure 7). Above providers, the predictive performance of the submodular optimization ILINet begins to decline with additional providers. As the number of providers approaches (the number of weeks in the training period), the estimated prediction models become overfit to the period. Thus, the slightly increased performance of the Random method over the submodular optimization after providers is spurious. For the values presented in Figure 7, the effect of noise and variable reporting are integrated out when calculating the expected provider reports. An alternative approach to out-of-sample validation is presented in Text S1; it yields the same rank-order of model performance.
Since the mid twentieth century, influenza surveillance has been recognized as an increasingly complex problem of global concern [22]. However, the majority of statistical research has focused on the analysis of surveillance data rather than the data collection itself, with a few notable exceptions [12], [17]. High quality data is essential for effectively monitoring seasonal dynamics, detecting anomalies, such as emerging pandemic strains, and implementing effective time-sensitive control measures. Using a new method for optimizing provider-based surveillance systems, we have shown that the Texas state ILINet would benefit from the inclusion of a few strategically selected providers and the use of Internet data streams.
Our method works by iteratively selecting providers that contribute the most information about influenza-like hospitalizations. We quantified the performance of various ILINets using the coefficient of determination resulting from a multi-linear regression between each provider's time series and state-wide influenza-like hospitalizations. Importantly, these simulated providers have reporting rates and error distributions estimated from actual ILINet providers in Texas (see Text S1). The result is a prioritized list of zip codes for inclusion in an ILINet that can be used for future ILINet recruiting. Although this analysis was specifically motivated by the Texas DSHS interest in predicting hospitalizations with ICD9 codes , , and , our method can be readily extended to design a network for any disease or influenza definition with the appropriate historical data. In general, the method requires both historical provider reports and historical time series of the prediction target. However, if one has reasonable estimates of provider reporting rates and informational noise from another source (e.g., estimates from a surveillance network in another region or for another disease), then historical provider reports are not necessary.
ILINet provider reports do not necessarily reflect true influenza activity. Rather they are supposed to indicate the number of patients that meet the clinical ILI case definition, which results in a substantial number of false positives (reported non-influenza cases) and false negatives (missed cases of influenza) [23]. The case definition for ILI is often loosely applied, further confounding the relationship between these measures and true influenza. Similarly, the ICD9 codes used in our analysis do not correspond perfectly to influenza hospitalizations: some influenza cases will fail to be classified under those codes, and some non-influenza cases will be. Nonetheless, public health agencies are interested in monitoring and forecasting the large numbers of costly hospitalizations associated with these codes. We find that ILINet surveillance data correlates strongly with this set of influenza-like hospitalizations, and that the networks can be designed to be even more informative.
Although we provide only a single example here, this optimization method can be readily applied to designing surveillance networks for a wide range of diseases on any geographic scale, provided historical data are available and the goals of the surveillance network can be quantified. For example, surveillance networks could be designed to detect emerging strains of influenza on a global scale, monitor influenza in countries without surveillance networks, or track other infectious diseases such as malaria, whooping cough, or tuberculosis or non-infectious diseases and chronic conditions such as asthma, diabetes, cancer or obesity that exhibit heterogeneity in space, time or by population subgroup. As we have shown with Google Flu Trends, our method can be leveraged to evaluate the potential utility of incorporating other Internet trend data mined from search, social media, and online commerce platforms into traditional surveillance systems.
While optimized networks meet their specified goals, they may suffer from over optimization and be unable to provide valuable information for other diseases or even for the focal disease during atypical situations. For example, a surveillance network designed for detecting the early emergence of pandemic influenza may look very different from one optimized to monitor seasonal influenza. Furthermore, an ILINet optimized to predict influenza-like hospitalizations in a specific socio-economic group, geographic region, or race/ethnicity may look very different from an ILINet optimized to predict state-wide hospitalizations. When optimizing networks, it is thus important to carefully consider the full range of possible applications of the network and integrate diverse objectives into the optimization analysis.
The optimized Texas ILINets described above exhibit much less redundancy (geographic overlap in providers) than the actual Texas ILINet. Whereas CDC guidelines have led Texas DSHS to focus the majority of recruitment on high population centers, the optimizer only sparsely covered the major urban areas because of their synchrony in influenza activity. This is an important distinction between submodular optimization and the other methods considered (Geographic, Random and Greedy). The submodular method does not track population density and instead adds providers who contribute the most marginal information to the network. Consequently, it places far more providers in rural areas than the other methods (Figure 4b). There can be substantial year-to-year variation in spatial synchrony for seasonal influenza, driven by the predominant influenza strains and commuter traffic between population centers [24]. As long as the historical data used during optimization reflect this stochasticity, the resulting networks will be robust. However, synchrony by geography and population density does not occur for all diseases including emerging pandemic influenza [24]; thus the relatively sparse networks designed for forecasting seasonal influenza hospitalizations may not be appropriate for other surveillance objectives, like detecting emerging pandemic strains or other rare events. For example, a recent study of influenza surveillance in Beijing, PRC suggested that large hospitals provided the best surveillance information for seasonal influenza, while smaller provincial hospitals were more useful for monitoring H5N1 [12].
Although our method outperforms the Maximal Coverage Method (MCM), referred to as Geographic, proposed by Polgreen et al. (2009), there are several caveats. First, population densities and travel patterns within Texas are highly non-uniform. The two methods might perform similarly for regions with greater spatial uniformity. Second, our method is data intensive, requiring historical surveillance data that may not be available, for example, in developing nations, whereas the population density data required for MCM is widely available. However, the type of data used in this study is readily available to most state public health agencies in the United States. For example, the CDC's Influenza Hospitalization Network (FluSurv-NET) collects weekly reports on laboratory confirmed influenza-related hospitalizations in fourteen states. In addition, alternative internet-based data sources like Google Flu Trends are becoming available. Third, as discussed above, our networks are optimized towards specific goals and may thus have no expected level of performance for alternate surveillance goals. Important future research should focus on designing networks able to perform well under a range of surveillance goals. Fourth, neither ILINet data nor influenza-like hospitalizations correspond perfectly to actual influenza activity. One could instead optimize ILINets using historical time series of laboratory-confirmed cases of influenza. Although some provider locations and the estimated regression models may change, we conjecture that the general geospatial distribution of providers will not change significantly. Fourth, we followed Polgreen et al. (2009)'s use of Euclidean distances. However, travel distance is known to correlate more strongly with influenza transmission than Euclidean distance [24], and thus alternative distance metrics might improve the performance of the MCM method. Finally, while submodular optimization generally outperforms the other design methods in out-of-sample prediction of influenza-like hospitalizations, it suffers from overfitting when the number of providers in the network approaches the number of data points in the historical time series.
The impressive performance of Google Flu Trends leads us to question the role of traditional methods, such as provider-based surveillance networks, in next generation disease surveillance systems. While Texas Google Flu Trends alone providers almost as much information about state-wide influenza hospital discharges as the entire 2008 Texas ILINet, an optimized ILINet of the same size contains more information than Google Flu Trends alone. Adding Google Flu Trends to this optimized network as a virtual provider increases its performance by an additional . Internet driven data streams, like Google Flu Trends, may have age and socio-economic biases that over-represent certain groups, a possible explanation for the difference in providers selected when Google Flu Trends is included, Figure 4a. Given the relatively low cost of voluntary provider surveillance networks, synergistic approaches that combine data from conventional and Internet sources offer a promising path forward for public health surveillance.
This optimization method was conceived through a collaboration between The University of Texas at Austin and the Texas Department of State Health Services to evaluate and improve the Texas ILINet. The development and utility of quantitative methods to support public health decision making hinges on the continued partnership between researchers and public health agencies.
The Texas Department of State Health Services (DSHS) provided (1) ILINet data containing weekly records from reporting the number of patients with influenza-like-illness and the total number of patients seen by each provider in the network, and (2) individual discharge records for every hospital in Texas from (excluding hospitals in counties with less than inhabitants, in counties with less than total hospital beds, or those hospitals that do not seek insurance payment or government reimbursement). We classified all hospital discharges containing ICD9 codes of , , or as influenza-related. Google Flu Trends data was downloaded from the Google Flu Trends site [21] and contains estimates of ILI cases per physician visits determined using Google searches [25]. Data on population size and density was obtained from the census [26].
The first step in the ILINet optimization is to build a data-driven model reflecting actual provider reporting rates and informational noise, that is, inconsistencies between provider reports and true local influenza prevalence.
We model reporting as a Markov process, where each provider is in a “reporting” or “non-reporting” state. A provider in the reporting state enters weekly reports, while a provider in the non-reporting state does not enter reports. At the end of each week, providers independently transition between the reporting and non-reporting states. Such a Markov process model allows for streaks of reporting and streaks of non-reporting for each provider, which is typical for ILINet providers. We estimate transition probabilities between states from actual ILINet provider report data. For each provider, the transition probability from reporting to non-reporting is estimated by dividing the number of times the transition occurred by the number of times any transition out of reporting is observed. The probabilities of remaining in the current reporting state and transitioning from non-reporting to reporting are estimated similarly.
We model noise in reports using a standard regression noise model of the form(1)where (i) denotes the number of ILI cases reported by the provider in week ; (i) denotes the estimated prevalence of ILI in the provider's zip code in week ; and are regression constants fixed for the provider; and is a normally distributed noise term with variance also fixed for the provider. For existing providers, we use empirical time series (their past ILINet reporting data matched with local ILI prevalence, described below) to estimate the constants and using least squares linear regression. This noise model has the intuitive interpretation that each provider's reports are a noisy reading of the percent of the population with ILI in the provider's zip code.
We use the Texas hospital discharge data to estimate the local ILI prevalences ((i)) for each zip code. Given an estimate of the influenza hospitalization rate [27] and assuming that each individual with ILI is hospitalized independently, we can obtain a distribution for the number of influenza-related hospitalizations in a zip code, given the number of ILI cases in the zip code. Using Bayes rule, a uniform prior, and the real number of influenza-related hospitalizations (from the hospital discharge data), we derive distributions for the number of ILI cases for each zip code and each week. We then set (i) for each zip code equal to the mean of the distribution of ILI cases in that zip code for week , divided by the population of the zip code.
The second step in the ILINet optimization is to generate a pool of mock providers. For each actual provider in the Texas ILINet, we estimate a reporting profile specified by [1)] transition probabilities between reporting and non-reporting (Markov) states, and the constants and , modeling noise in the weekly ILI reports. To generate a mock provider in a specified zip code, we select a uniformly random reporting profile out of all reporting profiles estimated from existing providers. The generated mock providers are thereby given reporting characteristics typical of existing providers. We can then generate an ILI report time series for a mock provider, by 1) generating reports only during reporting weeks, and calculating reports using equation (1) with the constants given in the provider's reporting profile and estimates of (i) for the mock provider's zip code.
We select providers from pools consisting of a single mock provider from each zip code. Zip codes offer a convenient spatial resolution, because they have geographic specificity and are recorded in both the Texas ILINet and hospital discharge data. The optimization algorithm is not aware of a mock provider's reporting profile when the provider is selected (discussed below).
The final step in our ILINet design method is selecting an optimized subset of providers from the mock provider pool. We seek the subset that most effectively predicts a target time series (henceforth, goal), as measured by the coefficient of determination () from a least squares multilinear regression to the goal from the report time series for all providers in the subset. Specifically, the objective function is given bywhere is the goal random variable; is a subset of the mock provider pool; are provider reports for provider ; and the are the best multilinear regression coefficients (values that minimize the second term in the numerator).
There are several advantages to this objective function. First, it allows us to optimize an ILINet for predicting a particular random variable. Here, we set the goal to be state-wide influenza-related hospitalizations for Texas. This method can be applied similarly to design surveillance networks that predict, for example, morbidity and/or mortality within specific age groups or high risk groups.
Second, the objective function is submodular in the set of providers, [28], implying generally that adding a new provider to a small network will improve performance more than adding the provider to a larger network. The submodular property enables computationally efficient searches for near optimal networks and guarantees a good level of performance from the resulting network [29]. Without a submodular objective function, optimization of a provider ILINet may require an exhaustive search of all subsets of providers from the provider pool, which quickly becomes intractable. For example, an exhaustive search for the optimal provider Texas ILINet from our pool of approximately mock providers would require roughly regressions.
Taking advantage of the submodular property, we rapidly build high performing networks (with providers) according to the following algorithm:
This is guaranteed to produce a network that performs within a fraction of of the optimal network [28]. The submodularity property also allows us to compute a posterior bound on the distance from optimality, which is often much better than . Finally, even if implemented naively, the algorithm only requires approximately regressions to select providers from a pool of .
When optimizing, it is important to consider potential noise (underreporting and discrepancies between provider reports and actual ILI activity in the zip code). However, we assume that one cannot predict the performance of a particular provider before the provider is recruited into the network. To address this issue, the optimization's objective function is an expectation over the possible provider reporting profiles. Specifically, we define as a random variable describing the provider reporting profile for the entire pool of mock providers. If is a specific reporting profile, then the objective function can be written asTo design the ILINet, we solve the following optimization problemThe objective function is a convex combination of submodular functions, and thus is also submodular. This allows us to use the above algorithm along with its theoretical guarantees to design ILINets using a realistic model of reporting practices and informational stochasticity, without assuming that the designer knows the quality of specific providers a priori.
We implemented the Maximal coverage model (MCM) following Polgreen et al. (2009). Briefly, a greedy algorithm was used to minimize the number of people in Texas who live outside a pre-defined coverage distance, , of at least one provider in the selected set, . A general version of this algorithm was developed by Church and Re Velle (1974) to solve this class of MCM's [30]. As per Polgreen et al. (2009), we assumed that the population density of each zip code exists entirely at the geographic center of the zip code and used Euclidean distance to measure the distance between zip codes. Using a matrix of inter-zip code distances we select providers iteratively, choosing zip codes that cover the greatest amount of population density within the pre-defined coverage distance, . We considered , 10, 20, and 25 miles, and found that miles yielded the most informative networks.
We used two naive methods to model common design practices for state-level provider-based surveillance networks.
To analyze similarities in ILI hospitalizations across different zip codes, we apply principal component analysis (PCA) [31]. Specifically, we perform PCA on the centered (mean zero), standardized (unit variance) hospitalization time series of all zip codes in Texas. We first compute a time series for the first principal component, and then compute an for each zip code, based on a linear regression from the first principal component to the zip code's centered, standardized hospitalizations. Zip codes with high values have hospitalization patterns that exhibit high temporal synchronicity with the first principal component.
To validate our method, we first use submodular optimization to create a provider network of providers, using only data from 2001 to 2007, and then evaluate the performance of the network in predicting 2008 influenza-like hospitalizations. Specifically, after creating the -provider network (), we use actual hospitalization data and mock provider reports for the 2001–2007 period to fit a multilinear regression model of the form where is time series of state-wide influenza-like hospitalizations at week for weeks in 2001 to 2007, is the mock report time series of provider during week for weeks in 2001 to 2007, and is the best multilinear regression coefficient associated with provider .
We then use the estimated multilinear regression function to forecast state-wide influenza-like hospitalization during 2008 from mock provider reports of 2008, and compare these forecasts to actual 2008 hospitalization data. This simulates a real-world prediction, where only historical data is available to create the provider network () and estimate the prediction function ('s), and then the most recent provider reports ('s) are used to make predictions. We evaluate the 2008 predictions using a variance reduction measure similar to , except that the multilinear prediction model uses coefficients estimated from prior data, as given bywhere is the hospitalization time series in 2008, is the provider noise profile, and are the mock provider reports in 2008. Importantly, we first calculate an expected value for the provider reports, , given the noise profiles , before calculating . We also considered an alternative validation method in which we first calculate an for each provider report and noise-profile combination, and then analyze the resulting distribution of values (see Text S1 for results).
|
10.1371/journal.pgen.1003438 | Mondo/ChREBP-Mlx-Regulated Transcriptional Network Is Essential for Dietary Sugar Tolerance in Drosophila | Sugars are important nutrients for many animals, but are also proposed to contribute to overnutrition-derived metabolic diseases in humans. Understanding the genetic factors governing dietary sugar tolerance therefore has profound biological and medical significance. Paralogous Mondo transcription factors ChREBP and MondoA, with their common binding partner Mlx, are key sensors of intracellular glucose flux in mammals. Here we report analysis of the in vivo function of Drosophila melanogaster Mlx and its binding partner Mondo (ChREBP) in respect to tolerance to dietary sugars. Larvae lacking mlx or having reduced mondo expression show strikingly reduced survival on a diet with moderate or high levels of sucrose, glucose, and fructose. mlx null mutants display widespread changes in lipid and phospholipid profiles, signs of amino acid catabolism, as well as strongly elevated circulating glucose levels. Systematic loss-of-function analysis of Mlx target genes reveals that circulating glucose levels and dietary sugar tolerance can be genetically uncoupled: Krüppel-like transcription factor Cabut and carbonyl detoxifying enzyme Aldehyde dehydrogenase type III are essential for dietary sugar tolerance, but display no influence on circulating glucose levels. On the other hand, Phosphofructokinase 2, a regulator of the glycolysis pathway, is needed for both dietary sugar tolerance and maintenance of circulating glucose homeostasis. Furthermore, we show evidence that fatty acid synthesis, which is a highly conserved Mondo-Mlx-regulated process, does not promote dietary sugar tolerance. In contrast, survival of larvae with reduced fatty acid synthase expression is sugar-dependent. Our data demonstrate that the transcriptional network regulated by Mondo-Mlx is a critical determinant of the healthful dietary spectrum allowing Drosophila to exploit sugar-rich nutrient sources.
| Diet displays extreme natural variation between animal species, which range from highly specialized carnivores, herbivores, and nectarivores to flexible dietary generalists. Humans are not identical in this respect either, but the genetic background likely defines the framework for a healthy diet. However, we understand poorly the genetic factors that define the spectrum of healthy diet for a given species or individual. Here we have explored the genetic basis of dietary sugar tolerance of Drosophila melanogaster. D. melanogaster is a generalist fruit breeder that feeds on micro-organisms on decaying fruits and vegetables with varying sugar content. However, mutants lacking the conserved Mondo-Mlx transcription factor complex display striking intolerance towards dietary sucrose, glucose, or fructose. This is manifested in the larvae by the inability to grow and pupate on sugar-rich food, including red grape, which belongs to the normal diet of wild D. melanogaster. Larvae lacking Mondo-Mlx show widespread metabolic imbalance, including highly elevated circulating glucose. Genome-wide gene expression analysis combined with systematic loss-of-function screening of Mlx targets reveal that the genetic network providing sugar tolerance includes a secondary transcriptional effector as well as regulators of glycolysis and detoxification of reactive metabolites.
| Mono- and disaccharides, i.e. sugars, are an important source of nutritional energy, but animal species display marked differences in the degree of sugar utilization and tolerance. While the diet of carnivores is typically low in sugars, nectarivores, like hummingbirds, feed primarily on sugar-rich nectar [1], [2]. Sugars from fruits and honey have been part of the ancestral human diet. However, the large quantities of refined sugars consumed by modern humans far exceed those available in natural sources [3]. In fact, it has been proposed that excessive added sugar in the diet, especially fructose, might contribute to the development of metabolic syndrome [3]–[5]. Yet the genetic factors governing the delicate balance between healthful dietary sugar utilization and the sugar overload-induced metabolic disturbance are poorly understood.
Drosophila is a well-suited model for exploring the physiological consequences of sugar intake. Drosophila melanogaster is a generalist fruit breeder naturally performing well on a broad range of dietary sugars [6]. However, excessive intake of sugars has been shown to cause diabetes-like metabolic changes in D. melanogaster, including insulin resistance, elevated circulating glucose and increased adiposity [7], [8]. Dietary sugars have also been shown to shorten Drosophila lifespan [9]. The sugar-induced insulin resistance has been attributed to the JNK-regulated lipocalin Neural Lazarillo [8]. Moreover, high sugar induced gene expression has been previously analysed [7], [10]. However, beyond these observations, the functional interactions between genotype and dietary sugar have remained poorly understood.
Elevated systemic glucose levels cause cellular stress and tissue damage [11], [12]. Animals therefore rapidly adapt their metabolism to fluctuating sugar intake, maintaining circulating glucose levels constant. A postprandial increase in circulating glucose triggers the release of insulin, which induces the rapid uptake of excess glucose by metabolic tissues including muscle, adipose tissue, and liver [13]. Intracellular glucose is immediately converted into glucose-6-phosphate and further metabolized into glycogen and lipids or catabolised to release energy. Metabolic tissues are exposed to large variations in the flux of intracellular glucose and therefore need to regulate their metabolism accordingly.
The basic helix-loop-helix transcription factor paralogs ChREBP (Carbohydrate Response Element Binding Protein) and MondoA act together with their common binding partner Mlx (Max-like protein X) to mediate transcriptional responses to intracellular glucose in mammals [14]. The ChREBP/MondoA-Mlx complex is activated by glucose-6-phosphate and other phosphorylated hexoses, and regulates gene expression by binding to target promoters containing a carbohydrate response element (ChoRE) [15]–[19]. ChREBP and MondoA regulate the majority of the global glucose-induced transcriptional responses and many of their target genes encode enzymes in glycolytic and lipogenic pathways [15], [16], [20], [21]. ChREBP and MondoA play differential tissue-specific roles in mammals: ChREBP functions in the liver, adipose tissue and pancreatic beta cells [22]–[26], while MondoA is predominantly expressed in the skeletal muscle [27].
Of the mammalian ChREBP/MondoA-Mlx complex, the role of ChREBP has been studied in a physiological setting using loss-of-function mice. While ChREBP is nonessential in terms of survival, the mutant mice display a number of metabolic phenotypes, including elevated plasma glucose and liver glycogen as well as reduced adiposity [25], [28]. ChREBP−/− mice survive poorly on a diet with high levels of sugars, but the underlying reasons have remained unexplored [25]. ChREBP is known to regulate a range of metabolic genes, including those involved in de novo lipogenesis [15], [21]. Which target genes contribute to the various physiological phenotypes and what is the causal interrelationship between the physiological phenotypes, are questions that require powerful genetics and are therefore challenging to address in vivo in mammals. Moreover, existence of another Mondo paralog, MondoA, might mask some phenotypes in the ChREBP−/− mouse. To better understand the physiological roles of the Mondo/ChREBP/-Mlx complex and its target genes, we have explored their role in Drosophila melanogaster. The Drosophila genome encodes one ortholog for each of ChREBP/MondoA and Mlx, which we call Mondo (alternative identifiers: CG18362, Mlx interactor, ChREBP) and Mlx (alternative identifiers: CG3350, Bigmax), respectively [27], [29], [30].
We have generated mlx null mutant flies, which displayed lethality in the late pupal stage. D. melanogaster larvae can normally utilize high levels of dietary sugars [6], but loss of Mlx or knockdown of Mondo caused striking intolerance towards sucrose, glucose and fructose. The mlx null mutant larvae also displayed extensive metabolic changes, with strongly elevated circulating glucose, signs of amino acid catabolism and altered lipid and phospholipid profiles. Systematic functional analysis of Mlx-regulated genes revealed three genes contributing to dietary sugar tolerance: cabut, encoding a Krüppel-like transcription factor, phosphofructokinase 2, a regulator of the glycolytic pathway, and Aldehyde dehydrogenase type III, which is linked to detoxification of reactive aldehydes.
To study the physiological role of Drosophila Mlx, we generated a mutant allele using imprecise P-element (P{XP}bigmaxd07258) excision. We recovered one mutant allele, mlx1, which lacked the entire coding region of mlx as well as 17 C-terminal amino acids of the neighbouring gene CG3368 (Figure 1A). As controls, we recovered lines from which the P-element had been excised precisely, leaving mlx intact. If not differently stated, the precise excision line is used as a control throughout the study. As expected, mlx1 mutant larvae expressed neither Mlx protein nor mRNA (Figure 1B; Figure S1A). The mlx1 mutants displayed lethality at the late pupal (pharate) stage (Figure S1B), and only a small number of adult flies could be recovered. mlx1 mutant flies also displayed a modest developmental delay when raised on our standard fly food (Figure S1C).
Experiments with defined nutrients revealed that mlx1 mutant larvae failed to survive on a diet with 20% sucrose as the sole nutrient source (Figure 1C). To test if the mutant lethality was due to either the inability to utilize carbohydrates as energy source or intolerance towards sucrose, we supplemented protein-rich diet (20% yeast) with increasing levels of sucrose. While mlx1 mutant larvae developed with similar kinetics as control animals on a high protein/low sugar diet, increasing the sucrose concentration gradually slowed down larval development of mlx1 mutants (Figure 1D,E). At higher sucrose levels, mlx1 mutants failed to pupate and died as larvae, while control animals displayed no apparent change in pupation kinetics with respect to 0–15% sucrose.
To confirm that the observed phenotypes were due to loss of mlx function, we used RNAi-mediated knockdown and transgenic rescue. Ubiquitous knockdown of Mlx by RNAi led to significantly slower pupation, and increased pupal lethality on protein rich food supplemented with 15% sucrose, while displaying no visible phenotype in the absence of added sucrose (Figure 1F). Driver line without RNAi was used as a control. Moreover, sugar intolerance and pupal lethality of the mlx1 mutants were efficiently rescued by ubiquitous expression of transgenic mlx (tub-GAL4>UAS-mlx) (Figure 1G, 1H). To further rule out the possibility that sugar intolerance was due to disturbed function of the neighbouring gene CG3368, we used RNAi for ubiquitous knockdown. CG3368 was efficiently silenced, but no sugar intolerance was observed (Figure S1D–S1F). Thus, mlx gene function is essential for tolerance to dietary sucrose.
To test for specific intolerance towards glucose or fructose, we supplemented the protein-rich food with 10% of either monosaccharide. Both caused clear developmental delays of mlx1 mutants (Figure 1I). Drosophila melanogaster is a dietary generalist, feeding on micro-organisms on decaying fruits and vegetables that have varying sugar content. To test whether the sugar intolerance of mlx1 mutants was relevant within the natural spectrum of D. melanogaster's diet, we allowed larvae to develop on pieces of red grape with baker's yeast inoculum. Indeed, mlx1 mutants were unable to pupate in these conditions, while >50% of the control larvae reached the pupal stage (Figure 1J).
In mammals, Mlx forms a functional complex with the Mondo paralogs, MondoA and ChREBP. We tested if the heterodimeric function of Mlx is conserved in Drosophila and essential for the sugar tolerance. Drosophila Mlx co-immunoprecipitated with Mondo when expressed in Drosophila S2 cells, suggesting heterodimeric function (Figure 2A). Notably, Mlx from S2 cells runs as two distinct bands (Figure 2A), which correspond to the two upper bands present in the in vivo sample (Figure 1B and Figure S2). The nature of these bands has remained unclear, no alternative splicing has been reported and both bands are resistant to alkaline phosphatase treatment (not shown). Ubiquitous RNAi knockdown of Mondo (tub-GAL4>Mondo RNAi) led to delayed pupation (Figure 2B) and reduced pupal survival on high sugar diet (20% yeast-15% sucrose) (Figure 2C). In sum, the biochemical and genetic evidence implies conservation of the Mondo-Mlx heterodimer function in Drosophila.
The diet-dependent phenotype of mlx1 mutants led us to perform a comprehensive survey of their metabolic status using mass-spectrometry based lipidomics and metabolomics. Lipidomics analysis revealed significant downregulation of key phospholipid groups, such as phosphatidylethanolamines (PE) and lysophosphatidylcholines (LysoPC) (Figure 3A). Total triglyceride (TG) levels showed a lower trend in mlx1 mutants, but the difference to the controls was not statistically significant (Figure 3B). However, mlx1 mutants showed significant enrichment in triglyceride species with long fatty acid tails (Figure 3B). At the same time, mlx1 mutants showed strong downregulation of certain fatty acids, such as myristoleic acid and lauric acid (Figure 3C). On the other hand, ceramide (Cer) levels were elevated in mlx1 mutants compared to controls (Figure 3D). Together, mlx1 mutants display signs of severely altered lipid and phospholipid metabolism. In addition to the changes in lipid profiles, total amino acid levels were significantly reduced in mlx1 mutants (Figure 3E), while concentration of urea was dramatically increased (Figure 3F). This implies that mlx1 mutants might catabolise amino acids for energy.
To study changes in glucose metabolism, we measured glucose and trehalose levels in the hemolymph of larvae raised on diets with varying sucrose content. Trehalose is a disaccharide released by gluconeogenesis and glycogenolysis in insects [31]. The levels of circulating glucose were moderately elevated in mlx1 mutant larvae raised on a low-sugar diet (20% yeast) (Figure 4A). However, increasing the dietary sucrose to 5%, which still sustains larval development of mlx1 mutants, led to a prominent increase of circulating glucose in mlx1 mutants while remaining constant in control animals (Figure 4A). Trehalose levels were also significantly elevated in mlx1 mutants, but unlike glucose, trehalose levels were little affected by the dietary sucrose levels (Figure 4B). Furthermore, glycogen levels were significantly elevated in mlx1 mutants (Figure 4C), indicating that glucose catabolism, not cellular glucose intake, is limiting glucose clearance from circulation in mlx1 mutants.
To verify that the elevated glucose levels were due to loss of Mlx function, we performed a transgenic rescue, which normalized circulating glucose levels (Figure S3A). Further, RNAi-mediated knockdown of Mlx led to a clear increase in circulating glucose, trehalose and glycogen (Figure S3B–S3D). In line with the expectation of heterodimeric function for Mondo and Mlx proteins, Mondo RNAi knockdown led to a prominent increase in circulating glucose and trehalose (Figure 4D,E). Also the glycogen levels were significantly increased in Mondo RNAi larvae (Figure 4F). Knockdown of CG3368, the neighbouring gene of mlx, had no influence on circulating glucose or trehalose (Figure S3B, S3C). In conclusion, mlx null mutant and mondo knockdown larvae display a reduced capacity to utilize circulating glucose leading to poor homeostatic adaptation to elevated dietary sugar.
As mammalian MondoA-Mlx and ChREBP-Mlx appear to display tissue-specific functions, we wanted to determine the functionally important tissues for Drosophila Mondo-Mlx. We first analyzed their mRNA expression using quantitative RT-PCR. Expression of Mondo and Mlx were highly correlated; highest levels were detected in the fat body, gut and Malpighian tubules (Figure 5A). To functionally dissect the contribution of different tissues to the sugar sensitivity, we used tissue-specific transgenic rescue. Restoring Mlx expression in neurons (Elav-GAL4) or muscle (Mef2-GAL4) did not significantly improve the sugar tolerance or survival of mlx1 mutants (Figure 5B). However, targeted expression in the fat body with two independent driver lines (ppl- and r4-GAL4) efficiently rescued survival on high sugar diet. Moreover, rescue of Mlx in the fat body, but not in muscle, was sufficient to normalize the levels of circulating glucose in mlx1 mutants (Figure 5C). Notably, transgenic Mlx expression by tub-GAL4 (Figure 1G) or Mef2-GAL4 (Figure 5B) caused moderately reduced survival, which was also observed in the presence of intact endogenous mlx (Figure S4).
To identify Mlx target genes, we performed a microarray gene expression profiling specifically in the fat body. Comparing gene expression between mlx1 mutant and control fat bodies from 3rd instar prewandering larvae raised on a moderate sucrose level diet (20% yeast-5% sucrose) revealed 97 down- and 96 up-regulated genes (>2-fold change and adjusted p-value<0.05) (Figure 6A; Table S1). As expected for a deletion mutant, mlx was identified as the most strongly downregulated gene on the microarray (Figure 6A). Gene Set Enrichment Analysis (GSEA) revealed a significant enrichment of KEGG categories involved in metabolic regulation (Figure 6B). For example, KEGG categories of fatty acid metabolism and nitrogen metabolism were strongly downregulated (Figure 6C), which is in good agreement with the metabolomics data (Figure 3). Many of the genes downregulated in mlx1 mutant fat body also showed reduced expression during earlier larval stages in whole larval samples (Figure 6D). Mlx-regulated genes include several key metabolic genes, such as glycerol-3-phosphate dehydrogenase-1 (Gpdh, CG9042), stearoyl-CoA 9-desaturase-1 (desat1, CG5887), Glutamine synthetase-1 (Gs1, CG2718), and 3-hydroxybutyrate dehydrogenase (sro, CG12068). The mean expression levels of three known ChREBP and MondoA targets in mammals, fatty acid synthase (Fas, CG3523), acetyl-CoA carboxylase (ACC, CG11198) and phosphofructokinase 2 (PFK2, 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase, CG3400), [20], [32], were also reduced, although they did not pass our strict microarray cut-offs (Figure 6D). Interestingly, one of the most strongly downregulated genes in mlx1 mutant fat body was that encoding the Krüppel-like transcription factor Cabut (cbt; CG4427) (Figure 6A, 6D).
To test in an unbiased way if any of the genes downregulated in mlx1 mutants had an essential role in maintaining organismal sugar tolerance, we systematically targeted 103 candidate genes by RNAi (Table S2). Intriguingly, ubiquitous knockdown of two Mlx-regulated genes identified on the microarray led to significant sugar intolerance. Transcription factor Cabut was among the most highly Mlx-regulated genes in the microarray. Ubiquitous knockdown of Cabut expression by RNAi during the larval stage caused a modest delay of pupation on low sugar diet. However, on high sugar diet (20% yeast-15% sucrose) Cabut knockdown led to prominent developmental delay and impaired survival (Figure 7A). This suggests that Mondo-Mlx activates a hierarchical transcriptional network to regulate dietary sugar tolerance with Cabut as an essential downstream effector.
Another Mlx-regulated gene, which caused sugar intolerance upon ubiquitous knockdown, was Aldehyde dehydrogenase type III (Aldh-III, CG11140; Figure 7B). Most Aldh-III knockdown animals reached the pharate stage on a low sugar diet, but died during early pupal stages on a high-sugar diet (Figure S5A). Similar early pupal lethality was observed in mlx1 mutants on sugar concentrations that allowed pupation (data not shown). Survival on a 20% sucrose-only diet was also significantly reduced upon Aldh-III knockdown (Figure 7C). We also tested whether restoring Aldh-III activity by transgenic expression would be sufficient to rescue impaired mlx1 mutant survival on high sugar diet. While transgenic expression of Aldh-III did not rescue mlx1 mutant pupation on 20% yeast-15% sucrose diet (data not shown), larval survival of mlx1 mutants on 20% sucrose-only diet was significantly improved by transgenic Aldh-III (Figure 7D). The same was true when Aldh-III expression was rescued only in the fat body (Figure S5B). In conclusion, Aldh-III is essential and sufficient for providing dietary sugar tolerance.
We also explored whether the sugar intolerance observed upon knockdown of Cabut and Aldh-III was associated with elevated circulating glucose levels. Surprisingly, knockdown of either gene did not result in a significant increase in circulating glucose (Figure 7E, 7F). This implies that impaired clearance of circulating glucose is not an essential prerequisite for intolerance to dietary sugars. Instead, these two Mondo-Mlx-regulated parameters can be uncoupled at the level of the downstream target genes. Based on these phenotypes, the Cabut-dependent branch of the transcriptional network mediates only a subset of Mondo-Mlx functions. We also studied the possibility that Cabut could be a direct regulator of Aldh-III, however the mRNA levels of Aldh-III were unchanged in Cabut RNAi larvae (Figure S5C). Also the mRNA levels of Mondo and Mlx were unchanced in Cabut RNAi larvae, indicating that Cabut is not a feedback regulator of Mondo-Mlx. Notably, it is possible that Cabut is regulating a subset of common genes with Mondo-Mlx.
Perhaps the best established function of mammalian ChREBP-Mlx is promotion of de novo lipogenesis in response to high carbohydrate intake [14]. This function is mediated through upregulation of lipogenic genes and it appears to be conserved in Drosophila (Figure 6D; [33]). Two key targets of Mondo-Mlx involved in de novo lipogenesis are acetyl-CoA carboxylase (ACC) and fatty acid synthase (Fas). Thus exploring their function in respect to dietary sugar will reveal whether de novo lipogenesis is functionally coupled to sugar tolerance. While knockdown of ACC was embryonic lethal (data not shown), Fas knockdown animals displayed some degree of survival until pupal stage. Strikingly, Fas knockdown larvae displayed early larval lethality on high protein diet (20% yeast paste), but diet supplementation with 15% sucrose partially rescued the lethality allowing pupation (Figure 7G). Thus, Mondo-Mlx-mediated regulation of fatty acid synthesis is a non-essential function for dietary sugar tolerance. In contrast, high dietary sugars promote survival upon compromised fatty acid synthesis.
As our systematic analysis of sugar tolerance genes did not reveal a mechanism by which Mondo-Mlx maintains low circulating glucose, we analyzed glucose levels on additional Mondo-Mlx targets that have a metabolic function. Phosphofructokinase 2 (PFK2, Pfrx, CG3400) synthesizes and breaks down fructose-2,6-bisphosphate, which is an allosteric activator of phosphofructokinase 1 and hence promotes glycolysis. PFK2 expression was downregulated in mlx1 mutants (Figure 6D). Knockdown of PFK2 led to elevated circulating glucose, showing that PFK2 is a Mondo-Mlx downstream target, which contributes to circulating glucose levels (Figure 7H). Interestingly, PFK2 knockdown also reduced pupation on high sugar diet (20% yeast-15% sucrose) (Figure 7I), implying that Mondo-Mlx-mediated activation of the glycolytic pathway contributes to dietary sugar tolerance.
Our study demonstrates that interfering with the function of the Mondo-Mlx complex severely affects Drosophila energy metabolism, rendering animals highly intolerant to sugars in their diet. The sugar intolerance is likely due to a combined effect of several downstream effectors, since our systematic loss-of-function analysis revealed three Mlx target genes that are essential for survival on high sugar diet. Sugar tolerance is influenced by glycolysis, which also contributes to clearance of glucose from circulation. However, circulating glucose levels and sugar tolerance are phenotypes that can be uncoupled, as in the case of two Mlx targets, Cabut and Aldh-III, which only contribute to sugar tolerance.
Mondo-Mlx shows a high degree of functional conservation between flies and mammals, as orthologs of many Drosophila Mlx-regulated genes are known targets of ChREBP/MondoA-Mlx in mammals. These include glycerol-3-phosphate dehydrogenase-1, stearoyl-CoA 9-desaturase-1, fatty acid synthase, acetyl-CoA carboxylase, and phosphofructokinase 2 [15], [21], [25]. Drosophila Mlx displayed an essential role in the fat body, which is the counterpart of mammalian adipose tissue and liver. Thus, it is conceivable that the liver and adipose tissue-specific ChREBP-Mlx, instead of the muscle-specific MondoA-Mlx, represents the ancestral function of the heterodimer.
This study identifies Aldh-III as a novel gene contributing to dietary sugar tolerance. Aldh-III is the ortholog of mammalian Aldh3 family. Aldehyde dehydrogenases (Aldhs) are highly conserved NAD(P)+ -dependent enzymes that oxidize aldehydes to the corresponding carboxylic acid and act on a broad range of substrates [34]. Aldehydes are highly reactive compounds forming adducts with nucleic acids and proteins, thus disturbing cellular functions. Reactive aldehydes can originate from exogenous sources or be products of cellular metabolism. Aldhs have been shown to provide protection against a number of ectopic stresses, including toxic chemicals, heat stress, and UV irradiation [34]–[37]. One of the best-established functions of Aldh proteins is neutralization of acetaldehyde, a toxic metabolite of ethanol. Acetaldehyde elimination is mainly mediated by Aldh2 [34], [38]. Polymorphism in the Aldh2 gene is common is Asian populations and it leads to poor ethanol tolerance [39]. Our finding that another member of the Aldh family, Aldh-III, is essential for dietary sugar tolerance is intriguing, since there are multiple parallels between the hepatic pathophysiologies related to excessive ethanol and fructose consumption in humans [3]. Of note, we observed that Aldh-III is sufficient in rescuing the sugar intolerance of mlx1 mutants in the fat body, the insect counterpart of liver, suggesting that metabolic stress-induced dysfunction of the fat body contributes to sugar intolerance. Future studies should be aimed at understanding the generation of sugar-derived reactive aldehydes and their molecular targets.
We also provide evidence that PFK2 expression is positively regulated by Mondo-Mlx in Drosophila and that the PFK2 levels are critical in managing circulating glucose levels and providing sugar tolerance. Thus, our data suggests that regulation of PFK2 gene expression might be a suitable strategy to manage hyperglycemia in diabetes. Elevated PFK2 expression has also been associated with the high rate of glycolytic flux in neoplastic tumors [40]. Exploring the contribution of Mondo and Mlx proteins in this setting is therefore warranted. In addition to being regulated transcriptionally, PFK2 activity in mammals is known to be posttranslationally regulated by insulin signalling through protein kinase AKT [41], [42]. It will be interesting to learn, what is the contribution of insulin signalling pathway activity on the dietary sugar tolerance.
The finding that Cabut is an essential secondary effector of Mondo-Mlx is intriguing. In fact, cabut expression has been reported to respond to other metabolism-related signals: it is upregulated upon inhibition of TOR complex 1 signalling [43], which is likely mediated by activation of the Forkhead transcription factor, FoxA [44]. While the developmental role of Cabut has been studied [45], [46], its metabolic functions have remained unexplored. Our finding showing that Cabut plays an essential metabolic role in providing dietary sugar tolerance implies this topic deserves an in-depth survey in the future. Notably, the closest mammalian homologs of Cabut, Klf-10 and Klf-11, have been linked to metabolic regulation. Mutations in the klf-11 locus are associated with risk of diabetes [47], while Klf-10 appears to negatively regulate lipogenic genes in hepatocytes [48]. klf-10 expression is regulated both by circadian signals [49] as well as by ChREBP upon high glucose [48]. Thus, Klf-10 might be the functional ortholog of Cabut. Future studies should be aimed at identifying the Cabut target genes involved in its metabolic functions.
Dietary sugar tolerance displays a wide natural variation, even in closely related animal species. For example, two Drosophila species, D. melanogaster and D. mojavensis, have strikingly different tolerance to dietary sugars [6]. In contrast to the fruit generalist D. melanogaster, D. mojavensis is a cactus breeder, which does not naturally encounter high levels of simple sugars and displays poor survival on sugar-rich diet [6]. Based on our data, it is possible to hypothesize that Mondo-Mlx-regulated transcriptional network contributes to the natural variation in sugar tolerance. Genetic differences in sugar tolerance are also observed in humans. For example, hereditary fructose intolerance (HFI) is caused by mutations in the aldolase-B gene [50]. A fructose-restricted diet renders HFI relatively benign, but ingestion of fructose or sucrose leads to strong symptoms, including nausea and vomiting as well as a risk of liver and kidney damage. It will be important to explore whether genetic changes in the ChREBP/MondoA-Mlx network influence individual's risk for sugar overload-induced metabolic disturbance.
This study highlights the usefulness of Drosophila as a model for systematically exploring the genetic factors defining the range of healthful nutrient intake. Notably, the sugar intolerance in mlx1 mutants is not a pleiotropic consequence of a generally disturbed energy metabolism. Knockdown of other key transcriptional metabolic regulators, such as SREBP, did not cause notable sugar intolerance (unpublished observation). The availability of genome-scale reagents, including in vivo RNAi lines, offers the possibility for the systematic dissection of genes contributing to dietary sugar tolerance. These genes might include novel members of the Mondo-Mlx-regulated genetic network, but they will also uncover whether parallel regulatory pathways are involved. The Drosophila model is also particularly useful in dissecting the function of transcription factors that are master regulators of several gene groups that contribute to distinct physiological outputs. While this study has focused on uncovering those Mondo-Mlx targets that contribute to sugar tolerance and circulating glucose levels, uncovering the downstream effectors behind the other metabolic phenotypes of mlx1 mutant fly awaits for future studies.
P{XP}bigmaxd07258 were obtained from Bloomington stock center. For generating UAS-mlx flies, the coding region of mlx cDNA was amplified by PCR and cloned into pUAST vector using BglII and XbaI restriction sites. FLAG-tag was incorporated into the C-terminus. Aldh-III coding region was cloned into pUAST vector using NotI and XhoI restriction sites. RNAi lines were obtained from Vienna Drosophila RNAi Center and from NIG-FLY Stock Center. The following GAL4 driver lines were used in this study: tub-GAL4 [51], ppl-GAL4 [52], r4-GAL4 [53], Elav-GAL4 [54] and Mef2-GAL4 [55].
In standard conditions flies were maintained at 25°C on medium containing agar 0.6% (w/v), semolina 3.2% (w/v), malt 6.5% (w/v), dry baker's yeast 1.8% (w/v), propionic acid 0.7% (v/v) and Nipagin (methylparaben) 2.4% (v/v). For defined nutrient studies, larvae were grown on food containing 20% (w/v) dry baker's yeast, 0.5% (w/v) agar, and 2.5% (v/v) Nipagin (methylparaben) in PBS supplemented with varying concentrations of sucrose, glucose or fructose. 1st instar larvae were collected from apple juice plates (apple juice 33.33% (v/v), agar 1.75% (w/v), sugar 2.5% (w/v) and Nipagin (methylparaben) 2.0% (v/v)) and larvae were grown at controlled density (30 larvae per vial).
For generating anti-Mlx antibodies full-length Drosophila mlx cDNA was cloned into pGEX-4T2. Recombinant GST-Mlx was purified using Glutathione-agarose (Sigma). Anti-Mlx antiserum was raised by immunizing a guinea pig (Storkbio Ltd).
Drosophila S2 cells were grown at 25°C in standard Shields and Sang M3 medium (Sigma) containing 2% of fetal bovine serum (Gibco), 1× insect medium supplement (Sigma) and penicillin/streptomycin (Gibco). The transfections were performed using Effectene (Qiagen), according to manufacturer's protocol. Expression of transfected genes was induced with 1.2 µM CuSO4 24 h post-transfection.
For detecting endogenous Mlx in vivo, 3rd instar prewandering larvae were homogenized in Laemmli sample buffer and boiled for 5 min. Samples were resolved on SDS-PAGE and detected by Western blotting using anti-Mlx antibodies. For the pulldown experiment, cells were lysed in IP lysis buffer (10 mM Tris-HCl pH 8.0, 150 mM NaCl, 0,1% NP40) and lysates were cleared by centrifugation. Lysate protein concentration was adjusted to 1 µg/µl. 1 ml of lysate was incubated o/n with 25 µl Strep-Tactin beads (IBA). The beads were washed 5 times with IP lysis buffer, after which Laemmli sample buffer was added and samples were boiled for 5 min. Pulldown and lysate samples were resolved on SDS-PAGE, transferred to nitrocellulose and analyzed by Western blotting using anti-V5 (Invitrogen), anti-Mlx and anti-Kinesin (Cytoskeleton).
Metabolomics was performed using prewandering 3rd instar larvae grown on 20% yeast-5% sucrose diet. Analysis was done in four biological replicas. Data was processed using Guineu [56] and MZmine 2 [57] software packages for small polar metabolites and molecular lipids, respectively. Detailed description is available in Protocol S1.
Metabolite assays were done using prewandering 3rd instar larvae grown on 20% yeast or 20% yeast-5% sucrose diet. All analyses were done at least in four biological replicas. Glucose, trehalose and glycogen measurements were conducted as described [58], [59].
Staged 3rd instar prewandering control and mlx1 mutant larvae were grown on a 20% yeast-5% sucrose diet. RNA was extracted from 3–4 larval fat bodies per sample in four biological replicas using the Nucleospin RNA XS kit (Macherey-Nagel). The Amino Allyl MessageAmp II aRNA Amplification kit (Ambion) was used for aRNA synthesis. Hybridization to Agilent Drosophila Gene Expression Microarray, 4×44K was performed according to the manufacturer's instructions. Data normalization and analysis was performed using R, taking advantage of packages from the Bioconductor repository. A full documentation of the data analysis is available in Protocol S1. For quantitative RT-PCR, the RevertAid H Minus First Strand cDNA Synthesis Kit (Fermentas) with random hexamer primers was used for first strand cDNA synthesis. PCR was performed using Maxima SYBR Green qPCR Master Mix (2X) (Fermentas) and analyzed on StepOnePlus (Applied Biosystems) real-time PCR system. Primer sequences are available as Table S3.
All expression profiling data is available under accession E-MTAB-699 at the ArrayExpress repository.
Statistical significance for each experiment (excluding metabolomics and microarray) was determined with unpaired Student's t-test with unequal variance. All quantitative data are presented as mean ± SEM for a minimum of three independent biological replicates.
|
10.1371/journal.ppat.1003554 | Stabilization of Myc through Heterotypic Poly-Ubiquitination by mLANA Is Critical for γ-Herpesvirus Lymphoproliferation | Host colonization by lymphotropic γ-herpesviruses depends critically on expansion of viral genomes in germinal center (GC) B-cells. Myc is essential for the formation and maintenance of GCs. Yet, the role of Myc in the pathogenesis of γ-herpesviruses is still largely unknown. In this study, Myc was shown to be essential for the lymphotropic γ-herpesvirus MuHV-4 biology as infected cells exhibited increased expression of Myc signature genes and the virus was unable to expand in Myc defficient GC B-cells. We describe a novel strategy of a viral protein activating Myc through increased protein stability resulting in increased progression through the cell cycle. This is acomplished by modulating a physiological post-translational regulatory pathway of Myc. The molecular mechanism involves Myc heterotypic poly-ubiquitination mediated via the viral E3 ubiquitin-ligase mLANA protein. EC5SmLANA modulates cellular control of Myc turnover by antagonizing SCFFbw7 mediated proteasomal degradation of Myc, mimicking SCFβ-TrCP. The findings here reported reveal that modulation of Myc is essential for γ-herpesvirus persistent infection, establishing a link between virus induced lymphoproliferation and disease.
| Being obligatory intracellular parasites, it is not surprising that viruses have evolved mechanisms to induce cellular proliferation to promote their own life cycle. This is notorious in the case of γ-herpesviruses, such as Epstein-Barr virus (EBV) and Kaposi's sarcoma virus (KSHV), which are human pathogens associated with lymphoproliferative disease and several tumors. Host colonization by γ-herpesviruses is critically dependent on the ability to expand latent infection in proliferating B-cells. Virus-induced cellular proliferation is a process mediated by the expression of specific viral proteins. One of such proteins is the latency-associated protein (LANA) of KSHV. In this study, we use murid herpesvirus-4 (MuHV-4) as a mouse model of γ-herpesvirus pathogenesis. We show that the MuHV-4 LANA (mLANA) stabilizes the cellular oncogene Myc, increasing its half-life, thus promoting its activity as a potent inducer of cellular proliferation. The molecular mechanism involves heterotypic poly-ubiquitination of Myc mediated via mLANA. The findings here reported demonstrate that modulation of Myc is essential for γ-herpesvirus persistent infection, establishing a link between virus induced lymphoproliferation and disease. The implication is that revealing a critical function of a viral protein possibly allows the development of small molecule probes to disrupt mLANA-Myc interaction, therefore inhibit virus induced lyhophoproliferative disease.
| Myc is a transcription factor that enhances the expression of genes involved in cellular growth and proliferation. Hence, it is not surprising that viruses have evolved mechanisms to modulate Myc to promote their own life cycle. Myc heterodimerizes with Max, through a basic region/helix-loop-helix/leucine-zipper domain, to regulate the transcription of specific E-box-containing genes in response to mitogenic stimuli. Myc functions as a universal amplifier of gene expression by promoting the transcriptional elongation of RNA polymerase II driving biomass accumulation and enhanced cellular bioenergetic pathways [1], [2], [3]. The expression of c-myc is tightly regulated with extremely short half-lives for mRNA and protein. In non-transformed cells, Myc is continuously subjected to ubiquitination and proteasomal-degradation, resulting in a highly unstable protein with a half-life of about 15–20 minutes [4]. Several mechanisms of Myc regulation have been identified that operate at the level of protein stability. The best characterized mechanism involves the interplay between phosphorylation at two specific residues and ubiquitination. Phosphorylation at serine (S) 62 by extracellular signal-regulated kinase (ERK) stabilizes Myc resulting in enhancement of its transcription activity. In contrast, phosphorylation of Myc at threonine (T) 58 by glycogen synthase kinase 3 (Gsk-3), which is dependent on previous phosphorylation of Myc at S62, leads to proteasomal degradation of Myc [5]. The mechanism involves the assembly of homotypic poly-ubiquitin chains on Myc specifically dependent on lysine (K) 48 linkage by SCF (Skp1/Cul/Fbox)Fbw7 [6], [7]. Myc turnover by SCFFbw7 is antagonized by polymerization of mixed heterotypic poly-ubiquitination chains via SCFβ-TrCP on the N-terminus of Myc [8]. Thus, SCFFbw7 and SCFβ-TrCP assemble different K-linkage poly-ubiquitin chains with functionally distinct outcomes on Myc stability, i.e., degradation versus stability. The physiological relevance of regulating Myc activity through protein stability is underscored by observations that point mutations at or near T58, which render Myc resistant to proteasomal degradation, occur with high frequency in B-cell lymphomas [9].
Examples of viruses that modulate Myc activity include Kaposi's sarcoma associated herpesvirus (KSHV) and Epstein-Barr virus (EBV). Infection by these γ-herpesviruses is characterized by the establishment of latent infection in memory B-cells. Access to this cell type is gained by virus-driven proliferation of germinal centre (GC) B-cells [10], where virus genomes replicate and segregate in step with normal cell division. This process is mediated by episomal maintenance proteins, which include EBV nuclear antigen-1 (EBNA-1) [11] and latency associated nuclear antigen (LANA) encoded by ORF73 of γ-2-herpesviruses [12]. Given the essential role of Myc for the initiation and maintenance of GCs [13], [14], it is not surprising that γ-herpesviruses have evolved mechanisms to modulate Myc activity. In the case of KSHV-associated primary effusion lymphoma, Myc was shown to be abnormally stabilized [15], [16]. The mechanism involves the direct interaction of the viral protein LANA with Gsk-3 resulting on reduced levels of Myc T58 phosphorylation [17]. Another strategy appears to be employed by EBV encoded EBNA-3C protein. This viral protein was shown to increase the transcriptional activity of Myc through an interaction with both Myc and SCFSkp2. Surprisingly the mechanism proposed does not involve poly-ubiquitination but rather the action of SCFSkp2 functioning as a transcription co-factor for Myc [18].
Here, we utilized murid herpesvirus-4 (HuHV-4) infection of mice as a model system to address the role of Myc in the pathogenesis of γ-herpesviruses. We show that Myc expression is required for the expansion of MuHV-4 infection in GC B-cells. The mechanism involves heterotypic poly-ubiquitination of Myc mediated through the ElonginC/Cullin5/SOCS (supressors of cytokine signaling) (EC5S) E3 ubiquitin-ligase activity of mLANA encoded by ORF73 of MuHV-4. EC5SmLANA mimics SCFβ-TrCP by antagonizing SCFFbw7-mediated proteasomal turnover of Myc but unlike the cellular E3 ubiquitin-ligases its activity is not dependent on the phosphorylation status of Myc. Our results underscore the importance of modulating Myc activity during γ-herpesvirus driven lymphoproliferation providing a link between persistent infection and lymphoproliferative disease.
Experiments were designed to investigate the role of Myc expression on gammaherpesvirus pathogenesis. We utilized MuHV-4 infection of laboratory mice as the model, which is characterized by the expansion of latently infected B-cells in GCs and virus persistence in memory B-cells [19]. We analysed the transcription of several Myc target genes in infected versus non-infected GC B-cells, purified from the same pool of splenocytes. We utilized a recombinant MuHV-4 expressing a yellow fluorescent protein (YFP) [20] to segregate infected (CD19+CD95hiGL7hiYFP+) from non-infected (CD19+CD95hiGL7hiYFP−) GC B-cells derived from C57BL/6 mice, at day 13 post-infection. We analysed the transcription of Myc signature genes involved in cell cycle entry (encoding for cyclins B1, D1, D2 and E, and cyclin-dependent kinase 4) and GC B-cell activation (encoding for IL-10, B-ATF, MIF and CD70). Cyclin D3 was included as a gene non-regulated by Myc. The transcription of Myc target genes was significantly increased in infected GC B-cells when compared with their non-infected counterparts (Figure 1A). These data show that during the expansion of latent infection in GC B-cells, MuHV-4 induces a transcription profile compatible with increased Myc transcriptional activity.
To define the impact of Myc expression on gamma-herpesvirus pathogenesis, we generated mice with conditional deletion of c-myc in B-cells undergoing GC reaction. This was achieved by breeding homozygous c-mycfl/fl mice, where second and third exons of the c-myc locus are flanked by two loxP sites [21] to heterozygous Cγ1-cre mice, in which expression of Cre recombinase is induced by transcription of the Ig-γ1 constant region gene segment early in GC development during immunoglobulin class-switch recombination [22]. Resulting progeny Cγ1-creKI/WT;c-mycfl/fl, hereafter designated GC Myc KO, is expected to have specific deletion of c-myc in class-switched GC B-cells, after immunization with T-dependent antigens. Thus, we next investigated GC responses in the absence of Myc expression. GC Myc KO mice were immunized with the Th-2 cell-dependent antigen 4-hydroxy-3-nitrophenylacetyl (NP)–chicken γ-globulin (CGG) adsorbed to alum. Frequencies of GC B-cells (CD19+CD95hiGL7hi) were analysed at day 10 post-immunization. Compared to control litter mates Cγ1-creWT/WT;c-mycfl/fl, hereafter designated control mice, GC Myc KO mice showed a marked impairment in GC development (Figure 1B), accompanied by a strong reduction in IgG1+ B-cell numbers (CD19+IgD−IgM−IgG1+) (Figure 1C). Immunization with ovalbumin (OVA) emulsified in complete Freund's adjuvant (CFA) further confirmed that GC Myc KO mice were defective to mount a normal GC response, though less deficient than upon immunization with NP-CGG, and presented reduced levels of IgG1 expressing B-cells (Figure 1D and 1E, respectively). However frequencies of IgG2a/2b+ B-cells (CD19+IgD−IgM−IgG2a/2b+), whose class-switching is not strictly dependent on Ig-γ1 promoter, revealed that GC Myc KO mice are competent in developing IgG2 class-switched B-cells (Figure 1F). These data demonstrate a clear requirement for Myc expression in order to generate GC reactions to T cell dependent antigens. Our results are in direct agreement with those recently published that demonstrate the lack of GCs in mice in which c-myc is ablated early during GC induction [13]. Since we obtained the two transgenic mice lines carrying the alleles c-mycfl/fl and Cγ1-cre from these authors, and independently generated GC Myc KO mice, our data are directly comparable.
The requirement of Myc for gammaherpesvirus pathogenesis was next investigated by infecting GC Myc KO mice with MuHV-4. Analysis of the percentages of GC B-cells revealed no significant differences between control and GC Myc KO mice (Figure 2A), with the majority of infected GC B-cells falling into dark zone (Figure 2B) as previously described for infection of wild type mice [23]. To determine if MuHV-4 infected GC B-cells had been or not exposed to Cre-mediated c-myc deletion, we FACS purified infected cells from GC Myc KO mice, at day 14 post-infection. A PCR assay was employed to detect floxed and deleted c-myc alleles in DNA from infected GC B-cells, compared to DNA from uninfected total B-cells (CD19+YFP−), purified from the same pool of splenocytes. Infected cells were found to have undeleted (floxed) c-myc allele and no c-myc rearrangement could be detected (Figure 2C). This contrasted with total non-infected B-cell population where c-myc deletion could be readily detected (Figure 2C). To further confirm the integrity of the c-myc locus in infected GC B-cells in GC Myc KO mice, we quantified Myc mRNA levels. Comparison of GC B-cells from GC Myc KO with wild type infected mice revealed no significant differences in Myc transcription (Figure 2D). These data imply that MuHV-4 is expanding exclusively in GC B-cells where the c-myc locus did not undergo Cre-mediated deletion. Accordingly, GC Myc KO mice infected with MuHV-4 were unable to class-switch to IgG1, which is consistent with Myc deficiency in these cells (Figure 2E). However, these infected mice showed high numbers of IgG2a/2b positive B-cells, equivalent to infected control mice counterparts (Figure 2F). Quantification of the frequency of viral DNA positive cells in GCs, assessed by limiting dilution combined with real time PCR, showed an approximately 10-fold deficit of latent infection at 14 days post-infection in GC Myc KO mice (Figure 2G). This deficit was likely a reflection of infection of B-cells that did not result in the expansion in GC reactions due to Ig-γ1 Cre mediated c-myc deletion. However, at day 21 post-infection latent viral loads in control and conditional KO mice were equivalent (Figure 2G), indicating a recovery from the early deficit in expansion. Collectively data obtained with GC Myc KO mice demonstrated that infection with MuHV-4 does not compensate for Myc loss in GC B-cells and virus is found to amplify exclusively in c-myc intact cells. Therefore, Myc is essential for the expansion of latently infected GC B-cells, thus critical for the establishment of persistent infection.
We have shown before that the ORF73 protein encoded by MuHV-4, designated mLANA by homology with the latency associated nuclear antigen encoded by KSHV, is selectively transcribed in GC B-cells [24]. Thus mLANA was a strong candidate to be responsible for the observed increased transcription of Myc target genes in MuHV-4 infected GC B-cells. To address this hypothesis, we analysed the transcription of Myc target genes in mLANA expressing cells. When compared to control transfected cells, mLANA expression induced the transcription of all Myc target genes analysed (Figure 3A). Expression of mLANA had no effect on c-myc mRNA levels indicating that its putative modulatory effect on Myc was post-transcriptional (Figure 3B). We have also shown before that mLANA acts as the substrate recognition factor of an ElonginC-Cullin5-SOCS (suppressor of cytokine signalling) (EC5S) E3 ubiquitin-ligase towards the p65/RelA cellular transcription factor NF-κB [25]. The mechanism involves the assembly of an EC5S -like complex, mediated by a viral unconventional SOCS-box motif present in mLANA. Hence, we analysed if mLANA-mediated modulation of transcription of Myc target genes could be attributed to its function as an E3 ubiquitin-ligase. To this end, we utilized a previously characterized mLANA mutant, designated mLANA-SOCS where residues V199, L202, P203 and P206 were substituted by alanines abrogating E3 ubiquitin-ligase function [25]. This mutant was no longer able to modulate the expression of Myc target genes when compared with intact mLANA (Figure 3A). To define if the observed mLANA modulatory effect on cellular transcription was Myc specific, we carried out gene reporter assays using a synthetic Myc reporter plasmid containing three copies of E-box sequences driving the expression of luciferase. As expected, overexpression of Myc was translated into a significant increase on luciferase activity (Figure 3C). Cells expressing mLANA exhibited comparable levels of luciferase activity, which increased further when Myc was concomitantly expressed. In contrast, mLANA-SOCS expression showed no effect on luciferase levels. We next proceeded to analyse the modulatory effect of mLANA on Myc in B-cells, which are physiological more relevant given the tropism of MuHV-4. When compared to control and mLANA-SOCS transfected cells, mLANA expression induced the transcription of all Myc target genes analysed (Figure 3D). As before, expression of mLANA in A20 B-cells had no effect on c-myc mRNA levels confirming that its modulatory effect on Myc was post-transcriptional (Figure 3E). Myc transcriptional activation of genes encoding proteins involved in cell cycle entry results in transition from G0-G1 to S phase. Thus, we analysed cell cycle profiles in B-cells expressing mLANA in comparison to control or mLANA-SOCS. These experiments showed a clear decrease in the number of mLANA expressing cells in G1 phase, with a concomitant increase in the number of cells in S and G2-M phases (Figure 3F). Combined these data indicate that mLANA is modulating Myc-dependent transcription and progression through cell cycle in B-cells through its activity as an EC5S E3 ubiquitin-ligase. Co-immunoprecipitation experiments also showed that mLANA and Myc exist in the same heteromolecular complex in a context of virus infection. This was demonstrated using a murine B-cell lymphoma-derived cell line latently infected with MuHV-4, designated S11 cells (Figure 3G). We also showed that this interaction was reduced for mLANA-SOCS (Figure 3H, compare lanes 3 and 5). This reduction could be due to lower levels of Myc in mLANA-SOCS expressing cells, when compared to mLANA, indicating that the interaction is independent of the SOCS-box motif. Alternatively it is plausible that the SOCS-box is participating in mLANA-Myc interaction.
We next set out experiments to investigate if EC5SmLANA was able to mediate poly-ubiquitination of Myc. We started by performing a nickel-nitrilotriacetic acid (Ni-NTA) pull-down in the presence of histidine-tagged ubiquitin. Upon culture and cell lysis, ubiquitinated proteins were extracted from total cellular lysates with Ni-NTA beads and resolved by SDS-PAGE. The levels of ubiquitinated Myc present in each condition were analysed by immunoblotting. We observed that when mLANA was expressed, the levels of ubiquitinated Myc were significantly enhanced (Figure 3I, compare lanes 3 and 4). To confirm this activity in a more relevant biological context, we evaluated the ability of mLANA to promote the ubiquitination of endogenously expressed Myc. In comparison with control transfected and mLANA-SOCS transfected cells, higher levels of ubiquitinated Myc were detected in the presence of intact mLANA (Figure 3J, compare lane 2 with 1 and 3). We have previously demonstrated that mLANA E3 ubiquitin-ligase activity towards p65/RelA required the E2 ubiquitin-conjugating enzyme UbcH5 [25]. Here we showed that EC5SmLANA requires UbcH5 as its E2 conjugating partner to mediate poly-ubiquitination of Myc (Figure S1). Myc protein has 25 lysine residues that can be potentially ubiquitinated. To define if the ubiquitination ladder observed was due to the ability of mLANA to mediate poly-ubiquitination or multiple mono-ubiquitination of Myc in different lysine residues, we made use of a previously described lysine-free (K−) version of Myc, which can only be ubiquitinated on its N-terminal residue [8]. By performing an in vivo ubiquitination assay with K−Myc, we observed the following. First, mLANA was able to mediate poly-ubiquitination of the N-terminal residue of Myc (Figure 3K, compare lanes 3 and 4). Secondly, K−Myc in the presence of mLANA exhibited a ladder of ubiquitination indicating the assembly of poly-ubiquitin chains (Figure 3K, compare lanes 3 and 4). Finally, the ubiquitination pattern of wild type Myc and K−Myc are distinctive, implicating that other lysine residue(s) are targets for mLANA-mediated poly-ubiquitination of Myc.
The addition of ubiquitin chains to a target protein can occur with different moieties that are emerging as determinants of biological outcome. These include, mono-ubiquitination and poly-ubiquitination. Ubiquin chains can be assembled using seven internal lysine residues (K) and thus define homotypic poly-ubiquitination (same K-linkage) or heterotypic poly-ubiquitination (mixed K-linkages) [26]. To investigate the type of lysine (K)-linkage involved in mLANA-mediated poly-ubiquitination of Myc we utilized of a series of ubiquitin mutants with every single lysine residue, of the possible seven, substituted by an arginine. By performing in vivo ubiquitination assays in mLANA expressing cells we observed that in the absence of K33, K48 or K63 residues the ability of mLANA to promote Myc poly-ubiquitination was suppressed (Figure 4A). Next, Myc transcription reporter assays were carried out in cells expressing the same ubiquitin mutants. Consistent with the poly-ubiquitination pattern, replacement of K33, K48 and K63 in ubiquitin rendered mLANA unable to positively modulate Myc transcriptional activity (Figure 4B). Measurement of Myc cellular levels in extracts from mock transfected control or mLANA-transfected cells co-expressing each ubiquitin mutant, revealed that in the presence of wild type ubiquitin, mLANA-expressing cells exhibit augmented Myc levels (Figure 4C, upper panel, compare lanes 1 and 2). The same result is observed when residues K6, K11, K27 and K29 of ubiquitin were substituted by arginines (Figure 4C, upper and lower panels). However, in good agreement with the previous data, substitution of K33, K48 and K63, diminished or abolished the ability of mLANA to promote the increase in Myc cellular levels (Figure 4C). Our observations demonstrate the involvement of different lysine-linkages in EC5SmLANA Myc poly-ubiquitination. However, these data do not distinguish between homotypic poly-ubiquitination at different K residues in Myc from mixed K-linkage poly-ubiquitination. Hence, we next utilized K−Myc where ubiquitination is only possible at the first metionine. In vivo ubiquitination assays and transcription reporter assays with K−Myc, in combination with each of the K ubiquitin mutants, essentially recapitulated the above observed results with wild type Myc (Figure 4D and E). Hence, heterotypic poly-ubiquitination of the N-terminus of Myc is sufficient for mLANA modulatory activity. Collectively, these data show that EC5SmLANA requires different K-linkages to ubiquitinate Myc. Moreover, they demonstrate a direct correlation between the ability of mLANA to mediate Myc poly-ubiquitination, to increase Myc cellular levels, and to promote its transcriptional activity, supporting that all three activities are directly linked.
Although classically associated with protein degradation, ubiquitination is now emerging as regulator of a wide variety of non proteolytic cellular signalling functions [26]. Therefore, and in agreement with a positive effect on Myc activity, we further confirmed that mLANA-mediated poly-ubiquitination of Myc was non-degradative. To that end, we performed an in vivo ubiquitination assay in the presence of the proteasome inhibitor MG132. In control-transfected cells, the presence of MG132 favoured the accumulation of Myc ubiquitinated species (Figure 5A, top panel, compare lanes 1 and 2), as well as the cellular levels of Myc protein (Figure 5A, bottom panel, compare lanes 1 and 2). In contrast, in cells expressing mLANA, inhibition of proteasomal degradation had a negligible influence on both Myc poly-ubiquitination and Myc protein levels (Figure 5A, compare lanes 3 and 4). The importance of these results is threefold. Firstly, they confirm that under physiological conditions Myc turnover is highly regulated by the proteasome. Secondly, they show that treatment with the proteasome inhibitor MG132 has no effect on mLANA-mediated poly-ubiquitination of Myc, thus mLANA in not directing Myc for proteasomal degradation. Thirdly, the increase in Myc levels in response to mLANA expression was not altered under conditions of proteasomal inhibition, suggesting that expression of mLANA is preventing the proteasomal degradation of Myc. To assess this hypothesis, we next compared the half-life of Myc in control and mLANA expressing cells. Cells were treated with the protein synthesis inhibitor cycloheximide (CHX), and Myc protein levels were analyzed by immunoblotting at different time-points post-treatment. Under these experimental conditions, expression of mLANA led to a pronounced raise in Myc stability with an increase of half-life from ≈20 minutes to over 4 hours (Figure 5B). Collectively these results demonstrate that mLANA expression has a positive effect on Myc cellular levels by preventing its proteasomal turnover, thus prolonging its half-life, which is associated with increased transcriptional activity.
Under physiological conditions Myc half-life is tightly regulated through poly-ubiquitination by two distinct Skp1/Cul1/F-box (SCF) E3 ubiquitin-ligases. That is, poly-ubiquitination of Myc by SCFβ-TrCP antagonizes SCFFbw7-mediated proteasomal dependent turnover [6], [8]. Therefore, we hypothesised if mLANA could be modulating Myc stability by counteracting degradation by Fbw7 and mimicking the activity of β-TrCP. We overexpressed Fbw7 and analysed Myc cellular levels in the presence of co-expressed mLANA. As expected, overexpression of Fbw7 led to a decrease in Myc levels (Figure 5C, compare lanes 1 and 2). Overexpression of Fbw7 had no effect on Myc levels when mLANA was concomitantly expressed (Figure 5C, compare lanes 3 and 4). Notably, the antagonizing activity of mLANA towards Fbw7 was more pronounced when compared with that afforded by β-TrCP (Figure 5C, compare lanes 4 and 6). We further characterized the mLANA effect on Fbw7 and β-TrCP interplay by depletion of the expression of the two cellular E3 ubiquitin-ligases and analysis of Myc levels. When Fbw7 was depleted Myc levels increased further by the presence of mLANA (Figure 5D, compare lanes 3 and 4). In agreement, the turnover effect on Myc levels caused by depletion of β-TrCP was counteracted by concomitant expression of mLANA (Figure 5E, compare lanes 3 and 4). Together these results demonstrate that mLANA mimics SCFβ-TrCP by antagonizing SCFFbw7-mediated proteasomal turnover of Myc.
Fbw7 control of Myc turnover is dependent on the interaction between the ubiquitin-ligase and its substrate. Fbw7 recognizes Myc when phosphorylated on threonine (T) 58 and catalyses its poly-ubiquitination resulting in Myc proteasomal degradation [6], [7]. Phosphorylation of Myc on T58 is sequentially preceded by phosphorylation on serine (S) 62, which activates and promotes Myc stability [5]. Thus T58 and S62 are key phospho-residues that regulate Myc activity at the protein level. Therefore, we set to investigate the influence of Myc phosphorylation on the modulatory activity of mLANA. Using phospho-specific antibodies we observed that under mLANA expression Myc is phosphorylated on both S62 and T58 (Figure 6A, lane 2, first and second panels, respectively). To analyse if sequential phosphorylation of Myc was intact on mLANA expressing cells we proceed to substitute S62 or T58 to alanines (A) on Myc and assess the phosphorylation status of those Myc mutants. Compatible with the model of sequential phosphorylation, in which phosphorylation of S62 precedes phosphorylation of T58, MycT58A was readily phosphorylated on S62, whereas MycS62A was not phosphorylated (Figure 6A, lanes 3–6, first and second panels, respectively). Remarkably, analysis of total Myc cellular levels revealed that, regardless of Myc phosphorylation on either S62 or T58, co-expression of mLANA led to increased Myc levels (Figure 6A, third panel). These data not only demonstrate that mLANA is not interfering with Myc phosphorylation, but also indicates that mLANA modulatory functions override cellular pathways that control Myc activity. Consistent with this hypothesis, mLANA is co-immunoprecipitated by both Myc mutants (Figure 6B), and it is able to mediate their poly-ubiquitination (Figure 6C). Transcriptional activities of MycT58A and MycS62A were also increased by expression of mLANA (Figure 6D), further supporting that mLANA targets Myc independently of cellular mechanisms of Myc regulation.
In this study, we describe the first example of a viral protein activating Myc transcriptional activity through increased protein stability by mimicking a physiological post-translational regulatory pathway. Modulation of Myc function was shown to be essential for the lymphotropic γ-herpesvirus MuHV-4 biology as infected cells exhibit increased expression of known Myc target-genes, and using a genetic approach, virus was found to amplify exclusively in intact Myc GC B-cells. The molecular mechanism involved heterotypic poly-ubiquitination of Myc mediated via the mLANA protein encoded by ORF73. This was reminiscent of a newly described pathway of Myc regulation through poly-ubiquitination. Popov et al. showed that the cellular E3 ubiquitin-ligase SCFβ-TrCP uses UbcH5 ubiquitin-conjugating enzyme to form heterotypic poly-ubiquitin chains on the N-terminus of Myc. Poly-ubiquitination of Myc by SCFβ-TrCP leads to Myc stabilization and was shown to antagonize SCFFbw7-mediated proteasomal turnover of Myc [8]. Like SCFβ-TrCP, EC5SmLANA uses UbcH5 and antagonizes SCFFbw7. Furthermore, as previously demonstrated for SCFβ-TrCP, single substitutions of K33, K48 or K63 of ubiquitin reduced or eliminated the ability of EC5SmLANA to poly-ubiquitinate, stabilize or increase the transcriptional activity of Myc. However, our results suggest that the molecular mechanism of mLANA modulation of Myc activity is not limited to Fbw7 antagonism. This is supported by the fact that mLANA protective effects on Myc stability, in conditions of Fbw7 over expression, are significantly more pronounced than observed for β-TrCP. Moreover, mLANA was able to increase the stability and activity of MycT58A, a Myc version that is not recognized by Fbw7. This contrasts with what has been reported for β-TrCP that albeit being able to poly-ubiquitinate MycT58A it did not impact on its turnover [8]. Thus, the ability of mLANA to increase Myc transcriptional activity through poly-ubiquitination, independently of the phosphorylation status of Myc on S58 and T62, indicates that this novel viral modulatory mechanism does not rely on post-translation cellular regulation.
Modulation of host B-cell biology is of vital importance for γ- herpesviruses, as they depend critically on the expansion of latently infected B-cell in GC reactions for host colonization. GC reactions exhibit two distinct morphological areas. These include the dark zone (DZ), where centroblasts are rapidly dividing, and the light zone (LZ), where B-cells exit the cell cycle to differentiate into plasma cells or memory B-cells or reenter the DZ for additional rounds of cell division. Recently two studies have established the importance of Myc for GC biology [13], [14]. These studies show that Myc expression is restricted to minute clusters of B-cells that initiate GCs and a small fraction of LZ B-cells. Furthermore, genetic interference with Myc expression or activity blocks GC formation and results in the collapse of mature GCs. Combined these studies demonstrate that expression of Myc is essential for the initiation and maintenance of GCs preceding B-cell proliferation in the DZ. It has been proposed that the lack of Myc expression in DZ B-cells, in conjunction with its short half-life, settles strict limits on the number of cell divisions afforded by centroblasts [27]. Therefore, by overriding the post-translational physiological control of Myc and significantly prolonging its half-life, mLANA favors an increase in the number of cell divisions during the expansion of MuHV-4 infected B-cells in the GC competitive niche. This modulation is likely to operate in infected B-cells at the initiation of a GC reaction and re-entry into the DZ. This interpretation is consistent with the observation that around 70% of MuHV-4 infected GC B-cells are rapidly dividing centroblasts versus approximately 20% infection in centrocytes. However, we show here that MuHV-4 is not able to latently expand in B-cells depleted of Myc expression. Hence, it is dependent on previous expression of Myc in B-cells. This is supported by our observation that MuHV-4 infected cells do not exhibit increased levels of Myc mRNA. Combined, our data is in good agreement with a post-translational mechanism of mLANA-mediated Myc stabilization in latently infected B-cells to increase their proliferative potential within GC reactions. This modulation is compatible with the selective expression of mLANA within GC B-cells [24] and its viral episomal maintenance properties [28].
By increasing the proliferative potential of latently infected GC B-cells through the modulation of Myc, γ-herpesviruses are effectively promoting host colonization. However, different γ-herpesviruses have evolved distinct mechanisms to accomplish Myc modulation. KSHV achieves this via LANA through a mechanism that involves targeting of Gsk-3 [17] whereas EBV appears to increase the transcription activity of Myc via interaction of EBNA3C with SCFSkp2 [18]. However, how vital is Myc modulation for γ-herpesvirus host colonization? We have previously reported that EC5SmLANA mediates poly-ubiquitination-dependent proteosomal degradation of the NF-κB family member p65/RelA [25]. In that study we demonstrate that a recombinant MuHV-4 with a disrupted SOCS-box motif in mLANA loses the ability of the virus to expand in GC B-cells and persist in the mouse. Given that this recombinant lacks ubiquitin-ligase activity we cannot ascribe its phenotype to NF-κB or Myc modulatory effects since both activities depend on an intact SOCS-box motif. We show here that mLANA interacts with Myc through a motif independent of the SOCS-box and have shown before that this also applies to interaction with p65/RelA [25]. We have mapped both interactions to the N-terminal half of mLANA but have been unable to identify any discrete binding motif to both cellular targets (unpublished observations). Since structural modeling of the N-terminal half of mLANA predicts it to be unstructured we envisage that conformational rather than linear binding motifs may be required for interaction of mLANA with p65/RelA and Myc. Current mLANA structural studies in our laboratory are addressing this question. However, the property of mLANA to modulate both Myc and p56/RelA supports that maintenance of a proliferative GC reaction through Myc stabilization requires simultaneous inhibition of NF-κB signaling.
The molecular basis for EC5SmLANA-mediated poly-ubiquitination of Myc and p65/RelA resulting in opposed outcomes is not known. Effector proteins with ubiquitin binding domains (UBDs) trigger specific cellular responses by recognizing different types of ubiquitin topologies [26]. Hence, the decoration of Myc and p65/RelA with distinct K-linkages and lengths by EC5SmLANA may determine distinct ubiquitin-mediated cellular functions. It is interesting to note that in this respect a parallel exists between EC5SmLANA and SCFβ-TrCP, as the latter also targets IκBα for proteasomal-mediated degradation [29] whereas it promotes Myc stabilization. The identification of UBD containing proteins that discriminate poly-ubiquitin chains topologies, which determine different biological outcomes is a field under intensive investigation. mLANA, therefore, provides as a good in vivo model for future studies.
Herein, we described a novel viral mechanism of stabilization of Myc through heterotypic poly-ubiquitination mediated by mLANA. The findings presented sustain the interpretation that increasing Myc stability is critical for the amplification of γ-herpesviruses in GC B-cells, thus persistence in the host. Therefore, this study provides a pathogenesis link between Myc and γ-herpesviruses associated lymphoproliferative disease.
This study was carried out in strict accordance with the recommendations of the Portuguese official Veterinary Directorate (Portaria 1005/92). The Portuguese Experiments on Animal Act strictly comply with the European Guideline 86/609/EEC and follow the FELASA. Animal experiments were approved by the Portuguese official veterinary department for welfare licensing under the protocol number AEC_2010_017_PS_Rdt_General and the IMM Animal Ethics Committee.
For reporter gene assays, cells were transiently transfected with 500 ng of reporter vector, 1 µg of Myc and mLANA/mLANA–SOCS expression plasmids. In all transfections, a Renilla luciferase plasmid (10 ng) was used to normalise luciferase activity. Firefly and Renilla luciferase activities were assayed using Dual-Luciferase (Promega). Results are shown as fold induction relative to firefly luciferase activity measured in control-transfected cells.
Cell cycle distribution profiles were analysed using Vybrant DyeCycle Violet Stain (Invitrogen) according to the manufacter's instructions. Briefly, 24 h post-tranfection with GFP, GFP-mLANA or GFP-mLANA-SOCS expressing plasmids, 1×106 A20 cells were incubated with 1 µl of Vybrant DyeCycle in complete RPMI for 30 minutes, at 37°C. The percentage of cells in the various phases of cell cycle was determined using FlowJo software (Tree Star, Inc), implementing the Dean-Jett-Fox model. Three independent experiments were performed for each experimental condition and a representative experiment is shown.
Cells were transiently transfected with plasmids encoding Myc or Myc phospho- mutants (2 µg), ubiquitin or ubiquitin with specific lysines mutated to arginines (4 µg), UbcH5 (2 µg) and/or mLANA or mLANA-SOCS (2 µg). Cells were disrupted in 10 mM Tris–HCl (pH 7.5), 150 mM NaCl, 1% Triton X-100, 1 mM NaF, 100 mM Na3VO4 and protease inhibitors (Complete; Roche). Supernatants were processed for immunoprecipitation as described [30]. Analysis of Myc or K− Myc ubiquitination was performed under denaturing conditions (20 mM Tris–HCl (pH 7.5), 5 mM EDTA, 1% SDS, 10 mM dithiothreitol and Complete). Lysates were boiled for 10 minutes at 100°C, diluted 1/10 in lysis buffer and proceeded to immunoprecipitation.
Levels of in vivo ubiquitinated Myc were determined by pull-down using Ni-NTA agarose beads. Cells were transfected with plasmids carrying His6-ubiquitin (4 µg), Myc (2 µg), and/or mLANA (2 µg). When indicated, cells were incubated for 8 hr in 10 µM MG132 (Calbiochem). Transfected cells were lysed in urea buffer (8M urea, 50 mM Tris-HCl (pH 7.5), 300 mM NaCl, 1% Triton X-100, 10 mM imidazole, 1 mM Na3VO4 and Complete), incubated for 2 hr at 4°C with Ni-NTA beads that were collected and washed with urea buffer. Proteins were eluted, denatured by boiling in Laemmli's buffer and analyzed by immunoblotting.
Total RNA from FACS-purified uninfected or infected GC B-cells, or transfected HEK 293T cells or A20 B cells was extracted with Trizol (Invitrogen). RNA (500 ng) was used for cDNA synthesis (DyNAmo, Finnzymes). qPCR was performed using DyNAmo Flash SYBR Green (Finnzymes). Primer sequences are available in Table S1. All reactions were run in duplicates. Amplification efficiencies and threshold cycle values were defined by the fractional cycle number at which fluorescence crosses the fixed threshold. Relative mRNA values, normalized to GAPDH, were calculated by the Pfaffl method [31].
Cγ1-Cre mice were provided by Dr. Kurosaki (Japan), with the agreement of Dr. Rajewsky and Dr. Casola. c-myc floxed mice were a gift from Dr. Moreno de Alborán (Spain). Cγ1-creKI/WT; c-mycfl/fl mice were generated by breeding heterozygous Cγ1-cre mice [22], with homozygous c-mycfl/fl mice [21]. Cγ1-creWT/WT; c-mycfl/fl mice littermates were used as controls. C57BL/6 mice were obtained from Charles River Laboratories International Inc. Mice were bred and housed at IMM. All experimental protocols were performed in animals with 6–8 weeks of age.
Immunizations were performed via intraperitoneal injection with 100 µg NP-CGG (Biosearch Technologies, Inc.) adsorbed to 3 mg of aluminium hydroxide (SERVA Electrophoresis GmbH), or 100 µg ovalbumin (OVA) grade V in CFA (Sigma). OVA immunized mice were challenged 7 days post-primary immunization with the same antigen/adjuvant combination. Inoculation of MuHV-4 was performed intranasally with 104 p.f.u. in 20 µl of PBS under halothane. Frequencies of MuHV-4 genome-positive cells in GC B-cells were determined by limiting dilution combined with real-time PCR as previously described [32]. GC B cells were FACS purified from pools of five spleens using a BD FACSAria Flow Cytometer (BD Biosciences) and serially two-fold diluted. Eight replicates of each dilution were analysed by real time PCR (Rotor Gene 6000, Corbett Life Science). The primer/probe sets were specific for the MuHV-4 M9 gene (5′ primer: GCCACGGTGGCCCTCTA; 3′ primer: CAGGCCTCCCTCCCTTTG; probe: 6-FAM-CTTCTGTTGATCTTCC–MGB). Samples were subjected to a melting step of 95°C for 10 min followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. Real-time PCR data was analyzed on the Rotor Gene 6000 software. The purity of sorted cells was always greater than 97%, as analyzed by flow cytometry.
Information provided in protocols S1.
|
10.1371/journal.pgen.1007903 | Genomic insights into neonicotinoid sensitivity in the solitary bee Osmia bicornis | The impact of pesticides on the health of bee pollinators is determined in part by the capacity of bee detoxification systems to convert these compounds to less toxic forms. For example, recent work has shown that cytochrome P450s of the CYP9Q subfamily are critically important in defining the sensitivity of honey bees and bumblebees to pesticides, including neonicotinoid insecticides. However, it is currently unclear if solitary bees have functional equivalents of these enzymes with potentially serious implications in relation to their capacity to metabolise certain insecticides. To address this question, we sequenced the genome of the red mason bee, Osmia bicornis, the most abundant and economically important solitary bee species in Central Europe. We show that O. bicornis lacks the CYP9Q subfamily of P450s but, despite this, exhibits low acute toxicity to the N-cyanoamidine neonicotinoid thiacloprid. Functional studies revealed that variation in the sensitivity of O. bicornis to N-cyanoamidine and N-nitroguanidine neonicotinoids does not reside in differences in their affinity for the nicotinic acetylcholine receptor or speed of cuticular penetration. Rather, a P450 within the CYP9BU subfamily, with recent shared ancestry to the Apidae CYP9Q subfamily, metabolises thiacloprid in vitro and confers tolerance in vivo. Our data reveal conserved detoxification pathways in model solitary and eusocial bees despite key differences in the evolution of specific pesticide-metabolising enzymes in the two species groups. The discovery that P450 enzymes of solitary bees can act as metabolic defence systems against certain pesticides can be leveraged to avoid negative pesticide impacts on these important pollinators.
| Bees have evolved sophisticated metabolic systems to detoxify the natural toxins encountered in their environment. Recent work has shown that specific enzymes (cytochrome P450s) in these biotransformation pathways can be recruited to protect honey bees and bumblebees against certain synthetic insecticides, including some neonicotinoids. However, it is unclear if solitary bees that carry out important pollination services have equivalent enzymes that play a key role in defining their sensitivity to insecticides. In this study we show that the genome of the solitary bee, Osmia bicornis, lacks the subfamily of cytochrome P450 enzymes that break down certain neonicotinoids in eusocial bees. Despite this, O. bicornis exhibits marked tolerance to the neonicotinoid thiacloprid as a result of efficient metabolism by a P450 enzyme from an alternative subfamily. The discovery that O. bicornis has key detoxification enzymes that determine its sensitivity to neonicotinoids can be leveraged to safeguard the health of this important pollinator.
| Bee pollinators encounter a wide range of natural and synthetic xenobiotics while foraging or in the hive, including phytochemicals, mycotoxins produced by fungi, and pesticides [1]. Understanding the toxicological outcomes of bee exposure to these chemicals, in isolation or combination, is essential to safeguard bee health and the ecosystem services they provide. Like other insects, bees have sophisticated metabolic systems that mediate the conversion of harmful xenobiotics to less toxic forms, and these detoxification pathways can be critically important in defining their sensitivity to xenobiotics including pesticides [2]. In an important recent example of this cytochrome P450 enzymes belonging to the CYP9Q subfamily were shown to play a key role in determining the sensitivity of honey bees and bumblebees to neonicotinoid insecticides [3]. Prior work on honey bees showed that the same P450s also provide protection against the toxic effects of certain insecticides from the pyrethroid and organophosphate classes that are used for the control of parasitic Varroa mites [4]. Taken together these studies suggest CYP9Q P450s may be important generalist detoxification enzymes. To date our understanding of bee biochemical defence systems stems from work on eusocial species, namely honey bees and bumblebees, with much less attention given to solitary species. However, the majority of bee species are solitary, and there is increasing awareness of the importance of solitary bees as pollinators of wild plants and certain crops [5–8]. It is currently unknown to what extent the discoveries on the metabolic systems of honey bees and bumblebees extend to solitary bees, and thus if the use of eusocial species as a proxy for solitary species in ecotoxicological studies is reliable.
The red mason bee, Osmia bicornis (syn. O. rufa) (Hymenoptera: Megachilidae) is the most abundant and economically important solitary bee species in Central Europe [9]. This species pollinates a range of wild plants and is also used for commercial pollination, particularly of fruit crops (almond, peach, apricot, plum, cherry, apple and pear). Understanding O. bicornis-pesticide interactions is particularly important as it has been recommended as a solitary bee model for the registration of pesticides in Europe [10]. However, to date, investigations on this topic have been hampered by a lack of genomic and transcriptomic resources for this species.
In this study we addressed this knowledge and resource gap by generating a high quality genome assembly of O. bicornis. We then exploited this genomic resource to compare the complement of P450 genes in O. bicornis with that of other bee species, and identify P450 enzymes that are important determinants of O. bicornis sensitivity to neonicotinoid insecticides.
To generate a high-quality genome assembly of O. bicornis we sequenced genomic DNA extracted from a single haploid male bee using a combination of Illumina paired-end and mate-pair libraries. Additional RNA sequencing (RNAseq) of male and female bees was also performed in order to improve the quality of subsequent gene prediction. DNAseq data was assembled to generate an O. bicornis genome of 212.9 Mb consistent with genome size estimates derived from k-mer analysis of the raw reads (S1 Table). The final assembly comprised 10,223 scaffolds > 1 kb with a scaffold and contig N50 of 604 kb and 303 kb respectively (S2 Table). Structural genome annotation using a workflow incorporating RNAseq data predicted a total of 14,858 protein-coding genes encoding 18,479 total proteins (S3 Table). The completeness of the gene space in the assembled genome was assessed using the Benchmarking Universal Single-Copy Orthologues (BUSCO) pipeline [11] with greater than 99% of Arthropoda and Insecta test genes identified as complete in the assembly (S4 Table). Approximately 78% of the predicted genes could be assigned functional annotation based on BLAST searches against the non-redundant protein database of NCBI (S1 Fig).
The gene repertoire of O. bicornis was compared with other colony forming (Apis mellifera, Apis florea, Bombus terrestris and Bombus impatiens) and solitary bee species (Megachile rotundata) by orthology inference (Fig 1A). The combined gene count of these species was 101,561 of which ~90% were assigned to 11,184 gene families. Of these 8,134 gene families were present in O. bicornis and all other species, and a total of 163 gene families were specific to O. bicornis compared to 21–97 in the other bee species (Fig 1A). Genes encoding cytochrome P450s were identified from orthogroups, and individual bee genomes (see methods), and the complete complement of P450s in each bee genome (the CYPome) was curated and named by the P450 nomenclature committee (S5 Table). The genome of O. bicornis contains 52 functional P450s (Fig 2B and S2 Fig), a gene count consistent with the other bee species and reduced in comparison to other insects, even including other hymenoptera [2]. As for other insect species bee P450 genes group into four main clades (CYP2, CYP3, CYP4 and mitochondrial clans) of which by far the largest (comprising 33 P450s in O. bicornis) is the CYP3 clan of CYP6, CYP9 and CYP336 (Fig 1A and 1B, S2 Fig). Phylogenetic comparison of the CYP9 family within this clade in O. bicornis and 11 other bee species [12] revealed that O. bicornis lacks the CYP9Q subfamily found in eusocial bee species that has been shown to define the sensitivity of honey bees and bumblebees to neonicotinoids (Fig 1C, S3 Fig) [3]. The most closely related subfamily in O. bicornis was CYP9BU (represented by CYP9BU1 and CYP9BU2), a newly described subfamily, that appears to share a relatively recent common ancestor with the CYP9Q subfamily (Fig 1C, S3 Fig).
In the absence of the CYP9Q subfamily of P450s it might be expected that O. bicornis would be more sensitive to neonicotinoids (especially N-cyanoamidine compounds) than honey bees and bumblebees. To test this we performed acute contact insecticide bioassays using imidacloprid and thiacloprid as representatives of N-nitroguanidine and N-cyanoamidine neonicotinoids respectively. Significant differences were found in the tolerance of O. bicornis to the two compounds with adult female bees >2,000-fold more sensitive to imidacloprid (LD50 of 0.046 μg/bee) than thiacloprid (LD50 of >100 μg/bee) (Fig 2A). These values are similar to those reported for honey bees and bumblebees [3,13,14] with imidacloprid classified as ‘highly toxic’ to O. bicornis according to the categories of the U.S. Environmental Protection Agency, but thiacloprid classified as ‘practically non-toxic’ upon contact exposure (Fig 2A). Thus these results clearly show that, despite the lack of CYP9Q P450s, O. bicornis has high levels of tolerance to the N-cyanoamidine neonicotinoid thiacloprid.
The molecular basis of the profound variation in the sensitivity of O. bicornis to imidacloprid and thiacloprid could reside in differences in: a) their affinity for the target-site, the nicotinic acetylcholine receptor (nAChR), b) their speed of penetration through the cuticle, or c) the efficiency of their metabolism. We first examined the affinity of the two compounds for the target-site using radioligand binding assays performed on O. bicornis head membrane preparations, and examined the displacement of tritiated imidacloprid by both unlabelled imidacloprid and thiacloprid. Both compounds bound with nM affinity—IC50 of 8.3 nM [95% Cl 4.6, 15.1] for imidacloprid and 2.4 nM [95% Cl, 1.4, 4.1] for thiacloprid (Fig 2B). These values suggest that thiacloprid binds with higher affinity than imidacloprid, however, no significant difference was observed between the slopes of the regression lines of the two compounds (p = 0.3). This finding clearly demonstrates that the tolerance of O. bicornis to thiacloprid relative to imidacloprid is not a consequence of a reduced affinity of the former for the nAChR.
To explore the rate of penetration of these two compounds through the cuticle of O. bicornis the uptake of [14C]imidacloprid and [14C]thiacloprid after application to the dorsal thorax was compared. No significant differences were observed in the amount of radiolabelled thiacloprid and imidacloprid recovered from the cuticle or acetone combusted whole bees at any time point post-application (the final uptake through the cuticle after 24h was 27% of [14C]imidacloprid and 28% of [14C]thiacloprid, Fig 2C). Thus, the differential sensitivity of O. bicornis to imidacloprid and thiacloprid is not a result of variation in their speed of penetration through the cuticle.
Insecticide synergists that inhibit detoxification enzymes have been used to explore the role of metabolism in the tolerance of honey bees and bumblebees to certain neonicotinoids. Specifically, the use of the P450 inhibitor piperonyl butoxide (PBO) provided strong initial evidence that P450s underpin the tolerance of both bee species to N-cyanoamidine neonicotinoids [3,15]. We therefore examined the effect of PBO pre-treatment on the sensitivity of O. bicornis to thiacloprid and imidacloprid in insecticide bioassays. No significant difference was observed in the sensitivity of O. bicornis to imidacloprid with or without PBO, however, bees pre-treated with PBO became >7-fold more sensitive to thiacloprid (Fig 2D), suggesting that P450s play an important role in defining the sensitivity of O. bicornis to neonicotinoids.
As detailed above, based on phylogeny, CYP9BU1 and CYP9BU2 are clearly the most closely related P450s in O. bicornis to the Apidae CYP9Q subfamily which metabolise thiacloprid in honey bees and bumblebees (Fig 1C, S3 Fig). We therefore examined the capacity of these P450s to metabolise thiacloprid and imidacloprid in vitro by individually coexpressing them with house fly cytochrome P450 reductase (CPR) in an insect cell line. Incubation of microsomal preparations containing each P450 and CPR with either thiacloprid or imidacloprid, and analysis of the metabolites produced by liquid chromatography tandem mass spectrometry (LC-MS/MS), revealed that both CYP9BU1 and CYP9BU2 metabolise these compounds to their hydroxylated forms (5-hydroxy thiacloprid and 5-hydroxy imidacloprid respectively) (Fig 3A). Both P450s metabolised thiacloprid with significantly greater efficiency than imidacloprid (Fig 3A) consistent with the relative sensitivity of O. bicornis to these compounds. To provide additional evidence that these P450s confer tolerance to N-cyanoamidine neonicotinoids in vivo, we created transgenic lines of Drosophila melanogaster expressing CYP9BU1, or CYP9BU2 and examined their sensitivity to imidacloprid and thiacloprid. Flies expressing the CYP9BU1 transgene were ~4 times less sensitive to thiacloprid than control flies of the same genetic background without the transgene in insecticide bioassays (Fig 3B, S6 Table). In contrast flies expressing CYP9BU2 showed no significant resistance to thiacloprid. In bioassays using imidacloprid no significant differences in sensitivity were observed between flies with either of the two transgenes and control flies. These results demonstrate that the transcription of CYP9BU1 confers intrinsic tolerance to thiacloprid in vivo.
Characterising when and where the neonicotinoid-metabolising P450s identified in this study are expressed is an important step in understanding their capacity to protect O. bicornis in vivo. To investigate this, we 1) explored changes in their expression in response to exposure to sublethal doses of imidacloprid and thiacloprid, and 2) examined their expression in tissues that are involved in xenobiotic detoxification, or are sites of insecticide action.
To investigate if the expression of any genes encoding P450s could be induced by neonicotinoid exposure RNAseq was performed on adult female O. bicornis 24 h after exposure to the LD10 of thiacloprid, imidacloprid or the solvent used to dissolve insecticides alone (as a control). Differentially expressed genes (corrected p value of <0.05) between control and treatments were identified and are shown in full in S7 Table and S8 Table. In general, changes in gene expression were modest with just 27 genes significantly upregulated after imidacloprid exposure and 16 genes upregulated after thiacloprid exposure. The function of these differentially expressed genes was either unknown or is unrelated to xenobiotic detoxification, and no P450 showed a significant increase in expression upon exposure to either neonicotinoid (Fig 4A, S7 Table and S8 Table). These findings suggest that constitutive rather than induced expression of the P450s identified in this study is more important in their role in pesticide detoxification.
The expression of neonicotinoid-metabolising P450s in the brain, midgut and Malpighian tubules of O. bicornis was assessed by quantitative PCR (Fig 4B). CYP9BU1 was found to be highly expressed in the Malpighian tubules, the functional equivalents of vertebrate kidneys, consistent with a primary role in xenobiotic detoxification. In contrast CYP9BU2 was expressed at equivalent levels in the Malpighian tubules, the midgut and the brain (Fig 4B).
The genomes of all bee species sequenced to date have a considerably reduced complement of cytochrome P450s compared to those of other insect species [12,16]. This suggests that, like humans [17], bees may depend on a relatively small subset of generalist P450s for the detoxification of xenobiotics [2]. An emerging body of work on eusocial bees has provided strong support for this hypothesis, with P450s of the CYP9Q subfamily identified as metabolisers of insecticides from three different classes [3,4], and key determinants of honey bee and bumble bee sensitivity to neonicotinoids [3]. In this study we examined the extent to which these findings apply to solitary bees, using the red mason bee, O. bicornis as a model. Consistent with data from honey bees and bumblebees sequencing of the O. bicornis genome revealed a reduced P450 inventory in comparison to most other insects, however, in contrast to these species no members of the CYP9Q P450 subfamily were present in the curated CYPome. We interrogated the recently published genomes of several other solitary and eusocial bee species [12] and confirmed that the CYP9Q subfamily is ubiquitous in the CYPome of sequenced social bees (represented by 2–3 genes in most species) but missing in all solitary bee genomes apart from Habropoda laboriosa, a species in the family Apidae, which has a single CYP9Q gene (CYP9Q9) (S3 Fig). Solitary bees are the ancestral state from which social bees evolved [12] suggesting the CYP9Q subfamily expanded after social bees diverged from solitary bees. A rapid birth–death model of evolution is characteristic of xenobiotic-metabolizing P450s, in contrast to P450s with endogenous functions [18], and the expansion of the CYP9Q subfamily in social bees may have occurred to allow xenobiotics specifically associated with this life history to be detoxified. In relation to this, recent analysis of the CYPomes of ten bee species has suggested that the expansion of the CYP6AS subfamily in perennial eusocial bees resulted from increased exposure to phytochemcials, as a result of the concentration of nectar into honey, pollen into beebread and plant resins into propolis [19].
The finding that most solitary bees lack the CYP9Q subfamily raises important questions about their capacity to metabolise and, by extension tolerate, certain pesticides. Thus, a key finding from our study is that despite the absence of the CYP9Q subfamily O. bicornis exhibits similar levels of sensitivity to the neonicotinoids imidacloprid and thiacloprid as honey bees and bumblebees, and, like these species, marked tolerance to the latter compound. We show that the observed variation in the sensitivity of O. bicornis to thiacloprid and imidacloprid does not result from differences in their affinity for the nAChR, or speed of cuticular penetration, but rather variation in their speed/efficiency of metabolism by cytochrome P450s. Functional characterisation revealed that, in the absence of the CYP9Q subfamily, O. bicornis employs P450s from the CYP9BU subfamily to detoxify the N-cyanoamidine neonicotinoid thiacloprid. While the CYP9BU subfamily is currently unique to O. bicornis phylogeny shows it is more closely related to the CYP9Q subfamily, with which it appears to share a recent common ancestor, than any other bee P450 subfamily. Given that we show that CYP9BU1 appears to be particularly effective in metabolising N-cyanoamidine neonicotinoids it will be important to explore which P450s other solitary bee species, such as the economically important leafcutter bee, Megachile rotundata, use to detoxify pesticides in the absence of this subfamily (S3 Fig).
Work on other insect species has shown that insecticide-metabolising P450s may be constitutively expressed or induced upon exposure to xenobiotic substrates [20]. We found no evidence of induction of any O. bicornis P450s in response to exposure to sublethal concentrations of thiacloprid or imidacloprid suggesting that constitutive expression of these enzymes provides protection against neonicotinoids. Their detoxification capacity may be further enhanced by expression in tissues with specialised roles in metabolism/excretion, and it is notable that CYP9BU1 is expressed at particularly high levels in the Malpighian tubules. The overexpression of CYP9BU1 in these osmoregulatory and detoxifying organs is highly consistent with a primary role in xenobiotic metabolism and parallels the high expression of CYP9Q3 in this tissue—the primary metaboliser of neonicotinoids in honey bees [3].
In summary, we show that the solitary bee O. bicornis is equipped with key biochemical defence enzymes that provide protection against certain insecticides. Together with previous work this demonstrates that while the underlying P450s involved may be different in O. bicornis and eusocial bees, the overarching detoxification pathways used by these species to metabolise neonicotinoids is conserved. Identification of the P450s responsible for the observed tolerance of O. bicornis to N-cyanoamidine neonicotinoids can be used to support ecotoxicological risk assessment and safeguard the health of this important pollinator. For example, the recombinant enzymes developed in our study can be used to screen existing pesticides to identify and avoid synergistic pesticide-pesticide interactions that inhibit these enzymes [21], and to examine the metabolic liability of future lead compounds as part of efforts to develop pest-selective chemistry. The genomic resources, tools and knowledge generated in this study are particularly timely as O. bicornis has recently been proposed as a representative solitary bee species for inclusion in future risk assessment of plant protection products in Europe [10].
Genomic DNA was extracted from a single male bee using the E.Z.N.A Insect DNA kit (Omega Bio-Tek) following the manufacturer’s protocol. DNA quantity and quality was assessed by spectrophotometry using a NanoDrop (Thermo Scientific), Qubit assay (ThermoFisher) and gel electrophoresis. Sufficient DNA from a single male bee was obtained for the preparation of a single PCR-free paired-end library and 5 long mate pair Nextera libraries that were sequenced on an Illumina HiSeq 2500 using a 250bp read metric at Earlham Institute, Norwich, UK. To improve the quality of subsequent gene prediction RNA sequencing was also performed. For this RNA was extracted from female and male O. bicornis 24 h after emergence using the Isolate RNA Mini Kit (Bioline) according to the manufacturer’s instructions. The quantity and quality of RNA was checked as described above. RNA was used as a template for the generation of barcoded libraries (TrueSeq RNA library preparation, Illumina) and RNA samples sequenced to high coverage on an Illumina HiSeq2500 flowcell (100 bp paired-end reads). All sequence data have been deposited under NCBI BioProject PRJNA285788.
Reads were assembled using DISCOVAR_de-novo–v 52488 [22] using default parameters. All sequences >500 bp from the initial draft assembly were used in scaffolding with 5 Illumina Nextera mate-pair libraries using Redundans–v 0.12a [23] with default parameters. To further increase the contiguity of the draft genome we applied a third scaffolding step, making use of the RNAseq data. Transcriptome contig sequences of O. bicornis and protein sequences of a closely related species Megachile rotundata, were mapped sequentially using L_RNA_scaffolder [24] and PEP_scaffolder [25]. The first round of gene prediction was performed using BRAKER–v 2.1.0 [26] utilising RNAseq data to improve gene calling. To generate training sets for ab-initio gene modellers AUGUSTUS [27] and SNAP [28], we searched core eukaryotic and insecta orthologous genes in the O. bicornis assembly using CEGMA–v 2.5.0 [29] and BUSCO–v 3.0.0 [11] respectively. BUSCO gene models were used to train AUGUSTUS–v 2.5.5, and SNAP (https://github.com/KorfLab/SNAP) was trained using the CEGMA gene models. Another set of hidden markov gene models was generated by GeneMark-ES–v 4.32.0 [30]. In addition, a custom O. bicornis specific repeat library was built from the assembly using RepeatModeler–v 1.0.4 [31]. To make use of expression data and exploit spliced alignments in genome annotation, expressed transcripts assembled from RNAseq data were further mapped to the O. bicornis genome using PASA–v 2.3.3 [32]. We initially ran MAKER2 [33] with just the O. bicornis assembly and EST data, collected from NCBI, followed by three consecutive iterations with the draft genome sequence, transcriptome dataset, models from BRAKER, SNAP and GeneMark-ES, the O. bicornis specific repeat library and the Swiss-Prot database (accessed at May 23, 2016). Between iterations, the BRAKER and SNAP models were retrained. As BRAKER models are originally predicted from AUGUSTUS, we used AUGUSTUS to train BRAKER models in each successive MAKER2 iteration according to the best-practice MAKER2 workflow. Finally, BRAKER and MAKER2 prediction sets, including PASA alignments, alignment of M. rotundata proteins using exonerate–v 2.4.0 were combined to generate a non-redundant gene set using EvidenceModeler–v 1.1.1 [34]. The final annotation set for O. bicornis was compared to other bee genomes to characterize orthology. The proteomes of Apis mellifera, Apis florea, Bombus terrestris, Bombus impatiens, Megachile rotundata, were downloaded from NCBI, and OrthoFinder–v 1.1.8 [35] was used to define orthologous groups of genes between these peptide sets. P450 sequences were recovered from the bee species using three approaches: 1) Text searches of existing annotation, 2) mining P450 gene sequences from ortholog data generated above, and 3) iterative BLAST searches using A. mellifera curated P450 genes as queries. All obtained sequences were then manually inspected and curated to generate a final list of P450 genes for each species which were named by the P450 nomenclature committee. Accession numbers are provided in S5 Table.
For phylogenetic analysis, manually curated protein sequences of cytochrome P450 genes were aligned using MUSCLE v3.8.31 [36]. FMO2-like (Protein ID: XP_016772196.1) and CYP315A1 from A. mellifera were used as an outgroup for phylogenies displayed in Fig 1 and S3 Fig respectively. An initial likelihood phylogenetic tree was created using the R package “phangorn: Phylogenetic Reconstruction and Analysis” v.2.4.0 [37]. Parameters including proportion of variable size (I) and gamma rate (G) were optimized using amino acid substitution matrices JTT for Fig 1 and S3 Fig and LG for S2 Fig based on minimum Bayesian information criterion (S9 Table) [37]. Finally rooted (Fig 1 and S3 Fig) or unrooted (S2 Fig) consensus trees of 1,000x bootstrapping using nearest-neighbor interchange were created and visualized using the R package “ggtree” v1.12.0 [37,38].
O. bicornis cocoons were purchased from Dr Schubert Plant Breeding (Landsberg, Germany) and stored at 4°C in constant darkness. To trigger emergence cocoons were transferred to an incubator (25°C, 55% RH, L16:D8) with emerged bees fed ab libitum with Biogluc (62% sugar concentration consisting of 37.5% fructose, 34.5% glucose, 25% sucrose, 2% maltose, and 1% oligosaccharides) (Biobest), soaked into a piece of cotton wool inside a plastic dish. Males (which are usually first to emerge) were removed from cages and discarded to reduce any unnecessary stress to the females used in insecticide bioassays.
Acute contact toxicity bioassays on unmated 2 day old female O. bicornis were conducted following the OECD Honey Bee Test guidelines, with modification where necessary [39]. Bees were anaesthetised with CO2 for 5–10 seconds to allow application of insecticide. 1 μL of technical grade imidacloprid was applied to the dorsal thorax of each bee at concentrations of 0.0001, 0.001, 0.01, 0.1, 1, and 10 μg/μL. No mortality was observed using the same concentrations of thiacloprid so a limit test of 100 μg/bee was performed. Control bees were treated with 1 μL 100% acetone. Three replicates of 10 bees were tested for each concentration. Tested individuals were placed back into cages in the incubator (25°C, 55% RH, L16:D8), with five bees per cage. In piperonyl butoxide (PBO) synergist bioassays, bees were first treated with the maximum sublethal dose (in this case 100 μg/μL) of PBO followed by insecticide one hour later. Synergist bioassays included an additional control group treated only with PBO. Mortality was assessed 48 and 72 hours after application. Probit analysis was used to calculate the LD50 values, slope, and synergism ratio (where relevant) for each insecticide (Genstat v.18 (VSNI 2015)).
[3H]imidacloprid (specific activity 1.406 GBq μmol−1) displacement studies were conducted using membrane preparations isolated from frozen (−80°C) O. bicornis heads, following previously published protocols [13]. Briefly, bee heads weighing 10 g were homogenized in 200 ml ice-cold 0.1 M potassium phosphate buffer, pH 7.4 containing 95 mM sucrose using a motor-driven Ultra Turrax blender. The homogenate was then centrifuged for 10 min at 1200 g and the resulting supernatant filtered through five layers of cheesecloth with protein concentration determined using Bradford reagent (Sigma) and bovine serum albumin (BSA) as a reference. Assays were performed in a 96-well microtitre plate with bonded GF/C filter membrane (Packard UniFilter-96, GF/C) and consisted of 200 μl of homogenate (0.48 mg protein), 25 μl of [3H]imidacloprid (576 pM) and 25 μl of competing ligand. Ligand concentrations used ranged from 0.001 to 10 000 nM and were tested in triplicate per competition assay. The assay was started by the addition of homogenate and incubated for 60 min at room temperature. Bound [3H]imidacloprid was quantified by filtration into a second 96-well filter plate (conditioned with ice-cold 100 mM potassium phosphate buffer, pH 7.4 (including BSA 5 g litre−1)) using a commercial cell harvester (Brandel). After three washing steps (1 ml each) with buffer the 96-well filter plates were dried overnight. Each well was then loaded with 25 μl of scintillation cocktail (Microszint-O-Filtercount, Packard) and the plate counted in a Topcount scintillation counter (Packard). Non-specific binding was determined using a final concentration of 10 μM unlabelled imidacloprid. All binding experiments were repeated twice using three replicates per tested ligand concentration. Data were analysed using a 4 parameter logistic non-linear fitting routine (GraphPad Prism version 7 (GraphPad Software, CA, USA)) in order to calculate IC50-values (concentration of unlabelled ligand displacing 50% of [3H]imidacloprid from its binding site). Non-linear regression model fitting and statistical comparison of the slopes obtained was performed in the drc package in R [40].
Bees were anaesthetised with CO2 for 5–10 seconds to allow application of insecticide. 5,000 ppm of [14C]imidacloprid or [14C]thiacloprid was applied to the dorsal thorax of each bee using a Hamilton repeating dispenser. Three replicates of five bees were placed into cages and fed a 50% sucrose solution from vertically hanging 2 ml syringes. Control bees were treated with acetone. Radiolabelled insecticide was rinsed off of each group of bees at set time intervals (0, 2, 4 and 24 hours after application) with acetonitrile water (90:10) three times. The acetone-washed bees were then individually combusted at 900°C in an Ox 120c oxidizer (Harvey Instruments Co., USA) followed by liquid scintillation counting of the released 14CO2 in an alkaline scintillation cocktail (Ultima Gold, PerkinElmer) using a liquid scintillation analyser (Perkin Elmer Tri-Carb 2910 TR). The levels of excreted [14C]imidacloprid or [14C]thiacloprid, and/or metabolites, were measured by wiping cages with filter papers dipped in acetone and 0.5 mL aliquots of cuticular rinse or filter papers were added to 3 mL of scintillation fluid cocktail and the radioactivity was quantified by liquid scintillation analysis as above. An unpaired t-test was used to compare the penetration of the two compounds at each time point.
Sequences of O. bicornis candidate genes were verified by PCR as follows: Adult female O. bicornis were flash frozen in liquid nitrogen and stored at -80°C prior to extractions. RNA was extracted from a pool of 3–5 bees using the RNeasy Plus kit (QIAGEN) following the manufacturer’s protocol. The quantity and quality of RNA were assessed as described above. First-strand cDNA was synthesised at a concentration of 200 ng/μL by reverse transcription using SuperScript III Reverse Transcriptase (Invitrogen) according to the manufacturer’s protocol. 25μL reactions contained 1.5U DreamTaq DNA Polymerase (Thermofisher), 10mM of forward and reverse primers (S10 Table) and 200 ng of cDNA. PCR reaction temperature cycling conditions were 95°C for 2 minutes, followed by 35 cycles of 95°C for 20 seconds (denaturation), 60°C for 20 seconds (annealing), and 72°C for 7.5 minutes (elongation). PCR products were visualised on a 1% agarose gel and purified using QIAquick PCR purification kit (QIAGEN). Samples were sequenced by Eurofins (Eurofins Scientific group, Belgium) and analysed using Geneious v8.1.3 software (Biomatters Ltd, New Zealand).
O. bicornis P450 genes and house fly NADPH-dependent cytochrome P450 reductase (CPR) (GenBank accession no. Q07994) genes were codon optimised for expression in lepidopteran cell lines, synthesized (Geneart, CA, USA) and inserted into the pDEST8 expression vector (Invitrogen). The PFastbac1 vector with no inserted DNA was used to produce a control virus. The recombinant baculovirus DNA was constructed and transfected into Trichoplusia ni (High five cells, Thermo Fisher) using the Bac-to-Bac baculovirus expression system (Invitrogen) according to the manufacturer’s instructions. The titre of the recombinant virus was determined following protocols of the supplier. High Five cells grown to a density of 2 x 106 cells ml-1 were co-infected with recombinant baculoviruses containing each bee P450 and CPR with a range of MOI (multiplicity of infection) ratios to identify the optimal conditions. Control cells were co-infected with the baculovirus containing vector with no insert (ctrl-virus) and the recombinant baculovirus expressing CPR using the same MOI ratios. Ferric citrate and δ-aminolevulinic acid hydrochloride were added to a final concentration of 0.1 mM at the time of infection and 24 h after infection to compensate the low levels of endogenous heme in the insect cells. After 48 h, cells were harvested, washed with PBS, and microsomes of the membrane fraction prepared according to standard procedures and stored at −80°C [41]. Briefly, pellets were homogenised for 30 s in 0.1 M Na/K-phosphate buffer, pH 7.4 containing 1 mM EDTA and DTT and 200 mM sucrose using a Fastprep (MP Biomedicals), filtered through miracloth and centrifuged for 10 min at 680g at 4°C. The supernatant was then centrifuged for 1 h at 100,000g at 4°C, with the pellet subsequently resuspended in 0.1M Na/K-phosphate buffer pH 7.6 containing 1 mM EDTA and DTT and 10% glycerol using a Dounce tissue grinder. P450 expression and functionality was estimated by measuring CO-difference spectra in reduced samples using a Specord 200 Plus Spectrophotometer (Analytik Jena) and scanning from 500 nm to 400 nm [41]. The protein content of samples was determined using Bradford reagent (Sigma) and bovine serum albumin (BSA) as a reference.
Metabolism of thiacloprid and imidacloprid was assessed by incubating recombinant P450/CPR (5 pmol/well) or control virus/CPR (5 pmol/well) with each insecticide (25 μM) in the presence of an NADPH regeneration system at 30±1°C, shaking, for 1 hour. Three replicates were performed for each data point and the total assay volume was 200 μL. Samples incubated without NADPH served as a control. The reactions were terminated by the addition of ice-cold acetonitrile (to 80% final concentration), centrifuged for 10 min at 3000 g and the supernatant analyzed by tandem mass spectrometry as described previously [42]. LC-MS/MS analysis was performed on a Waters Acquity UPLC coupled to a Sciex API 4000 mass spectrometer and an Agilent Infinity II UHPLC coupled to a Sciex QTRAP 6500 mass spectrometer utilizing electrospray ionization. For the chromatography on a Waters Acquity HSS T3 column (2.1x50 mm, 1.8 μm), acetonitrile/water/0.1% formic acid was used as the eluent in gradient mode. For detection and quantification in positive ion mode, the MRM transitions 253 > 186, 269 > 202 (thiacloprid, OH-thiacloprid), and 256 > 175, 272 > 191 (imidacloprid, OH-imidacloprid) were monitored. The peak integrals were calibrated externally against a standard calibration curve. Recovery rates of parent compounds using microsomal fractions without NADPH were normally close to 100%. Substrate turnover was determined using GraphPad Prism version 7 (GraphPad Software, CA, USA).
CYP9BU1 and CYP9BU2 were codon optimised for D. melanogaster expression and cloned into the pUASTattB plasmid (GenBank: EF362409.1). These constructs were used to create transgenic fly lines, which were then tested in insecticide bioassays against imidacloprid and thiacloprid, as described previously [3].
To examine if P450 expression in O. bicornis is induced by exposure to sublethal concentrations of neonicotinoids, imidacloprid and thiacloprid were dissolved in acetone to the highest concentration possible, before being diluted to the LD10 of imidacloprid (0.0001 μg/bee) and thiacloprid (0.01 μg/bee) with 50% sucrose (w/v) in order to limit the amount of acetone consumed by bees. Prior to commencing oral bioassays bees underwent a 24 hour ‘training’ period in the Nicot cages to enable them to learn to feed from the syringes. This was followed by a 16h starvation period to encourage subsequent feeding. 15μL of the insecticide/sucrose solution was supplied orally to the bees in disposable plastic syringes. Control bees were fed 15μL of a sucrose solution containing the same volume of acetone used to make up the insecticide/sucrose solutions. When all of the solution had been consumed the bees were fed ab libitum with a 50% sucrose solution for 24 h. After this period for each condition four replicates comprising 5 bees per replicate were snap frozen in liquid nitrogen and RNA extracted from each replicate as above. RNA was used as a template for the generation of barcoded libraries (TrueSeq RNA library preparation, Illumina) which were sequenced across two lanes of an Illumina HiSeq2500 flowcell (100 bp paired end reads). Sequencing was carried out by Earlham Institute, Norwich, UK. To identify genes differentially expressed between control and the treatment the Tuxedo workflow was used to map with TopHat against the annotated reference genome, to estimate gene expression with Cufflinks and test for differential expression with Cuffdiff [43].
To examine the expression of candidate P450 genes in tissues with a known role in detoxification or the site of insecticide action the brain, midgut and Malpighian tubules were extracted from flash frozen adult female O. bicornis. RNAlater-ICE (Life technologies) was used to preserve RNA during dissections. RNA was extracted as above and first-strand cDNA synthesised using SuperScript III Reverse Transcriptase (Invitrogen) according to the manufacturer’s protocol. Quantitative RT-PCR was carried out using a Rotor Gene 6000 machine with the thermocycling conditions: 3 minutes at 95°C followed by 40 cycles of 95°C for 20 seconds (denaturation), 60°C for 20 seconds (annealing), and 72°C for 7.5 minutes (elongation). A final melt-curve step was included to rule out any non-specific amplification. 15μL reactions consisted of 6μL cDNA (10ng), 7μL of SYBR Green Master Mix (Thermofisher Scientific) and 0.25μM of the forward and reverse primers. All primers were designed using the Prime3 primer design tool (http://biotools.umassmed.edu/bioapps/primer3_www.cgi) and are listed in S10 Table. All primers were designed to amplify a ~200bp region of each target gene with low percentage identity to other target genes. The efficiency of each primer set was examined using a standard curve (concentrations 100–0.01ng of cDNA). Elongation Factor α1 and elongation factor γ1 were used as housekeeping genes as these were found to exhibit stable expression between different tissues. Each data point consisted of three technical replicates and four biological replicates. Data were analysed using the ΔΔCT method [44] using the geometric mean of the two housekeeping genes to normalise data.
|
10.1371/journal.pgen.1000872 | PPS, a Large Multidomain Protein, Functions with Sex-Lethal to Regulate Alternative Splicing in Drosophila | Alternative splicing controls the expression of many genes, including the Drosophila sex determination gene Sex-lethal (Sxl). Sxl expression is controlled via a negative regulatory mechanism where inclusion of the translation-terminating male exon is blocked in females. Previous studies have shown that the mechanism leading to exon skipping is autoregulatory and requires the SXL protein to antagonize exon inclusion by interacting with core spliceosomal proteins, including the U1 snRNP protein Sans-fille (SNF). In studies begun by screening for proteins that interact with SNF, we identified PPS, a previously uncharacterized protein, as a novel component of the machinery required for Sxl male exon skipping. PPS encodes a large protein with four signature motifs, PHD, BRK, TFS2M, and SPOC, typically found in proteins involved in transcription. We demonstrate that PPS has a direct role in Sxl male exon skipping by showing first that loss of function mutations have phenotypes indicative of Sxl misregulation and second that the PPS protein forms a complex with SXL and the unspliced Sxl RNA. In addition, we mapped the recruitment of PPS, SXL, and SNF along the Sxl gene using chromatin immunoprecipitation (ChIP), which revealed that, like many other splicing factors, these proteins bind their RNA targets while in close proximity to the DNA. Interestingly, while SNF and SXL are specifically recruited to their predicted binding sites, PPS has a distinct pattern of accumulation along the Sxl gene, associating with a region that includes, but is not limited to, the SxlPm promoter. Together, these data indicate that PPS is different from other splicing factors involved in male-exon skipping and suggest, for the first time, a functional link between transcription and SXL–mediated alternative splicing. Loss of zygotic PPS function, however, is lethal to both sexes, indicating that its role may be of broad significance.
| In Drosophila the sex-specific ON/OFF regulation of Sex-lethal (Sxl) is controlled by an autoregulatory splicing mechanism that depends on the SXL protein interacting with general splicing factors. Here we identify PPS as a novel component of the machinery required for Sxl splicing autoregulation by showing that the lack of pps function interferes with Sxl expression and that the PPS protein is physically linked to the Sxl pre–mRNA, the SXL protein and components of the general splicing machinery. PPS, however, stands apart from all other proteins known to control Sxl splicing because it is not a general splicing factor. Furthermore, PPS has a distinct pattern of accumulation along the Sxl transcription unit that suggests PPS is loaded onto the RNA at the promoter. Together with the observation that the PPS protein contains four signature motifs typically found in proteins that function in transcriptional regulation, our data suggest that linking transcription to splicing regulation is important for controlling Sxl expression. This idea is especially intriguing because it indicates that the coupling of transcription and splicing seen in vitro and in cell culture studies is likely to be pertinent to developmentally controlled patterns of gene expression in the living animal.
| Understanding tissue- and stage-specific gene regulation remains one of the central issues in developmental biology. Studies of developmentally important genes, such as those that specify and maintain cell fate, have revealed that many genes are regulated post-transcriptionally. The Drosophila sex-determination gene Sex-lethal (Sxl) is a prime example of a developmental switch gene regulated by alternative splicing. Throughout most of development and in adult tissues, Sxl is controlled by sex-specific alternative splicing to produce mRNAs with different coding potentials [1]. In males, all transcripts include the translation-terminating third exon leading to the production of mRNAs that encode truncated, inactive proteins. In females, the third exon is always skipped to generate protein encoding mRNAs. The mechanism leading to exon skipping is autoregulatory and depends on the SXL protein binding to multiple intronic sites located both upstream and downstream of the regulated exon. Current models, based on both biochemical and genetic studies, suggest that SXL forces the male exon to be skipped by interacting with and antagonizing a set of general splicing factors, including the U1 snRNP, the U2AF heterodimer, FL(2)d and SPF45 [2]–[4]. Because Sxl controls both its own expression and the expression of a set of downstream target genes, this autoregulatory splicing loop serves as a heritable and irreversible molecular switch for the developmental pathways controlling both somatic sex determination and X-chromosome dosage compensation.
Initiation and stable engagement of the Sxl autoregulatory splicing loop requires the coordinated use of two alternative promoters [5]–[7]. Throughout most of development, Sxl is expressed from the non-sex specific “maintenance” promoter, SxlPm. SxlPm is first expressed during the maternal to zygotic transition, but prior to that time Sxl is transiently expressed from the female-specific “establishment” promoter, SxlPe. The SxlPe-derived transcripts, unlike the transcripts produced from SxlPm, are spliced by default to produce SXL protein. Thus the SXL protein present in XX embryos when SxlPm is first activated serves to drive the initiating round of exon skipping which leads to a self-sustaining splicing loop. In XY animals, on the other hand, SxlPe is not activated, there is no SXL protein, and all SxlPm-derived transcripts are spliced in the male mode. While coordinated promoter switching is critical for successful establishment of the Sxl autoregulatory splicing loop in early embryogenesis, it has been generally assumed that transcription plays little, if any, role in sex-specific regulation after this point.
Here we report the identification and analysis of a previously uncharacterized protein, named Protein Partner of Sans-fille (PPS, CG6525), as a novel component of the machinery that controls Sxl alternative splicing. PPS, a large multidomain protein classified as a transcription regulator based on the presence of 4 distinct and conserved sequence motifs, was identified in a yeast two hybrid screen for proteins that interact with Sans-fille (SNF), the Drosophila homolog of the U1 snRNP protein, U1A. We provide compelling evidence that PPS has a direct role in Sxl male exon skipping by showing first that the loss of pps function interferes with Sxl function, and second that PPS can form a complex with the U1 snRNP, SXL and the Sxl pre-mRNA. In addition, we mapped the association of PPS, SXL and SNF along the Sxl gene by chromatin immunoprecipitation (ChIP), providing evidence that these proteins, like many other splicing factors, bind their RNA targets while in close proximity to the DNA. While we found that SXL and SNF associate with their predicted binding sites, PPS has a distinct pattern of accumulation along the Sxl gene which suggests that PPS is loaded onto the RNA at the promoter. Finally, we show that PPS function is not restricted to Sxl splicing regulation, indicating that PPS is likely to be more broadly involved in development.
CG6525 was identified in a yeast two hybrid screen for SNF-interacting proteins, giving the gene its name protein partner of sans-fille (pps; Figure 1A). To demonstrate that the PPS/SNF interaction also occurs in Drosophila cell extracts, we assayed for complex formation by pull-down experiments in which a GST fusion protein containing the C-terminal end of PPS (amino acids 1370–2016) was expressed in E. coli, bound to glutathione sepharose beads, and incubated with protein extracts made from embryos. The presence or absence of SNF in the complex formed on the beads was assayed by Western blot analysis (Figure 1B). In control studies, we used a GST::SXL fusion protein since it is known to form a complex with SNF [2]. As predicted by the two hybrid data, we found that GST::PPS, but not GST alone, was capable of selecting SNF out of extracts as efficiently as GST::SXL. These data therefore confirm that PPS and SNF associate in vivo.
PPS is located on the 3rd chromosome (87B) and, in agreement with the predicted gene structure, we found that the pps transcription unit extends over 6.7 kb. and the 11 constitutively spliced exons are predicted to encode an uncharacterized 2016 amino acid protein (Figure 1C and 1D). The pps open reading frame contains 4 conserved motifs: PHD finger (plant homeodomain), BRK (Brahama and Kismet), TFS2M (transcription elongation factor S-II middle) and SPOC (Spen paralogue and orthologue C-terminal). According to the Gene Ontology Database, which assigns functions to uncharacterized proteins based the presence of sequence motifs, PPS is likely to function in transcriptional regulation (see discussion).
To gain insight into the biological role of PPS, we generated a molecular null allele using an FRT-based targeted deletion strategy [8],[9]. Briefly, we induced recombination in animals heterozygous for two FRT-bearing piggyBac insertions with controlled expression of the FLP recombinase and identified a deletion with the desired endpoints using a PCR based strategy. The resulting deletion, depicted in Figure 1D, removes the entire coding sequence of pps as well as the adjacent gene, Scg-β. Animals homozygous for this two gene deletion die during the third instar larval stage. Two critical experiments demonstrate that the lethality is due to the loss of pps and not Scg-β. First, lethality was fully rescued by one or two copies of P{pps+}, a genomic transgene that carries just the pps gene (90%, n = 554). Second, all aspects of the mutant phenotype remained unchanged by the addition of multiple copies of the adjacent P{Scg-β+} genomic transgene (see Materials and Methods for details). Thus, these data provide strong evidence that disruption of PPS is responsible for the larval lethal phenotype and the two gene deletion we have isolated behaves as a pps null allele. Based on these genetic data, we have named this deletion pps1.
Homozygous pps1 mutant animals fail to survive to adulthood, although all animals reach the third instar larval stage. Consistent with the failure to pupate, mutant third instar larvae were found to have a number of defects, including small, underdeveloped imaginal discs, abnormal polytene chromosome morphology and melanized patches of tissue that resemble melanotic tumors (data not shown).
Although pps null mutants complete embryogenesis without any apparent defects, we cannot rule out an earlier function in embryogenesis. PPS is a maternally provided protein and the extended stability common to many maternally provided proteins typically result in the rescue of homozygous mutant animals into the larval stages. Thus, pps mutant animals may survive until the maternal stores of protein are depleted, masking a potential requirement in embryogenesis.
During the course of this analysis, we noted that, while either one or two copies of the P{pps+} transgene was sufficient to rescue the lethality of pps1 homozygous mutant females, two copies were necessary to rescue the females to fertility. An examination of the ovaries isolated from these sterile mutant females revealed that the ovaries contained tumors (Figure 2A). Ovarian tumor phenotypes are also observed in partial loss of function snf mutant backgrounds, where the phenotype is caused by defects in Sxl splicing regulation [2],[10]. To investigate the possibility that the pps tumor phenotype is also correlated with Sxl misregulation, we used RT-PCR to assay the Sxl RNA products present in isolated ovarian tissue. Using a single primer pair capable of detecting the female and the larger male spliced products, we found that in ovarian tissue isolated from sterile mutant females, a significant proportion of the spliced products contained the male-specific exon (Figure 2B and 2C). Thus, based on these partial loss of function mutant phenotypes, we conclude that pps, like snf, is required to achieve stable Sxl activity in the female germline.
Activation of Sxl in the embryo is a multi-step process, starting with the coordinated use of two promoters and culminating with successful engagement of the autoregulatory splicing loop. Thus, perturbation of any single step in the process can lead to a defect in alternative splicing. As a consequence, embryos heterozygous for the normally recessive null allele of Sxl (Sxlf1/+) are particularly sensitive to the supply of specific splicing and transcription factors deposited into the egg by the mother (e.g. [2]–[4]). We therefore reasoned that if maternally provided PPS protein is important for any aspect of Sxl regulation, we might expect the viability of Sxlf1/+ females to be affected if their mothers were heterozygous for pps (pps1/+). However, we found that these Sxlf1/+ females were as viable as their control siblings (data not shown). To increase the sensitivity of this assay, we introduced a mutant allele of daughterless (da2) into the genetic background. da encodes a maternally supplied transcription factor required to activate Sxl [11],[12]. We chose da2 to sensitize the genetic background because we have previously shown that the genetic interaction between snf and da is particularly strong [13]. In control crosses, we found that 57% of the expected Sxlf1/+ daughters from da2/+ mothers survived to adulthood (n = 275; Figure 3). However, when the mothers were heterozygous for both pps1 and da2, there was a significant reduction in viability with only 7% of the expected Sxlf1/+ daughters surviving to adulthood (n = 222). Restoration of female viability by the genomic rescue construct P{pps+} indicates that this female-lethal synergistic interaction is due to the loss of pps function (26%; n = 517).
To confirm the genetic relationship between pps and Sxl, we looked for synergistic interactions with mutant alleles of fl(2)d, U2af38 and spf45. Mutations in these three genes were picked because they encode core spliceosomal proteins known to play an important role in Sxl autoregulation [2]–[4]. These data show that pps1 in combination with mutations in each of these spliceosomal genes exerts a detrimental synergistic effect on the viability of Sxlf1/+ females (Table 1). Together, these data indicate that the maternally provided PPS protein contributes, in some way, to Sxl regulation.
Previous studies have shown that SXL interacts with SNF in the context of the U1 snRNP [2]. We reasoned, therefore, that if pps has a direct role in Sxl splicing autoregulation, then we might be able to detect physical interactions between PPS, the U1 snRNP and SXL. To test this, we generated an antibody against the C-terminal end of PPS (amino acids 1370–2016) for co-immunoprecipitation assays. PPS is predicted to encode a single polypeptide of 222 kD, and as predicted, we found that on Western blots, the wild type protein migrates at about 220 kD in extracts made from adults of both sexes, embryos and third instar larvae (Figure 4A, and data not shown). In contrast, no immunoreactivity was detected in extracts made from third instar larvae homozygous for pps1, demonstrating the specificity of this antibody. Using this antibody for co-immunoprecipitation experiments, we were able to confirm that PPS and SNF associate in vivo (Figure 4B). As expected, RNase addition did not abrogate the SNF/PPS interaction, even though the RNase treatment was sufficient to disrupt the known RNase-sensitive interaction between SNF and U2A' (Figure 4B).
To test whether PPS associates with SNF as a component of the U1 snRNP, we asked whether we could detect an interaction between PPS and another core U1 snRNP protein, U1-70K. Our data shows that we were able to co-immunoprecipitate PPS and U1-70K (Figure 4C), although we noted that PPS seems to preferentially associate with the more slowly migrating U1-70K species, among the major U1-70K isoforms observed in whole cell extracts. U1-70K is a phosphorylated protein, and studies in mammalian cells that have shown that dephosphorylation of U1-70K is necessary for the splicing reaction to proceed [14]. Thus, if PPS does in fact preferentially associate with the highly phosphorylated form of U1-70K, our data would lead to the conclusion that PPS, unlike SNF, only transiently associates with the U1 snRNP. Direct support for this conclusion comes from our more detailed analysis of PPS's role in Sxl splicing autoregulation described below.
Finally, we asked whether PPS associates with the SXL protein and found that antibodies against the PPS protein can in fact immunoprecipitate SXL (Figure 5A). Interestingly, this interaction was weakened when we carried out these experiments in the presence of RNase. This suggests that the SXL/PPS interaction is mediated and/or stabilized by RNA.
Because the SXL protein exerts its effect by binding directly to its own pre-mRNA, we postulated that PPS might also associate with the unspliced Sxl pre-mRNA. To test this idea, we asked whether Sxl pre-mRNA is detectable in PPS immunoprecipitates. The results of these RNA immunoprecipitation assays (RIP), which were carried in nuclear extracts without fixation, clearly shows that the unspliced Sxl RNA is detectable by RT-PCR using an intron 3-exon 4 primer pair (Figure 5B). In control reactions, we found that Sxl RNA was also detected in SXL immunoprecipitates, but not in extracts treated with antibodies against the chromatin binding protein Polycomb (PC) or in pre-immune serum.
To determine whether the SXL protein is required for the association between PPS and the Sxl pre-mRNA, we carried out RIP assays in nuclear extracts made from embryos collected from mothers homozygous for a viable allele of daughterless, da1. da1 mutant mothers produce eggs that lack SXL protein because SxlPe is not activated [12]. SxlPm, however, is activated, and the resulting transcripts are therefore spliced in the male mode. As illustrated in Figure 5C, PPS was able to co-immunoprecipitate unspliced Sxl RNA in these SXL-deficient mutant extracts. In control reactions, we found that Sxl RNA was detected in SNF immunoprecipitates, but not in controls. Thus, we conclude that the PPS/Sxl pre-mRNA association does not depend on the presence of SXL protein in the extract.
To gain a better understanding of the functional relationship between PPS, SXL and SNF, we compared the dynamics of their recruitment to the nascent Sxl transcript by combining genetic analysis with chromatin immunoprecipitation (ChIP) assays (Figure 6). Splicing factor-ChIP assays, which have been used in both yeast and mammalian cells, are possible because many splicing factors are recruited to their RNA targets while still in close contact with template DNA [15]–[17].
To validate this approach, ChIP analysis was first carried out with antibodies against SNF in a sexually mixed population of wild type 8–12 hour embryos. ChIP studies in mammalian cells have shown that U1 snRNP proteins specifically target regions of genes that include 5′ splice sites of recognized exons [17]. This predicts that SNF will accumulate on a region that includes the male-specific third exon (Ex3), but not on the SXL binding site which is located ∼250 bp away in the third intron (In3). As a specificity control, we assayed for SNF accumulation on the first exon of the SxlPe transcripts (E1) because in 8–12 hour embryos E1 is treated as an intron, and thus should not be recognized by the splicing machinery. In agreement with our expectations, we found that SNF was present at the third exon (Ex3), but not at the other two locations. Additional controls for specificity include our demonstration that these three regions of the Sxl gene were not precipitated in controls or in ChIP assays carried out with the DNA binding Heat Shock Factor (HSF). As a final control for specificity, ChIPs were also carried out with the 8WG16 antibody against the hypophosphorylated form of RNA polymerase II (Pol IIa), because previous studies have shown that Pol IIa does not accumulate within the body of actively transcribed genes [18],[19].
Having shown that recruitment of SNF to the Sxl gene can be detected by ChIP, we next asked whether we could use this methodology to view SXL and PPS recruitment. In agreement with in vitro RNA binding assays [20], we found that SXL was present at the intronic SXL binding site, In3. PPS, on the other hand, was not only present on the third exon (Ex3) but also localized to the intronic E1 and In3 regions. Together these results argue that PPS, in contrast to both SNF and SXL, is uniformly distributed across the Sxl transcription unit.
Next we asked whether the pattern of recruitment is different on nascent transcripts destined to be spliced in the female or male mode. Males do not express SXL protein; therefore, SXL-ChIP of chromatin isolated from a mixed sex population of embryos resulted in the analysis of only female embryos. PPS and SNF, on the other hand, are expressed in both male and female embryos, thus the analysis of chromatin from wild type embryos would mask any sex-specific differences, should they exist. To circumvent this issue, we repeated the SNF and PPS ChIP experiments in two mutant populations of embryos. To exclusively assay Sxl transcripts destined to be spliced in the female mode, chromatin was prepared from embryos collected from a stable stock in which all females carry an attached X chromosome and all males carry Sxl7BO, an X-linked deletion allele of Sxl. As there is no Sxl DNA present in the male embryos, this analysis is limited to Sxl chromatin isolated from female embryos. To generate a population of embryos where all nascent Sxl transcripts are destined to be spliced in the male mode, we prepared chromatin from embryos from da1 mothers. As described above, maternal DA protein is required to initiate SxlPe transcription early in embryonic development, therefore all eggs laid by homozygous mutant females fail to produce SXL protein. As shown in Figure 6, we found that the pattern of PPS and SNF accumulation was not dependent on the source of the chromatin: PPS accumulated at all three sites, whereas SNF was only detected on the third exon. We therefore conclude that the recruitment pattern of PPS and SNF along the Sxl gene is the same in males and females.
The uniform distribution of PPS on the Sxl transcription unit, together with its classification in the Gene Ontology Database as a protein involved in transcription, suggested to us that PPS might initially be recruited near SxlPm. We therefore repeated the ChIP experiments using two different primer sets targeting sequences upstream of the SxlPm transcription start site (P1 and P2) and one that includes the first exon (P3). ChIP studies in Drosophila and mammalian cells have shown that the hypophosphorylated form of RNA polymerase II (Pol IIa), detected by the 8WG16 antibody, is highly concentrated at the start of actively transcribed genes [18],[19]. In agreement with these studies, we found that Pol IIa specifically accumulates at P1, P2 and P3 (Figure 7). SNF, as expected, only accumulates on P3, the region that overlaps with the first exon. As shown in Figure 7, we found that PPS accumulates on P1, P2 and P3 and that this distribution is not sex-specific. Taken together, these results suggest that PPS associates with the Sxl promoter.
In addition to its autoregulatory function, the SXL protein also binds the tra pre-mRNA to regulate its sex-specific expression [21]. To determine whether PPS is involved in tra pre-mRNA splicing, we first carried out RIP assays and found that tra pre-mRNA is detectable in PPS immunoprecipitates, as well as in control SXL and SNF immunoprecipitates (Figure 8A). We then carried out ChIP experiments to determine whether PPS is recruited to the tra promoter region (Figure 8B). To demonstrate that we had targeted the promoter region, ChIP experiments with antibodies against the hypophosphorylated form of RNA polymerase II (Pol IIa) 8WG16 were used as a positive control. Antibodies against SNF are used here as a negative control. In accordance with our expectations, we found that PPS does in fact associate with the tra promoter region.
While these studies clearly suggest that PPS has an additional role in tra splicing regulation, it is unlikely that PPS is globally associated with all actively transcribed genes, as we fail to detect associations with the intronless U2A gene and the intron containing snf gene (Figure 8A and 8B). On the other hand, PPS is clearly not limited to SXL-mediated splicing events because loss of PPS function is lethal to both sexes. What these additional functions are remains to be determined.
Genetic studies have established that SXL protein is both necessary and sufficient to engage the Sxl autoregulatory splicing loop [22]. Mechanistically, however, SXL does not act alone and collaborates with components of the general splicing machinery, including the U1 snRNP, to block inclusion of the male exon [2]. In this study, ChIP assays showed that SNF and SXL are specifically recruited to their predicted binding sites on the nascent transcript: SNF to 5′ splice sites and SXL to its intronic binding sites. These data, together with our observation that the recruitment of SNF is not influenced by the presence or absence of SXL, support the current model in which SXL blocks male exon inclusion by interacting with general splicing factors bound to authentic splice sites (Figure 9). Splicing could be blocked immediately, or spliceosome assembly could continue, stalling only later in the pathway. The U1 snRNP, however, is only transiently associated with the spliceosome as it assembles on the splicing substrate and is released before the spliceosome is catalytically active [23]. Therefore it is likely that SXL acts by interrupting spliceosome assembly at some point after splice site recognition by the U1 snRNP, but before catalysis begins.
In studies begun by screening for SNF-interacting proteins, we identified PPS, a conserved and previously uncharacterized Drosophila protein, as a novel component of the machinery required for skipping the Sxl male exon. We were able to establish this connection by demonstrating that (1) animals carrying loss of function pps mutations are compromised in their ability to regulate Sxl splicing, (2) PPS associates with the U1 snRNP via a direct interaction with SNF and (3) PPS associates with the SXL protein and the unspliced Sxl RNA.
Although physically associated with the U1 snRNP, PPS does not appear to be a general splicing factor because it does not associate with all spliced transcripts (this study), it is not found in affinity-purified Drosophila spliceosomal complexes [23] and it is not a homolog of a previously identified human splicing protein [24]. Thus, PPS stands apart from the other proteins known to facilitate proper Sxl splicing, all of which are known to be components of the splicing machinery.
The results of our ChIP analysis also distinguishes PPS from known splicing factors, as it reveals a strikingly distinct pattern of accumulation along the Sxl gene, including occupancy at the SxlPm promoter region. This pattern of accumulation suggests that PPS is loaded onto the RNA at the promoter and/or that it has a role in transcription. Numerous studies have documented physical interactions between the transcriptional machinery and splicing factors [25]. Thus, PPS may well act in concert with the transcription machinery to promote SXL-mediated exon skipping (Figure 9). For example, PPS could serve as a bridging protein to accelerate recruitment of SXL to the nascent transcript, or it could facilitate the formation of the inhibitory SXL/U1 snRNP interaction.
Whether PPS is physically coupled to the transcription machinery and/or has a role in controlling transcription will require additional studies. However, the fact that PPS contains 4 signature motifs typically found in proteins with known functions in transcription adds credence to this idea. Of these 4 motifs, the PHD finger is the most extensively studied. Numerous studies have shown that PHD fingers have histone methylation binding activity. Indeed, PPS is likely to have histone binding activity, as the PHD domains of both the S. cerevisiae (BYE1) and mammalian (DIDO) PPS homologs preferentially bind to tri-methylated H3K4 (H3K4me3) in vitro [26],[27]. The possibility of a PPS–histone link is further strengthened by the presence of the metazoan specific BRK motif, a domain that is found in only two other Drosophila proteins–Brahma and Kismet–both of which are known to be chromatin binding proteins [28],[29]. A connection to transcription is also suggested by the presence of the TFS2M motif. This motif is named after its founding member located in the center of the transcription elongation factor S-II, where it is essential for binding Pol II [30]. Finally, SPOC domains have been identified in a variety of proteins linked to transcription, the best characterized of which is the human SHARP nuclear hormone co-repressor [31],[32]. A conserved function in transcription is particularly compelling in light of the current view that transcription and splicing are mechanistically coupled. In this regard, there are a few well-documented examples of mammalian chromatin binding proteins that affect alternative splicing [33]. For example the H3K4me3 binding protein, CHD1, associates with the spliceosome and is required for efficient splicing [34]. In another example the BRK domain containing chromatin remodeling protein, BRAHMA/BRG1, influences the alternative splicing of several transcripts [35].
Although still speculative, a mechanism linking transcription to splicing regulation is likely to be of major importance in early embryogenesis. Engagement of the autoregulatory splicing loop requires that the initiating source of SXL protein, produced from the transiently expressed SxlPe derived transcripts, be present when SxlPm is activated so that its transcripts can be alternatively spliced to produce more SXL protein. The changeover from SxlPe to SxlPm is tightly coordinated and uncoupling these events leads to disruptions in Sxl regulation [6],[7]. While these studies suggest that transcriptional regulation of SxlPm is important for the switch to autoregulation, our studies lead us to propose that PPS contributes to the success of this switch by concurrently facilitating SxlPm transcription and promoting male-exon skipping.
PPS function is not restricted to Sxl splicing regulation. In studies designed to test for specificity, we discovered that PPS also associates with the SXL-regulated tra pre-mRNA. In addition, we found that pps function is essential for viability of both sexes, indicating that pps function is not limited to SXL-mediated splicing events and is involved in other developmental pathways. In humans, the PPS homolog DIDO has been linked to a blood disorder called myeloproliferative disease (MPD) [36]. The relevance of this connection is suggested by our finding that homozygous pps mutant larvae contain melanotic tumors, tumors that often result from over-proliferation and aggregation of blood cells [37]. Thus, the discovery of PPS' role in controlling alternative splicing may be of significance to additional developmental pathways.
Using the entire SNF protein as bait, we screened 9.8×107 clones from Drosophila embryonic and adult cDNA libraries and identified 78 positive clones, all of which included the C-terminal end of the pps (CG6525) gene. PPS was also reported to be a binding partner of CDK7 (CG3319) [38]. However, we have not been able to verify the authenticity of this interaction (data not shown), and suspect that this interaction is based on an annotation error because the snf and cdk7 genes partially overlap [39].
Mutant alleles and deficiencies used in this study include: Sxlf1, Sxl7BO da1, da2, fl(2)d2, U2af38ΔE18, Df(2Lh)D1 (designated as spf45Δ in Table 1), Df(3R)Exel7316, PBac{WH}Dip-Cf00706 and PBac{WH}CG17202f01979 [2]–[4], [8], [12], [40]–[42]. We generated pps1 by FRT-mediated recombination between PBac{WH}Dip-Cf00706 and PBac{WH}CG17202f01979 using the conditions described previously [8],[9]. Throughout this analysis we found that the phenotypes of pps1/pps1 and pps1/Df(3R)Exel7316, animals to be identical, indicating the absence of confounding background mutations on the pps1 mutant chromosome. The P{pps+} and P{Scg-β+, CG17202+} genomic rescue constructs were generated by standard methods in the pCaSpeR4 transformation vector and transgenic flies were produced at Genetic Services (http://www.geneticservices.com). Functional P{Scg-β+, CG17202+} transgenes (abbreviated as P{Scg-β+} in the text) were selected based on their ability to complement a known point mutation in CG17202. Each transgenic line was then tested for its ability to rescue the different pps mutant phenotypes, including the lethality of pps1/Df(3R)Exel7316 and pps1/pps1 animals. The data presented in this paper are obtained with P{pps+} line # 10. Additional marker mutations and balancers used in this study are described on Flybase (http://www.flybase.org).
The antibody against PPS was raised in guinea pig by Covance (http://www.covance.com) against a glutathione S-transferase (GST) tagged C-terminal domain PPS fragment (amino acids 1370–2016) purified from bacteria. We note here that this PPS antibody has not proven to be useful for immunohistochemistry. The other antibodies used in this study include mouse anti-SNF-4G3 [43],[44], guinea pig anti-U2A' [45], rabbit anti-U170K-151 [2], mouse anti-SXL-M114 [46], guinea pig anti- HSF [47], rabbit anti-PC [48], and mouse anti-RNA Pol IIa-8WG16 (Millipore, #05-952). Crude extracts for GST-pull down experiments (Figure 1) and Western blots (Figure 4) were prepared from 3–8 hour old embryos, sexed and genotyped third instar larvae or sexed adults in NET buffer (150 mM NaCl, 50 mM Tris, pH 7.5, 5 mM EDTA) supplemented with 0.5% NP-40 and Complete Mini Protease Inhibitor Cocktail Tablets (Roche). Nuclear extracts for co-immunoprecipitation experiments were prepared from 3–18 hour old embryos as described previously [49] using NET buffer supplemented with 0.5% NP-40 for the co-IPs in Figure 4 and 0.05% NP-40 for the co-IPs in Figure 5. For experiments in which the extracts were pretreated with RNase, 1/10 volume of RNase A (10 mg/ml) and 1/20 volume of RNase T1 (100,000 units/ml) were added directly to the extract and incubated overnight at 4°C. Co-immunoprecipitations, Western blot analysis and GST pull down assays were carried out according to standard protocols, using the conditions described previously [2],[4],[50].
Total RNA was isolated from ovaries, adults or embryos using TRIzol (Invitrogen) as directed by the manufacturer. To analyze the endogenous Sxl splicing products, the first strand synthesis was carried out with 1 µg of RNA, 500 ng/µl random hexamers with the SuperScript II Reverse Transcriptase System (Invitrogen). The PCR reactions, using the High Fidelity Taq system (Roche), were performed in 50 µl volume with 2 µl of the RT reaction with the following primers: GTGGTTATCCCCCATATGGC and GATGGCAGAGAATGGGAC. The PCR conditions were as follows: 94°C for 1 min, followed by 30 cycles of 94°C for 1 min, 55°C for 1 min, and 72°C for 2 min, and a single final step at 72°C extension for 10 min. Products were detected on a 2% agarose gel by staining with ethidium bromide.
RNA/protein complexes were immunoprecipitated from nuclear extracts and diluted to 5 µg/µl in NET buffer (150 mM NaCl, 50 mM Tris, pH 7.5, 5 mM EDTA), supplemented with 0.05% NP-40, Complete Mini Protease Inhibitor Cocktail Tablets (Roche) and RNase inhibitor (100 U/ml) using the conditions described previously [50]. RNA was isolated from the RNA/protein complexes using TRIzol (Invitrogen) as directed by the manufacturer. RNA was resuspended in 20 µl RNase-free water and DNase-treated. cDNA was synthesized with the SuperScript II Reverse Transcriptase System (Invitrogen) using 4 µl of the eluted RNA with a Sxl gene specific primer to exon 4 (GATGGCAGAGAATGGGAC; Figure 6) or random hexamers (Figure 8). The PCR reactions, using the High Fidelity Taq system (Roche), were performed in 50 µl volume with 2 µl of the RT reaction with the following primers–Sxl: GAGGGTCAGTCTAAGTTATATTCG and GATGGCAGAGAATGGGAC; snf: GGGATGTGCGAATGACTAG and GACTGGAGTTGCGTTCAC; tra: GATGCCGACAGCAGTGGAAC and GATGGCACTGGATCAGAATCTG; U2A: GGTGAAACT AACGCCGGAGC and CTCAGCTCCTGCAGGTTGTTG. PCR conditions were as follows:: 94°C for 1 min, followed by 30 cycles of 94°C for 1 min, 55°C for 1 min, and 72°C for 2 min, and a single final step at 72°C extension for 10 min. 2 µl of the first-round PCR amplification was subjected to a second round of PCR. . Products were detected on a 2% agarose gel by staining with ethidium bromide.
Live embryos were dechorionated with 50% bleach and fixed for 15 min in a 1.8% paraformaldehyde/heptane fixative solution. Chromatin was prepared from 1–2 gram of fixed 8–12 hour old embryos using the conditions described previously [51] and sonicated for a total of 80 seconds (20 sec pulses with a 1 min rest on ice) to produce sheared products of 300 to 400 bp. ChIP assays were performed with a commercially available ChIP assay kit (#17–295; Millipore). Antibodies used for the IP step were diluted 1∶40 (Pol IIa, HSF, PC and PPS) and 1∶20 (SXL and SNF). After purification, the ChIPed DNA samples were resuspended in 30 µl water. Enrichment of specific DNA fragments was analyzed by PCR on 2 µl ChIP material with the following primer sets: For Sxl–P1: CGGGGCTCAAAAGACATAAA and GCGTTAGTTAAGACTCAC TCCATTT; P2: CCGTTACGAATCAAGCGAAG and GGCTGGTCACAC TGTTCATT; P3: CAGCCGAGTGCCTAGAAAAA and ACTTTCCTTCTTCGGCAACA; E1: CAAGTCCAACTTGTGTTCAGA and TCGAACAGGGAGTCACAGTAT; Ex3: CGAAAAGCGAAAGACACTC and GTG TCCTCGATTCAAAAACAT; In3: ACATCATGCTTTTCTTAAGTGC and AACGATCCCCCAGTTATATTC. For U2A–GGCAGCGAATTG TTTTTCTG and GAATCTTATAGCCGCGCAAA; For tra–TGGTCTCCATGGAAAACGAG and TGCAAACACGGTTTCATTTC; For snf–AAACACCGGTGCGATAACAT and CGTTTGGTTGGGTAGCATCT. The PCR conditions for Sxl primers P1, P2, P3, E1 and Ex3, tra and snf were as follows: 94°C for 2 min, followed by 25 cycles of 94°C for 30 sec, 53°C for 30 sec, and 72°C for 1 min. The PCR conditions for Sxl In3 and U2A were as follows: 94°C for 2 min, followed by 25 cycles of 94°C for 30 sec, 55°C for 30 sec, and 72°C for 1 min. Products were detected on a 3% agarose gel by staining with ethidium bromide.
|
10.1371/journal.pgen.1003191 | Comprehensive Methylome Characterization of Mycoplasma genitalium and Mycoplasma pneumoniae at Single-Base Resolution | In the bacterial world, methylation is most commonly associated with restriction-modification systems that provide a defense mechanism against invading foreign genomes. In addition, it is known that methylation plays functionally important roles, including timing of DNA replication, chromosome partitioning, DNA repair, and regulation of gene expression. However, full DNA methylome analyses are scarce due to a lack of a simple methodology for rapid and sensitive detection of common epigenetic marks (ie N6-methyladenine (6 mA) and N4-methylcytosine (4 mC)), in these organisms. Here, we use Single-Molecule Real-Time (SMRT) sequencing to determine the methylomes of two related human pathogen species, Mycoplasma genitalium G-37 and Mycoplasma pneumoniae M129, with single-base resolution. Our analysis identified two new methylation motifs not previously described in bacteria: a widespread 6 mA methylation motif common to both bacteria (5′-CTAT-3′), as well as a more complex Type I m6A sequence motif in M. pneumoniae (5′-GAN7TAY-3′/3′-CTN7ATR-5′). We identify the methyltransferase responsible for the common motif and suggest the one involved in M. pneumoniae only. Analysis of the distribution of methylation sites across the genome of M. pneumoniae suggests a potential role for methylation in regulating the cell cycle, as well as in regulation of gene expression. To our knowledge, this is one of the first direct methylome profiling studies with single-base resolution from a bacterial organism.
| DNA methylation in bacteria plays important roles in cell division, DNA repair, regulation of gene expression, and pathogenesis. Here, we use a novel sequencing technique, Single-Molecule Real-Time (SMRT) sequencing, to determine the methylomes of two related human pathogen species, Mycoplasma genitalium G-37 and Mycoplasma pneumoniae M129. Our analysis identified two novel methylation motifs, one of them present uniquely in M. pneumoniae and the other common to both bacteria. We also identify the methyltransferase responsible for the common methylation motif and suggest the one associated with the M. pneumoniae unique motif. Functional analysis of the data suggests a potential role for methylation in regulating the cell cycle of M. pneumoniae, as well as in regulation of gene expression. To our knowledge, this is one of the first genome-wide approaches to study the biological role of methylation in a bacterial organism.
| Among a few documented mechanisms, methylation of specific DNA sequences by DNA methyltransferases provides one way by which epigenetic inheritance can be orchestrated [1]. For instance, in many eukaryotes, methylated cytosine residues at 5′-CG-3′ (CpG) sequences are recognized by methyl-CpG binding proteins that usually repress the transcription of local DNA regions [2]–[5]. In the bacterial world, methylation is most commonly associated with restriction-modification (R-M) systems that provide a defense mechanism against invading foreign genomes [6]. In addition, it is known that a variety of enzymes capable of methylating DNA at adenine [7] and cytosine [8], [9] play functionally important roles, including timing of DNA replication, chromosome partitioning, DNA repair, transposition and conjugal transfer of plasmids, and regulation of gene expression [7], [10]–[16]. Phenomena involving inheritance of DNA methylation patterns are also known in bacteria. These systems use DNA methylation patterns to pass on information regarding the phenotypic expression state of the mother cell to the daughter cells. Methylation can alter the DNA structure and affect the binding of regulatory protein(s) to its DNA target site, thereby controlling gene expression [17], [18]. Notably, most adhesion genes in Escherichia coli are regulated by DNA methylation patterns [19], [20]. Little is known about how widespread heritable epigenetic control is in the bacterial world or the roles that epigenetic regulatory systems play in bacterial biology, including pathogenesis. For instance, it has been shown that DNA methylation in Streptococcus mutans up-regulates the expression of virulence factors like gbpC and bacteriocins [21]. It has also been shown that in E. coli, the expression of the Type IV secretion gene cluster is regulated by a non-stochastic epigenetic switch that depends on methylation of the Fur binding box [22].
In some gram-positive and gram-negative species that have been studied, adenine methylation plays a critical role in regulating chromosome replication. Adenine is generally methylated by members of the Dam family of methyltransferases, such as Dam in E. coli and DpnII in Streptococcus pneumoniae, that recognize the sequence motif 5′-GATC-3′ [23]. In these bacteria, the protein SeqA binds to hemi-methylated DNA target sites (5′-GATC-3′) clustered at the origin of replication (oriC) and sequesters the origin from replication initiation. SeqA also binds to hemi-methylated 5′-GATC-3′ sites in the dnaA promoter, blocking the synthesis of DnaA protein, which is necessary for replication initiation [24]–[27]. All of these events use the hemi-methylated state of newly replicated DNA as a signal. This hemi-methylated DNA is generated by semi-conservative replication of a fully methylated DNA molecule. Because of the transient nature of the hemi-methylation state, none of these phenomena are heritable. However, this mechanism is not universal, and other bacteria, like Bacillus subtilis, lack the Dam methyltransferase and SeqA proteins that E. coli employs to repress (sequester) its oriC during replication [28].
While there are many studies demonstrating the potential roles of methylation in epigenetic control of bacteria, the number of studies is significantly smaller than those for eukaryotes. This dearth of studies on bacterial epigenetics is partly due to a lack of a simple methodology that would allow rapid and sensitive detection of common epigenetic markers, such as N6-methyladenine (6 mA) and N4-methylcytosine (4 mC), in these organisms. Through bisulfite treatment, 5-methylcytosine (5 mC) was the only base modification detectable with efficiency and sensitivity suitable for genome wide epigenetic studies [29], [30]. Recently, Single-Molecule, Real-Time (SMRT) sequencing was described to provide the capability of directly detecting different base modifications beyond the canonical A, C, G, and T bases, in addition to yielding the sequence information [31]. The technique has been successfully demonstrated to identify methyltransferase specificities on plasmids [32].
Here, we use SMRT sequencing to comprehensively determine the methylomes of two mycoplasma species, Mycoplasma genitalium and Mycoplasma pneumoniae, with single-base and -strand resolution. M. pneumoniae and M. genitalium are closely related human pathogens that cause atypical pneumonia and non-gonococcal urethritis, respectively [33], [34]. These bacteria are members of the Mollicutes class characterized by the lack of a cell wall and by their reduced genomes with a low GC content. The genome sizes of M. pneumoniae and M. genitalium are 816 kb and 580 kb, respectively [35], [36]. M. genitalium is widely considered to have the smallest genome of any bacteria that can be grown in a test tube in the absence of host cells [37]. Our analysis identified a widespread 6 mA methylation sequence motif common to both bacteria (5′-CTAT-3′, with m6A in italics), as well as a more complex Type I m6A sequence motif in M. pneumoniae (5′-GAN7TAY-3′/3′-CTN7ATR-5′). Analysis of the chromosome distribution pattern of the first motif in M. pneumoniae suggests that methylation is involved in regulating cell division. To our knowledge, this work is one of the first comprehensive methylome analysis of bacteria.
We analyzed the genomes of M. pneumoniae and M. genitalium for all the putative methyltransferase genes using comparative sequence analysis and our previous functional assignment [38]. In the M. pneumoniae genome, we identified different putative Type I and Type II restriction modification systems. Type I involves a complex consisting of three polypeptides: R (restriction), M (modification), and S (specificity). The resulting complex can both cleave and methylate DNA. The S subunit determines the specificity of both restriction and methylation [39]. M. pneumoniae Type I system includes a methyltransferase (mpn342), a DNA specific recognition protein that brings the methyltransferase to the target DNA (HdsS, mpn343), and a restriction enzyme that cleaves unmethylated DNA (HdsR, mpn345). The restriction protein HdsR gene contains three frameshift mutations which likely make it inactive (additional protein fragments could be coded by mpn346 and mpn347). There are also some isolated genes encoding duplicated copies of the specificity determining subunit HdsS (mpn089, mpn289, mpn290, mpn365, mpn507, mpn615, and mpn638). In the Type II, methyltransferase and endonuclease are typically encoded as two separate proteins and act independently [39]. In M. pneumoniae, Type II systems could consist of the methyltransferase protein (HsdM, mpn107, mpn108 or mpn111) and the restriction enzyme (HsdR, mpn109 or mpn110). Additionally, a putative uncharacterized methyltransferase (mte1; mpn198), annotated as an EcoRI-like methylase in Uniprot and not associated with any R-M system, was identified. EcoRI restriction/modification system (R/M) is a Type II system that has been well characterized in vivo and in vitro [40], [41]. M. genitalium has an orthologous of mpn198 (mg184) and only one of the Type II-specificity determining subunits HdsS, mpn638 (mg438) (Table 1).
We looked at the transcript and protein levels for the putative genes involved in methylation systems by using information of the transcriptome [42], [43] and proteome [44] of M. pneumoniae (Table 1). Although we could detect transcripts in the tiling array for all genes, albeit at very low level for many of them, we could identify in multiple MS experiments unique peptides for only six of them: mpn109, mpn198, mpn342, mpn343, mpn615, and mpn638 (Table 1). Of these, mpn198, mpn342, mpn615 and mpn638 were found to bind DNA by doing affinity chromatography with a DNA column followed by salt elution and MS analysis (manuscript in preparation). Only mpn198 (mte1; EcoRI-like) and mpn342 (Type I) are putative DNA adenine methyltransferases.
Identification of methylated bases in M. pneumoniae and M. genitalium genomes was performed by SMRT sequencing at exponential (6 h) and stationary phases (96 h). Figure 1A shows the results of the genome-wide base modification detection analysis for the M. pneumoniae genome in stationary phase. The inner and outer most tracks in the Circos plot are the modification values (Qmod) of polymerase kinetics for the reverse and forward strands of the genome relative to an unmodified WGA (whole genome amplification) control. Qmod is the −10log(Pvalue) from a t-test and described in further details in the Materials and Methods section. The plot shows many significant peaks which correspond to methylated template positions. Figure 1B shows examples of the IPD (interpulse duration) ratios of a representative genomic section, highlighting both the base and strand resolutions of the technique. The statistically significant peaks, which were defined as Qmod >100 (Figure 1C; see Methods), were clustered as a function of sequence context to determine the recognition motifs of the methyltransferases responsible for the observed signals. The clustering results for M. pneumoniae identified >99.9% of all detected genomic positions as falling into two distinct sequence motifs: 5′-CTAT-3′ and 5′-GAN7TAY-3′/3′-CTN7ATR-5′ (Y = T or C and R = A or G, with m6A in italics). The first motif is found in both bacteria and is methylated on only one of the two DNA strands. In the second motif, the first adenines in the plus and minus strands are methylated (Figure 1B). The stretch of degenerate bases that separates the two recognition elements in the motif is characteristic of Type I methyltransferase signatures (Figure 1C) [45]. Despite the fact that the second sequence motif appears 1825 times per strand in M. genitalium (Table 2), there was no instance where it was detected as methylated. In contrast, this motif appears 1681 times in the genome of M. pneumoniae and 1678 are methylated (99,8%, Table 2). Approximately 1–2% of the assigned peaks were secondary peaks of the primary detected m6A and treated as redundant information for the tabulation in Table 2 [31].
Analysis of two biological replicates of M. pneumoniae grown for 96 hours showed a reproducibility of 99.88% in the assignment of methylated positions.
Putative Type II independent methyltransferases (HsdM) (mpn198, mpn107, and mpn108) without an associated DNA recognition partner (HsdS), considered as possible candidates for the methylation of 5′-CTAT-3′ motif, were cloned into pRSS vector and then transformed into a methyltransferase-free E. coli ER2796 (DB24) [46] (Table S9b) following procedures described previously [32]. Mpn111 was discarded because it is a duplication of mpn108.
After cloning, the different plasmids were isolated and analyzed by SMRT sequencing. Of the three putative single proteins with methyltransferase activity, only mpn198 was capable of modifying the 5′-CTAT-3′ sequence. Interestingly, this is the only one of this group of methyltransferases that was found to be expressed by mass spectroscopy (MS) analyses (Table 1). As expected, no methyltransferase was identified by this approach for the Type I 5′-GAN7TAY-3′/3′- CTN7ATR-5′motif, since Type I motifs also require the DNA recognition protein HsdS [45]. These results agree with the finding that both mycoplasma species are methylated at the same motif (5′-CTAT-3′) and share a common methyltransferase, namely, mpn198 in M. pneumoniae and mg184 in M. genitalium.
The fact that our MS analysis in M. pneumoniae detected protein expression only for DNA methylases MPN198 and MPN343, together with the lack of a mpn343 ortologue and the absence of the 5′-GAN7TAY-3′/3′-CTN7ATR-5′ methylated motif in M. genitalium, suggest that MPN343 could be responsible for the methylation of the 5′-GAN7TAY-3′/3′-CTN7ATR-5′ motif. These results validated the motifs observed for M. genitalium and M. pneumoniae and identified them as the recognition sequences of previously unassigned methyltransferases.
The new identified methyltransferases have been submitted in the REBASE and re-named using the standard nomenclature (mpn198: M.MpnI, mpn342: M.MpnII, mpn343: S.MpnII, mg184: M.MgeI). M indicates methyltransferase; S refers to the specificity subunit for Type I system; Mpn indicates M. pneumoniae and Mge indicates M. genitalium.
We next focused on M. pneumoniae to study the role of methylation in regulating gene expression and DNA replication, since the transcriptome and proteome data are currently available for it [42], [43].
To study the putative role of methylation in DNA replication, we analyzed the density distribution of the 5′-CTAT-3′ methylation motifs in a sliding window of 1 kb along the M. pneumoniae genome (Figure 2A). The mean number of 5′-CTAT-3′motifs per 1 kb window is two (±1.6 standard deviation). Regions with more than five 5′-CTAT-3′ motifs (Pvalue<0.01) were considered to be “hot spots of methylation” for 5′-CTAT-3′ (Table S2b). A functional enrichment analysis of all the genes in M. pneumoniae present at the 5′-CTAT-3′ hotspots showed two functional categories of clusters of orthologous groups (COGs) over-represented: defense mechanisms (Pvalue = 0.025) and genes coding for membrane proteins or lipoproteins (Pvalue = 9×10−4) (Table S4a). Of the hot spots, there are three regions that have more than 10 motifs/kb. Interestingly, these regions are symmetrically distributed around the first kb of the genome (Figure 2B). This region of the genome comprises an intergenic region of 687 bp with three non-coding RNAs (MPNs200, MPNs201, and MPNs381) that frame eight repetitive 5′-TATTA-3′ sequences (identified as DnaA boxes based on Chip-seq analysis; Yus et al manuscript in preparation; Figure 2C [47]). There are three 5′-CTAT-3′ methylation motifs, two of them in overlapping and opposite strands of the region with the putative DnaA boxes suggesting that DNA methylation, although different from E. coli, could play a role in DNA replication. The other two regions are located at approximately 105 kb to the left and right from the putative origin of replication (Figure 2B). Search of common motifs in these two methylation hot spots revealed a common motif of 14 bp (5′-GATAG/ACCAAGG/AAGC-3′) (Figure 2D). This motif is found at opposite strands in the two regions, but only the left side region contains the 5′-CTAT-3′ sequence overlapping.
We also analyzed the genome-wide distribution of the Type I motif. The average distribution for the 5′-GAN7TAY-3′/3′- CTN7ATR-5′motif in 1 kb is 1 motif/kb (±1.1 standard deviation), and hot spot regions were considered to be those with more than 3 motifs within 1 kb (Pvalue<0.01). Most of the genes that overlap with these hotspots are of unknown function with a Pvalue of 0.04 (Table S4b). There are four 1 kb regions in the genome that have more than five instances of 5′-GAN7TAY-3′/3′-CTN7ATR-5′methylation (Figure 3A, and Table S2a). Interestingly, this highly methylated region with the most motifs (6 in 582 bp), is within mpn140, the first gene of the cytadherence operon that contains one of the main virulence factors of M. pneumoniae (Figure 3B). These motifs are located just upstream of the transcriptional start site (TSS) of an antisense transcript (MPNs383) that could be involved in regulating the expression of mpn140 (Figure 3C). The other three enriched regions correspond to mpn684 (that encodes a conserved hypothetical protein), mpn357 (DNA ligase), and mpn358 (conserved hypothetical protein) and, surprisingly, to the region containing mpn342 (M.MpnII) and mpn343 (S.MpnII). As mentioned above, M.MpnII is the putative methyltransferase responsible for 5′-GAN7TAY-3′/3′-CTN7ATR-5′ methylation.
The genome-wide access to methylation information allows for the interrogation of genomic locations which match the methyltransferases sequence targets, but are kept in an unmethylated state by the bacterium. The results in Table 3 show 5′-CTAT-3′ and 5′-GAN7TAY-3′/3′-CTN7ATR-5′ sites that are always unmethylated, two examples are shown in Figure 4. Only one unmethylated 5′-CTAT-3′ site was identified (genome position: 466475). This motif is overlapping with the stop codon of the mpn390 gene that codifies for the dihydrolipoamide dehydrogenase (PdhD). This gene together with mpn391 (PdhC, dihydrolipoamide acetyltransferase) constitute an operon involved in pyruvate metabolism. Also, three 5′-GAN7TAY-3′/3′-CTN7ATR-5′ unmethylated sites were detected. One is located in an intergenic region and the other two sites are located inside mpn493 (UlaD, 3-keto-L-gulonate-6-phosphate decarboxylase) involved in ascorbate and aldarate metabolism and mpn503 (cytadherence protein) (Table 3). We hypothesize that these unmethylated sites indicate the presence of an interacting protein or a DNA structure that is protecting from methylation along the different phases of growth.
Recent identification of TSSs in M. pneumoniae [42] allowed us to study methylation patterns in promoter regions. We analyzed the regions comprising 40 bp upstream from the TSS (e.g. the promoter region) for 663 transcripts with TSS assigned and found 197 that were methylated in the promoter region (Table S5), with a total of 162 5′-CTAT-3′ and 74 5′-GAN7TAY-3′/3′-CTN7ATR-5′ motifs (located on both strands at the context site). Of these 197 transcripts, 103 are for non-coding RNAs (MPNs) and 89 correspond to ORFs. Fisher's exact test shows that there is a strong enrichment in methylation of MPNs promoters, with a Pvalue of 8.98×10−11. No functional enrichment is found for genes or MPNs (considering coding genes that overlap) methylated at the promoter regions (Table S6a).
Figure 5 shows the distribution in promoter regions of the distances from the methylation site (located upstream) to the TSS. Both motifs show that the highest frequency of methylation is at positions near the TSS and the Pribnow box (∼10–12 bases) (Pvalue of 0.03 for the 5′-CTAT-3′ motif, and of 0.005 for the Type I motif). These results could suggest the methylation has a potential role in transcription by affecting interaction of the sigma70, or of specific transcription factors, with the promoter.
We have also investigated the methylation pattern of 5′UTR regions encompassing the DNA sequences between the TSS and the translational start codon longer than 40 bp (long 5′UTR). Ninety two of 154 ORFs that have long 5′UTR regions showed methylation (Table S7). COG analysis of genes showing methylation in long 5′UTRs (Table S6b) revealed that genes involved in defense mechanism were three times more represented, with a Pvalue of 0.02. Interestingly, mpn342 gene (M.MpnII) has a 56 bp 5′UTR with two5′-GAN7TAY-3′/3′-CTN7ATR-5′ motifs, with 11 bp distance between the TSS and the motifs. As mentioned above, this gene could be responsible for methylating the 5′-GAN7TAY-3′/3′-CTN7ATR-5′motif, suggesting an autoregulatory gene expression mechanism.
Although the majority of the 5′-CTAT-3′ sites were methylated in both exponential (6 h) and stationary (96 h) phases, using the conservative Qmod threshold of 100, a few sites were identified as having significantly different Qmod values which would suggest a change in methylation fraction at the given sites. Figure 6 illustrates the decrease in the 5′CTAT-3′ Qmod distributions from stationary to exponential growth samples, while the 5′-GAN7TAY-3′/3′-CTN7ATR-5′ Qmod distributions remain unchanged. This drop in the Qmod values points to a potential decrease in the methylation fraction at some 5′-CTAT-3′ sites at exponential growth as compared to stationary phase. To address this question of methylation changes at any given 5′-CTAT-3′ site between the growth phases at 6 h vs 96 h, we performed a direct comparison analysis between M. pneumoniae 6 h and 96 h. From this analysis, there are 35 5′-CTAT-3′ sites that were unmethylated at 6 h but became methylated by 96 h (Qmod≥60), indicating a change in methylation status between exponential and stationary phases of growth (Table S3). Twenty-five of the 35 methylation motifs are inside genes coding for membrane proteins, one in a 5′UTR, and the rest in intergenic regions. Analyzing the transcriptome for these 25 genes at 6 h and 96 h showed that their expression levels did not significantly change (Table S3), suggesting that this change in methylation state inside the genes is not related to the regulation of gene expression at different phases of growth. It was also observed that the fraction of methylation increased from 6 h to 96 h but not vice versa, further suggesting that the methylation in these regions are dependent on the phase of growth. It is noteworthy that M.MpnI reaches its maximal level of expression at exponential growth [39]. No general increase or decrease in gene expression was found associated with methylation. However, some specific cases, such as MPNs111, displayed an increase in promoter methylation with a significant decrease in transcript levels (fold change log2 = 2.93) (Table S8).
Previous analysis of DNA methylation in several mycoplasma species by HPLC revealed the presence of 6 mA in all of them, and of 5 mC in Mycoplasma hyorhinis [48]. Further studies performed in Mycoplasma arthritidis, to increase the efficiency of transformation, revealed methylated cytosine residues at 5′-AGCT-3′ and 5′-GCGC-3′ sites [49], [50]. Our current bioinformatic analysis in M. pneumoniae and M. genitalium did not find any evidence for 5 mC and only detected 6 mA. The study of proteome data (Table S1), together with a comparative analysis of gene conservation between these two species, suggest that there is an adenine methyltransferase (M.MpnI in M. pneumoniae, and M.MgI in M. genitalium) common to both genomes, and a putative Type I system in M. pneumoniae (mpn342 for HsdM (M.MpnII), mpn343 for HdsS (S.MpnII), and mpn345 for HdsR). It also revealed other putative methyltransferases in M. pneumoniae, and parts of the Type I system identified at the genome level, but these were not detected by proteome analysis of extracts from the bacteria exposed to different stresses or along the growth curve [44], or from SDS gels. These results suggest that there are two functional methylation systems in M. pneumoniae, and one in M. genitalium.
We employed SMRT sequencing to test these hypotheses by comprehensively characterizing the methylomes of M. pneumoniae and M. genitalium. The unique capability of SMRT sequencing to have both base and DNA strand specificities in base modification detection enable whole microbial methylome profiling with unprecedented resolution. We identified an asymmetric adenine methylation motif common to both bacteria, 5′-CTAT-3′, and a Type I motif with methylated adenines in both strands (5′-GAN7TAY-3′/3′-CTN7ATR-5′) found only in M. pneumoniae. The role of M.MpnI in the methylation of the 5′-CTAT-3′ motif was experimentally validated by expressing the methyltransferase in an E. coli strain devoid of methyltransferases [32].
The 5′-CTAT-3′motif was found enriched at the putative origin of replication (ORI) in M. pneumoniae as well as at two sites ∼100 kbs distant on both sides of the ORI which could be putative replication checkpoints, like the ψ sites described in B. subtilis [51]. The presence of two methylated 5′-CTAT-3′sites on the top and bottom strands at the mid-position of the putative DNA boxes at the ORI suggests a role for methylation in regulating DNA replication by M.MpnI. This hypothesis is reinforced by the fact that we did not find a restriction enzyme associated to this gene like in a classical EcoRI Type II system, similar to Dam methyltransferase in E. coli. The oriC of E. coli also contains an enriched region of methylated motifs (5′-GATC-3′). SeqA preferentially binds to clusters of two or more hemimethylated 5′-GATC-3′sites, delaying re-methylation and preventing binding of DnaA, which controls the initiation of DNA replication [52], [53]. No orthologous to E. coli SeqA protein has been identified in M. pneumoniae. However, a fundamental difference is found between the Dam system of E. coli and the M.MpnI methyltransferase of M. pneumoniae: in M. pneumoniae, only the one strand harboring the motif at any given genomic position is methylated, while in E. coli, both strands of the 5′-GATC-3′ motif can be methylated. Thus, it is not expected that M. pneumoniae will use a similar system with SeqA as E. coli to control DNA duplication. In fact, the M pneumoniae firmicute relative B. subtilis also lacks seqA and dam orthologous but contains several other proteins, like Spo0, that regulate oriC [54], [55]. Interestingly, analysis of transcript levels along the growth curve shows that M.MpnI correlates with genes involved in transcription like mpn515 (rpoC) and mpn516 (rpoB), DNA duplication (mpn003 [gyrB] and mpn004 [gyrA]) and growth (ribosomal proteins like mpn538, mpn539, and mpn540) (Figure S1). This suggests a coordination between expression of M.MpnI and other genes involved in cell division and growth. Additionally, M.MpnI is the only methyltransferase that is essential for M. pneumoniae growth reinforcing its key role in cell cycle regulation.
Analysis of COG categories for ORFs located in regions enriched for 5′-CTAT-3′ showed that these are involved in virulence, similar to previously described adhesion genes regulated by DNA methylation in E. coli [19], [20]. We also found genes in M. pneumoniae methylated at their promoter or 5′UTR regions that have orthologous known to be regulated by methylation in other bacteria, such as trpS [56] and the SOS regulon [57] in E. coli, and ClpB in Streptococus mutans. However, no relationship between methylation and transcription levels was observed when we studied the correlation between M.MpnI and ORFs with methylation in their regulatory sequences. Nonetheless, this apparent lack of correlation may be due to the lack of synchrony in the bacterial population, which may therefore exhibit different phenotypic properties. The high number of antisense RNAs that show methylation in promoter regions could imply that in the absence of regulatory proteins, methylation could serve as a mechanism to regulate the expression of the antisense strand and, consequently, any overlapping genes.
In most active R-M systems, all sites recognized by the restriction enzyme are protected by methylation in order to prevent the microbe's own defense mechanism from damage to its genome. However, there are incidences in which a protein protects certain sites from restriction digestion or methylation. For example, a 5′-GATC-3′sequence within the regulatory region of the car operon in E. coli was found to be protected from Dam methylation [58]. Indeed, CarP and IHF were shown to bind in this regulatory region and protect the 5′-GATC-3′ site from methylation [59].We have detected unmethylated 5′-GAN7TAY-3′/3′-CTN7ATR-5′ and 5′-CTAT-3′ sites, which could indicate that there is a protein interacting with these regions. A comparative study of the transcriptome at 6 h and 96 h in M. pneumoniae did not reveal any difference in transcription of genes containing unmethylated motifs when they are compared with the rest of the genes in the genome. Thus, these regions could be interaction sites for DNA-binding proteins that protect the DNA from methylation; in this case, methylation could play a role in transcription when the interacting protein is not occupying the region [60], [61]. However, interaction of structural elements that determine the structure of chromosome cannot be ruled out. Studies of protein occupancy could help to reveal why these regions are protected from methylation.
In conclusion, using SMRT DNA sequencing, we were able to directly observe and analyze with single-base and strand resolution the genome-wide methylomes of M. genitalium and M. pneumoniae. The two strains share an analogous methlytransferase that targets the sequence 5′-CTAT-3. M. pneumoniae additionally has a Type I methyltransferase with a 5′-GAN7TAY-3′/5′-CTN7ATR-3′ specificity. Together, these 2 motifs correspond to more than 99.9% of all sites directly detected by SMRT sequencing as modified. While ongoing work involving methyltransferase knock-out and over-expression studies are underway to help establish the relationship, this work demonstrates the unique capability of SMRT sequencing to directly sequence and profile the methylome of a whole microbial genome, allowing for unprecedented progress towards understanding the role of epigenomics in the world of prokaryotes.
Escherichia coli TOP 10 strain (Invitrogen) and E. coli ER2796 (DB24) [46] deficient in methyltransferases, also called DB24 (New England Biolabs), were grown at 37°C in LB broth or LB agar plates containing 100 µgml−1 ampicillin. The M. genitalium G-37 WT and M. pneumoniae M129 strains were grown in SP-4 and Hayflick media, respectively [62] at 37°C under 5% CO2 in tissue culture flasks (TPP). Cells were grown for 96 h for the stationary phase of growth. Alternatively, after 96 h of growth, the media was removed and replaced by fresh media, and the cells were scraped and re-grown for 6 h (exponential phase of growth).
Genomic DNA of M. genitalium and M. pneumoniae was isolated using the Illustra bacteria genomic Prep Mini Spin Kit (GE Healthcare). Plasmid DNA was obtained using the QIAprep Spin Miniprep Kit (Qiagen). All primers and plasmids used in this work are summarized in Table S9a and S9b. PCR products and digested fragments from agarose gels were purified using the QIAquick PCR purification Kit (Qiagen).
Genomic and plasmid samples of M. genitalium and M. pneumoniae were prepared for SMRT sequencing following standard SMRTbell template preparation protocols for base modification detection on the PacBio RS [63]. In brief, each genomic sample was used to construct two SMRTbell template libraries: a ∼500 bp randomly sheared insert library of native genomic DNA, and a whole-genome-amplified (WGA) library of the same insert size to remove any existing base modifications in the genomic DNA. The WGA sample served as a control. SMRT sequencing was performed using C2 chemistry. At 2–4 SMRTCells each, all samples achieved ∼500× average sequencing coverage across the genome.
The principle of base modification detection using SMRT sequencing by synthesis was detailed in previous publications [31], [32]. The technique relies on the sensitivity of the polymerase kinetics to the DNA template structure as DNA synthesis is recorded in real time. It was observed that the time between base incorporations, or interpulse duration (IPD), is on average longer when the nucleotide incorporation occurs opposite of a methylated base in the DNA template, as compared to an incorporation opposite of a canonical base.
In previous studies, the analysis involved computing the ratio of the mean IPD of the native sample to the mean IPD of the WGA control sample for every reference template position, and setting a threshold to call certain template positions as methylated. The data analysis implemented here uses a t-test with a log-normal distribution model for the IPDs and associated Pvalue at every position for identifying the methylated sites. The null hypothesis in this analysis is that the IPDs from the native and WGA samples are part of the same population, and the alternate hypothesis is that the native set of IPDs stems from a population with larger IPDs, namely from incorporations opposite of a methylated rather than canonical template base. A threshold value of 100 for the log-transformed Pvalue from the t-test (called Qmod = −10log(Pvalue)) at each reference position was used for assigning the given position as methylated. The value of 100 was chosen based on the Qmod distribution observed in the data, where there was a clear bimodal distribution arising from unmodified background and modified positions. Furthermore, a Qmod≥100 corresponds to better than the Bonferroni corrected Pvalue of 0.0001 for the 816 kb genome.
To detect relative changes of the methylation status between samples grown for different time periods, the two native samples were directly compared against each other, rather than against a WGA control sample, thus highlighting the methylome difference between those samples. This analysis is performed after whole methylome analysis of the genome of interest. Hence, all sites of the discovered motifs were used as the n independent test sites giving a Bonferroni corrected Pvalue of better than 0.01 (0.0067) at Qmod≥60. Plots were made using Circos [64].
Both modes of analysis were carried out using SMRT Portal (http://www.smrtcommunity.com/SMRT-Analysis/Software/SMRT-Portal), while sequence motif cluster analysis was done using Pacific Biosciences's Motif Finder (http://www.smrtcommunity.com/CodeShare_Project?id=a1q70000000GtatAAC). Data sets containing kinetic values for each reference position and DNA strand are available at http://www.pacbiodevnet.com/Share/Datasets/Senar-et-al.
M. pneumoniae mpn107 gene was obtained by PCR using genomic DNA as template and specific primers (Table S9a). 5′-end oligonucleotides incorporated a PstI site followed by the sequence 5′-TTAAGG-3′ (to terminate translation of the lac α-peptide reading frame of the pRSS plasmid vector and to reinitiate translation of the cloned methyltransferse (MTase) genes, followed by an eight nucleotide spacer sequence 5′-TTAATCAT-3′ and sequences complementary to the 5′-end of the relevant MTase coding sequence. 3′-end oligonucleotides were complementary to the 3′-end of the MTase coding sequences, including translation termination codons and a BamHI restriction site. Since the TGA codon encodes tryptophan in Mycoplasma but an opal stop codon in E. coli, the mpn198 and mpn108 genes having several opal codons were codon-transformed and synthesized by GeneScript. After PCR amplification, the different genes were cloned into a PstI-BamHI digested pRSS vector. The resulting vectors were termed pRSS107, pRSS198, and pRSS108 (Table S9b).
The vectors described above were used to transform the E. coli deficient in methyltransferases ER2796 strain (kindly provided by R. Roberts, NEB). The plasmid DNA of every transformed strain was analyzed by SMRT sequencing as described previously [32].
Transcriptional start sites of the M.pneumoniae transcriptome have been described recently [42]. This information was used to define the 5′-UTR (RNA sequences from transcriptional start site to translational start codon). Transcription levels of M. pneumoniae genes at 6 h and 96 h were previously determined by tiling and ultrasequencing [43]. These data were used to study the relation between methylation and transcription in M. pneumoniae (Table S1).
|
10.1371/journal.pntd.0005917 | Local selection in the presence of high levels of gene flow: Evidence of heterogeneous insecticide selection pressure across Ugandan Culex quinquefasciatus populations | Culex quinquefasciatus collected in Uganda, where no vector control interventions directly targeting this species have been conducted, was used as a model to determine if it is possible to detect heterogeneities in selection pressure driven by insecticide application targeting other insect species.
Population genetic structure was assessed through microsatellite analysis, and the impact of insecticide pressure by genotyping two target-site mutations, Vgsc-1014F of the voltage-gated sodium channel target of pyrethroid and DDT insecticides, and Ace1-119S of the acetylcholinesterase gene, target of carbamate and organophosphate insecticides. No significant differences in genetic diversity were observed among populations by microsatellite markers with HE ranging from 0.597 to 0.612 and low, but significant, genetic differentiation among populations (FST = 0.019, P = 0.001). By contrast, the insecticide-resistance markers display heterogeneous allelic distributions with significant differences detected between Central Ugandan (urban) populations relative to Eastern and Southwestern (rural) populations. In the central region, a frequency of 62% for Vgsc-1014F, and 32% for the Ace1-119S resistant allele were observed. Conversely, in both Eastern and Southwestern regions the Vgsc-1014F alleles were close to fixation, whilst Ace1-119S allele frequency was 12% (although frequencies may be underestimated due to copy number variation at both loci).
Taken together, the microsatellite and both insecticide resistance target-site markers provide evidence that in the face of intense gene flow among populations, disjunction in resistance frequencies arise due to intense local selection pressures despite an absence of insecticidal control interventions targeting Culex.
| Culex quinquefasciatus is a primary vector of arboviruses such as West Nile virus (WNV) and St. Louis encephalitis virus (SLEV) in temperate regions, and in many tropical/sub-tropical regions is implicated in the transmission of lymphatic filariasis (LF). Insecticide-based approaches are one of the most important interventions to mitigate disease burden; nevertheless, increased resistance of vectors to insecticides poses a challenge for sustainability and effectiveness of both current and future vector control interventions. Here, Cx. quinquefasciatus collected across a transect from Eastern to South Western Uganda were utilized to infer the likely pattern of the evolution of insecticide resistance among populations through a combination of putative selective loci (the target-site mutations Vgsc-1014F and Ace1-119S) and 26 neutral microsatellite markers. Taken together, the results yielded by these markers provide evidence that Ugandan Cx. quinquefasciatus mosquitoes are under heterogeneous selection pressure imposed by insecticides from distinct classes.
| Contemporary evolution of insecticide resistance in mosquitoes, which threatens the effectiveness of insecticide-based approaches to reduce the burden of vector-borne disease in endemic regions, results from a complex interaction of biological traits (e.g. genetic mechanisms of resistance) and operational implementation of control programs (e.g. choice of insecticide class and selection strength see [1, 2]).
Target-site mutations in either the acetylcholinesterase gene (Ace-1) or the voltage-gated sodium channel (Vgsc) gene, as well as metabolic resistance mediated by upregulation of, or mutations in, detoxification gene families such as the cytochrome P450 (P450s), esterases and glutathione S-transferases (GSTs) are the most well studied genetic mechanisms underpinning resistance [3–5]. Pyrethroids or DDT bind to the VGSC neuronal ion channel resulting in the channel pore being unable to close, so causing repetitive nerve firing, paralysis and death [6]. A number of resistance mutations have been identified in this gene in a range of insect species with an L to F change at codon 1014 frequently identified [7]. By contrast, carbamate and organophosphate insecticides target the enzyme acetylcholinesterase (AChE) at cholinergic synapses, blocking transmission of nerve impulses by inhibition of AChE. A G to S mutation at codon 119 (Torpedo numbering system) of the Ace-1 gene results in resistance and is widespread in organophosphate and carbamate selected populations [8, 9]. In Culex and other vectors and pest insects the intense use of insecticides has been associated with sharp increases of Vgsc-1014F –also called kdr (knock-down resistance) [10–12] and an increased frequency of the Ace1-119S resistant allele.
Despite the recognized role of such target-site mutations in the adaptability of mosquitoes under insecticide selection, many studies have also shown that such resistant-associated mutations can have a range of deleterious effects and fitness disadvantages (see [13]). In Culex and other resistant insect populations where such fitness detriments are seen, the directional selection of adaptive target-site mutations has been linked with an increase in gene copy number. The most well documented example of this is for the duplication of Ace-1, which has been reported in diverse arthropod species such as Aphis gossypii, Tetranychus evansi, Anopheles gambiae as well as in Culex from a range of geographic regions [14–17]. Many studies have proposed that duplication of Ace-1 acts as a compensatory adaptive mechanism to restore the disrupted function of the nervous system driven by the resistant allele so reducing the fitness costs [18–20]. Additionally, duplication of the Vgsc gene was also recently reported in both Cx. quinquefasciatus and Aedes aegypti, indicating a likely adaptive role of gene duplication in the evolution of resistance to pyrethroids [21, 22].
In addition to the underlying genomic basis of resistance, the evolutionary pattern of insecticide resistance can also be influenced by the heterogeneity of environmental selection pressures encountered by mosquito populations. For example, mosquitoes from urban and rural settlements could experience contrasting patterns of selection imposed by direct or indirect vector control interventions, and/or the indirect impact of pesticides and environmental pollution encountered by vectors in rural settings [23]. From a vector control perspective, developing selective control approaches targeting only the primary species of interest is challenging since in most endemic regions, species with roles in vector-borne disease transmission can be co-endemic, heightening the risk of insecticide selective pressure on non-target vector species.
On the African continent, lymphatic filariasis (LF) is predominantly transmitted by Anopheles mosquitoes in rural areas, while in urban and coastal areas of East Africa Culex species are the main vector [24]. In Uganda, a country with high rates of malaria transmission and LF prevalence, Anopheles mosquitoes are incriminated as the primary vector of both diseases [25, 26]. Therefore, vector control operations using insecticide-based approaches, either through insecticide treated nets (ITN) or indoor residual spraying (IRS), chiefly target Anopheles. Whilst there is no state vector control programme targeting Culex populations across Uganda, insecticide resistance may be indirectly selected through the use of insecticides for the control of Anopheles, other vectors or pest species [23, 27].
In this study, we used collections of Cx. quinquefasciatus from Uganda as a model to determine if it is possible to infer heterogeneities in insecticide selection on non-target species through combined study of putatively selected resistant alleles (Vgsc-1014F and Ace1-119S) and neutral markers (microsatellites). To this end, microsatellite markers were developed and utilized to detect likely influence of geographic and colonization events on Cx. quinquefasciatus population diversity and structure [28, 29]. By contrast, changes in the pattern of resistant alleles (Vgsc-1014F and Ace1-119S) was used a proxy of the strength of insecticidal selection across Uganda.
Ethical approval was obtained from the Makerere University School of Medicine Research and Ethics Committee. This was an experimental setting for research purposes and was not linked to a malaria control program. Adult household representatives gave written informed consent with the Ethics Statement before entry into their houses for mosquito aspiration. This study used solely mosquito samples, while no human participants were involved.
Cx. quinquefasciatus mosquitoes were collected across a transect from Eastern to Southwestern Uganda, with collection points located in four districts (Fig 1A, S1 and S2 Tables). Collection points were located in Tororo and Kanungu districts (located in the Eastern and Western administrative regions) with populations of 517,000 and 252,000 people respectively and where more than 80% of dwellings are rural, and in Jinja and Kampala (Central region) with populations of 76,000 and 1.5 million people respectively [30, 31]. Samples were from a peri-urban area in Jinja (63% of houses in rural areas) and in the urban centre of Kampala.
From each collection site adult male and female mosquitoes were collected by aspiration from inside 10–15 houses (up to 10 mosquitoes per house) between July and August 2012. Then, a sub-set of the samples (S2 Table) were used for microsatellite and target-site mutation genotyping as described below.
Collected mosquitoes are representative of three Ugandan Demographic and Health Survey (UDHS) regions; Central, South-western and Mid-eastern with similar levels of insecticide-treated net (ITNs) coverage i.e. percentage of householders with at least one ITN ranging from 56 to 58.6% reported in 2011 [30] (S1 Table). However, the insecticide usage pattern in Uganda has changed rapidly in recent years, therefore the data provided (S1 Table) reflects very recent use and it does not absolutely reflect the last 5–10 years when resistance may have developed [32, 33]. Therefore, the ITN and IRS figures from 2011 for each UDHS are used herein as a proxy of likely selection strength among the collection sites, although they may not represent precisely the sources and strength of selection in each location. Indeed, recently Abeku et al. [34] have demonstrated that the effect of ITN usage on Kdr allele frequency is impacted by annual rainfall.
Across Uganda, two main pyrethroid based ITNs have been distributed by the government; Permanet 2.0 (polyester coated with deltamethrin) and Olyset Nets (polyethylene with permethrin incorporated). Additionally, a distinct level of ITN usage and Indoor Residual Spraying (IRS) is reported throughout the UDHS regions from where samples were collected (S1 Table). While major IRS programmes have not been reported in Jinja or Tororo, in Kanungu, approximately 45,000 households covering a population of 190,000 were sprayed with λ-cyhalothrin in 2007 as part of the malaria control programme [35]. In Kampala, a programme of environmental management was undertaken in the early 2000s (http://health.go.ug/programs/national-malaria-control-program) but no sustained insecticidal vector control has been used.
Genomic DNA from individual mosquitoes was isolated using a DNeasy kit (Qiagen) following the manufacturer’s recommendations. All samples were confirmed as Cx. quinquefasciatus by a diagnostic PCR assay which types diagnostic length polymorphisms in the second intron of the acetylcholinesterase-2 gene to differentiate Cx. quinquefasciatus from Cx. pipiens [36].
To facilitate design of target-site mutation genotyping assays, partial genomic fragments spanning the location of the G119S mutation in the Ace-1 gene and the L1014F mutation in the Vgsc gene were amplified from genomic DNA isolated from mosquitoes collected from the four populations. A TaqMan assay targeting the Ace-1 gene was designed to genotype the SNP (GGC/AGC) in the first base of the codon at position 119, while a Vgsc-pyrosequencing assay to type the L1014F mutation in exon 20 [6] of the Vgsc gene was developed to detect two synonymous resistant alleles (TTC and TTT) and the wild-type allele (TTA) (S1 Methods). We chose the pyrosequencing method for genotyping Vgsc-L1014F instead of applying the widely used Taqman assay due to the existence of three alleles at this codon, which limits the use of Taqman for genotyping in a single reaction. Further details for PCR amplification of the partial fragment of both genes, as well as the target-site assay primer and probe sequences and reaction conditions are provided in the S1 Methods.
In the studied populations, genotyping of both Ace-1 and the Vgsc indicated the presence of gene duplication. Such gene duplications hinder calculation of many population-genetic statistics such as allele frequencies, linkage disequilibrium (LD) and population differentiation especially when there is polymorphism for copy number. Indeed, the absence of straightforward genotyping and data analysis methods limited our ability to address further the implications of the gene duplication in the evolution of resistance, with similar drawbacks also reported in other studies [15, 37]. Due to the limitations of the genotyping methods (e.g. it is not possible to differentiate the genotypes RRSS from RRRS), our analysis is based solely on the presence/absence of resistant and susceptible alleles as applied to a single copy gene. This assumption could result in an underestimation of insecticide associated-resistant alleles within the studied populations whilst still allowing us to infer a proxy of increased resistance allele frequency in a population under selection pressure.
The absence of Ace-1 homozygous resistant genotypes (AGC/AGC) observed through Taqman (see Results) indicated a likely gene duplication of Ace-1 in all four populations studied. To investigate the possible presence of multiple gene copies, a partial fragment of Ace-1 was PCR amplified from 12 individuals from Kampala. After cloning (S1 Methods), between six and eight colonies from each individual were sequenced in order to detect the presence of >2 alleles, indicative of gene duplication [15].
Sequences from each individual were aligned in CodonCode Aligner software version 4.2.2 with ClustalW [38] and visualized using Jalview [39]. MEGA 5.1 [40] was used to analyse haplotype variability by calculating the number of polymorphic sites and nucleotide diversity (π). Frequencies and relationships between haplotypes were visualised by a minimum spanning network tree generated using the program PopArt available at http://popart.otago.ac.nz.
To date few microsatellite markers have been described for Cx quinquefasciatus [41–44], limiting the scope of population genetic studies. Herein, the microsatellite genotyping was conducted using a novel panel of 30 microsatellites loci developed for this study (S2 Methods), which have been combined into five multiplex reactions, thereby enhancing the throughput screening for genetic diversity in Cx. quinquefasciatus.
The newly designed microsatellite markers were isolated by scanning 180 Cx. quinquefasciatus supercontigs downloaded from VectorBase [45] with SciRoKo [46]. Individual mosquitoes were typed using five six-plex PCR reactions followed by genotyping using a Beckman-Coulter CEQ8000 capillary electrophoresis system with a 400 size standard kit. Genotypes were sized using the Beckman-Coulter CEQ 2000 DNA analysis system software and manually verified. Microsatellite genotype data were analysed with the program Micro-Checker [47] to detect possible scoring errors (stutter peaks and allele drop-out) and null alleles.
Genotyping of two target-site mutations, Ace1-119S and Vgsc-1014F, was conducted utilizing TaqMan allelic discrimination and pyrosequencing assays, respectively (S1 Results). The Ace-1 resistant allele (AGC) was observed in all populations with a frequency ranging from 12–32% (Fig 1A). Despite the relatively high frequency of the 119S allele, we observed a complete absence of homozygous resistant mosquitoes (S1 Fig). The frequency of heterozygotes was almost three times higher in Kampala and Jinja compared to the other two populations (Fig 1B). Among the four populations a moderate He (0.323 ± 0.065) was observed with a marked excess of heterozygotes (FIS = -0.362), whilst within populations, genotype frequencies were not significantly different from Hardy-Weinberg expectation for both the Kanungu and Tororo populations (Table 1). Additionally, no significant difference in resistance allele frequencies was detected between Jinja and Kampala (P = 1.0) and between Kanungu and Tororo (P = 1.0).
The Vgsc-1014F locus in Cx. quinquefasciatus has two alternative non-synonymous substitutions (TTT and TTC) at the third position of the codon (L1014F) [60]. For the Vgsc-1014F mutations we detected a high frequency of resistant alleles ranging from 62–100% (Fig 1A). The pyrosequencing genotyping identified 11 individuals harbouring all three kdr alleles instead of the two expected for diploid organisms (S3A Fig). Mosquitoes with tri-allelic genotypes were observed in all populations with the exception of Kanungu, with a frequency of 4.0–10.8% (S3B Fig). Excluding individuals with this tri-allelic pattern to allow the use of standard methods to infer departures from Hardy-Weinberg, our analysis detected significant deviation from HWE only in mosquitoes from Tororo (Table 1).
Significant differences in Vgsc resistance allele frequencies were detected between all populations with the exception of the Jinja and Kampala comparison (P = 0.63). The TTC resistant allele was observed at a high frequency (41–69%) in all populations with the exception of Kanungu, where a higher frequency of the TTT resistant allele (55%) was detected, coupled with an absence of the wild-type allele (Fig 1A). In all populations, we observed a low frequency of susceptible homozygotes (ranging from 0 to 14%), while the majority of resistant alleles were observed in heterozygotes (Fig 1B).
A 535 bp fragment of the Ace-1 gene was sequenced from 12 individuals (N = 57 sequences with between six and eight colonies sequenced from each individual; GenBank accession numbers: KT591708-KT591765) with nine haplotypes detected displaying a total of 10 polymorphic sites (S3 Table) and haplotype frequencies ranging from 0.017–0.397. The resistant allele was detected in four haplotypes: A, E, H and I accounting for 40% of total haplotypes identified. The minimum spanning network tree (S3C Fig) shows that three resistance bearing haplotypes (A, H, I) differ from each other by a single mutational step with Haplotype A the most common. The remaining 119S haplotype (E) is equally distant from both haplotypes A and B differing by three mutational steps, while it differs by only two mutational steps from haplotype D. For wildtype Ace1-119G, three haplotypes B, C and G were detected with similar frequencies whereas haplotypes D and F were the least frequent. From 12 individuals studied through cloning and sequencing we identified three individuals with more than two distinct Ace-1 haplotypes, indicative of duplication. These three mosquitoes displayed from three to five haplotypes in the neighbor-joining dendrogram (S3D Fig).
Resistance alleles at the target sites of insecticides have known deleterious effects [13, 62] and are therefore not expected to persist in the absence of selection. Our data which indicate copy number variation at both the Vgsc-L1014F and Ace1-G119S loci and allele frequency differences between populations for both loci suggest differential insecticide selection pressure on Cx. quinquefasciatus populations across Uganda. The significantly higher frequency of Vgsc-1014F in comparison to Ace1-119S for all populations has been reported previously in Culex and other vector mosquitoes on the African continent [63–65], and might result from past and ongoing reliance on pyrethroid-based vector control approaches to mitigate the burden of vector-borne diseases. However, it is important to bear in mind that our allele frequency estimates are likely under-reporting true frequencies, particularly for the Vgsc, as the duplication observed in Ugandan populations imposes limitations on available methods to infer allele frequency as reported elsewhere [37].
Whilst recognising these limitations, our data show a contrasting pattern of Ace1-119S and Vgsc-1014F allele frequencies across populations, likely indicating heterogeneous selection pressure. For both markers, we were not able to identify a clear spatial geographic spread, but we did detect heterogeneity in allele frequencies of these selected markers when comparing Uganda Central to Eastern and Southwest population. Indeed, the highest frequency of Vgsc-1014F was detected outside the Central region, while the highest frequency of Ace1-119S was observed with a pattern of geographic distribution opposite to that of Vgsc-1014F. Therefore, these figures highlight the challenges for implementation of efficient control intervention in highly fragmented habitats including overlapping of urban and rural patches or due to heterogeneous urbanization structure, which could result in a mosaic of landscapes with sharp shifts of insecticide selection pressure and frequency of insecticide-resistance associated alleles [66–68].
Markedly, although our study demonstrates high frequencies of both target-site mutations (Vgsc-1014F and Ace1-119S) in Ugandan Cx. quinquefasciatus populations no vector control programme targeting this species has been conducted. Therefore, these results support a hypothesis that the evolution of insecticide resistance in these populations is driven by control interventions targeting other vector species or pest insects. Indeed, previous studies have suggested that the evolution of insecticide resistance in Cx. quinquefasciatus on the African continent might result from intervention approaches targeting sympatric Anopheles species [27, 69]. For instance, a study of Cx. quinquefasciatus from Zambia showed a significant increase of the Vgsc-1014F mutation and increase of detoxification enzyme activity assays after introduction of ITNs for malaria control [27].
In Uganda in particular, we suggest that variation of Vgsc-1014F frequency between districts might reflect the combination of approaches applied to reduce Anopheles populations, including national distribution of ITN and IRS only in highly endemic and epidemic-prone regions. For example, across Uganda irregular spraying with insecticides such as lambda-cyhalothrin, DDT, alpha cypermethrin and latterly bendiocarb [70] has been reported since 2006 when IRS was re-introduced for vector control [71]. Regional differences in ITN coverage may also contribute to the Vgsc-1014 allele distribution e.g. the Uganda Bureau of Statistics [72] report that householders that own at least one ITN differ by 44% between the East Central and West Nile regions in 2011.
Markedly, we have also noticed a distinct pattern of ITN usage among UDHS, which differ by 14% between Central-South western and 8.7% Central-Mid-eastern, despite a similar ITN coverage of around 58% for all three regions (S1 Table). Additionally, the contrasting pattern of Vgsc-1014 markers among the studied districts might also reflect the drastic shift of ITN distribution, which increased by 26% (Central region), 41.6% (Mid-eastern) and 32.8% (South-western) between 2006 and 2009 [32, 33].
Nevertheless, with our study design it is not possible make a direct association of the UNMCP (Uganda National Malaria Control Program) interventions with the evolution of resistance in Culex populations. Indeed, we recognise that pinpointing the particular source underlying the evolution of resistance remains a difficult task. For example, increased resistance in Cx. quinquefasciatus populations from Ghana and Benin, with high frequencies of Ace1-119S was attributed to the use of organophosphates and/or carbamate insecticides by farmers for pest control or pollutants in mosquito breeding sites [66, 67]. Heavy use of domestic insecticide was also associated with increased insecticide resistance in Cx. quinquefasciatus populations from Ghana [73].
Regardless of the strength of selection imposed by diverse insecticide-based approaches such as ITNs, IRS, pesticides and environment pollutants [23], the evolutionary pattern of increased resistance in Cx. quinquefasciatus from Uganda could also be modulated by variation in the repertoire of adaptive genetic mechanisms, as in the studied population we identified gene duplication for both target-site loci and allelic variation of the Vgsc resistant allele. Interestingly, we identify two alternative resistant alleles (TTT and TTC) in the Vsgc-1014 codon observed at high frequency in all populations, contrasting with previous reports on the geographic distribution of Vgsc-1014F in Culex mosquitoes worldwide [11], which exclusively detect the TTT allele [74–76]. To date, the TTC variant has been detected at high frequency only in Cx. quinquefasciatus from Sri Lanka, which was previously argued as the most likely geographic origin [60], and was only recently detected at very low frequency in African Cx. quinquefasciatus from Zanzibar [69]. Although we identified co-occurrence of both resistant variants of Vgsc it was not possible to determine if heterozygote TTT/TTC increases a mosquito’s level of resistance, or offsets possible deleterious effects.
We also identify a significant difference in frequency of the Vgsc-1014F resistant alleles among populations, with the TTC variant being predominant in all populations with the exception of Kanungu. We do not have a conclusive explanation for the difference in frequency of TTT and TTC among Ugandan populations and between Uganda and other African populations. Nevertheless, a few potential mechanisms could explain this. One possible explanation involves a colonization process with the TTC allele introduced to Uganda from another geographical region with a large founder effect followed by genetic drift that shifted the TTT and TTC allele frequency, although a de novo origin and dispersion cannot be discounted. Alternatively, it is also possible that the TTC allele arose on the African continent but has not been detected in other populations due to the application of specific Culex species Vgsc-1014 genotyping assays, which were mostly designed for detecting the TTT and TTA alleles [77].
Our target-site genotyping of the Ace1-119 and Vgsc-1014 variants also indicated the presence of gene duplication for both genes, suggesting that CNV could be among the mechanisms driving the evolution of insecticide in these populations. Indeed, gene duplication of insecticide-associated genes such as Ace-1, esterase estα21 and estβ21, Rdl (resistance to dieldrin) and detoxification genes (e.g. GSTs, P450s) have been reported as adaptive resistance mechanisms in diverse arthropod species including Tetranychus urticae, Aphis gossypii and A. gambiae [3, 14, 16, 37, 78].
The likely duplication of the Vgsc was indicated by the presence of 6.25% of mosquitoes genotyped by pyrosequencing displaying three alleles simultaneously for Vgsc-1014. For Ace-1, duplication was evidenced by a haplotype analysis, which identified three out of 12 individuals exhibiting either 3 or 5 distinct haplotypes simultaneously. Whilst contamination or DNA from spermatheca could be mistaken for duplications we have shown that the duplication of Vgsc is present in males [79] and we see no triallelic genotypes in the microsatellite data. This duplication could explain the large variation observed within resistant and susceptible alleles as demonstrated by the minimum spanning Network (S3C Fig), which contrasted with an expected low haplotype diversity due to a selective sweep effect on a locus under selection.
The absence of Ace-119S homozygotes in all four populations, was not surprising as the same pattern has been reported previously in both Culex and Anopheles populations [16, 20, 80]. In both species, this pattern has been linked to the deleterious effects of the homozygous resistant genotype due to changes in acetylcholinesterase kinetic properties and mutation fitness cost in an insecticide-free or reduced insecticide exposure environment. Alternatively, the absence of resistant homozygotes might also arise from duplication of the Ace-1 gene involving resistant and susceptible alleles, creating a permanent heterozygosis to partially normalise AChE enzyme activity levels [37, 81]. In contrast to the proposed compensatory adaptive mechanism of the Ace-1 duplication to overcome deleterious effects and fitness disadvantage of the resistant Ace1-119S [19, 20], the likely adaptive role played by the recently reported Vgsc duplication in Cx. quinquefasciatus from the USA (Xu et al. 2011) and in A. aegypti from Brazil [22] is currently unknown.
Our study also detected a contrasting pattern between microsatellite and insecticide selected markers. For both markers linked to resistance, it was not possible to identify a clear spatial geographic pattern, but we did detect heterogeneity in allelic frequencies when comparing Ugandan Central (largely urban) to Eastern and Southwest populations (largely rural). By contrast, no significant difference in genetic diversity indices was observed at neutral markers among the populations, although these data did indicate a structured pattern. Although the low levels of genetic variation of microsatellites across Ugandan populations from distinct geographic locations have also been reported in Cx. pipiens from California, USA [82], heterogeneity in diversity levels across Uganda was expected since vector control interventions have been applied annually, and might be expected to result in population bottlenecks. Such reductions of genetic diversity between populations under different strengths of selection pressure have been reported in Cx. quinquefasciatus from Recife, Brazil after vector control intervention with Bacillus sphaericus, a bio-larvicide [83]. As indicated by these authors, discrepancies between neutral and selected markers might be linked to differences in the marker’s genetic properties that modulate changes in allele frequencies, such as drastic bottleneck effects and differences in intensity of local selection with frequent gene flow [84, 85].
The microsatellite data also indicated significant genetic structure among populations with the three clusters identified by the STRUCTURE and DAPC analyses corresponding to Eastern, Central and Southwest Uganda regions. The level of population structure among our study sites was consistent with findings from similar studies in Culex mosquitoes using molecular markers as diverse as ISSR (Inter-Simple Sequence Repeat) and microsatellites, which reported that population structure reflects geoclimatic, geographic distance and environmental variables [86–88]. Additionally, based on microsatellite markers an intense migration among Ugandan populations with significant patterns of isolation by distance (IBD) was detected, which could prevent loss of genetic diversity and consequently could explain the homogenous microsatellite genetic diversity observed across the studied populations.
The distinct pattern of IBD observed for microsatellite markers and selected markers, also indicated that factors other than geographic distance, such as environmental heterogeneity as well as serial sequential founder effects, may also influence the pattern of spatial autocorrelation of resistance-associated markers [89]. For example, in Aedes rusticus, comparisons among areas treated and not-treated with Bacillus thuringiensis israelensis (Bti) show more intense IBD among treated than among non-treated sites [90]. These authors argue that the difference in observed IBD instead of spatial distribution could result from differences in population size after insecticide application or lower migration capability of selected mosquitoes imposed by a fitness cost of the resistance mechanism.
Therefore, despite the intense gene flow identified in the studied populations, it is possible that local differences in strength of insecticide selection might have created heterogeneous “insecticide adaptive islands” across the geographic regions. Analysis of microsatellite locus neutrality indicated that all markers indeed adhered to a neutral model with the exception of the MCQ21 across all populations and MCQ11 when comparing only Ugandan Central to Eastern populations. For both markers, further analysis of the genomic location of markers does not indicate likely linkage with known insecticide associated genes such as Ace-1, Vgsc or genes encoding detoxification enzymes (e.g. esterase, GSTs and P450s). However, this analysis was limited by the incomplete Cx. quinquefasciatus genome assembly, which restricted the genomic window capable of being applied in the analysis. Due to the contrasting patterns of neutrality for marker MCQ21, we speculate that this marker could be hitchhiking with an insecticide-associated gene nearby this marker, with a genomic location beyond the supercontig resolution.
In conclusion, our study shows that the contemporary pattern of local selection is driven by the evolution of insecticide resistance in Ugandan Cx. quinquefasciatus occurring despite the absence of vector control intervention targeting this species. This study demonstrates that although it is challenging to pinpoint the direct source of insecticide exposure in a background of selection pressure, studies of the evolution of insecticide in non-target species can provide important information about the nature of the insecticidal challenges to which vectors are exposed.
|
10.1371/journal.ppat.1005415 | E2F/Rb Family Proteins Mediate Interferon Induced Repression of Adenovirus Immediate Early Transcription to Promote Persistent Viral Infection | Interferons (IFNs) are cytokines that have pleiotropic effects and play important roles in innate and adaptive immunity. IFNs have broad antiviral properties and function by different mechanisms. IFNs fail to inhibit wild-type Adenovirus (Ad) replication in established cancer cell lines. In this study, we analyzed the effects of IFNs on Ad replication in normal human cells. Our data demonstrate that both IFNα and IFNγ blocked wild-type Ad5 replication in primary human bronchial epithelial cells (NHBEC) and TERT-immortalized normal human diploid fibroblasts (HDF-TERT). IFNs inhibited the replication of divergent adenoviruses. The inhibition of Ad5 replication by IFNα and IFNγ is the consequence of repression of transcription of the E1A immediate early gene product. Both IFNα and IFNγ impede the association of the transactivator GABP with the E1A enhancer region during the early phase of infection. The repression of E1A expression by IFNs requires a conserved E2F binding site in the E1A enhancer, and IFNs increased the enrichment of the E2F-associated pocket proteins, Rb and p107, at the E1A enhancer in vivo. PD0332991 (Pabociclib), a specific CDK4/6 inhibitor, dephosphoryles pocket proteins to promote their interaction with E2Fs and inhibited wild-type Ad5 replication dependent on the conserved E2F binding site. Consistent with this result, expression of the small E1A oncoprotein, which abrogates E2F/pocket protein interactions, rescued Ad replication in the presence of IFNα or IFNγ. Finally, we established a persistent Ad infection model in vitro and demonstrated that IFNγ suppresses productive Ad replication in a manner dependent on the E2F binding site in the E1A enhancer. This is the first study that probes the molecular basis of persistent adenovirus infection and reveals a novel mechanism by which adenoviruses utilize IFN signaling to suppress lytic virus replication and to promote persistent infection.
| Interferons play important roles in both innate and adaptive immunity, and have broad antiviral properties. We demonstrate that type I (IFNα) and type II (IFNγ) IFNs inhibit the replication of divergent adenoviruses via an evolutionally conserved E2F binding site. IFNs augment the association of the tumor suppressors Rb and p107 with the E1A enhancer region in vivo to repress viral immediate early transcription. By comparing the properties of wild type and E2F site mutant viruses, we show that the IFN–E2F/Rb axis is critical for restriction of adenovirus replication to promote persistent viral infection. Relief of E2F/Rb repression counteracts IFN signaling whereas enforcement of E2F/Rb interaction mimics IFN signaling. These results reveal a novel mechanism by which adenoviruses utilize IFN signaling to suppress lytic virus replication and promote persistent infection.
| Interferons (IFNs) are widely expressed cytokines that have pleiotropic effects on cells. IFNs play important roles in both innate and adaptive immunity [1,2]. There are three types of IFNs: I, II and III. Type I IFNs (α, β, ε, κ and ω) are produced by multiple cell types following the activation of pathogen pattern recognitions receptors and function in both an autocrine and paracrine manner. Type II IFN (γ) is produced by T cells and natural killer cells, as well as other cells of the immune system. Type III IFNs (λs) play an important role in mucosal cell immunity. All three types of IFNs bind to cell surface receptors that activate Janus kinases to phosphoryate STAT (Signal Tranducer of Activated Transcription) proteins [1,2]. STAT proteins homo- and heterodimerize and induce the expression of numerous IFN-stimulated genes (ISGs) that have antimicrobial properties [3]. IFNs have broad antiviral properties and function by different mechanisms.
Adenoviruses (Ad) are ubiquitous pathogens infecting a wide range of vertebrates. Ad infection is generally associated with mild respiratory, ocular, and gastrointestinal diseases, but Ads have been recognized in recent years as significant pathogens in immunocompromised patients [4]. IFNs fail to inhibit wild-type Ad replication in established cancer cell lines [5–7]. The resistance of wild-type Ad to the effects of IFNs is due to multiple counteracting effects of viral gene products. The Ad E1A proteins block IFN signaling by binding STAT proteins and preventing the activation of interferon-stimulated gene factors 3 (ISGF3) complex by type I IFNs and IFNγ activation factor (GAF) complex by type II IFN [8]. The E1A proteins also bind and disrupt the hBre1 transcription complex and prevent IFN-induced histone H2B monoubiquitination and associated ISG expression [9,10]. Both actions of E1A lead to the global suppression of ISG expression. Analogously, the Ad E1B-55K protein inhibits the expression of cellular ISGs through its transcriptional repression domain [11,12]. Numerous studies have shown that promyelocytic leukemia nuclear bodies (PML-NB) play an important role in cellular intrinsic and IFN-induced antiviral immunity [13]. The Ad E4-ORF3 protein antagonizes the functions of PML-NB by disrupting these structures and sequestering antiviral components including PML and Daxx [7,14]. The Ad E1B-55K:E4-ORF6 ubiquitin ligase complex also targets Daxx for proteasome degradation [15]. Finally, Ad VA RNA-I inactivates PKR to prevent IFN-induced phosphorylation of the eIF2α translation factor which inhibits global protein translation during the late phase of viral infection [6].
Current models of interplay between Ad infection and IFN signaling have mostly been conducted in cancer cell lines. Such cells are coupled with abnormal signal transduction, unlimited proliferation, and evasion of apoptosis, and are compromised in many normal signaling pathways. Indeed, it has been shown that the Ad E1B-55K protein was able to inhibit a set of ISG expression in response to type I IFN signaling in primary human cells, which has not been reported in established cell lines [12]. Moreover, a recent study showed wild-type Ad exhibits an enhanced virus load in the organs of the STAT2-knockout Syrian hamsters compared to wild-type animals, revealing an important role of type I IFNs in controlling Ad replication in vivo [16]. To understand Ad pathogenesis in a natural context, this study focused on understanding the regulation of Ad replication by IFNs in normal human cells.
The E1A protein is the first Ad product expressed following infection and it is indispensible for virus growth [17]. In addition to antagonizing IFN signaling, the E1A proteins directly interact with a number of cellular proteins to regulate viral and cellular gene expression, and promote cell cycle progression [17]. E1A gene expression is regulated by an upstream enhancer region primarily via the activity of the cellular transcription factor GA-binding protein (GABP)[18–21]. As a tetrameric transcriptional complex, GABP (also known as E4TF-1 and NRF-2) is composed of two GABPα subunits which bind DNA, and two GABPβ subunits which transactivate gene expression [22,23]. Deletion of the two GABP binding sites in the E1A enhancer region dramatically decreases E1A expression [19,20]. A separate E1A enhancer segment, located between the GABP binding sites, regulates transcription from the entire Ad genome [21]. Finally, both GABP binding sites in the E1A enhancer are located adjacent to sequences that bind E2F transcription factors, although these E2F sites are dispensable for E1A expression [19,21].
In this study, we analyzed the effects of IFNs on Ad replication in normal human cells. Our data demonstrate that both IFN‹ and IFNγ blocked the replication of divergent adenoviruses in these cells. The inhibition of Ad5 replication by IFN‹ and IFNγ is the consequence of repression of transcription of the E1A immediate early gene product. The repression of E1A expression by IFNs is associated with the binding of E2F/Rb complexes to a conserved E2F site in the E1A enhancer. Finally, we established an Ad persistent infection model in vitro and demonstrated that IFNγ supresses productive Ad replication in a manner dependent on the conserved E2F binding site in the E1A enhancer. These results reveal a novel mechanism by which adenoviruses utilize IFN signaling to inhibit virus replication and to promote persistent infection.
To evaluate Ad replication in normal human cells during an IFN response, primary human bronchial epithelial cells (NHBEC) were used, in comparison to the established epithelial adenocarcinoma cell line A549 (Fig 1A and 1B). Strikingly, Ad5 replication in NHBEC was inhibited by IFNα and IFNγ 40-fold and 70-fold, respectively. Neither IFNα nor IFNγ was able to block replication of Ad5 in A549 cells. NHBEC only survive for a few passages in culture and vary from lot to lot. Therefore, we sought other primary human cells where IFNs inhibit Ad replication. We utilized a normal, non-transformed human diploid fibroblast cell line immortalized by human telomerase (HDF-TERT) that is permissive for Ad infection. Ad5 had an infectious particle/PFU ratio of ~1000:1 in HDF-TERT cells compared to ~20:1 in A549 cells due to reduced infectivity of HDF-TERT cells. We established the time line of the Ad5 life cycle in HDF-TERT cells. The majority of incoming viral genomes (detected by immunofluorescence using an antibody against the viral core protein VII) were still in the cytoplasm at 2 hr post-infection; complete entry of viral genomes into the nucleus required ~ 6 hr (S1A Fig). By 16 hr post-infection, the number of core protein VII foci had significantly declined indicating viral early gene transcription had occurred [24]. Analyses of Ad immediate early (E1A) and early mRNA levels (E1B, E2A, E2B and E4) showed an exponential increase in early gene expression from 6 to 24 hr post-infection (S1B Fig). The kinetics of viral DNA replication was determined (S1C Fig); viral DNA replication began during the 18–24 hr period and increased substantially thereafter with a 4-log total increase in genome copy number. Viral early gene (E4-ORF3) and intermediate gene (IVa2) expression was evident as DNA replication occurred and late gene products continued to accumulate through 144 hr post-infection (S1D and S1E Fig). There was no cytopathic effect observed up to 6 days post-infection even at a multiplicity of infection (MOI) of 1000 virus particles/cell where nearly all cells were infected.
To evaluate Ad5 replication in HDF-TERT cells in the presence or absence of IFNs, cells were incubated with IFNα, IFNγ or left untreated for 24 hr, followed by Ad5 infection at low MOI (25 virus particles/cell). Viral DNA replication was quantified at 48 hr post-infection (Fig 1C). Consistent with the results observed in NHBEC, both IFNα and IFNγ significantly inhibited virus replication (20- and 50-fold, respectively). The effect of IFNs on Ad replication was moderately diminished at higher MOIs but Ad replication was still inhibited in IFN-treated cells. IFNα and IFNγ reduced infectious virus production 2 and 3 logs, respectively (Fig 1D). IFNs did not block the entry of viral genomes into the nucleus and the majority of viral genomes did not colocalize with PML-NB with or without IFN treatment (Fig 1E and S1F Fig). We examined Ad5 early mRNA levels with or without IFN treatment. Ad immediate early (E1A) and early gene expression (E1B, E2A, E2B and E4) was significantly suppressed in IFN-treated cells compared to untreated cells (Fig 1F). Similar results were observed in NHBEC (Fig 1G). The reduction in viral mRNA levels correlated with decreased early, intermediate and late gene expression (Fig 1H).
IFNs may repress early gene expression directly or IFNs may inhibit viral DNA replication resulting in a corresponding decrease in early gene expression. To examine these possibilities, HDF-TERT cells were infected with a replication-defective mutant virus ΔTP-GFP. ΔTP-GFP contains a disruption in the coding sequences for Ad5 terminal protein, a protein that is absolutely essential for viral DNA replication [25]. Even though the relative copy number of ΔTP-GFP viral genomes in the presence and absence of IFNs did not increase, E1A mRNA levels were reduced by both IFNα and IFNγ throughout the time course of the experiment (Fig 2A and 2B). These results demonstrate that IFNs inhibit Ad early gene expression in a replication-independent manner. E1A is the first viral protein expressed after infection and it is essential for efficient activation of all Ad gene expression [17]. It is possible that IFNs suppress E1A transcription directly, resulting in the observed phenotype, rather than having a global effect on Ad early gene expression. To test this hypothesis, we generated an E1A-expressing HDF-TERT cell line (HDF-TERT-E1A); the major E1A isoforms were expressed in these cells (Fig 2C, top panel) at levels similar to that found in HDF-TERT cells infected with Ad5 (S2A Fig, lanes 2 and 3). Since E1A can inhibit IFN signaling [8], we examined IFN signaling in HDF-TERT-E1A cells (Fig 2C). E1A expression did not impede IFN signaling in these cells; increases in STAT1 protein levels and STAT1 phosphorylation as well as the induction of IGS54 and ISG15 expression were observed in HDF-TERT-E1A cells following IFNα and IFNγ treatment. Ad DNA replication and early gene expression were almost completely restored in HDF-TERT-E1A cells infected at low MOI and treated with IFNs (Fig 2D and 2E). The levels of all Ad early mRNAs were completely restored in HDF-TERT-E1A cells infected at high MOI and treated with IFNα or IFNγ (S2C Fig).
As a viral oncogene, E1A is able to transform cells [17]. Stable expression of E1A might lead to transformation of HDF-TERT losing the biological property of normal human cells. We evaluated Ad replication following IFN treatment when E1A was only transiently expressed. We generated an Ad expression vector (in340-Δ2-CMV) which contains the CMV promoter/enhancer in place of the E1A enhancer region and E1A/E1B coding sequences. The complete E1A coding region was inserted downstream of the CMV promoter. HDF-TERT cells were infected with the Ad-CMV-E1A virus or the control virus for 1 hr, followed by the addition of IFNα or IFNγ. Twenty four hr later, cells extracts were prepared and E1A expression was analyzed, or the cells were super-infected with Ad5 and Ad5 replication was assayed at 48 hr post-infection. Neither IFNα nor IFNγ reduced E1A expression in cells infected with the Ad-CMV-E1A virus (Fig 2G, left). Ad5 replication was significantly reduced in IFN-treated cells coinfected with the control Ad-CMV virus. Consistent with the results obtained using HDF-TERT-E1A cells, ectopic expression of E1A partially rescued this defect in IFN-treated cells (Fig 2G, right). As observed with HDF-TERT-E1A cells, transient expression of E1A did not impair IFN signaling in these assays (Fig 2F) even though the E1A proteins were significantly over-expressed compared to infection with Ad5 (S2A Fig, lanes 3 and 4). Collectively, these results demonstrate that both IFNα and IFNγ inhibit Ad replication by repressing E1A gene expression.
The cellular transcription factor GABP binds to two sites in the E1A enhancer region and GABP is the major regulator of E1A transcription [18,19,21]. These two sites have synergistic effects on E1A transcription; when both sites are deleted, E1A expression is dramatically diminished. Therefore, we examined the interaction of GABP with the E1A enhancer region in vivo by ChIP. At 18 hr post-infection, prior to the onset of viral DNA replication (S1C Fig), IFNα or IFNγ decreased the association of GABP with the E1A enhancer in vivo 3- and 2.5-fold, respectively (Fig 3A). Expression of both GABPα and GABP β in IFN-treated cells was the same as that found in untreated cells (Fig 3C). In HDF-TERT cells, GABPα and β were primarily nuclear localized and their localization were not altered by either IFNα or IFNγ (Fig 3D). Finally, we performed coimmunoprecipitation assays to determine if IFNs impaired endogenous GABPα:GABP β interaction. These results did not reveal any differences in IFN-treated cells (Fig 3E).
The cellular transcription factor Sp1 binds to several sites in the Ad5 inverted terminal repeat (ITR)[26] that are adjacent to the E1A enhancer region. We analyzed if Sp1 binding to the ITR was altered by IFN signaling in vivo. Sp1 bound to the Ad5 ITR at similar levels in the presence and absence of IFNs (Fig 3B). This result suggests that reduced binding of GABP to the E1A enhancer following IFN treatment is target specific and not due to reduced global accessibility of the left-end of the Ad5 genome.
We hypothesized that IFNs may induce the binding of a transcriptional repressor to the E1A promoter/enhancer region. We examined the effect of IFNs on E1A expression using a panel of existing [20,21,27] and newly created E1A enhancer region mutants (Fig 4A and 4B). E1A expression of deletion mutant in340-B1, but not of the adjacent deletion mutant in340-A5, was completely restored in IFNγ-treated cells, and partially restored in IFNα-treated cells (Fig 4C). This finding indicated that an IFN-induced repressor binding site is located in the downstream half of the E1A enhancer region, corresponding to Ad5 nt 270–358. We also noted that basal E1A expression with mutant in340-B1 was significantly augmented compared to the parent virus in340. We next screened a set of mutants carrying smaller deletions within the in340-B1 interval of the E1A enhancer. Interestingly, dl309-21 (Δ271–301) and dl309-Δ273/317 exhibited significantly increased basal E1A protein levels and were refractory to IFN-mediated repression. In contrast, E1A expression with the adjacent mutants dl309-3 (Δ288–336) and dl309-Δ317/358 was still fully repressed by IFN treatment (Fig 4C). We conclude that Ad5 sequences located within nt 271–288 are the target of IFN-mediated repression of E1A expression.
Mutations were introduced into the core of the nt 271–288 interval (Ad5-mut1, Ad5-mut2) and in the adjacent GABP binding site (Ad5-mut3)(Fig 4B). E2F/Rb family proteins bind to sequences positioned at nt 280 (TTTCGCGGGAAA) in vitro [18,28]. Ad5-mut1 disrupts this entire sequence and Ad5-mut2 disrupts the core of the E2F binding site (CGCG). Ad5-mut1 and Ad5-mut2 exhibited efficient E1A expression in the presence of IFNs compared with Ad5-WT (Fig 4C). Basal E1A expression was increased with both mutants minus IFN, and this reflected increased levels of E1A mRNAs in infected cells (S2B Fig). Efficient E1A expression with Ad5-mut1 in the presence of IFNs also was observed in NHBEC (Fig 4D). Ad5-mut3 had normal basal expression levels and E1A expression was fully repressed by IFNs. The compelling insensitivity of Ad5-mut2 to IFNα and IFNγ strongly suggests that E2F/Rb family protein binding to the IFN-induced repressor site mediates suppression of E1A expression.
DNA sequence alignment of the E1A enhancer regions of divergent Ad serotypes revealed that the E2F binding site located between nt 270–290 is highly conserved (Fig 5A). Thus, we hypothesized that IFNs would repress viral replication of evolutionarily divergent Ads. IFN-treated and untreated HDF-TERT cells were infected with Ad3, Ad4, Ad5, Ad9, and Ad12 (subgroups B, E, C, D and A, respectively) at MOIs that resulted in similar levels of infection. DNA replication of all of these Ad serotypes was inhibited by IFNα and IFNγ treatment (Fig 5B).
The E2F/Rb pathway is disrupted in nearly all human cancers [29] and all previous analyses of the E1A enhancer region were conducted using nuclear extracts prepared from established cancer cell lines [18,28]. Thus, we examined E2F binding to the E1A repressor element using HDF-TERT cells. Nuclear extract was prepared from HDF-TERT cells and analyzed by EMSA using a radiolabeled probe corresponding to the IFN-induced E1A repressor site (Ad nt 271 to 288)(Fig 6A). Multiple DNA-protein complexes were observed that were identified using specific antibodies to E2F/Rb family proteins. The fastest migrating complex contained E2F-4 and DP-1 while slower migrating complexes contained E2F3/DP1/Rb and E2F1/DP1/Rb (lanes 2–4, 8, 12 and 13). The complexes with the slowest migration contained E2F-4/DP1 with p107 or p130 (lanes 5–8). The specificity of each antibody was ensured by competition using the corresponding peptides used to generate the antibodies. Identical complexes were observed using nuclear extracts from IFN-treated HDF-TERT cells (Fig 6A, lanes 14–16).
We performed ChIP experiments to evaluate the enrichment of Rb family proteins at the E1A enhancer early after infection with or without IFN treatment (Fig 6B and 6C). These results showed a 2-fold increase in Rb binding following IFNα treatment with no significant effect of IFNγ, and a 5-fold increase in p107 binding following IFNγ treatment with no significant effect of IFNα. The binding of Rb and p107 to the E1A enhancer of Ad5-mut1 was reduced 4-fold compared to wild-type Ad5 without IFN treatment, and no significant increase in binding of either of these proteins to Ad5-mut1 was observed following IFN treatment (Fig 6B and 6C). These results demonstrate that IFN treatment induces the binding of E2F/Rb complexes to the E1A enhancer. The reduction in Rb and p107 binding to near background levels with Ad5-mut1 suggests that E2F family proteins primarily associate with the downstream E2F site centered at the IFN-induced repressor site rather that at the upstream E2F binding site. This conclusion is consistent with the E1A enhancer mutant screen where a mutant in the upstream E2F binding site showed repression of E1A expression following IFN treatment (Fig 4, dl309-A5).
PD0332991 (Palbociclib) is a specific CDK4/6 inhibitor and can cause cell cycle arrest through dephosphorylation of Rb family proteins [30]. To avoid changes in Rb family protein phosphorylation by contact inhibition, low density HDF-TERT cells were incubated with PD0332991 and Rb, p107 and p130 phosphorylation was evaluated. Rb, p107 and p130 phosphorylation was reduced within 6 hr of PD0332991 addition (Fig 6D, left, 0 hpi) although growth arrest was not evident until later times (Fig 6E). PD0332991 inhibited Ad5-WT replication whereas Ad5-mut1 replication was increased (Fig 6F); these results correlated with E1A and E2A DNA binding protein (DBP) expression levels (Fig 6D, right). Thus, PD0332991 treatment mimicked the effects of IFNs. It is well established that the E1A 12S protein can directly interact with Rb family proteins, leading to their dissociation from E2Fs and activation of E2F target gene expression [17]. Ectopic expression of the E1A12S protein partially rescued Ad5 replication in the presence of IFNs (Fig 6H). IFN treatment did not reduce E1A 12S protein expression with the vector used in these experiments (Fig 6G); the E1A 12S protein was over-expressed compared to cells infected with Ad5 (S2A Fig, lanes 3 and 5). Collectively, these experiments reveal that IFNs inhibit Ad replication and E1A expression via E2F/Rb family proteins and induction of transcriptional repressor activity.
PML nuclear bodies (PML-NB) play important roles in intrinsic and IFN-mediated immunity against various viruses [13]. Three major components of PML-NB are PML, Daxx and Sp100. We previously reported that PML and Daxx, but not Sp100, mediated an IFN-induced antiviral response during Ad infection [14]. Interestingly, it has been shown that PML can induce cellular senescence in an Rb-dependent manner [31–33]. In these studies, PML relocalized Rb and induced heterochromatin formation and silencing of E2F target genes, leading to cell cycle arrest. Thus, we analyzed if PML-NBs participate in repression of Ad replication and E1A expression in HDF-TERT cells following IFN treatment. We generated a series of PML, Daxx and Sp100 knockdown HDF-TERT cell lines using shRNA expression. PML and Sp100 protein levels were completed depleted in the corresponding knockdown cells and Daxx expression was completely depleted in two out of four shDaxx-cells and significantly reduced in the other two lines; normal levels of all three proteins were found in cells expressing control shRNA (S2A Fig, left). Ad5 replication was inhibited by both IFNα and IFNγ in single knockdown cells, as well as in the control cells, despite high MOI infection (S2B Fig, left). RT-qPCR confirmed that E1A mRNA levels were correspondingly reduced (S3C Fig). A recent study revealed that PML, Daxx and Sp100 have synergistic effects on the suppression of herpesvirus replication with greater effects of multiple knockdown compared to single knockdown [34]. We generated HDF-TERT cells in which two or three PML-NB proteins were simultaneously depleted (Daxx plus PML, shDP, PML plus Sp100, shPS, and all three protein, shDPS; S2A Fig, right). IFNs repressed Ad replication in these double- and triple-knockdown cells to the same extent as with parental or control knockdown cells (S2B Fig, right).
IFI16, a HIN-200 family protein, senses intracellular viral DNA leading to IRF3 activation and IFNβ expression [35]. Ectopic expression of IFI16 in prostate cancer cell lines increases p21 expression and inhibits E2F-stimulated transcription [36–38]. Taken together, these results suggested that IFI16 might regulate global E2F transcription activity upon IFN treatment in HDF-TERT cells. We generated three independent pools of IFI16 knockdown cells, shIFI16-1-1, shIFI16-1-2 and shIFI16-3-1 (S3A Fig). Although treatment of the IFI16 knockdown cells with IFNs slightly increased IFI16 protein levels (S3B Fig), Ad5 replication was still inhibited by IFNα and IFNγ treatment to the extent observed in control cells (S3C Fig). We conclude the neither PML-NB or IFI16 are required for IFN-mediated repression of Ad5 replication.
Previous studies demonstrated that T lymphocytes in tonsil and adenoid tissues are the primary reservoir for latent Ad [39,40]. As important cytokines mediating innate and adaptive immunity, type I and type II IFNs could inhibit Ad replication and promote the establishment of persistent infection. The conserved IFN-induced repressor element in the E1A enhancer region may be involved in Ad latency by repressing E1A expression in response to IFNs and allow Ad to evade immune surveillance and clearance. To test this hypothesis, HDF-TERT cells were infected with Ad5 at low MOI in the presence or absence of IFNs. Infected cells were cultured for an extended period of time and the production of infectious virus quantified by plaque assay (Fig 7A). During the first 5 days of infection, no cytopathic effect was observed in any infected cells. In the absence of IFNs, peak virus yield was reached at 10 days post-infection (~108 PFU/ml). Cytopathic effect was observed by day 8–10 (S5 Fig,–IFN) and full cytopathic effect was observed at 15 days post-infection. In contrast, in IFN-treated cells, virus yields gradually increased over 25–30 days, giving a peak of 1-2x107 PFU/ml. IFNα postponed the onset of cytopathic effect until 45 days post-infection. In IFNγ-treated cells, no cytopathic effect was observed throughout the entire period (up to 100 days post-infection). These results demonstrate that both IFNα and IFNγ can promote establishment of long-term Ad infection. An intermediate amount of infectious virus was produced throughout the course of infections with IFNγ treatment (106–107 PFU/ml) indicating the establishment of persistent, not latent, infection.
We examined the role of the E2F binding site in the E1A enhancer region in an IFN response during long-term infection of HDF-TERT cells. HDF-TERT cells were infected with Ad5-WT or Ad5-mut1 in the presence of absence of IFNγ. Both infectious virus yield and viral genome copy number were measured during the course of infection (Fig 7B and 7C). In the absence of IFN treatment, both Ad5-WT and Ad5-mut1 reached peak virus production at 10 days post-infection, and full cytopathic effect was observed at 15 days post-infection. In IFNγ-treated cells, virus production of Ad5-WT and Ad5-mut1 gradually increased from 0 to 30 days post-infection. Cells infected with Ad5-mut1 exhibited cytopathic effect from days 30 through 55 when the experiment was terminated, whereas Ad5-WT infection continued for 100 days with minimal cell death (Fig 7B and S5 Fig). Viral genome copy number correlated well with infectious virus yields throughout the course of these infections (Fig 7C). In the presence of IFNγ, Ad5-mut1-infected cells showed reduced cell viability and proliferation from approximately day 30 onward in comparison to Ad5-WT-infected cells (Fig 7D and 7E) as well as enhanced cytopathic effect (S5A Fig, +IFNγ). As expected, Ad5-mut1 also showed enhanced E1A expression compared to Ad5-WT of the course of infection (S5B Fig). IFNγ-treated cells infected with Ad5-WT were analyzed by immunofluorescence at 113 days post-infection to examine viral early and late gene expression in individual cells (Fig 7F). These result showed that ~40% of the infected cells were positive for viral gene expression with some cells expressing only early proteins and other cells expressing both early and late proteins.
We also examined the fate of long-term infected cells upon the withdrawal of IFNγ (Fig 7G and 7H), There were no significant changes in viral genome copy number and infectious virus yield 5 days after IFNγ removal compared to IFNγ-treated cells. However, significant increases in viral genome copy number and infectious virus yield were observed by 10 and 15 days following IFNγ removal; full cytopathic effect was observed by day 15. Collectively, we conclude that IFNγ treatment represses the Ad5 lytic cycle in infected HDF-TERT cells which promotes persistent infection, and that this effect requires the E2F binding site in the E1A enhancer region.
Adenoviruses establish two different types of infection in the host. Primary infection occurs in epithelial cells, e.g., the nasopharyngeal mucosa with Ad5, resulting in lytic infection and the production of progeny virus. Following acute infection, Ad5 establishes latent infection in the mucosa-associated lymphoid tissue, preferentially T-lymphocytes in tonsil and adenoid tissues [40,41]. Cellular mechanisms that regulate lytic Ad infection are well understood, but the molecular basis for the control of persistent Ad infections was not understood. Here, we demonstrate that both type I and type II IFN signaling leads to the repression of Ad5 immediate early gene expression in normal human cells. This reduction in E1A expression leads to suppression of all subsequent aspects of the virus life cycle. Our results show that repression of E1A expression by IFNs requires a conserved E2F binding site in the E1A enhancer region. E2Fs can transcriptionally activate or repress gene expression depending on their interactions with Rb family proteins [42]. IFNs augmented the binding of the tumor supressors Rb and p107, well characterized E2F binding partners and transcriptional repressors [42], to the E1A enhancer region in vivo. Mutation of the conserved E2F binding site in the E1A enhancer abrogated the effects of IFNs on E1A expression and Ad replication. Collectively, our results demonstrate that the IFN–E2F/Rb axis is critical for restriction of adenovirus replication during type I and type II IFN responses.
Given the negative role that the E2F binding site in the Ad5 E1A enhancer region has on both basal and IFN-regulated immediate early gene expression (Fig 5), it was surprising that this E2F site is conserved across divergent Ad serotypes. Indeed, both IFNα and IFNγ treatment of cells repressed the replication of Ads in five evolutionarily distinct subgroups. These observations suggested that Ads may utilize the conserved E2F site in the E1A enhancer to suppress E1A expression in certain infection contexts, for example during persistent Ad infections. There are a number of reports of persistent Ad shedding in individuals following primary infection [39,40,43–47] consistent with our results.
We established a persistent Ad infection model in vitro and demonstrated that IFNγ suppresses productive Ad replication in a manner dependent on this conserved E2F binding site (Fig 7). Relief of this repression by removal of IFN resulted in a transition from persistent to lytic Ad infection. A viral mutant that lacks the conserved E2F binding in the E1A enhancer region was resistant to the effects of IFNs and was unable to establish extended persistent infection in vitro. IFN signaling reduced the binding of the major E1A enhancer transactivator complex GABPα/β to the E1A enhancer in vivo. The E2F binding site is located immediately adjacent to the GABPα/β binding site in the E1A enhancer. It is possible that GABPα/β and E2F/Rb proteins compete for binding to the enhancer given the proximity of their binding sites. Alternatively, IFNs may regulate the binding of GABPα/β and E2F/Rb family proteins to the E1A enhancer through independent mechanisms. In additional to the E1A enhancer region, E2Fs bind to sites in the Ad5 E2 promoter [48]. In NHBEC and HDF-TERT cells, IFNs exhibited stronger inhibition of transcription from the E2A and E2B regions compared with other early regions (Fig 1F and 1G) consistent with the idea that E2Fs also directly negatively regulate E2 expression in addition to E1A. This effect also was observed in HDF-TERT-E1A cells (Fig 2E).
These results reveal a novel mechanism by which adenoviruses utilize IFN signaling to inhibit lytic virus replication and promote persistent infection. It is well established that IFNs fail to inhibit wild-type Ad replication in established cell lines [5–7]. The resistance of wild-type Ad to the effects of IFNs is due to multiple counteracting effects of viral gene products. Additionally, this is due to the nature of established cancer lines that contain alterations in different signaling pathways. The same experimental conditions were utilized in experiments that demonstrated significant effects of IFNs in normal human cells but a distinct lack of effect in an adenocarcinoma cell line (Fig 1). Given the association of IFN signaling with E2F/Rb family protein function shown in our studies, we attribute defects observed in IFN signaling in cancer cell lines likely due to perturbations in regulation of E2F/Rb family members since the E2F pathway is mutated in numerous cancer cells [29]. In previous experiments, IFNα caused a modest reduction in Ad5 E1A and DBP expression and a 5–10 fold reduction in viral replication and virus yield in Ad5-infected NHBEC [12]. The mechanism of IFNα activity was not determined. We attribute the moderate effects observed in these assays to the profound effects of IFNs in our studies to potential differences in cell populations and/or the amount of IFNα used to treat the cells (250U/ml in [12], 500 U/ml here).
PML-NBs have established intrinsic and IFN-induced activity against many herpesviruses [49,50] which prompted us to examine if resident PML-NB proteins PML, Daxx or Sp100 inhibited Ad replication in HDF-TERT cells plus or minus IFN signaling. These activities were knocked-down individually and combinatorially using shRNAs. Depletion of these proteins singly or in combination did not alter the inhibitory effect of IFNs on Ad replication in a significant manner (S2 Fig). There was some variation among the cell lines on Ad5 infection minus IFN that we attribute to clonal variation. The lack of effect of these proteins on Ad replication, plus or minus IFN signaling, in comparison to herpesviruses may reflect the nature of the viral genomes within the virus particle. Herpesviruses contain naked viral DNA that may be easily recognized by cellular pattern recognition receptors to trigger IFN signaling and suppress virus infection [35]. In contrast, the Ad genome is coated by a histone-like core protein that may protect the genome from such recognition [51]. Consistent with this idea, we previously showed that Ad core protein VII protects the viral genome from recognition by the DNA damage machinery [52]. Thus, Ad may be resistant to the effects of PML-NBs in specific contexts.
The murine protein p202, a HIN-200 family member, is an IFN-inducible gene product that represses gene expression via E2F4/DP1 in transient expression assays [53,54]. These results and the known role of IFI16 in regulating cellular proliferation via CKIs which impact E2F-Rb activity [36–38] prompted us to examine if the human p202 homolog, IFI16, is involved in IFN-mediated repression of Ad replication in HDF-TERT cells. Depletion of IFI16 using shRNAs did not abrogate the negative effect of IFNs on Ad E1A expression and viral DNA replication in HDF-TERT cells (S3 Fig). Murine p202 regulates the DNA binding activity of E2F4/DP1 [53] and we found no effect of IFNs on E2F DNA binding properties in vitro plus or minus IFNs (Fig 6A).
We found that HDF-TERT cells infected with wild-type Ad5 continued proliferating without lysis over 100 days of maintenance in the presence of IFNγ (Fig 7). Viral DNA replication was restricted and maintained at a steady state level (1,000–10,000 viral genomes/cell). This number is comparable to persistent Ad infection in the BJAB and Ramos B cell lines [41]. In contrast, HDF-TERT cells infected with wild-type Ad5 and treated with IFNα initially maintained persistent Ad infection but succumbed by day 45 (Fig 7A). It is not clear why cells treated with IFNα did not maintain a persistent infection. Ad5-infected HDF-TERT cells continuously produced low amounts of infectious virus in these experiments suggesting that Ad established a persistent infection, rather than a latent infection, in the presence of IFNs. We attempted to study the effect of IFNs on the properties of Ad infection in lymphocytes (Jurkat and PM1 T cells and BJAB B cells). Even though these cells responded to IFN signaling (STAT1 phosphorylation was detected 1 hr after IFN treatment, STAT1 expression was induced by 24 hr), IFNα and IFNγ only had a minimal effect on Ad5 replication. These cell lines were derived from leukemias and lymphomas and likely contain alterations in the E2F/Rb pathway.
E1A protein expression did not block IFN signaling in HDF-TERT cells in our experiments (Fig 2C and 2F) in contrast to the inhibitory effect of E1A on Jak/STAT signaling and IFN-induced gene expression in previous experiments ([8–10] and references therein). In all but one of these reports, established cancer cell lines were used and we believe that IFN signaling may be altered in these cells compared to normal human cells used in our experiments. E1A was shown to inhibit IFN signaling in primary human tracheal cells [55] but using 10-fold less IFNγ than used here.
This is the first study that probes the molecular basis of lytic versus persistent adenovirus infection. We established a persistent Ad infection model in vitro and demonstrated that IFNγ suppresses productive Ad replication in a manner dependent on a conserved E2F binding site in the E1A enhancer region. These results reveal a novel mechanism by which adenovirus utilizes IFN signaling to suppress virus replication and promote persistent infection, and demonstrate that the IFN-E2F/Rb axis plays a critical role in this process.
Normal human bronchial epithelial cells (NHBEC) were purchased from Lonza and maintained in Bronchial Epithelial Cell Growth Medium containing BulletKits (Lonza) according to manufacturer’s instructions. HDF-TERT cells [56] were kindly provided by Dr. Kathleen Rundell (Northwestern University, Chicago, IL) and maintained in Dulbecco’s Modified Eagle’s Medium (DMEM) containing 10% fetal bovine serum (HyClone Laboratories). A549 cells (ATCC) were grown in DMEM containing 10% bovine calf serum. 293FT cells (Life Technologies) were used for the generation of lentivirus stocks and were maintained in DMEM containing 10% fetal bovine serum according manufacturer’s instructions. 293-TP cells [57] were used to propagate mutant virus ΔTP-GFP and were maintained in DMEM containing 10% fetal bovine serum and 400 μg/ml G418. All cell growth media were supplemented with 100 U/ml penicillin and 100 μg/ml streptomycin. Cells were treated with 500U/ml IFNα or 1000 U/ml IFNγ. The Ad5 E1A coding region was inserted into lentiviral expression vector pLenti6/v5-D-TOPO (Life Technologies). A lentivirus stock was generated by cotransfection of pLenti-E1A plasmid with ViraPower packaging mix (Life Technologies) into 293FT cells. HDF-TERT cells were transduced with the E1A-expressing lentivirus according to the manufacturer’s instructions. Pools of E1A-expressing HDF-TERT cells were obtained after selection with 1μg/ml blasticidin.
The replication defective adenovirus, ΔTP-GFP, was generously provided by Dr. Jerry Schaack (University of Colorado, Denver, CO). With ΔTP-GFP, the Ad5 terminal protein coding region is disrupted by the GFP gene. Mutant viruses, in340, in340-A5, in340-B1, dl309-21, dl309-3 and dl309-317/358, were previously described [20,27]. In340-Δ2-CMV, dl309-273/371, and pTG3602 mutants 1 through 3 (mut1–mut3) were generated by PCR and recombination using parent viruses in340 [20], dl309 [58], and TG3602 [59]. In340-Δ2-CMV contains a deletion of the E1A enhancer region which was replaced by the CMV promoter/enhancer region; Ad5 packaging sequences were inserted at the right-end of the viral genome [20].
HDF-TERT cells were incubated with IFNs or left untreated for 24 hr, followed by adenovirus infection at 37°C for 1 hr at the multiplicities of infection indicated in figure legends. Nuclear DNA and total cell DNA were purified at 6 and 48 hr post-infection, respectively, using a QIAGEN DNeasy Blood & Tissue Kit. Both viral and cellular genome copy numbers were determined by qPCR using primer pairs that recognize either the Ad5 genome or cellular GAPDH gene with DyNAmo HS SYBR Green qPCR Kit (Thermo). The relative viral copy numbers of each time point were normalized to GAPDH. The fold-increase of viral copy number was calculated by normalizing to input viral DNA in 6 hr post-infection samples. Relative viral replication efficiency in IFN-treated cells was presented as the relative value compared to untreated cells. In the case of infection in A549 and NHBEC cells, nuclear and total DNA was harvested at 2 and 24 hr post-infection, respectively.
Total RNA from infected cells was isolated using a QIAGEN RNeasy kit. Equal amounts of RNA from each sample were used for synthesis of first-strand cDNA using SuperScript II reverse transcriptase (Life Technologies) and quantified by qPCR using primer pairs that recognize different Ad5 early mRNAs or cellular GAPDH mRNA with DyNAmo HS SYBR Green qPCR Kit (Thermo). The Pfaffl method of relative quantification was used to convert the resulting threshold cycle data for each sample to relative fold change information [60]. Viral mRNA levels were normalized to cellular GAPDH mRNA.
ChIP was performed as described previously [61] with modification. HDF-TERT cells were pre-treated with IFNs or left untreated for 24 hr, followed by Ad infection at 37°C for 1 hr at 200 virus particles/cell. At 18 hr post-infection, cells were cross-linked by adding serum free DMEM containing 1% formaldehyde and incubated at 37°C for 10 min. Cross-linking was quenched by the addition of glycine to a final concentration of 125 mM. Cells were harvested and cell pellets were resupended in 1 ml SDS lysis buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA, 1% SDS, 1 ug/ml aprotinin, 1 ug/ml pepstatin, 1 mM phenylmethanesulfonyl fluorid) per 107 cells, followed by incubation on ice for 10 min. Lysed cells were sonicated to yield chromatin fragments of 200–1000 bp. Cellular debris was removed by high speed centrifugation. Cell lysates containing 100 μg chromatin were diluted 10-fold using dilution buffer (20 mM Tris-HCl, pH 8.0, 140 mM NaCl, 1.2 mM EDTA, 0.01% SDS, 1.1% Triton X-100, plus protease inhibitors) and pre-cleared by incubation with protein A-agarose/salmon sperm DNA slurry (Millipore) for 1 hr at 4°C with rotation. Samples were clarified by centrifugation and pre-cleared lysates incubated with 10 μg of anti-GABPα, anti-Rb, anti-p107, or anti-HA antibody overnight at 4°C. Immune complexes were captured using protein A-agarose/salmon sperm DNA for 2 hr at 4°C with rotation and pelleted by centrifugation. Immunoprecipitates were washed once with low-salt wash buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 0.1% SDS, 1% Triton X-100, plus protease inhibitors), once with high salt wash buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 2 mM EDTA, 0.1% SDS, 1% Triton X-100, protease inhibitors), once with LiCl wash buffer (10 mM Tris-HCl, pH 8.0, 0.25 M LiCl, 1 mM EDTA, 1% SDS, 0.5% Triton X-100, 1% sodium deoxycholate, plus protease inhibitors), and twice with TE buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA, plus protease inhibitors). Immune complexes were eluted using 100 mM NaHCO3, 1% SDS at room temperature. Formaldehyde cross-links were reversed with 50 mM Tris-HCl, pH 7.5, 200 mM NaCl, 10 mM EDTA, and 0.5 mg/ml proteinase K at 65°C overnight. DNA was recovered by standard phenol/chloroform extraction and ethanol precipitation and resuspended in 50 μl 10 mM Tris-HCl, pH 7.5, 1 mM EDTA. 2 μl of DNA sample was subjected to qPCR. One μg pre-cleared chromatin was used to measure DNA input levels. The fold-enrichment of specific DNA target was presented as percentage of input DNA.
All numerical values represent mean ± sd. Each experiment was done in three replicates, and a representative replicate is shown for each blot. Statistical significance of the differences was calculated using student’s t-test.
|
10.1371/journal.pmed.1002206 | Genomic Analysis of Uterine Lavage Fluid Detects Early Endometrial Cancers and Reveals a Prevalent Landscape of Driver Mutations in Women without Histopathologic Evidence of Cancer: A Prospective Cross-Sectional Study | Endometrial cancer is the most common gynecologic malignancy, and its incidence and associated mortality are increasing. Despite the immediate need to detect these cancers at an earlier stage, there is no effective screening methodology or protocol for endometrial cancer. The comprehensive, genomics-based analysis of endometrial cancer by The Cancer Genome Atlas (TCGA) revealed many of the molecular defects that define this cancer. Based on these cancer genome results, and in a prospective study, we hypothesized that the use of ultra-deep, targeted gene sequencing could detect somatic mutations in uterine lavage fluid obtained from women undergoing hysteroscopy as a means of molecular screening and diagnosis.
Uterine lavage and paired blood samples were collected and analyzed from 107 consecutive patients who were undergoing hysteroscopy and curettage for diagnostic evaluation from this single-institution study. The lavage fluid was separated into cellular and acellular fractions by centrifugation. Cellular and cell-free DNA (cfDNA) were isolated from each lavage. Two targeted next-generation sequencing (NGS) gene panels, one composed of 56 genes and the other of 12 genes, were used for ultra-deep sequencing. To rule out potential NGS-based errors, orthogonal mutation validation was performed using digital PCR and Sanger sequencing.
Seven patients were diagnosed with endometrial cancer based on classic histopathologic analysis. Six of these patients had stage IA cancer, and one of these cancers was only detectable as a microscopic focus within a polyp. All seven patients were found to have significant cancer-associated gene mutations in both cell pellet and cfDNA fractions. In the four patients in whom adequate tumor sample was available, all tumor mutations above a specific allele fraction were present in the uterine lavage DNA samples. Mutations originally only detected in lavage fluid fractions were later confirmed to be present in tumor but at allele fractions significantly less than 1%. Of the remaining 95 patients diagnosed with benign or non-cancer pathology, 44 had no significant cancer mutations detected. Intriguingly, 51 patients without histopathologic evidence of cancer had relatively high allele fraction (1.0%–30.4%), cancer-associated mutations. Participants with detected driver and potential driver mutations were significantly older (mean age mutated = 57.96, 95% confidence interval [CI]: 3.30–∞, mean age no mutations = 50.35; p-value = 0.002; Benjamini-Hochberg [BH] adjusted p-value = 0.015) and more likely to be post-menopausal (p-value = 0.004; BH-adjusted p-value = 0.015) than those without these mutations. No associations were detected between mutation status and race/ethnicity, body mass index, diabetes, parity, and smoking status. Long-term follow-up was not presently available in this prospective study for those women without histopathologic evidence of cancer.
Using ultra-deep NGS, we identified somatic mutations in DNA extracted both from cell pellets and a never previously reported cfDNA fraction from the uterine lavage. Using our targeted sequencing approach, endometrial driver mutations were identified in all seven women who received a cancer diagnosis based on classic histopathology of tissue curettage obtained at the time of hysteroscopy. In addition, relatively high allele fraction driver mutations were identified in the lavage fluid of approximately half of the women without a cancer diagnosis. Increasing age and post-menopausal status were associated with the presence of these cancer-associated mutations, suggesting the prevalent existence of a premalignant landscape in women without clinical evidence of cancer. Given that a uterine lavage can be easily and quickly performed even outside of the operating room and in a physician’s office-based setting, our findings suggest the future possibility of this approach for screening women for the earliest stages of endometrial cancer. However, our findings suggest that further insight into development of cancer or its interruption are needed before translation to the clinic.
| Endometrial cancer is the most common gynecologic malignancy in the United States.
When detected early, endometrial cancer survival rates are improved.
There are no screening methods that can detect either pre-malignant lesions or early-stage cancers.
We conducted a prospective study in 107 women where uterine lavage fluid was analyzed for genetic mutations using ultra-deep, panel-based next-generation sequencing (NGS).
Seven of the 107 women were identified by gold-standard histopathology as having endometrial cancer. All seven, even those with microscopic cancer foci, had significant cancer-associated gene mutations detected in their lavage fluid.
Fifty-one women without histopathologic evidence of cancer had high allele fraction cancer-associated mutations.
This study provides the first evidence for the ability of an NGS-based approach to prospectively detect early-stage, microscopic endometrial cancer.
Our study identified a previously unknown but highly prevalent landscape of driver and potential driver mutations in women who did not have histopathologic evidence of endometrial cancer.
These findings suggest that while NGS-based analysis of uterine lavage can achieve the necessary sensitivity for endometrial cancer screening, further insights into the steps leading to endometrial cancer development and/or its interruption are still needed before this goal can be achieved.
These findings, based on ultra-deep sequencing, may also have implications for understanding the development and clonal expansion of somatic cancer-driver mutations in apparently non-diseased tissues.
| Endometrial cancer is the most common gynecologic malignancy in the United States, with 60,000 incident cases and greater than 10,000 deaths estimated for 2016. Alarmingly, both the incidence and associated mortality are rising [1]. By 2030, endometrial cancer is projected to surpass colorectal cancer to become the third most common cancer among women in the United States [2]. Despite its already high prevalence and increasing morbidity and mortality, no effective screening exists for endometrial cancer. Specifically, no screening methods can effectively detect either pre-malignant lesions (primary prevention) or early-stage cancers (secondary prevention). The lack of screening is particularly significant because when detected early, endometrial cancer survival rates are dramatically improved. The 5-year survival for localized disease is 95%, whereas it is <20% for disease that has metastasized [3].
Postmenopausal bleeding is the most common presenting symptom for women with endometrial cancer. Abnormal bleeding is reported in ~90% of cases [4] and simultaneously is one of the most common reasons for an office gynecology visit [5]. Conversely, and dependent upon risk factors, less than 10% of these women will have endometrial cancer [6,7]. Uterine fibroids, adenomyosis, polyps, and ovulatory dysfunction represent the most common causes of bleeding. Currently, the direct visual inspection of the uterine cavity through hysteroscopy combined with curettage of tissue or complete hysterectomy are considered “gold standards” for evaluating endometrial pathology and diagnosing endometrial cancer [8]. Both procedures entail the use of an operating room setting, patient anesthesia, some degree of patient discomfort, and high costs. The optimal screening test would avoid many of these issues while offering the ability to reliably detect all endometrial cancers at the earliest stage.
For more than 70 years, it has been appreciated that endometrial cancer and its precursors exfoliate cells into the uterine cavity [9], and, for almost as long, attempts have focused on obtaining these cancerous/precancerous cells for diagnostic purposes [10]. The success in marked mortality reduction in cervical cancer, another gynecologic cancer, through the use of the Papanicolaou (Pap) test provides the driving rationale for endometrial cancer screening. In this relatively simple test, a simple brushing or scraping of the cervix provides a sampling of cells for histologic evaluation of premalignant changes [11].
The first description of using uterine lavage for endometrial cancer detection was in 1957 [12]. Saline was introduced into the uterine cavity and then returned via aspiration. Cells within this lavage were centrifuged, smeared onto slides, and then evaluated by a cytopathologist [12]. A number of issues, including overall accuracy in cancer identification, difficulties in handling and processing of the aspirate, and requirement for cytopathology expertise, all limited the clinical adoption of this technique [10]. Nonetheless, a number of investigators have used variations upon this uterine lavage theme for attempts at developing screening and/or diagnostic tests. These include: collecting cells for cytology during ultrasound evaluation of the uterus [13,14], evaluating matrix metalloproteinase levels from women with endometrial cancer [15], and measuring DNA microsatellite instability in Lynch syndrome patients with endometrial cancer [16].
Recently, two exciting proof-of-principle studies demonstrated the use of next-generation sequencing (NGS) of DNA of uterine shed cells to identify somatic mutations in patients with known gynecologic cancers [17,18]. In already established endometrial and ovarian cancer cases, the authors demonstrated that panel-based, targeted sequencing of shed cells, retrieved either by brushing of the cervical canal [17] or through uterine lavage [18], could detect somatic mutations consistent with these two Müllerian duct-derived cancer types. In part, these two studies were made possible by the in-depth genetic characterization of endometrial cancers by The Cancer Genome Atlas Research Network (TCGA). TCGA-derived data facilitated the design of targeted sequencing panels for more sensitive mutation detection given the succinct nature of the panels. [19] In general, the classification of endometrial cancers by mutational characteristics is reproducible and potentially an improved method of classification over traditional pathological diagnosis, which is known to be subjective and prone to error [20,21].
In this study, we sought to provide the first prospective analysis of uterine lavage fluid from women taken at the time of their evaluation for a definitive tissue-based diagnosis to assess the use of targeted NGS for detecting endometrial carcinomas. The women in this study were primarily either experiencing abnormal uterine bleeding or had abnormal pelvic ultrasound findings and were being evaluated by hysteroscopy and curettage for a tissue diagnosis. We obtained both cellular DNA present in uterine lavage fluid and cell-free DNA (cfDNA), which itself has never previously been described from the uterine cavity. Using first a pan-cancer 56-gene panel and then a TCGA-guided 12-gene endometrial cancer panel, we detected somatic mutations in all women who were later diagnosed with stage IA endometrial cancer. In addition, we determined that half of the women in our study who did not have clinical evidence of cancer nonetheless possessed a significant landscape of driver mutations at relatively high allele fractions. Our findings therefore suggest the apparently opposing possibilities of a genomics-based approach for endometrial cancer screening and the discovery of prevalent driver mutations in clinically defined non-cancerous tissue. Ultimately, these results may lead to further insights into the steps distinguishing between endometrial cancer development and its interruption.
The study was conducted from September 2015 to November 2016. Patient samples were collected during the months of September 2015 to April 2016, with DNA extraction being performed concurrently with sample collection. NGS, Sanger sequencing, and digital droplet PCR were performed on these samples and validation sets from February 2016 to October 2016. Data analysis was performed once all samples were sequenced and histopathology results confirmed.
All uterine lavage, blood, and tumor samples were collected in accordance with the Institutional Review Board of the Icahn School of Medicine at Mount Sinai at the time of the diagnostic procedure (GCO# 10–1166). All clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki. Written informed consent was obtained from each enrolled patient. All patients undergoing hysteroscopy and dilation and curettage at our institution were offered the opportunity to enroll in the study. A total of 111 patients were enrolled and 107 samples collected from September 2015 to April 2016. Four participants did not undergo the scheduled procedure due to difficulties unrelated to our study but which precluded the surgeon from performing the procedure. Final diagnoses were available after completion of the molecular analyses, and seven patients were diagnosed with endometrial cancer by histopathology. Tumor tissue was available in sufficient amounts for research-based analysis from four of these seven patients.
Uterine lavage specimens were collected in the operating room at the time of hysteroscopy. Hysteroscopy was performed under either general or laryngeal mask airway anesthesia as deemed appropriate by the anesthesiologist. After induction of anesthesia, patients were placed in dorsal lithotomy position. A vaginal surgical prep with iodine was performed. Next, a speculum was placed in the vagina and the cervix was visualized. A tenaculum was used to grasp the cervix. If stenotic, dilators were used to dilate the cervix. The hysteroscope was advanced into the cervix and subsequently into the uterine cavity aided with either saline or glycine inflow. Immediately upon entering the uterine cavity with the hysteroscope, the initial 20–30 mL of fluid was collected using a 40 mL specimen trap device (Medline Mucus Specimen Trap 40cc, No. DYND44140 Venture Respiratory Inc, Brooklyn, NY) attached to suction. Following this collection, the patient underwent the remainder of their procedure as per their surgeon’s discretion.
All uterine lavage samples in the specimen trap device were placed on ice and taken to the laboratory and processed within 1 hour of collection. In the laboratory, the lavage specimens were transferred to 50 mL centrifuge tubes (No. C1061, Denville Scientific Inc, Holliston, MA) and centrifuged at ~3,200 g for 20 min at 4°C. The acellular supernatant was separated from the cell pellet using a pipette and recentrifuged for an additional 10 min to remove any remaining cellular material and debris. This fraction was then collected and stored at -80°C until final DNA extraction.
The cell pellet was washed with red blood cell lysis solution (5 Prime, No. 2301310, Gaithersburg, MD) by adding 1 mL of the solution to the cell pellet, resuspending by gentle pipetting, incubating at room temperature for 5 min, then centrifuging at 420 g in a table top centrifuge for 5 min. The RBC lysis supernatant was then discarded, leaving behind the cell pellet. This was repeated until the cell pellet was cleared of visible red cell contamination. The cell pellet was stored at -80°C until DNA isolation was performed.
Cell-free DNA (cfDNA) was first concentrated from the acellular portion using a centrifugal filter (Amicon Ultra-15 30 kiloDalton Filter Units, EMD Millipore, No. UFC903096, Darmstadt, Germany) into smaller volumes ranging from 0.5 to 2 mL using the manufacturer’s protocol. The concentrated cfDNA was then extracted (Circulating Nucleic Acid Kit, Qiagen, Hilden, Germany) and eluted with 105 uL of AVE buffer.
The efficiency of the cfDNA extraction process was initially tested by spiking each acellular lavage sample with a known concentration of HindIII digested lambda DNA (Qiagen, Hilden, Germany) prior to concentration and cfDNA isolation. Quantitative PCR was then used to quantify the DNA fragments. Based on the calculated amount of spiked DNA, extraction efficiency was estimated to be between 46% and 87%.
Cellular DNA was extracted from the cell pellets (ArchivePure DNA Kit, 5 Prime, Gaithersburg, MD) with a modified protocol to account for low cell density. Briefly, total centrifugation times were increased for the two DNA precipitation/wash steps. The centrifugation times were increased to 10 min and 5 min each for the isopropanol and ethanol washes, respectively. The precipitated cellular DNA was eluted in 35 uL of AVE buffer.
Germline DNA was isolated from 10 ml blood samples collected from each patient at the time of their hysteroscopy (K3 EDTA tubes, Fischer Scientific, Pittsburgh, PA). Germline DNA was isolated (ArchivePure DNA Kit, 5 Prime, Gaithersburg, MD) according to the DNA purification protocol for whole blood, as per the manufacturer’s protocol.
The DNA concentrations of all fractions were determined by QuBit fluorometry (ThermoFischer Scientific, Waltham, MA).
For each patient, a set of sample trios from germline PBMC DNA and DNA isolated from the lavage cellular and acellular fractions were sequenced to an average of 5,000X coverage using a targeted amplicon panel. DNA sample quantity and integrity were assessed with an ALU repeat qPCR assay (Swift Biosciences, Ann Arbor, MI), and 10 ng qPCR quantified DNA was used as input into the Accel-Amplicon Panel. To establish baseline performance for the different sample types, sample trios from nine patients were initially sequenced using the Accel-Amplicon 56G Oncology Panel (Swift Biosciences, Ann Arbor, MI). Using the TCGA dataset for endometrial tumors and their associated mutational profiles, a smaller custom endometrial tumor amplicon panel was developed to cover the 12 genes with the highest mutation frequencies. These 12 genes included PTEN, PIK3CA, TP53, CTNNB1, KRAS, FGFR2, FBXW7, RB1, ATM, APC, ARID1A, and PIK3R1. This 12-gene panel includes 102 amplicons with an average length of 138 bp to maintain sensitivity with short, acellular DNA. The genomic target regions were designed to cover both hotspot loci and contiguous full-coding exons, including the full exonic coverage of TP53 (See S1 Table for a complete list of genomic loci covered).
To confirm patient identity and preserve proper sample assignments for each trio, a spike-in of a germline SNP panel was included in the 102-amplicon endometrial tumor panel, requiring 4% of sequencing reads. This collection of high minor allele fraction SNP variants provided robust discrimination among samples [22]. The low concentration spike-in enabled a 200X sequencing depth of SNP targets for germline variant calling while the oncology targets were simultaneously sequenced to a 5,000X sequencing depth for somatic variant calling.
Resulting targeted NGS libraries were quantified using qPCR and sequenced on an Illumina MiSeq using v2 chemistry. For data analysis, amplicon primers were trimmed using Cutadapt [23] and trimmed reads were aligned to the GRCh37 build of human genome using BWA [24]. Somatic variant calling was performed using MuTect, Varscan, and Lofreq after following GATK Best Practices. A target of 5,000x coverage and 10 ng inputs enabled the lower limit of detection to be set to the 1% fraction. The average performance metrics for each sample was 91% on target and 97% coverage uniformity as defined by 20% of the mean.
To minimize the potential for NGS-based sequencing errors, our protocol selected 30% of NGS-identified variants for validation using an orthogonal methodology. We used either digital droplet PCR, for those NGS-identified variants with allele fractions <10%, or Sanger sequencing, for those variants with allele fractions ≥10%. Custom TaqMan Assays (see S2 Table) were designed using the Life Technologies web-based design tool (http://www.thermofisher.com/order/custom-genomic-products/tools/genotyping/). Assays contained VIC or FAM labeled probes, which probed for the wild-type and mutant variants, respectively. Specificity of each assay was first validated by quantitative PCR. Next, sensitivity and lower limits of detection were established by digital droplet PCR (RainDance Technologies, Billerica, MA), as we have previously described [25]. When no tumor DNA was available, positive controls were synthesized and used for these reactions as 300–500 bp gBlocks Gene Fragments (Integrated DNA Technologies, Coralville, IA).
To examine the possibility that sequence artifacts were introduced during lavage cfDNA concentration, isolation, and purification, sheared control genomic DNA from well-characterized single (NA12878, Coriell Institute, Camden, NJ) and multiplexed (HD701, Horizon Diagnostics, Cambridge, UK) cell line references were processed through all of the steps as lavage cfDNA starting from dilution into 15 ml of saline. These control sample replicates were then sequenced at coverage levels greater (range: 12,000–26,000x) than the lavage samples.
To rank affected patients, mutations were classified into three groups: “drivers,” “potential drivers,” and “passengers,” or “mutations of unknown significance.” For mutation classification, we used TCGA-defined endometrial cancer mutation statistics [26], the inclusive TCGA mutation statistics provided by the CBIO Cancer Genomics portal [27], and the COSMIC database [28]; the functional impact of sequence variants was assessed by Mutation Assessor [29]. The group of driver mutations included those in activating or inactivating hotspots of oncogenes or tumor suppressors in major endometrial cancer driver genes of our panel, as well as truncating mutations in tumor suppressor genes of endometrial cancer. Thus, all mutations nominated as drivers were previously observed in endometrial cancer [26] and are also recurrent pan-cancer mutations [30]. In the group of “potential drivers,” we included predicted functional missense mutations in major endometrial cancer genes from the 12-gene panel. Mutation-drivers are typically observed in evolutionarily conserved positions and therefore assessed as functional by a Mutation Assessor score [29]. This result justifies nomination of driver mutation using a shorter list of predicted functional mutations, rather than the “long tail” of all mutations [31,32].
An overview of the study pipeline is presented in Fig 1. The first 111 women who were scheduled to be evaluated by hysteroscopy and curettage for a tissue-based diagnosis were enrolled in the study. Endometrial lavage samples were collected from 107 patients (Table 1). The most frequent preoperative diagnoses were abnormal uterine bleeding (n = 50, 46.7%), uterine polyp (n = 30, 38.0%), and thickened endometrium (n = 10, 9.3%). The most common indications for hysteroscopy in the general population include abnormal bleeding and structural uterine abnormalities [5]. In our cohort, this was also the case with the majority of patients undergoing hysteroscopy for abnormal bleeding, followed by various suspected structural abnormalities, including polyps, thickened endometrium, and fibroids suspected through initial ultrasound evaluation. Patients ranged in age from 29 to 85 y, with an average age of 57.5 y. The majority of patients were white (n = 70, 66.0%), with BMIs >25 (n = 66, 61.7%), and were post-menopausal (n = 59, 57.3%), parous (one or more children; n = 59, 57.8%), and non-smokers (n = 85, 80.2%) (Table 1).
All patients underwent uterine tissue curettage as part of the hysteroscopy, and the final diagnoses were determined by histopathologic assessment. The most frequent final diagnoses were polyp or polypoid fragment (n = 60, 56.1%), normal endometrium (n = 17, 15.9%), and fibroids (n = 13, 12.1%). Seven patients were diagnosed with cancer based on tissue analysis by histopathology (Table 2, Fig 2). A specialized gynecologic pathologist (T.K.) reviewed and verified each of these seven cases to confirm the diagnosis. Six of seven had stage IA cancer, and four of these were grade 1. One of these cancers was identified as a microscopic focus within a polyp and classified as <1 mm in size (Fig 2A). Four of seven were diagnosed with grade 1 endometrioid type cancer. The other three diagnoses were grade 2 endometrioid type, mixed grade 3 serous and grade 2 endometrioid type, and grade 3 carcinosarcoma. The clinicopathologic correlates of these cases are shown in Table 2.
Lavage samples were collected from 107 patients and processed into cellular and acellular fractions following centrifugation. We had reasoned, based on previous studies, that lavage fluid should contain not only tumor cells [10,33,34] but also, given the origin of circulating free (cfDNA) and circulating tumor DNA (ctDNA) from apoptosing cells [35,36], cfDNA and ctDNA shed from the epithelial surface of the uterus containing normal, premalignant, and endometrial cancer cells. Endometrial cancers arise from cells in the inner lining of the uterus [37]. We therefore extracted DNA from the post-centrifuged cell pellet and acellular supernatant fractions. The average amount of DNA extracted from each cell pellet was 2,255 ng (range 1 ng–38,875 ng). In the acellular fraction, the average DNA amount was 2,046 ng (range 0.4 ng–44,572 ng). We quantified the extracted acellular DNA (2100 Expert Bioanalyzer, Agilent Technologies, Santa Clara, CA). A majority of the cfDNA fraction was approximately 175 bp in size (S1 Fig). This suggested not only size uniformity of the isolated DNA, inconsistent with contamination by randomly sheared genomic DNA arising from cells possibly within the collected lavage sample, but a size profile consistent with apoptotic fragmentation of genomic DNA at nucleosome ends.
Samples from 102 patients passed all quality metrics, and only these were further analyzed. Nine patient samples were selected for sequencing using a targeted 56-gene, clinically relevant general oncology-related panel. In this pilot testing, four patients were found to have somatic mutations in their cellular and/or cfDNA. Based on these pilot results, we refined our sequencing strategy such that a targeted 12-gene endometrial cancer panel was used for all remaining samples. A flowchart describing the analysis and outcomes of the sequencing analysis is shown in Fig 3. In total, and based on results from both panels, 58 patients were found to have 126 unique somatic mutations in either the lavage cellular DNA or cfDNA (S3, S4 and S5 Tables). These included 89 missense, 24 nonsense (9 stop, 9 frame-shift, and 6 in-frame), and 13 silent mutations. Fifty-seven of these 58 patients had mutations detected by the 12-gene panel. A summary of the genes mutated, overall gene mutation frequencies, and correlation with histopathologic diagnosis is shown in Fig 4. Fourty-four patients had no mutations detected in either lavage cell pellet or cfDNA.
In total, 75 unique mutations were nominated as drivers (Table 3, S3 Table). Twenty-three mutations were nominated as “potential drivers,” 11 of which are recurrent mutations in that they are observed in other cancers, but not in endometrial cancer; two mutations were observed in endometrial cancer, but not in other cancers; and 10 mutations are newly described. The remaining 28 mutations were classified as “passenger” mutations or those having unknown significance.
As a validation assessment of the NGS-identified mutations, we selected a cohort of nearly one-third of all the mutations (55/184), across high and low allele fractions, for confirmation using two orthogonal, independent technologies. In total, 58 cellular DNA and/or cfDNA samples were thereby analyzed (S6 Table). For those mutations with allele fractions as defined by NGS as being ≥10%, we used Sanger sequencing; for allele fractions <10%, we used droplet digital PCR (ddPCR).
Notably, all mutations originally identified by NGS were validated. This included those variants with allele fractions as low as 1.0%, which was our lowest threshold for reporting variants by NGS. In addition to this analysis, when using ddPCR to validate a mutation identified in either the cellular DNA or cfDNA, the other paired specimen was also tested, even if no mutation was originally detected by NGS. In 14/15 sample pairings, the mutation was validated and confirmed to be present in the other sample and at an allele fraction of <1.0%; again, our original threshold cutoff. As shown in Table 4, the mutations identified in the lavage cell pellet and cfDNA were well correlated when binned by genes (R2 = 0.92, Pearson correlation coefficient) (Table 4; S2 Fig). Another interesting observation in those patients diagnosed with cancer by histopathology, as compared to all patients, is the increase in lavage-identified mutations with higher allele fractions (Table 4, bottom row). Specifically, for allele fractions of <5%, these fractions are 9% (cell pellet) and 15% (cfDNA) and for allele fractions of 5%–10%, the fractions are 27% (cell pellet) and 25% (cfDNA). Most notably, when the allele fractions are >10%, the fractions have increased to 50% (cell pellet) and 75% (cfDNA), and this is significant for both cfDNA (p = 0.0007, Fisher test) and cell pellets (p = 0.02), when the allele fraction is >10%.
To rule out the possibility that sample preparation, occurring at any point starting from the actual uterine lavage collection, may have induced DNA artifacts, we attempted to replicate the same steps, including using the same data analysis pipeline, but using spiked-in control DNA samples. For this we chose two well-characterized sequencing reference controls. Genomic DNA isolated from the CEPH single cell line standard NA12878 and the multiplexed sample HD701, which represents a mixture of three different cell lines, was sheared (Covaris, Woburn, MA), size-selected, and then processed as all lavage cfDNA samples. NA12878 was processed in quadruplicate (i.e., the starting sample split into four and each sample processed and sequenced independently) and HD701 in duplicate. The sequencing coverage (range: 12,000–26,000x) was more than two-fold greater than that used for sequencing of the patient-derived lavage-isolated DNA samples. While all germline variants associated with these two control samples were identified at the appropriate allele fractions, no artifactual variants were identified across all replicates (S9 Table).
As noted above, seven of the patients in our cohort were diagnosed with clinical evidence of cancer following their hysteroscopy and tissue curettage (Table 2). All seven cases had somatic driver mutations identified in both the cell pellets and cfDNA isolated from their lavage fluid (S4 and S7 Tables). PT398 was diagnosed with stage 1A, grade 1 endometrioid endometrial adenocarcinoma, with a tumor measuring <1 mm in diameter contained within a polyp (Fig 2). Cellular DNA from the lavage fluid contained a total of six driver mutations, including three PTEN mutations (W111*, F337fs, G132D), one PIK3CA mutation (E545A), one CTNNB1 mutation (S45F), and one FBXW7 mutation (R505C). Two of these six driver mutations were also detected in the cfDNA. PT433, also diagnosed with stage 1A, grade 1 endometrioid endometrial adenocarcinoma, was one of the patients sequenced in the pilot study using the 56-gene panel. Six driver mutations were detected, five in RET (L773fs, E775_F776fs, F776L, V778fs, K780Q781fs) and one in CDH1 (K86fs). Another stage 1A grade 1 endometrioid endometrial adenocarcinoma patient, PT451, had two driver mutations detected in FGFR2 (S252W) and PIK3CA (M1043V) in both uterine lavage fractions. PT492, the fourth stage 1A, grade 1 endometrioid adenocarcinoma case, had one driver mutation detected in PIK3CA (G106V; cell pellet DNA). PT484, diagnosed with stage 1A, grade 2 endometrioid adenocarcinoma, had one driver mutation, PTEN (I67R), detected in both cellular DNA and cfDNA.
PT468 was diagnosed with a stage 1A mixed histology cancer, one component being high-grade serous adenocarcinoma and the other being grade 2 endometrioid adenocarcinoma. This patient had a total of ten driver mutations detected in the following four genes: ARID1A (R1722*, R1446*, R1989*), RB1 (R445*), PIK3R1 (R514C), PTEN (I33S, R130Q, C211Q), and PIK3CA (E81K, R88Q). With the exception of the ARID1A (R1722*), PIK3R1, and PTEN (C211Y) mutations, all other mutations were present in both cellular DNA and cfDNA fractions. In addition, there were six potential driver mutations detected in both or either one of the uterine lavage fractions. The diversity of mutations most likely reflects the diversity of the mixed histology tumor, distinguished by an aggressive high-grade serous component. PT488, the patient with stage 3A carcinosarcoma, had two TP53 mutations. One was classified as a driver (Q165fs) and the other a potential driver mutation (C176F). There was also an additional ARID1A driver mutation (E1444*). TP53 mutations have been shown to be present in aggressive endometrial adenocarcinomas, including high-grade serous types and carcinosarcomas [38]. ARID1A mutations have been shown to be associated with more aggressive endometrial adenocarcinomas [39].
Owing to the limited volume of three of these tumors (PT398, PT433, PT492), as per our IRB consent, there was no tumor tissue that could be made available for DNA isolation for research purposes. However, for four cases, we were able to isolate tumor DNA and compare paired tissue/lavage mutation profiles. DNA was extracted from fresh frozen tissue of these four tumor samples (PT451, PT468, PT484, PT488) and sequenced using the 12-gene panel. In all cases, at least one mutation present in the paired tumor DNA was detected in both the lavage cellular DNA and cfDNA (S7 Table). For some, but not all, of the tumor mutations, the allele fractions matched those from the lavage fractions. As noted in S7 Table, low allele fraction mutations in the tumor were not always detectable in the lavage fluid, and this varied by tumor.
Among these four cases in which the paired tumor was available, PT468 was unusual not only because of the large number of mutations identified (n = 20) but also because a large number of lavage-identified mutations had not been detected in the tumor (n = 14) and, conversely, one tumor-identified mutation was not detected in the lavage fluid (S7 Table). To determine the degree to which lavage-identified mutations could be present in the tumor, and again to investigate the possibility of artifact, we used ddPCR to interrogate the tumor DNA. We selected the eight mutations that were present in both cfDNA and cell pellet DNA (S8 Table) and designed probes for their analysis. There was a very high degree of concordance between the NGS- and ddPCR-defined allele fractions for all eight lavage mutations. Five of the eight mutations were confirmed to be present in the tumor (S8 Table, S3–S7 Figs), and these were present at allele fractions that would not have been detected by NGS, as our cutoff threshold for NGS was 1.0%. The allele fractions of these tumor variants ranged from 0.15% to 0.004%, whereas the cognate lavage fractions ranged from 2.0% to 21.0%.
In total, 95 women were diagnosed with benign or non-cancer conditions (Table 1). Sequencing of lavage fluid from 44 of these women identified no mutations. In marked contrast, 51 patients without a histopathologic diagnosis of cancer were identified as having somatic mutations in their uterine lavage samples. A total of 95 driver mutations were detected in this group, with 59 unique mutations.
The most frequent driver mutations detected among this group were KRAS G12C (eight patients), KRAS G12S (ten patients), and PIK3CA H1047R (eight patients) (S5 Table). The finding that a majority of women without a cancer diagnosis carried mutations, the relatively high allele fractions (range: 1.0% to 30.4%; average: 3.0%), and the projected oncogenic impact of these mutations was surprising. For example, PT395, with a histopathologic diagnosis of benign “polypoid fragments,” had 12 driver mutations detected in her uterine lavage. These affect a total of five genes: PTEN (I32del [AF: 2.1%], R130G [AF: 2.4%], G165E [AF: 2.3%]), PIK3R1 (Y463_L466del [AF: 5.5%], E558fs [AF: 1.6%]), PIK3CA (Q546K [AF: 5.1%], C420R [AF: 1.5%]), KRAS (G12S [AF: 1.2%], G12C [AF: 6.3%], G12C [AF: 1.4%]), and FGFR2 (S252W [AF: 5.8%]). Nine of these driver mutations were detected in cellular DNA, and five were also present in cfDNA. Two additional mutations were detected in the lavage cfDNA (KRAS 9G12S [AF: 1.2%], KRAS G12C [AF:6.3%]). To validate the existence of these mutations and exclude the possibility of sequencing artifacts, three of these 12 mutations were selected and tested by ddPCR (PTEN R130G, KRAS G12S, KRAS G12C; S5 and S6 Tables). All the mutations were confirmed. Another striking example is provided by PT485. She was diagnosed by tissue histopathology as having a benign polyp. We identified four driver mutations; three were detected in the cellular pellet ([KRAS G12S (AF: 1.1%)], PIK3CA [(H1047R (AF: 3.0%) and E542A (AF: 1.4%)]) and the fourth in the cfDNA (PIK3CA G106V; AF: 1.2%). Two of these four mutations (KRAS G12S and PIK3CA H1047R) were selected and validated by ddPCR (S6 Table).
A histogram of the mutation classifications, driver, potential driver, passenger, in all the patients is shown in Fig 5. The sum of driver and potential driver mutations, based upon the 12-gene panel, is plotted on the y-axis. The patients diagnosed by traditional histopathology cluster to the left of the graph.
To establish possible correlations between the presence of driver and/or potential driver mutations and clinical characteristics including age, race/ethnicity, BMI, diabetes, parity, smoking status, and menopausal status, we performed univariate analysis. Increasing age (p-value = 0.002; Benjamini-Hochberg [BH] adjusted p-value = 0.015; mean age no mutations = 50.35; mean age mutated = 57.96, 95% CI: 3.30–∞) and postmenopausal status (p-value = 0.004; BH-adjusted p-value = 0.015) were significantly associated with the presence of driver and potential driver mutations (S8 and S9 Figs). Increasing age was also found to be significantly associated with a diagnosis of cancer (p-value = 0.041) (S10 Fig).
We performed additional univariate analysis to further analyze our data and establish whether there is an association between mutations in each of the 12 genes on the targeted panel and the same clinical variables. Increasing age was significantly associated with presence of driver or potential driver mutations in PIK3CA (p-value = 0.008) and TP53 (p-value = 0.001) (S11 and S12 Figs).
This prospective study on women undergoing hysteroscopy and dilation and curettage as a diagnostic procedure demonstrated the ability of ultra-deep targeted sequencing of uterine lavage sampling to identify not only early-stage endometrial cancers but also a surprisingly high burden of driver mutations in a majority of women in our cohort who were without histopathologic evidence of cancer. As such, we believe these molecular findings simultaneously offer both the future promise of screening women for endometrial cancer while also providing a rare opportunity to explore the processes that are associated with the early steps of defining cancer development and its abrogation.
We specifically chose to examine all women, without any preselection for known genetic risk factors or family history of endometrial cancer, who were undergoing hysteroscopy for diagnostic evaluation. The most frequent reason for hysteroscopic evaluation in our cohort was abnormal uterine bleeding. Abnormal uterine bleeding is estimated to affect up 1.4 million women annually in the United States [40] and up to 14% of all women during their lifetime [41]. The most life-threatening etiology of this common gynecologic symptom is endometrial cancer. In women with abnormal uterine bleeding there is upwards of an ~10% risk of endometrial cancer, depending on age and menopausal status [42].Of the women in our cohort with a prediagnosis of abnormal uterine bleeding, 7% were found to have endometrial cancer and 49% were identified to be carrying driver/potential driver mutations but without a diagnosis of cancer.
The second most frequent reason for hysteroscopy in our cohort was the presence of an endometrial polyp. Endometrial polyps are associated with a small risk of cancer. In our cohort, Patient 484 was found to have a stage 1A, grade 2 endometrioid cancer arising from within a polyp. In general, upwards of 25% of women will be diagnosed with polyps at some point in their lives [43]. A number of studies have revealed that the overwhelming majority of polyps are clinically benign [43–45], 1.3%–6.0% are premalignant [45–48], 1.2%–3.9% display atypical hyperplasia [43,49,50], 10%–25.7% contain simple or complex endometrial hyperplasia [43,47,48], and 0.8%–3.5% are cancerous [43–45,51,52]. While polyps rarely become malignant, 10%–34% of cases of endometrial cancer in postmenopausal women are associated with polyps [53]. In our cohort, 34 women with polyps were not diagnosed with endometrial cancer by histopathology, while 31 possessed driver and/or potential driver mutations.
Endometrial hyperplasia or endometrial intraepithelial neoplasia is clinically relevant because it is a known precursor lesion to endometrial cancer [54]. An example of the associated risk for the development of cancer is provided by a study of 170 patients with all grades of endometrial hyperplasia, who did not undergo a hysterectomy for at least 1 y. Only 1% of patients with simple hyperplasia progressed to carcinoma, 2.0%–3.4% of patients with complex hyperplasia progressed to carcinoma, 10.5% of patients with complex hyperplasia progressed to atypical hyperplasia, and 23%–52% of atypical hyperplasia cases progressed to carcinoma [55,56]. The advancement from simple to complex and atypical hyperplasia takes years and is potentially influenced by factors such as specific genetic aberrations, patient age, BMI, and menopausal status [45]. Given the relative risks, it is therefore necessary to correctly distinguish between the grading categories, because these have direct relevance to cancer development risk and, thus, potential overtreatment or undertreatment. Based on our findings, we suggest that ultra-deep sequencing of lavage fluid is able to detect stage IA cancer and could also identify these premalignant lesions. Further studies will be necessary to test these hypotheses.
A decidedly unexpected outcome of these studies was the discovery that the majority of women without a cancer diagnosis in our cohort possessed somatic driver and candidate driver mutations within uterine lavage cells and cfDNA. Age (BH-adjusted p-value = 0.015) and postmenopausal status (BH-adjusted p-value = 0.015) were both positively associated with the likelihood of harboring these mutations. Therefore, from a clinical perspective, because of the prevalence of cancer-driver mutations identified in women without histopathologic evidence of cancer, our lavage screening protocol is not yet able to distinguish between women with and without clinically relevant evidence of cancer. However, our results seemingly provide a unique opportunity to gain insight into the mechanisms underlying selection and clonal expansion, as mutated cells evolve either towards a final cancer phenotype or are halted and eliminated in their progression. Based on the experience in our institution, it is expected that the overwhelming majority of women with a negative cancer diagnosis, based on the operating room procedure of combined hysteroscopy and tissue curettage, in the absence of continued symptoms will not develop clinically relevant endometrial cancer. The long-term surveillance of women in our study who possess these driver/candidate driver mutations will nonetheless provide clarification on the ultimate outcome.
Our assay was designed to have internal validation of NGS-based mutation calls using orthogonal detection methods, namely Sanger sequencing and digital PCR. Thus, the NGS-identified driver/candidate driver mutations in women without a diagnosis of cancer are not likely to represent technical artifacts because all variants that we tested by these two orthogonal technologies were validated. Nonetheless, because the possibility exists for “in vitro artifacts,” we do highlight this caveat and provide several arguments suggesting that, if they were present in our study, we do not believe they would undermine the main results. First, “in vitro artifacts” would be expected to affect both normal germline as well as “lavage” DNA. We did not observe any new, previously undocumented SNPs or deletions/insertions in germline controls. Second, while “in vitro artifacts” would be irrelevant to natural selection processes, the observed prevalence of detected mutations in common cancer hotspots (e.g., position G12 in KRAS) is more suggestive of an oncogenic selection process resulting in a typical cancer gene mutation distribution. Third, the comparison of the DNA mutation spectra in hotspot positions detected in this study with those reported by TCGA uterine cancer tumors is presented in S3 Table. The distribution of nucleotide mutations reveals that the most frequently observed TCGA mutations are (i) detected in our study and (ii) also detected as the most frequently occurring ones, e.g., mutation C>A, C>T in KRAS position 25398284; mutations A>G, G>A in PIK3CA positions 178952085 and 178936082, respectively; mutation C>G in PTEN position 89692904; and mutation G>C in FGFR2 position 123279677. Also, all hotspot mutations detected in our study are also reported in TCGA. The spectrum of TCGA mutations is, however, noticeably broader, which is not surprising given that the number of samples in their dataset was approximately ~3.5 times larger. In particular, percentages of DNA hotspot mutations reported in TCGA and not observed in our study are ~3% for KRAS; ~25% for PIK3CA; ~12% for PTEN; ~14% for FBXW7; and 33% for CTNBB1. For FGFR2 and ARID1A, we observed the same hotspot mutation types as reported in TCGA. Thus, our findings identify a clear tendency that the hotspot mutation spectra are basically similar for both studies. Finally, using spiked-in control DNA samples, we were not able to identify DNA artifacts introduced during the steps associated with lavage sample isolation and processing. Therefore, we believe it is unlikely that there is a strong in vitro artifact component affecting the observed spectrum of lavage mutations.
If "driver mutations" provide a selective growth advantage and can lead to cancer, how, then, does one explain the presence of high-frequency (allele fractions ranging from 1%–30%) driver/candidate driver mutations in half of our study population without cancer and who may have only a minimum risk of developing endometrial cancer. Three potentially instructive paradigms from recent genomic-based studies in other tissues from apparently healthy individuals may provide some insight. First, in normal blood and skin cells, driver mutations associated with clonal expansion have now been described [57]. Results from three nearly back-to-back whole-exome [58,59] and gene panel targeted sequencing studies [60] on nearly 34,000 individuals identified clonal hematopoiesis with leukemia-related, somatic driver mutations, most notably DNMT3A, in 10% of apparently healthy individuals >65 years of age and in nearly 20% in those between 90 to 108 years of age [58,59]. In individuals under <50 years of age, levels, while detectable, did not rise above 1%. While the absolute risk remained small, the individuals carrying these driver mutations were clearly at increased risk for developing future hematologic cancers, suggesting the premalignant nature of the detected clones [58,59].
Second, it has been appreciated for some time that clonal patches of skin contain TP53 mutations [61,62]. Recently, and using an ultra-deep sequencing strategy, it was shown that upwards of 32% of non-lesion-containing, sun-exposed epithelial cells from the eyelid contain mutations in key drivers of squamous cell carcinomas [63]. Finally, in a study searching for p53 mutations in peritoneal fluid, again using an ultra-deep sequencing strategy, all women in the study, 17 with ovarian cancer and 20 without evidence of cancer, were found to harbor TP53 mutations. For the women without cancer, the TP53 mutations were extremely low frequency (median mutant fraction <1/10,000) and associated with increasing age, but still were mostly deleterious and clustered in hotspots [64].
Taken together with our findings, genomic analysis is, thus, revealing a more complex genetic environment than previously believed to exist, in that some tissues, genetically-defined "driver mutations" are relatively frequent and prevalent in healthy individuals. Cancers do arise from clonal evolution and expansion of a single cell, and driver mutations confer a growth advantage to that cell [65,66]. But, as has been pointed out based on at least some of these findings [67], caution is needed prior to predicting clinical consequences or making patient-care decisions based solely upon gene mutations. Our NGS-based findings, and those by others [57–64], thus suggest that the evidence of potential malignancy may be determined well in advance of its pathologically defined appearance or clinical relevance. Given the limited size of the targeted panel used in our current study, it would seem reasonable to hypothesize that ultra-deep whole exome sequencing or whole genome sequencing would have identified an even higher percentage of women carrying cancer driver mutations but without a clinically defined cancer diagnosis. What may be most clinically relevant is identifying and understanding the mechanisms by which some clones continue to evolve and become cancer while others are halted. While successful screening for earlier detection of cancer will undoubtedly improve quality of life and improve survival, insights into halting the progression of steps linking somatic mutation and the evolution towards cancer provide the greatest benefits. These goals must be balanced by the harm that would currently be imposed by over-diagnosis.
In our study, we identified driver mutations in all seven women who were found to harbor endometrial cancers. Six of these were stage IA, the earliest cancer stage wherein the cancer either does not invade or does not invade beyond half of the myometrial tissue. Indeed, one of the stage IA cancers detected in our cohort was microscopic (Fig 2A), and for three others not enough tissue was available for additional research purposes. In detecting driver mutations in women with stage IA cancer, even in those cases in which only microscopic amounts of tissue were available and, thus, were a clinical challenge, we establish that adequate amounts of tumor shed cells and DNA are present in uterine lavage. Thus, at this time, ultra-deep sequencing represents a candidate screening methodology. Currently, there is no effective recommended screening method for endometrial cancer in the general population. Obtaining adequate endometrial tissue to establish a diagnosis is critically important. Endometrial sampling devices used in an office setting are currently the first line of evaluation. These methods are not always successful, particularly in postmenopausal women—exactly the population of women who are at highest risk of endometrial cancer. In these cases, hysteroscopy and curettage are used. Thus, novel methods are needed. Based on the results of this study, we propose that future clinical studies should address the possibility of uterine lavage as a potential screening test in women to detect endometrial cancers. Paradoxically, the same ability to detect mutations in women with endometrial cancer using ultra-deep sequencing, which promises the ability to screen for this cancer, also reveals a previously unknown prevalent landscape of driver mutations in women without clinically apparent evidence of cancer. This duality represents a diagnostic dilemma but an intriguing view towards new biologic questions. With the use of innovative technologies on the horizon, such as single-cell sequencing and further developments in NGS, we hope that these results will provide a catalyst for novel insights into endometrial cancer development and the protective mechanisms that limit the evolution of normal cells into cancer.
|
10.1371/journal.pntd.0005350 | Trypanosoma cruzi High Mobility Group B (TcHMGB) can act as an inflammatory mediator on mammalian cells | High Mobility Group B (HMGB) proteins are nuclear architectural factors involved in chromatin remodeling and important nuclear events. HMGBs also play key roles outside the cell acting as alarmins or Damage-associated Molecular Patterns (DAMPs). In response to a danger signal these proteins act as immune mediators in the extracellular milieu. Moreover, these molecules play a central role in the pathogenesis of many autoimmune and both infectious and sterile inflammatory chronic diseases.
We have previously identified a High mobility group B protein from Trypanosoma cruzi (TcHMGB) and showed that it has architectural properties interacting with DNA like HMGBs from other eukaryotes. Here we show that TcHMGB can be translocated to the cytoplasm and secreted out of the parasite, a process that seems to be stimulated by acetylation. We report that recombinant TcHMGB is able to induce an inflammatory response in vitro and in vivo, evidenced by the production of Nitric Oxide and induction of inflammatory cytokines like TNF-α, IL-1β and IFN-γ gene expression. Also, TGF-β and IL-10, which are not inflammatory cytokines but do play key roles in Chagas disease, were induced by rTcHMGB.
These preliminary results suggest that TcHMGB can act as an exogenous immune mediator that may be important for both the control of parasite replication as the pathogenesis of Chagas disease and can be envisioned as a pathogen associated molecular pattern (PAMP) partially overlapping in function with the host DAMPs.
| When an infection occurs, the innate immune cells recognize Pathogen Associated Molecular Patterns (PAMPs) through their Pattern Recognition Receptors. This triggers an inflammatory response intended to kill the foreign microbe. But inflammation can also be triggered by the recognition of endogenous molecules called “Danger (or Damage) Associated Molecular Patterns” (DAMPs) that are released by damaged or necrotic cells to “ring the alarm” of the immune system that repair is needed, so some of them are also known as “alarmins”. High Mobility group box 1 protein (HMGB1) is a prototypical alarmin molecule released by injured cells and it is also actively secreted by cells of the innate immune system in response to invasion as well as to sterile damage. Trypanosoma cruzi, the causal agent of Chagas Disease, has its own HMGB protein that we called TcHMGB. Using in vitro and in vivo experimental systems, we demonstrated for the first time that TcHMGB is able to mediate inflammation on mammalian cells, inducing the expression of both pro-inflammatory and anti-inflammatory cytokines. Our results suggest that the parasite´s protein could have a role in the immune response and the pathogenesis of Chagas disease, probably overlapping to some extent with the host cell DAMP molecules´ functions.
| Chagas disease is considered by the World Health Organization (WHO) as one of the neglected tropical diseases (NTD) causing illness and hampering economic development in poor populations. It is caused by the protozoan parasite Trypanosoma cruzi and, like other infectious diseases, it can be fatal. According to the latest reports of the WHO, it is estimated that nearly 6 to 7 million people are infected by T. cruzi worldwide. Most cases occur in Latin America where Chagas disease is endemic and the parasite is mainly transmitted by an insect vector. However, as a consequence of human migrations the distribution of many illnesses has been changing in the last decades and Chagas disease has been increasingly detected in the United States of America, Canada, and many European and Western Pacific countries (http://www.who.int/mediacentre/factsheets/fs340/en/). Humans usually acquire the infection when a triatomine insect releases metacyclic trypomastigotes with its feces after a blood meal. Metacyclic trypomastigotes can pass through the damaged skin or intact mucosa and, once inside the body, infect a variety of cells. Other routes of infection have been also described: oral, congenital, via blood transfusion or by organ transplantation. Inside the host cell, metacyclic trypomastigotes transform into amastigotes, the intracellular replicative form found in vertebrate hosts. After several cycles of binary division, amastigotes differentiate back to trypomastigotes which are released upon cellular lysis, invading nearby nucleated cells and being disseminated through the bloodstream to other organs and tissues. In most people, the infection has a self-limiting acute phase, which is usually asymptomatic. During this stage, parasites replicate into the cytoplasm of a variety of cell types including macrophages, muscle cells, adipocytes and cells of the central nervous system and they can be found in blood and tissues in high numbers [1,2]. This acute phase lasts about 50–60 days and is characterized by high parasitaemia and tissue parasitism and a strong activation of the innate immunity with the concomitant high plasma levels of inflammatory cytokines like Tumor Necrosis Factor-α (TNF-α), interleukin (IL)-12 and interferon-γ (IFN-γ) as well as nitrogen reactive intermediates. Also during the acute phase, B- and T- cell are activated leading to the establishment of the adaptive immune response. The immune response usually controls the T. cruzi infection but fails in the complete eradication of the parasite, so people remain infected for life establishing a dynamic equilibrium with the parasite in the chronic phase of the disease, where parasitaemia and tissue parasitism are very low [3,4]. Most chronic infected individuals remain asymptomatic, but some of them develop different complications after a decade or more [5]. About 20% to 30% of patients will experience chronic Chagasic myocarditis with sequelae including heart failure, arrhythmias, thromboembolism and eventually death. Another 15% to 20% will experience chronic digestive sequelae like megaesophagus and megacolon [6]. Is not yet fully understood why different patients develop different clinical forms of the disease ranging from asymptomatic to severe cardiac problems. Also it is noteworthy the high inflammatory response associated to a relative low parasite number during the chronic phase leading to the suggestion of an autoimmune component in the disease pathogenesis. Many questions are still unsolved, but it is well known that both parasite and host response to infection contribute to the pathogenesis of Chagas disease [7].
High Mobility Group B proteins (HMGBs) are highly abundant proteins that play important biological roles both inside and outside the cell. HMGBs are nuclear DNA binding proteins involved in chromatin remodeling and they are key players in the control of transcription, DNA replication, recombination and DNA repair [8,9]. Besides the nuclear functions of all HMGBs, HMGB1 of humans and other mammals has been largely studied because it is a well-recognized Damage Associated Molecular Pattern (DAMP) molecule that is secreted by immune cells or released by injured cells “alarming” the immune system to trigger an inflammatory response [10–12]. It has been implicated in the pathogenesis of several inflammatory disorders like sepsis, atherosclerosis and arthritis and also autoimmune diseases like systemic lupus erythematosus [13].
Proteins belonging to the HMGB family have been identified in wide range of organisms from yeast to human including several protozoan and helminth parasites [14–18]. We have previously characterized the DNA-binding functions of Trypanosoma cruzi HMGB (TcHMGB) suggesting it may be an important player in transcription control in this parasite [18]. In this report we evaluated the ability of TcHMGB to induce cytokine production in a first attempt to study its putative role as an immune-mediator in the pathogenesis of Chagas disease. Recombinant TcHMGB (rTcHMGB) was able to induce an inflammatory response in vitro and in vivo, evidenced by the production of Nitric Oxide and the induction of inflammatory cytokines like TNF-α, IL-1β and IFN-γ gene expression. Interestingly, Transforming growth factor-β (TGF-β) and IL-10, usually associated to the opposite effect (anti-inflammatory) and known to play key roles in chronic chagasic myocardiopathy, were also induced by rTcHMGB. Moreover, experimental T. cruzi infection in BALB/c mice, showed during acute infection numerous cardiomyocytes and macrophages heavily infected with amastigotes, which showed strong TcHMGB immunostaining. TcHMGB immunostaining was also observed in fibrin located intravascular and attached to the endocardium in coexistence with numerous inflammatory cells that showed strong immunostaining to TNF-α, IL-1β and IFN-γ. Thus, immunohistochemistry in mice hearts during acute experimental T. cruzi infection showed high production of TcHMGB by amastigotes that apparently is secreted and co-exist with inflammatory cells that are producing pro-inflammatory cytokines.
These results suggest that TcHMGB can act as an exogenous immune mediator that may be important for the pathogenesis of Chagas disease and can be envisioned as a pathogen associated molecular pattern (PAMP) partially overlapping in function with the host DAMPs.
All the animal work was done according to the guidelines of the Mexican constitution law NOM 062-200-1999, and approval of the Ethical Committee for Experimentation in Animals of the National Institute of Medical Sciences and Nutrition in Mexico (CINVA) (PAT1021), and all efforts were made to minimize suffering. All experiments were approved by the Institutional Animal Care and Use Committee of the School of Biochemical and Pharmaceutical Sciences, National University of Rosario (Argentina) (File 6060/227) and conducted according to specifications of the US National Institutes of Health guidelines for the care and use of laboratory animals.
Recombinant proteins expression in Escherichia coli and affinity-chromatography purification, particularly Glutathione S transferase (GST) fusion TcHMGB (rTcHMGB) and GST (rGST), have been previously optimized in our laboratory [18]. However, since the recombinant protein was expressed in E. coli, we had to rule out the possibility of endotoxin or lipopolysaccharide (LPS) contamination, which could mask the putative effect of TcHMGB as a mediator of inflammation. To achieve our purpose, we modified the previous protocol adding a Polymixin B incubation step, to bind contaminant endotoxin previous to rTcHMGB purification. The whole bacterial lysate was passed through a Polymixin B-affinity chromatography column (Detoxi-Gel Endotoxin Removing Gel, Thermo Scientific, Argentina) obtaining a “detoxyfied lysate”. Then, the rTcHMGB GST-fusion protein was purified with a Glutathione-agarose specific affinity chromatography column (Sigma Aldrich, Argentina). Recombinant GST was purified following the same protocol and this protein was used as negative control in the immune assays. The purity of the detoxyfied recombinant protein was determined by SDS–PAGE and Pierce LAL Chromogenic Endotoxin Quantitation Kit (Thermo Scientific, Argentina). Protein concentration was quantified using Pierce BCA Protein Assay kit (Thermo Scientific).
Rabbits were inoculated with recombinant TcHMGB protein to raise antibodies. Formal animal ethics approval was given for this work by the School of Biochemistry of Rosario National University, Argentina, Ethics Committee. Animals were housed and maintained according to the institution’s experimental guidelines for animal studies. Specific antibodies against the parasite HMGB protein were affinity purified from the antiserum by using recombinant proteins immobilized on an agarose matrix. Briefly, after binding the protein to the matrix, it was cross-linked with 1% formaldehyde in PBS for 15 min and extensively washed with at least 50 vol. of PBS. The antiserum was then passed three times through the column containing the immobilized protein and extensively washed with PBS to separate non-specific antibodies. Finally, anti-TcHMGB-specific antibodies were eluted with 0.1 M Triethylamine (pH 11.5) and immediately neutralized to pH 8 with HCl. Purified antibodies were then concentrated through filtration and conserved in 50% glycerol at -20°C. Purified antibody specificity was tested using immunoblotting assays.
Supernatants were obtained after 6 h of incubation in RPMI medium without FCS from: T. cruzi epimastigotes, trypomastigotes and infected Vero cells. Uninfected Vero cell cultures and recombinant TcHMGB protein were used as negative and positive controls respectively. At the beginning of the incubation period deacetylase inhibitors were added to half of the samples, Nicotinamide 100 μM (Sigma Aldrich), Sodium Butyrate 5 mM (Sigma Aldrich) and Trichostatin A 1 μM (Sigma Aldrich).
The supernatants were concentrated 20 times by ultrafiltration and 20 μl from each sample was blotted onto nitrocellulose filters in a slot blot devise (BioRad) following the manufacturer´s instructions. The blotted proteins were visualized with Ponceau S staining. The membranes were treated with 5% non-fat milk in PBS for 1 h and then incubated with specific antibodies diluted in PBS for 3 hs. The antibodies used were affinity-purified polyclonal rabbit anti-TcHMGB, monoclonal mouse anti-trypanosome a-tubulin clone TAT-1 (a gift from K. Gull, University of Oxford, UK) and rabbit polyclonal anti-SAPA (a gift from Dr. Oscar Campetella, IIB-INTECH, Argentina). Bound antibodies were detected using peroxidase-labelled anti-mouse or anti-rabbit IgGs (GE Healthcare, Buckinhamshire, UK) and ECL Prime (GE Healthcare) according to the manufacturer’s protocol.
Trypomastigotes and exponentially growing epimastigotes were incubated with deacetylase inhibitors Nicotinamide 100 μM (Sigma Aldrich), Sodium Butyrate 5 mM (Sigma Aldrich) and Trichostatin A 1 μM (Sigma Aldrich) for 6 hs. Then they were centrifuged, washed twice with PBS, settled on polylysine-coated coverslips, and fixed with 4% paraformaldehyde in PBS at room temperature for 20 min. The fixed parasites were washed with PBS and permeabilized with 0.2% Triton X-100 in PBS for 10 min. After washing with PBS, the parasites were incubated with rabbit anti-TcHMGB primary antibody diluted in 1% BSA in PBS for 3 h at room temperature. Nonbound antibodies were washed with 0.01% Tween 20 in PBS, and then the slides were incubated with fluorescence-conjugated anti-rabbit IgG (fluorescein; Jackson ImmunoResearch) and 2 μg ml-1 DAPI (4,6-diamidino-2-phenylindole) for 1 h. The slides were washed with 0.01% Tween 20 in PBS and finally mounted with VectaShield (Vector Laboratories). Images were acquired with a confocal Nikon Eclipse TE-2000-E2 microscope using Nikon EZ-C1 software or an epifluorescence Nikon Eclipse Ni-U microscope. Adobe Photoshop CS and ImageJ software were used to pseudocolor and process all images.
The murine macrophage cell line RAW 264.7 (ATCC TIB-71) was cultured in Dulbecco's Modified Eagle's Medium (DMEM, Invitrogen, Carlsbad, CA, USA), supplemented with 10% (v/v) Fetal Calf Serum (FCS, Natocor, Córdoba, Argentina) and 1% (v/v) of a mixture of antibiotics (10,000 units/ml penicillin and 10,000 g/ml streptomycin, Sigma, Argentina) in a humidified atmosphere containing 5% CO2 at 37°C.
Cells were then seeded in 24-well plates at a density of 4.5 × 105 cells/well and maintained in culture for 24 h in DMEM supplemented with 2% (v/v) FCS and 1% (v/v) antibiotics mixture. After 24 h culture, the medium was renewed with the addition of rTcHMGB (10 μg/ml), rGST (10 μg/ml) or LPS (100 ng/ml) in the corresponding wells. Only DMEM (supplemented with FCS and antibiotics) was added to non-treated control cells. After 3, 6 or 24 h culture supernatants were collected for Nitric Oxide (NO) production determination (see below). Cells were washed once in sterile Phosphate-buffered saline (PBS) and then collected with Versene Solution (Thermo Fisher Scientific, Life Technologies, Argentina) and transferred to a clean tube for RNA purification.
NO produced by RAW cells was determined by evaluating the nitrite content in the culture supernatants with the Griess reagent, a flourometric reagent used for the quantitative analysis of nitrite in solution [19]. Absorbance at 543 nm was determined, and NO concentration was calculated by comparison with Abs543nm of standard curve of NaNO2 solutions prepared in culture medium.
Cell viability after all the treatments was determined by the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltretazolium bromide (MTT) reduction assay. Briefly, RAW cells were incubated in 96 well plate in the presence of the different treatments (rTcHMGB, rGST, LPS or DMEM alone) for 24 h. Then 20 μl MTT solution (5mg/mL in PBS) were added to each well and incubated for 1 h at 37°C. After this incubation period, MTT solution was removed and precipitated formazan was solubilized in 200 μl dimethyl sulfoxide (DMSO). Optical density (OD) was quantified spectrophotometrically (measurement λ = 530 nm, reference λ = 630 nm) using a microplate reader LD-400 (Beckman Coulter, Brea, CA, USA). DMSO was used as blank. Each treatment was performed in triplicate.
Male BALB/c mice, 7 weeks old, were obtained from our own breeding facilities. Animals were feed ad libitum with a standard laboratory pellet diet and were allowed free access to water during treatment. Mice were randomly divided into four experimental groups. The different treatments were administered intraperitoneally (i.p.) in a unique dose in sterile saline solution as vehicle (rTcHMGB 1μg, rGST 50 μg, LPS 1μg), control group only received the vehicle. Groups of four animals were euthanized by ex-sanguination and under anesthesia with pentobarbital at 3, 6 and 24 h after the injection, the spleens were immediately removed, frozen in liquid nitrogen and maintained under -70°C until RNA extraction.
In order to detect by immunohistochemistry rTcHMGB and cytokines in cardiac lesions produced by T. cruzi infections, BALB/c mice were subcutaneously injected with 1000 viable trypomastigotes of the Tulahuen strain of T. cruzi. Animals had access to food and water ad libitum and were allocated in temperature-controlled rooms under light–dark 12-h cycles and handled according to institutional guidelines. After 14 (acute) or 100 (chronic) days of infection, mice were killed and hearts were removed and sliced transversally into three sections. Tissue samples were fixed in buffered formalin or embedded in Tissue-Tek (Miles Inc., Elkhart, USA) and frozen in liquid nitrogen.
For the histological study, the heart samples from infected mice were fixed by immersion in 10% formaldehyde dissolved in PBS. Transversal heart sections were dehydrated and embedded in paraffin (Oxford Labware, St Louis, MO, USA), sectioned and stained with hematoxylin and eosin (HE). The local expression of TcHMGB and the cytokines TNFα, IFN-γ, IL-1 and IL-10 was determined by immunohistochemistry (IHC). Tissue sections 5 μm-width were obtained and placed on slides loaded with poly l-lysine (Biocare Medical, Lake Concord, CA, USA). For dewaxing, the slides were placed at 60–70°C for 20 min and incubated for 5 min into xylene. The slides were changed five times into the medium in the following sequence: (i) xylene–alcohol (1:1), (ii) absolute alcohol, (iii) alcohol 96%, and (iv) distilled H2O. Once hydrated, endogenous peroxidase was blocked with methanol–10% H2O2. The washings were performed with HEPES-buffered saline (HBS)–Tween 20 (10 mM HEPES, 150 mM NaCl, 2 mM CaCl2, 0.05% Tween 20). The areas of tissue were delineated and then blocked with 100 μl of HBS with 2% Background Sniper (Biocare Medical), and incubated for 30 min in a humid chamber. Then the sections were incubated with appropriate dilutions of rabbit polyclonal specific antibodies to TcHMGB, TNF-α, IFN-γ, IL-1 and IL-10 (Santa Cruz Biotechnology, Santa Cruz, CA, USA) overnight at room temperature. Subsequently, slides were washed and 100 μl of goat anti-rabbit antibody-horseradish peroxidase (AB/HRP) (Vectastain ABC System, Burlingame, CA, USA) was added and incubated for 30 min to be revealed with 100 μl of diaminobenzidine/ H2O2 (0.004 g diaminobenzidine + 10 mL HBS + 4 mL H2O2). Slides were washed and contrasted with hematoxylin.
After treatment, total RNA was isolated from RAW 264.7 cells or mouse spleens using TRI Reagent RT (Molecular Research Center Inc., Cincinnati, USA) according to manufacturer's instructions. Quality and quantity of RNA were evaluated through agarose gel electrophoresis and spectrophotometry (Abs260nm/280nm), respectively. Samples were treated with RQ1 RNase-free DNase (Promega, Wisconsin, USA) to remove possible DNA contamination prior to Reverse Transcription (RT). Five micrograms of RNA were used in the RT reaction with oligo (dT) and the Omniscript kit (Qiagen, Inc.) according to manufacturer's instructions to obtain the cDNA. Real time PCR reactions were carried out on a StepOne (Applied Biosystem, ThermoFisher Scientific, Foster city, CA, USA) with Power SYBR Green PCR Master Mix (Invitrogen, Carlsbad, CA, USA) using specific primers (Table 1).
Results for cytokine mRNAs were normalized to the 60S acidic ribosomal protein P0 (RPLP0) mRNA as housekeeping gene, based on the standard curve quantitative method [20]. The specificity of each reaction was verified with a melting curve between 55°C and 95°C with continuous fluorescence measure.
For analytical evaluation of the real-time RT-PCR, gene specific PCR amplicons of each gen were prepared as standards, PCR products were analyzed by 1% agarose gel electrophoresis and purified (GFX PCR DNA and gel band Purification kit; GE Healthcare life sciences, Pittsburgh, PA, USA), cloned into pCR2.1TOPO (Invitrogen, Carlsbad, CA, USA) and sequenced. Concentration of the purified nucleic acid was calculated by measuring the absorbance at 260 nm and its corresponding concentration was converted into copies per microlitre by using the Avogadro constant (6.023 x 1023) and its molecular weight (number of bases of the PCR product plus number of bases of vector, multiplied by the average molecular weight of a pair of nucleic acids). Tenfold serial dilutions of the quantified construct solutions were kept in aliquots at -20°C and used throughout the study as external standards of known concentration for the real-time PCR reaction (range of the standards: 103–106 copies/μl). The calibration curve was created by plotting the threshold cycle (Ct) corresponding to each standard, versus the value of their corresponding log number of each gen concentration (expressed as copies/μl).
Statistical analysis was performed with GraphPad Prism 3.0 software (GraphPad Software, La Jolla, CA, USA) using the One-Way Analysis of variance (ANOVA) followed by Dunnett´s post-hoc test. Data are expressed as the fold change of the control group, presented as mean ± standard error of the mean (SEM). Significance was set at p<0.05.
Under normal conditions, TcHMGB is located in the nucleus in all T. cruzi life cycle stages [18]. Like other HMGB family members, TcHMGB lacks a leader sequence that would enable its secretion by the classical route, so we decided to investigate if acetylation can promote its redirection to the cytoplasm and eventual secretion, as already described for mammalian and Schistosoma HMGB proteins [21,22]. We treated the parasites with a combination of deacetylase inhibitors (DACis), namely sodium butyrate, Trichostatin A and Nicotinamide, which target different deacetylases, and then analyzed the effect over protein localization on epimastigote and trypomastigote forms of the parasite by immunofluorescence with specific anti-TcHMGB antibodies. As can be seen in Fig 1A, we observed a relocalization of TcHMGB from the nucleus to the cytoplasm upon DACi treatment, suggesting that acetylation promotes TcHMGB translocation to the cytoplasm.
In order to determine if TcHMGB can be secreted extracellularly, we tested the presence of TcHMGB in supernatants from T. cruzi-infected Vero cells as well as from trypomastigotes and epimastigotes cell-free media. Also, taking into account the relocalization observed after DACi treatment (Fig 1A), we performed the secretion assay comparing DACi-treated and non-treated parasites or infected cells. We observed an immunoreactive band corresponding to TcHMGB in epimastigote, trypomastigote and infected cells supernatants, but not in uninfected Vero cells supernatant (Fig 1B). As expected, the DACi treated samples, show a more intense band, consistent with the acetylation-induced secretion hypothesis. However, the presence of TcHMGB in the supernatants of non-treated parasites and infected cells suggests that acetylation may stimulate TcHMGB secretion, but it is probably not the only mechanism. Anti-tubulin gave no signal when incubated with the same supernatants, demonstrating that the TcHMGB immunoreactive band is due to the presence of the protein in the extracellular medium and not a consequence of parasite lysis. Also, the shed acute phase antigen (SAPA) was detected in the trypomastigote and infected cells supernatants, confirming the presence of a well-known secreted protein in the extracellular samples (S1 Fig) [23].
In order to analyze if TcHMGB has pro-inflammatory properties like other HMGB family members, in vitro cultured RAW 264.7 cells were treated with rTcHMGB for 3, 6 and 24 hours. Nitric Oxide production by RAW cells was determined by quantification of nitrite in the culture supernatants by the Griess reaction, which is usually used as an estimation of NO production as a consequence of macrophage activation [19]. Also, cells were collected and the expression of different cytokines was determined by quantitative RT-PCR (qRT-PCR). LPS-treatment was used as pro-inflammatory control whereas, as negative controls, we used both non treated cells and recombinant GST-treatment. None of the treatments affected cellular viability as determined by MTT reduction assay (S1 Table).
In general, poor effect was noted at shorter times (3 and 6 hours), but at 24 hours post-treatment, RAW cells treated with rTcHMGB and with LPS were clearly activated, showing the typical “star-like” activated macrophage morphology (Fig 2A) and an increase in the NO production compared to negative controls (Fig 2B).
Recombinant TcHMGB stimulation also induced TNF-α and IL-1β expression (Fig 3A and 3B respectively), to a lesser extent than LPS-treatment but higher than non-treated or GST-treated cells. TNF-α expression was detected on macrophages only at 24 hours post-induction with rTcHMGB but not at shorter times. In contrast, IL-1β expression was triggered much earlier. At 3 hours of rTcHMGB-treatment, we observed an important increase in IL-1β expression that slowly decreases as treatment time elapses, but staying above the negative controls expression levels during the whole duration of the experiment.
Notably, also gene expression of TGF-β and IL-10, which are normally associated to an anti-inflammatory effect, increased after rTcHMGB treatment (Fig 3C and 3D). TGF-β expression showed to be slightly higher in rTcHMGB-treated cells than in control cells at 3 h post-treatment and continued increasing with time reaching the maximum at 24 h. On the other hand, IL-10 expression was raised at 3h but decreased at longer times.
Notably, for all the cytokines analyzed, the expression pattern was similar after TcHMGB- and LPS-treatment, although in most cases, LPS induced more radical changes. This data show that, like LPS, TcHMGB can act as an inflammatory mediator in vitro.
The activity of TcHMGB as a potential mediator of the immune response was also tested in an in vivo system. We performed a similar experiment treating BALB/c mice with rTcHMGB and pro- and anti-inflammatory cytokines expression was evaluated by qRT-PCR from spleen samples at 3, 6 or 24 hours post-inoculation (p.i.). Interestingly, in this in vivo test we obtained similar results to the previous in vitro assays (Fig 4). Both TNF-α and IL-1β were induced after treatment with rTcHMGB. TNF-α induction was significantly higher in rTcHMGB-treated mice at 6 hours p.i., and continued increasing for at least 24 hours (Fig 4A). Like in the in vitro assay, IL-1β expression was induced in rTcHMGB-treated mice earlier than TNF-α, being significantly higher than the negative controls at 3 hours p.i., but then decreasing to reach levels similar to those of negative controls (Fig 4B). In this in vivo system, we also evaluated the expression of IFN-γ, which is a key cytokine in the pathogenesis of Chagas disease (but it is not expressed by macrophages). As can be seen in Fig 4C, IFN-γ expression clearly increased at 24 h p.i., reaching levels comparable to those of LPS-treatment.
As well as in the in vitro assay, mice treated with rTcHMGB showed increased TGF-β and IL-10 expression (Fig 4D and 4E). In this case, both TGF-β and IL-10 show a statistically significant difference from controls at 6 h, that become greater at 24 h p.i.
These results suggest that TcHMGB not only can induce an inflammatory response on the host, but also it may participate in the consequent control of inflammation, which can be related to the parasite need to persist in the host.
We finally performed a histological study on an experimental murine model of acute and chronic Chagas disease to support our observations. Small nodules of chronic inflammation, randomly distributed in the myocardium and epicardium layers (Fig 5A), characterized early cardiac lesions induced by T. cruzi in mice. These lesions showed numerous spherules that exhibited strong TcHMGB immunostaining located in the cytoplasm of macrophages or outside cells which correspond to T. cruzi amastigotes (Fig 5B). Indeed, these polyclonal rabbit antibodies were highly specific to TcHMGB permitting the identification of numerous cardiomyocytes infected with T. cruzi with minimal or absent inflammation (Fig 5C and 5D). Interestingly, some areas of the endocardium were covered by macrophages infected with strong TcHMGB-immunostained amastigotes that also showed a diffuse cytoplasmic TcHMGB immunostaining (Fig 5E, black asterisks). These infected cells were surrounded by granular or fibrilar material that also showed strong TcHMGB immunostaining (Fig 5E, arrow). These results suggest that TcHMGB protein can be released from amastigotes and can reach the cytoplasm of the infected cells as well as the extracellular space in infected tissues. Some small blood vessels also showed fibrillary or granular material that exhibited strong TcHMGB immunoreactivity (Fig 5F), which apparently corresponded to fibrin covered or associated to TcHMGB. Some monocytes or macrophages around this material inside blood vessel or around blood vessels wall, showed positive immunostaining to TNF-α, IL-1β and IFN-γ (Fig 5G, 5H and 5I). This observation is consistent with the above described results with rTcHMGB-stimulated mice, and suggest that during acute infection, T. cruzi could liberate TcHMGB, which in turn induce cytokines production. Interestingly, chronic Chagas lesions that were characterized by extensive inflammatory infiltrate, cardiomyocytes fragmentation and necrosis with calcification (Fig 5J and 5K) did not show TcHMGB staining (Fig 5L). Thus, it seems that only during early infection T. cruzi releases TcHMGB, which can contribute to induce inflammation.
Using in vitro and in vivo experimental systems, we demonstrated for the first time that TcHMGB, the trypanosome homolog of the prototypic mammalian DAMP molecule HMGB, can be released and it is able to mediate inflammation on host cells, suggesting that the parasite´s protein may have a role in the immune response or the pathogenesis of Chagas disease.
The sequence analysis of T. cruzi HMGB highlights interesting similarities and differences with its counterpart from mammalian hosts´ and other HMGB family members (S2 Fig). HMGBs from mammals and higher eukaryotes have two HMG-box domains in tandem named “A-box” and “B-box”, followed by a negative charged domain composed of about 30 glutamate and aspartate residues usually termed “C-terminal acidic tail”. In contrast, most unicellular eukaryotes have HMGB proteins with only one HMG-box, like yeast Nhp6 (ScNhp6A, ScNhp6B) [24], Plasmodium HMGBs (PfHMGB-1, PfHMGB-2) [14], Toxoplasma HMGB (TgHMGB1a) [25] or Entamoeba HMGB (EhHMGB) [17]. Interestingly, Trypanosoma cruzi and other kinetoplastid HMGB proteins have two HMG-box domains like mice and humans (TcHMGB, LmHMGB, TbTDP1) [18,26]. The C-terminal acidic tail typical from metazoan HMGBs, which can regulate the DNA binding properties of the HMGB-box domains [27,28], is absent from most unicellular organisms included Trypanosoma brucei and T. cruzi, but, the putative HMGB from the related kinetoplastid Leishmania (LmHMGB), bears a shorter C-terminal sequence rich in serine and acidic aminoacids quite similar to that from other parasites like Entamoeba and Schistosoma (EhHMGB, SmHMGB1) [15,17,29]. Probably the most striking difference that stands out in the alignment from S2 Fig is the presence of the additional N-terminal 110 aminoacid-long sequence present in kinetoplastid HMGBs, but absent from all other HMGB family members. This sequence is highly conserved between trypanosomatids and contains a nuclear localization signal (NLS) predicted by bioinformatic methods (S2 Fig, black box) and a divergent “DEK-C terminal” DNA-binding domain that seems to be an additional point of interaction with DNA (S2 Fig, green box). Our previous results showed that besides the differences, TcHMGB keeps the DNA-architectural functions typical of the HMGBs [18].
HMGBs are known to be subjected to different post translational modifications (PTMs), which impact not only on their activities but also in their localization [30–32]. Mammalian HMGB1 has a bipartite nuclear localization signal (NLS) that directs the protein to the nucleus, where the protein plays important functions (S2 Fig, grey boxes). When macrophages are activated with the concomitant hyperacetylation of lysines present in these NLSs, the protein translocates to the cytoplasm and from there to the extracellular milieu through a non-conventional secretory pathway [21]. None of the mammalian NLSs are conserved in kinetoplastids, however, as already mentioned, trypanosomal HMGB has a putative NLS at the end of the N-terminal region specific of these organisms [18]. Results from our lab suggest that this NLS is functional and necessary to direct the protein to the nucleus, since expression of a truncated form of the protein lacking the 110 N-terminal sequence in T. cruzi epimastigotes showed the protein distributed along the cytoplasm (unpublished results). The predicted NLS also has five lysine residues that can be subjected to acetylation and probably regulate the nucleus-cytoplasm shuttle and eventual secretion, similar to what happens with HMGBs of mammals or Schistosoma [21,22]. Even though mutational analysis should be performed to confirm if these lysine residues are indeed acetylated to direct the protein out of the nucleus, our results show that acetylation can alter TcHMGB localization in the different life cycle stages of the parasite. Deacetylase inhibitors treatment of infected cells or free epimastigotes and trypomastigotes, which induces the hyperacetylation of most acetylable proteins, caused the relocalization of TcHMGB to the cytoplasm and an increase in TcHMGB secreted to the extracellular medium. However, the fact that TcHMGB could be detected in DACi non-treated supernatants of infected cells or free trypanosomes, suggests that other PTM or signals could be responsible for the protein release. Our immunohistochemistry results in experimental acute myocarditis showing strong TcHMGB immunostaining in intracellular amastigotes and fibrin located intravascular or deposited on the endocardium surface, suggest active expression of TcHMGB by the parasite and exportation, which could contribute to induce cytokines expression considering the positive immunostaining of pro-inflammatory cytokines by inflammatory cells near to extracellular TcHMGB deposits.
Since its first description as a late mediator of endotoxin lethality in mice by Wang et al.[33], our knowledge regarding HMGB1 functions as an immune mediator has increased exponentially. Human HMGB1 is released by dying cells or it can be actively secreted from immune cells like macrophages or monocytes after stimulation with lipopolysaccharide (LPS), cytokines or nitric oxide through a non-classical secretory pathway. Outside the cell, it can promote inflammation in different ways: (1) it can act as a chemotactic agent recruiting leukocytes to the site of danger, (2) it can act as a DAMP stimulating innate immune cells via pattern recognition receptors (PRR), like the Receptor for Advanced Glycation End products (RAGE) and Toll-like receptor 4 (TLR4), and (3) it can also act in association to cytokines and other molecules (LPS, lipoteichoic acid, IL-1β, chemokine CXC receptor 4 (CXCR4), RNA and DNA) binding to its major transmembrane receptors and amplifying the response to PAMPs [32,34,35].The pro-inflammatory activity of Human HMGB1 has been mapped to a highly conserved 20-aminoacid-long sequence (residues 89 to 108) located in the B-box which corresponds to the TLR4 binding site (S2 Fig, blue-box) [34,36]. Concerning trypanosomatid HMGBs, most residues in this region are either identical or conservative changes relative to metazoan HMGBs, but there are some important substitutions in this region that may produce structural changes in the protein like the three proline residues 91, 97 and 98 from human HMGB1. The RAGE-binding site, which is also critical for the HMGB1-mediated cytokine induction, is located in aminoacids 150–183 in the human protein, and it is also very conserved in metazoan HMGBs (S2 Fig, purple-box). Unlike the TLR4 binding site, this region is less conserved in other HMGB family members. In the case of T. cruzi, only four aminoacids of this region are identical to the mammalian homolog. Another outstanding difference is the substitution of cysteine 106 for a valine in T. cruzi, a serine in T. brucei and an alanine in Leishmania HMGB. These changes are surprising because cysteine 106 is very important in mammals since the protein function depends on its redox state. Three forms of human HMGB1 have been described according to the redox state of cysteine residues 23, 45 and 106 (S2 Fig, black arrowheads). When all three cysteines are in the reduced state (Thiol HMGB1), HMGB1 forms a complex with the chemokine stromal cell-derived factor (SDF-1/CXCL12) and binds to the CXCR4 receptor, increasing SDF-1/CXCL12 chemoattractant potential [37,38]. Cysteine residues 23 and 45 can form a disulphide bond, that can be accompanied by Cys106 in thiol state (disulphide HMGB1), which is the HMGB1 form that binds TLR4 leading to NFκB activation and the consequent transcription of genes involved in inflammation included those of many cytokines and chemokines [39]. Finally, the oxidized HMGB1 (cysteines oxidized to sulfonate) loss both chemoattractant and inflammatory properties and can even induce immunosuppression by the recruitment and activation of T regulatory cells [11,40,41]. These cysteine residues are conserved in mammals, birds and some worms. In contrast, none of them are present in unicellular organisms included Plasmodium, Toxoplasma, Trypanosoma and Leishmania. Besides the differences at the sequence level, particularly regarding the cysteine residues that seem to be essential for determining the final function out of the cell, both Toxoplasma and Plasmodium HMGB proteins can induce TNF-α production by macrophages in vitro and in vivo [14,25]. These observations led us to investigate the effect of TcHMGB over the inflammatory process to study its putative role on anti-trypanosome immunity. Moreover, taking into account that both host and parasite factors are thought to be responsible for the Chagas disease pathogenesis which has an important immune component, it would be expected that the parasite homolog of HMGB alarmin protein, plays a role in the disease pathogenesis. We observed that the TcHMGB protein is able to activate cultured macrophages in the classical way, leading to the production of NO and the pro-inflammatory cytokines IL-1β and TNFα. The protein also induced splenocytes to produce these cytokines as well as IFN-γ, another key inflammatory mediator, after intraperitoneal administration in mice. Several parasite antigens are able to induce classical macrophage activation and the consequent increase in NO and pro-inflammatory cytokines production. This is the case of antigens like glycophosphatidylinositol-anchored mucin-like glycoproteins (GPI) [42], the TolA-like surface protein from trypomastigotes´ flagella [43] or Tc-52 which synergizes with IFN-γ to stimulate NO production signaling via TLR2 and conferring resistance against lethal infection in BALB/c mice [44,45]. The production of NO and pro-inflammatory cytokines like IL-1β, TNF-α and IFN-γ is critical for destroying intracellular microorganisms, included the protozoan parasite T. cruzi. However, in the case of Chagas disease, even though the parasite´s replication is controlled through pro-inflammatory cytokines and microbicidal mediators released by cells of the innate immunity and the subsequent lymphocyte subsets activated during the adaptive immunity response, not all parasites are killed and the infection persists in the host for life. There are several proposed mechanisms by which T. cruzi escapes from the immune attack to reach persistence. One of them is the poor PRR signaling because of the absence of strong PAMPs capable of activating an efficient innate immune response [1]. Kurup and Tarleton suggested that an optimal anti-pathogen immunity requires not only effective PAMPs but also the continuous expression of these PAMPs, allowing the enhancement of both innate and adaptive immune responses resulting in the final clearance of the pathogen [46]. In our study, we compared the rTcHMGB stimulation effect with that of the prototypical DAMP molecule: the LPS from Gram negative bacteria. In Fig 4A and 4B, we can see that TNF-α and IL-1β gene expressions follow similar patterns in TcHMGB- and LPS-treated macrophages, although the LPS effect is stronger. This suggests that TcHMGB could be considered a putative PAMP molecule. In the in vivo study, although both TcHMGB and LPS induced pro-inflammatory cytokines expression, the kinetics seem to be different, what is not surprising because of the complexity of the in vivo model compared to the cultured cell line and the different receptors and/or cells that the two molecules could target. Our results suggest that not only structural or secreted antigens from T. cruzi, but also DAMP-like nuclear proteins, are able to induce inflammation and immune cells activation. Our hypothesis is that the trypanosome HMGB protein may be able to recognize and bind PRRs of the host, thus triggering the inflammatory response probably overlapping, at least to some extent, with the host cell DAMP molecules´ functions. This hypothesis is supported by the high conservation of the TLR4-binding region (Fig 1), although further experimental evidence is needed to confirm the protein binding to this receptor. Other PRRs could also be recognized by TcHMGB contributing to the immune response. Besides the differences at the sequence level, particularly regarding the cysteine residues that seem to be essential for determining the final function out of the cell, TcHMGB, as well as its orthologs from other unicellular parasites like Toxoplasma and Plasmodium, can induce inflammatory cytokines production in vitro and in vivo. Thus, these unicellular parasites´ HMGBs may have a different way of action that might be independent of the protein redox-state, they could be influenced or regulated by different PTMs and probably they could even involve the activation of other receptors. Khan et al. described that the intracellular PRR TLR9, is activated by Leishmania and Trypanosoma DNA, and the inflammatory response triggered by TLR9 activation is enhanced when the parasite DNA is complexed with the host HMGB1 protein [47]. In this regard, it would be interesting to evaluate TLR9 as another candidate PRR, which could be activated by the parasite TcHMGB-DNA complex. Moreover, it seems likely that DNA-bound TcHMGB may have increased immunostimulatory properties, similar to Toxoplasma gondii homolog, where TgHMGB1a-induced TNF-α secretion by macrophages is partially dependent on the bound DNA [25]. Our previous report showing that TcHMGB is a nuclear protein capable of interacting with DNA [18] is consistent with this hypothesis.
Every inflammatory response is usually counteracted by an opposite anti-inflammatory response mediated by IL-10 and TGF-β that modulates the final effect, thus preventing from an excessive inflammation that would cause severe injury and even death. Indeed, IL-10 and TGF-β play critical roles in regulation of host immune response to T. cruzi. Specific parasite molecules have shown to induce secretion of these anti-inflammatory cytokines, which have a beneficial effect for the parasite. The major T.cruzi cysteine proteinase Cruzipain (Cz), for example, induces IL-10 and TGF-β secretion as well as arginase expression by macrophages, leading to alternative macrophage activation and allowing an increased intracellular replication of the parasite [48]. We also analyzed the expression of these cytokines, after rTcHMGB treatment. Interestingly, both TGF-β and IL-10 were induced by rTcHMGB. This observation may seem contradictory with the pro-inflammatory properties proposed for TcHMGB. However, HMGB proteins have shown to be very versatile molecules with functions ranging from chromatin architectural factors involved in transcription control, DNA replication and repair inside the cellular nucleus to a plethora of regulatory functions included cellular maturation, proliferation, motility, inflammation, survival and cell death, upon interaction with a large set of receptors out of the cell [32]. Thus, it could seem contradictory, but not surprising, the fact that TcHMGB -the parasite ortholog of HMGB1- can induce opposing effects on the immune system. As already mentioned, although human and mouse HMGB1 proteins are most frequently associated with inflammation, it is very well documented that radical changes in the protein extracellular functions occur as a consequence of PTMs and the redox state of the protein, leading not only to the lack of inflammatory induction but even to immunosuppression [11,40,41]. Indeed, some authors suggest that HMGB1´s primary role in the setting of chronic inflammation is to promote immunosuppression [49]. Anti-inflammatory or immunosuppression effects of either endogenous or therapeutically administered HMGB1 have been described in different in vivo systems. For example, the systemic administration of HMGB1 suppressed skin inflammation by inducing an accumulation of PDGFRα+ mesenchymal cells from bone marrow [50]. Moreover, changing effects of HMGB1 have been associated to different stages of a tuberculosis experimental model where at day 7 to 21 the oxidized HMGB1 was predominant, while during late infection only the reduced form was seen. Thus, liberated HMGB1 during experimental tuberculosis can promote or suppress the immune response and inflammation depending on its redox state [51]. Although the oxidized cysteines from HMGB1 are not conserved in T. cruzi homolog, the apparent dual effect of TcHMGB regarding its inflammatory properties, may be regulated by another kind of PTM, that deserves to be studied. Our results are solid in showing the host cells cytokine production could be influenced by the presence of a parasite HMGB protein, however, the simplification of the in vitro and in vivo models used here compared to a real Chagas disease case, could explain why TcHMGB seems to have such opposite effects. Indeed, no positive correlation between the inflammatory and anti-inflammatory cytokines expression was observed in most cases, suggesting that no double (opposite) effect is occurring in the same animal at the same time (S2 Table). Additionally, we observed that TcHMGB coincides with inflammatory TNF-α, IL-1β and IFN-γ but not with TGF-β or IL-10 in acute Chagas heart histological samples. A more detailed study of TcHMGB expression and release kinetics during T. cruzi infection would help to understand this apparent contradictory effect. Moreover, the observation that TcHMGB immunostaining was not detected in chronic heart tissues, in contrast to the strong signal in the acute samples, suggests that TcHMGB release could be differentially regulated in the different stages of Chagas Disease.
Besides the pivotal functions that TcHMGB may have, that most probably can be regulated in vivo by PTMs, subcellular location, parasite life cycle stage, interacting partners and disease evolution, it is important to note that TGF-β is not just an anti-inflammatory cytokine. Transforming growth factor beta is actually a pleiotropic cytokine that controls various biological processes including inflammation, fibrosis, immune suppression, cell proliferation, cell differentiation and apoptosis [52]. Regarding Chagas disease, TGF-β plays important roles at different time points from the invasion of host cells to the establishment of the chronic chagasic myocardiopathy. TGF-β secretion is activated by the parasite and is required for the invasion process. It is captured by the intracellular amastigotes and controls their proliferation and their subsequent differentiation into trypomastigotes or eventual death through apoptosis inside the host cells. Moreover, TGF-β is a key player in the development of chagasic myocardiopathy where it participates in the extracellular matrix protein production and consequent fibrosis as well as in the modulation of cardiomyocyte proliferation and death [53,54]. Fibrosis in the heart has been associated to TGF-β ability to induce expression of matrix components, to inhibit the secretion of several matrix-degrading proteases and to stimulate the synthesis of protease inhibitors. Also, extracellular matrix production by fibroblasts can be stimulated by TGF-β [55]. A correlation between increased fibrosis and increased activation of the Smad-2 pathway, through which gene expression of TGF-β-targeted genes is activated, has been documented both in Chagas patients and in an experimental model [56,57]. In this context, we can speculate that the parasite could induce TcHMGB-driven TGF-β and IL-10 secretion for its own benefit interfering with macrophage microbicide activity and probably facilitating the parasite persistence, and that this cytokines production could also have a role in the chronic phase of Chagas disease.
To our knowledge, this is the first report showing that Trypanosoma cruzi HMGB can be released in infected tissues and act as a mediator of the immune response in mammals. The results presented herein suggest that the contribution of TcHMGB to the protective immune response and immunopathology in Chagas disease should be significant and consequently it deserves to be studied in detail in the future. Moreover, TcHMGB share characteristics with other HMGB family members but it also seems to have some unique features and functions, that may be conserved in other trypanosomatids, so it would be interesting to evaluate if these results could be extrapolated to other kinetoplastid-diseases like Leishmaniasis or Sleeping sickness.
|
10.1371/journal.pntd.0005885 | Modeling the environmental suitability of anthrax in Ghana and estimating populations at risk: Implications for vaccination and control | Anthrax is hyper-endemic in West Africa. Despite the effectiveness of livestock vaccines in controlling anthrax, underreporting, logistics, and limited resources makes implementing vaccination campaigns difficult. To better understand the geographic limits of anthrax, elucidate environmental factors related to its occurrence, and identify human and livestock populations at risk, we developed predictive models of the environmental suitability of anthrax in Ghana. We obtained data on the location and date of livestock anthrax from veterinary and outbreak response records in Ghana during 2005–2016, as well as livestock vaccination registers and population estimates of characteristically high-risk groups. To predict the environmental suitability of anthrax, we used an ensemble of random forest (RF) models built using a combination of climatic and environmental factors. From 2005 through the first six months of 2016, there were 67 anthrax outbreaks (851 cases) in livestock; outbreaks showed a seasonal peak during February through April and primarily involved cattle. There was a median of 19,709 vaccine doses [range: 0–175 thousand] administered annually. Results from the RF model suggest a marked ecological divide separating the broad areas of environmental suitability in northern Ghana from the southern part of the country. Increasing alkaline soil pH was associated with a higher probability of anthrax occurrence. We estimated 2.2 (95% CI: 2.0, 2.5) million livestock and 805 (95% CI: 519, 890) thousand low income rural livestock keepers were located in anthrax risk areas. Based on our estimates, the current anthrax vaccination efforts in Ghana cover a fraction of the livestock potentially at risk, thus control efforts should be focused on improving vaccine coverage among high risk groups.
| Anthrax is a soil-borne zoonotic disease found worldwide. In the West African nation of Ghana, anthrax outbreaks occur annually with a high burden to livestock keepers and their animals. To control anthrax in both humans and animals, annual livestock vaccination is recommended in endemic regions. However, in resource poor areas distributing and administering vaccine is difficult, in part, due to underreporting, logistical issues, limited resources, and an under appreciation of the geographic extent of anthrax risk zones. Our objective was to model high spatial resolution anthrax outbreak data, collected in Ghana, using a machine learning algorithm (random forest). To achieve this, we used a combination of climatic and environmental characteristics to predict the potential environmental suitability of anthrax, map its distribution, and identify livestock and human populations at risk. Results indicate a marked ecological divide separating the broad areas of environmental suitability in northern Ghana from the southern part of the country, which closely mirrors the ecotone transitions from southern tropical and deciduous forests to the northern Sudanian and Guinea Savanna. Based on our model prediction, we estimated >3 million combined ruminant livestock and low income livestock keepers are situated in anthrax risk zones. These findings suggest a low level of annual livestock vaccination coverage among high risk groups. Thus, integrating control strategies from both the veterinary and human health sectors are needed to improve surveillance and increase vaccine dissemination and adoption by rural livestock keepers in Ghana and the surrounding region.
| Anthrax is a soil-borne, zoonotic disease found on nearly every continent (except Antarctica) that primarily infects herbivorous animals while secondarily infecting humans through the handling or ingestion of contaminated meat or animal by-products [1,2]. The geographic distribution of the disease appears to be limited by a combination of climatic (e.g. precipitation and temperature) and environmental (e.g. alkaline soil pH) conditions [3,4]. Under the appropriate ecological conditions, which remain poorly understood, the causative agent of anthrax, Bacillus anthracis, can survive for long-periods of time in the environment, perhaps years [1,4]. Although it has received much attention as a potential agent of bioterrorism, the World Health Organization (WHO) has listed anthrax as a neglected disease [5]. Poor livestock keepers and their animals often experience a disproportionate burden of anthrax in the hyper-endemic regions of Central Asia and West Africa [5,6]. Despite the effectiveness of regular animal vaccination and proper outbreak response following recommended guidelines in controlling anthrax in humans, underreporting of the disease often skews its true burden and geographic distribution making it difficult to implement adequate vaccination campaigns [1,7].
In Ghana, anthrax outbreaks have been reported annually in humans associated with contact with infected livestock and their contaminated animal by-products (e.g. meat or hides) [8]. Anthrax vaccine is manufactured locally by the Central Veterinary Laboratory in Pong-Tamale, Ghana and is fully subsidized by the government. Despite this, animal outbreaks are documented annually, and primarily affect cattle. Although both human and animal cases are reported, few human cases are linked to confirmed animal cases [9]. As a result, surveillance data alone provide limited information to efficiently plan prevention activities. Previous efforts to elucidate the environmental suitability of anthrax in Africa have been focused on southern countries, such as Zimbabwe [10], or national parks [11]. A recent study from West Africa also used a machine learning algorithm to map and model the distribution of anthrax and B. anthracis in Cameroon, Chad, and Nigeria, however, that effort was based on limited sample size and no comparable efforts have been carried out in Ghana [12].
To support Ghana’s national anthrax control and assessment, we our study had the following objectives: (1) model the environmental suitability of anthrax; (2) identify environmental and climatic factors associated with the occurrence of anthrax; (3) describe seasonal patterns; and (4) estimate populations at risk.
This work was performed on nationally available data on anthrax outbreaks in livestock from the Ministry of Food and Agriculture in Ghana.
We constructed a GIS of livestock anthrax outbreaks using data collected by the Ghana Field Epidemiology and Laboratory Training Program (GFELTP) and the Ministry of Food and Agricultural Veterinary Services. (Fig 1). Outbreaks were mapped using GPS coordinates collected by field personnel responding to outbreaks or the center of the village where the outbreak occurred. We included data on outbreaks from 2005 through the first 6-months of 2016 included information on the geographic coordinates, date, livestock species, and number of individual animals infected (periodically recording mortality and survival status) for each outbreak. However, total livestock populations on affected properties was rarely reported. For this study, an outbreak was defined as any location with one or more anthrax cases in animals. We plotted the seasonality of anthrax outbreaks in relation to the average rainfall during 1991–2015 using data obtained from the Climate Change Knowledge Portal (http://sdwebx.worldbank.org/climateportal/index.cfm?page=country_historical_climate&ThisCCode=GHA). We also obtained livestock anthrax vaccine administration data during 2005–2015 from the World Animal health Information Database Interface (OIE; http://www.oie.int/animal-health-in-the-world/the-world-animal-health-information-system/data-after-2004-wahis-interface/). Mapping and spatial analysis was performed in Q-GIS version 2.14 (www.qgis.org) and the R statistical package (https://www.r-project.org/). Final maps were produced in ArcGIS version 10.3.1 (ESRI, Redlands, CA, USA).
We used a combination of environmental and climatic variables at a spatial resolution of 30-arcseconds (approximately 1km x 1km) that followed, in part, recent studies in West Africa [13] and Central Asia [14] (Table 1). Five “bioclimatic” variables describing measures of temperature and precipitation were obtained from the WorldClim database (www.worldclim.org) [15]. WorldClim variables are interpolated monthly measurements recorded at weather stations located worldwide between 1950 and 2000. WorldClim produces bioclimatic variable grids to describe annual trends, seasonality, and ecological parameters such as temperature of the coldest and warmest quarters. We also used a combination of physical (sand content), chemical (soil pH), and taxonomic classifications of soil characteristics (cancerous vertisols and humults). Soil data were obtained from the SoilGrids1km database http://www.isric.org/explore/soilgrids) [16]. SoilGrid variables were created using spatial model predictions based on a global database of soil profiles and a combination of environmental covariates. Furthermore, we used two normalized difference vegetation index (NDVI) variables describing average conditions and the amplitude of vegetation greenness, which were obtained from the Trypanosomiasis and Land Use in Africa (TALA) research group (Oxford, United Kingdom) [17]. TALA variables were derived from temporal Fourier analysed (TFA) time series data of advanced very-high resolution radiometer (AVHRR) satellite measurements taken between 1992 and 1996 [17]. Mapped variables are shown in S1 Fig.
Random Forest (RF) modeling [18,19] was used to identify environmental characteristics associated with the occurrence of anthrax outbreaks using the ‘randomForest’ package for R. Previous studies have used this approach to map and model the distribution of Anopheles spp. mosquito vectors in Africa and Europe [20] and reservoirs of avian influenza [21]. RF modeling has been described and compared to other modeling approaches in detail elsewhere [18,22]. Briefly, RF is a non-parametric method derived from classification and regression trees that consists of a combination of trees built using randomly selected bootstrap samples of the training data (used to build the model), with the number of bootstrap samples equal to the number of trees (ntrees) selected. Each tree is split by randomly sampling a number of predictor variables to use (mtry) at each node and then choosing the best split. Model error estimates are obtained by internal splits of the training data (63.2% for model building) and then predicting the data not used to build a tree (out-of-bag or OOB) and aggregating these predictions for each ensemble of trees [18]. Since internal validation of the OOB data is performed, no external testing data is required to validate the model, but testing splits (external data withheld from the model) of the data are routinely utilized to assess model performance. Partial dependence plots and variable importance of RF models were assessed for covariates in the model.
We used an ensemble modeling approach that incorporated information from multiple random splits of our data into training (80%) and testing (20%) sets. Since our data consisted of presence only records of anthrax outbreaks, we generated pseudo-absence data from all available background data. Several studies have either relied on internal derivations of pseudo-absence in species distribution models [23] or user-defined generations such as in the modeling of the global distribution of dengue virus [24]. The required number of user-defined background pseudo-absence draws for every presence location is not standardized. It has been suggested that a 1:1 random draw of pseudo-absence to presence data in machine learning algorithms such as RF produces optimal results [25], although variations of this (2:1 or 3:1 draws) have been adopted successfully [24]. Similarly, pseudo-absence data creation has been shown to influence results; thus, research has recommended filtering pseudo-absence data from locations that are known to fall within suitable habitat or that occur within a defined proximity threshold [25,26].
We first filtered geo-located anthrax presence data in Ghana (n = 61) using a 5km x 5km proximity threshold in order to improve model performance and avoid overfitting [27]. We generated background pseudo-absence data (n = 200), from all available background [24], at a ratio of four absence points to every one filtered presence point (n = 50), restricting pseudo-absence data to exclude landscape within 5km of presence locations. We then generated 10 random draws each of 1:1, 2:1, and 3:1 pseudo-absence to presence data (30 total draws) with replacement. Each randomly generated pseudo-absence to presence draw (n = 30) was randomly divided into training and testing data splits to validate model performance. The final RF models were built using a mtry = 4 at each split and ntrees = 1000 with a combination of variables in which the ensemble list contributed to a mean decrease in accuracy >1%. The 30 individual RF models were then combined into an ensemble prediction at a spatial resolution of ~1km x 1km and scaled from 0 (low suitability) to 1 (high suitability); uncertainty in the model prediction was calculated by taking the range in the 95% confidence intervals of the ensemble model scaled from 0 (low uncertainty) to 1 (high uncertainty) following Deribe et al. [28].
The resulting output of our ensemble RF model represents the environmental suitability of anthrax in Ghana. To estimate the number of livestock and poor rural livestock keepers at risk in anthrax suitable areas, we dichotomized the modeled environmental suitability into a suitable versus not suitable prediction using a probability threshold that maximized sensitivity and specificity. We then overlaid a database of global livestock density at a spatial resolution of ~1km x 1km (http://www.livestock.geo-wiki.org/) [29] with the dichotomized anthrax prediction to estimate the livestock populations (cattle, sheep, goats, and swine) at risk. Livestock populations at risk were further stratified to estimate the population at risk within each of the livestock production zones of Ghana using the livestock production systems data version 5 (http://www.livestock.geo-wiki.org/) [29–31]. Furthermore, we estimated the number of low income rural livestock keepers at risk within each livestock production zone by overlaying the dichotomized anthrax suitable areas with estimates of the population of low income rural livestock keepers provided in Robinson et al. [31] and deriving the fraction of cells that were within our model prediction. Uncertainty in the populations at risk and 95% confidence intervals were calculated by using the 2.5% (lower) and 97.5% (upper) bounds of the ensemble RF model prediction [28].
Model performance and validation was conducted for each individual RF model and included the internal: OOB error classification, area under the receiver operating characteristics curve (AUC), sensitivity, and specificity. Additionally, we performed accuracy assessments on the external testing data, which consisted of thirty random subsets of 20% of the data sampled with replacement. Mean values and 95% confidence intervals were estimated for each accuracy metric. The AUC has been used extensively in species distribution modeling to measure the discriminatory performance of models [32]; an AUC value of 1 indicates a perfect discrimination while values of >0.9 are outstanding, 0.8–0.9 excellent, 0.7–0.8 acceptable, and <0.7 indicate poor discriminatory performance [28,33].
From 2005 through the first 6 months of 2016, there were 67 reported anthrax outbreaks in livestock (61 that were geo-located) (Fig 1). Nationally, there was a mean of 6 (95% CI: 4, 7) outbreaks per year with a peak in 2011 (n = 12) and lull in reporting in 2009 (n = 2) (Fig 2). The geography of outbreaks shows a higher frequency of anthrax in northern Ghana in the Upper East and Northern regions. Of the reported outbreaks, 4 (6%) were comprised of two or more livestock types. Domestic cattle were reported in 53% (35) of outbreaks, followed by sheep in 32% (21), goats in 11% (7), and swine in 5% (3). During 2005–2016, cattle anthrax cases were reported every year except in 2009. Sheep cases were ubiquitous annually and were characterized by a large number of deaths in 2012, the same year there was also a large number of swine cases (n = 500) (Table 2).
The seasonality of anthrax outbreaks nationally and regionally are illustrated in Fig 3. Nationally, outbreaks were reported, on average, across seasons and in every month (except November). There was an increase in outbreaks in the late winter and early spring months, with February through April having the highest reported number of outbreaks. On average, there outbreaks appeared to occur in the dry season before the onset of the rains.
Trends in livestock anthrax vaccination among livestock type are shown in Fig 4. From 2000–20015, there was a median of 17,957 doses [0–175 thousand] of anthrax livestock vaccine administered annually livestock vaccination occurred annually with a median number of doses administered of 19,709 [range: 0–175 thousand doses], followed by a decline in vaccine administration during 2008–2015. No vaccination was administered during the years 2010, 2012, and 2013. During 2008–2015, there was a median of 542 [range: 0–147 thousand doses] doses administered. In response to ongoing outbreaks, there was a vaccination campaign in 2014 that resulted in nearly an 8-fold increase in the number of doses administered compared to the previous six years. Among livestock types, cattle were most frequently administered vaccine, followed by sheep, goats and swine (Fig 4).
The ensemble RF model suggests a latitudinal gradient in the environmental suitability of anthrax in Ghana (Fig 5A). High environmental suitability was identified in the Northern, Upper East, and Upper West regions of Ghana that encompass seasonal livestock migration routes from Burkina Faso in the north. Conversely, low or no environmental suitability was identified in southern Ghana among the more acidic soils in the Western, Ashanti, Central, and Eastern regions. Uncertainty (range: 0–0.20) in the model prediction was scaled from 0 to 1 and showed it was highest in the Upper West and Northern regions (Fig 5B). The internal OOB model validation indicated excellent discrimination with an AUC = 0.88 (95% CI: 0.87, 0.89). The external validation of anthrax outbreak locations withheld from the model (testing data) also showed excellent discrimination (AUC = 0.87 [95% CI: 0.85, 0.90]).
The final list of variables used in the ensemble model are shown in Fig 6. A combination of bioclimatic, environmental, and soil characteristics had the greatest impact on the OOB prediction errors. The most important variables influencing accuracy were: soil pH, bio7 (annual temperature range), and bio14 (precipitation of the driest month) (S2 Fig). The probability of the occurrence of anthrax increased in a step like manner in response to soil pH, increasing as the soil became more alkaline, between 5.5 and 6.5, and again between 6.5 and 7.0. Annual temperature ranges between 16 and 20°C were also related to a greater probability of occurrence. The occurrence of anthrax showed an affinity for low values of precipitation during the driest month (0 to 10 mm) and then dropped off dramatically as precipitation increased from 10 to 40 mm. Furthermore, as average NDVI (wd0114a0) increased from 0.3 to 0.6 the probability of anthrax occurrence decreased linearly, with a more suitable range of vegetation greenness identified in the lower ranges between 0.1 and 0.3 (Fig 6).
To estimate livestock and human populations at risk, we dichotomized the environmental suitability prediction (on a continuous probability scale) into suitable versus non-suitable environments for anthrax based on the optimal threshold (0.46) that maximized sensitivity (0.78) plus specificity (0.89) (Fig 7). The dichotomized prediction shows a marked north-south demarcation in the suitability of anthrax, with a majority of northern Ghana predicted as suitable within the accompanying upper (97.5%) and lower (2.5%) confidence bounds.
The national livestock population located in areas environmentally suitable for anthrax was estimated to be ≈ 2.2 (95% CI: 2.0, 2.5) million (Table 3). More than 50% of the livestock populations at risk were sheep and cattle (650 [95% CI: 583, 745] thousand and 480 [95% CI: 434, 527] thousand, respectively). Among livestock production systems, semi-arid rain-fed, mixed crop livestock systems (MRA) contained the greatest number of livestock at risk > 1.2 (95% CI: 1.1, 1.3) million (Table 3).
Nationally, there are approximately 3 million low income rural livestock keepers in Ghana (Table 4). Our model suggests that ≈ 805 (95% CI: 519, 890) thousand are located in areas suitable for anthrax, with the majority located in a humid and sub-humid, mixed crop livestock system production zone (MRH).
Anthrax is a globally distributed neglected disease that is often underreported, particularly in West Africa where it is hyper-endemic [1,2,6,13]. Given the reliance of control on the vaccination of livestock, understanding the occurrence of anthrax is crucial for identifying populations at risk in order to disseminate limited resources. Here, we used data on the location of livestock outbreaks to identify seasonal patterns and model the environmental suitability of anthrax in Ghana. In keeping with previous studies, our findings indicate a defined outbreak season with a combination of ecological constraints on the potential geographic distribution of anthrax [3,34]. Our modeled prediction suggests a marked ecological divide separating the broad areas of environmental suitability in northern Ghana from the southern part of the country. Additionally, we estimated that populations characteristically at high risk for anthrax, which included >3 million combined ruminant livestock and poor rural livestock keepers are situated within the predicted anthrax risk zone. Based on our estimates, current anthrax vaccination efforts cover only a fraction of the livestock potentially at risk. Hence, these findings can be used to better direct public health intervention strategies and inform surveillance.
Official reports of livestock anthrax in endemic areas often go undocumented for a number of reasons, including the inability or unwillingness to report, limited surveillance capacity, and a lack of local knowledge about the disease [1]. In Ghana, livestock cases are likely underreported due to the slaughter and consumption of sick or dead animals [8,35], consistent with findings in the Caucasus and elsewhere [1,6,36,37]. This practice is often undertaken as a means of recouping economic losses from livestock mortality as well as providing food and a readily available source of protein [1,8,35]. The livestock anthrax outbreak data we used in this study were concordant with data reported to OIE during the same time frame suggesting Veterinary Services in Ghana are compliant with international reporting requirements (http://www.oie.int/wahis_2/public/wahid.php/Wahidhome/Home).
Despite the close proximity to the equator, we identified marked seasonality in anthrax reporting; outbreaks increased during the onset of the rainy season from February through April. Similar patterns of anthrax outbreaks associated with the rainy-season have also been reported in Namibia [34]. One hypothesis suggests that there is greater soil consumption among ruminants during with the rainy season [34], although soil exposure during the dry season has also been hypothesized as a cause of anthrax outbreaks [1]. Regardless, these findings suggest vaccination of livestock could be carried out in Ghana ahead of the peak outbreak season (September–November).
Livestock anthrax control in Ghana follows a similar trend in many endemic regions of reactively vaccinating in response to anthrax outbreaks [1,38]. In Ghana, the livestock population we identified at risk comprises approximately ≈ 25% of the total national livestock population [29]. Based on official vaccination reports (Fig 4), our estimates of the livestock populations at risk indicates poor vaccine coverage; this finding is consistent with ongoing outbreaks in endemic communities in Ghana where vaccination has not been officially documented for at least a decade [39]. Barriers to vaccine uptake such as practices of livestock keepers my also affect coverage [1,40]. However, Ghana faces additional control challenges with the potential presence of B. cereus biovar (bv) anthracis and West Africa strains (D and E Clades, respectively [41]). The West African strains have been hypothesized to evade the Sterne vaccine, which is the vaccine used in Ghana and throughout much of the world [13,42]. Further research is needed on vaccine efficacy and to understand what proportion of anthrax outbreaks are due to either insufficient application methods or the vaccine itself.
Research has suggested that soil pH >6.1 in conjunction with high calcium levels are a crucial component of B. anthracis spore survival [1,4,43]. Alkaline soils were also found to be associated with the persistence of anthrax transmission over several years [43,44]. In keeping with these findings, we identified an increasingly higher likelihood of anthrax occurrence in soils as pH increased from 5.5 to 7.0 and with an increasing level of calcareous vertisols. The association of anthrax suitability with lower levels of precipitation in our model is in line with reports that have documented soil nutrient leaching in regions with high precipitation, which may lead to soil acidification [45]. We predicted an area of environmental suitability for anthrax that encompasses ≈ 36% of Ghana’s total area (Fig 7); this is demarcated by a south (largely unsuitable) to north (highly suitable) divide, which closely mirrors the ecotone transitions from southern tropical and deciduous forests to the northern Sudanian and Guinea Savanna.
Our study had several limitations. As with all neglected zoonoses, our data likely represent an underestimation of the true burden of disease due to underreporting and limited resources for surveillance and testing. To better address issues with diagnostic testing and reporting we used a more contemporary dataset of anthrax outbreaks recorded during the last decade. Anthrax can also be transmitted from contaminated feed that is imported, and animal mortality may occur from livestock moved across long distances; however, we had no information on any outbreaks arising in these instances [1,46]. The use of machine learning algorithms to model the distribution of environmental pathogens has been well described, but such approaches, by their definition in conjunction with the use of averaged climate data, may over-generalize the landscape that supports the occurrence of anthrax outbreaks. Other factors not included in our models that may influence the occurrence of anthrax include the health and immune status of the livestock [47].
In conclusion, the current anthrax situation in West Africa, and in particular Ghana, remains a public and veterinary health threat due to challenges with reporting, surveillance, and control. Our findings suggest that broad areas of northern Ghana are environmentally suitable for anthrax. Furthermore, based on recent vaccination efforts, our estimates indicate that only a fraction of livestock at risk are being vaccinated. These findings can be used to help improve differential diagnostics, vaccine coverage estimates, and surveillance efforts. Given the reliance on agriculture and the large population of low income rural livestock keepers at risk in the northern part of the country where predicted suitability was highest, future control efforts should focus on improving livestock vaccination coverage and public awareness of the disease, prioritizing communities in the predicted anthrax zone.
|
10.1371/journal.pbio.1002119 | Integration of Shallow Gradients of Shh and Netrin-1 Guides Commissural Axons | During nervous system development, gradients of Sonic Hedgehog (Shh) and Netrin-1 attract growth cones of commissural axons toward the floor plate of the embryonic spinal cord. Mice defective for either Shh or Netrin-1 signaling have commissural axon guidance defects, suggesting that both Shh and Netrin-1 are required for correct axon guidance. However, how Shh and Netrin-1 collaborate to guide axons is not known. We first quantified the steepness of the Shh gradient in the spinal cord and found that it is mostly very shallow. We then developed an in vitro microfluidic guidance assay to simulate these shallow gradients. We found that axons of dissociated commissural neurons respond to steep but not shallow gradients of Shh or Netrin-1. However, when we presented axons with combined Shh and Netrin-1 gradients, they had heightened sensitivity to the guidance cues, turning in response to shallower gradients that were unable to guide axons when only one cue was present. Furthermore, these shallow gradients polarized growth cone Src-family kinase (SFK) activity only when Shh and Netrin-1 were combined, indicating that SFKs can integrate the two guidance cues. Together, our results indicate that Shh and Netrin-1 synergize to enable growth cones to sense shallow gradients in regions of the spinal cord where the steepness of a single guidance cue is insufficient to guide axons, and we identify a novel type of synergy that occurs when the steepness (and not the concentration) of a guidance cue is limiting.
| During development of the nervous system, axons are propelled by the growth cone, a motile structure that is specialized to detect the direction of concentration gradients of guidance cues. Although it is known that commissural axons—those that cross the midline from one side of the nervous system to the other—of the spinal cord are guided by multiple cues simultaneously, it is unclear whether they integrate multiple guidance cues and, if that is the case, the advantage of doing so. In the developing spinal cord, the gradients of the guidance cues are shallow and, thus, their direction is difficult to determine. We hypothesize that under these circumstances, a combination of cues could be used synergistically by the growth cone. To test this hypothesis, we built a microfluidic gradient generator capable of simulating the shallow gradients that growth cones encounter in the developing spinal cord. Using this guidance assay, we demonstrated that commissural axons are best at reorienting themselves in the steepest part of a gradient of Shh or Netrin, two guidance cues that these axons encounter in the spinal cord. We then challenged axons with combined gradients of both cues. At low gradient steepness, we observed synergy in their turning response and in the asymmetry of the shared downstream signaling molecules Src-family kinases (SFKs). We therefore propose a model in which SFKs integrate distinct signaling pathways, and we define this as steepness-limited synergy.
| During embryogenesis, axons grow through a complex environment to make specific connections with their targets. The growth cone follows concentration gradients of guidance cues by sensing a difference in receptor occupancy across its width, and it turns to align with its interpretation of the gradient direction. Moreover, multiple guidance cues are often needed to correctly guide axons. For example, commissural axons are initially repelled by bone morphogenic proteins (BMPs) in the dorsal half of the spinal cord [1,2]. They are then attracted by gradients of Netrin-1 [3], Sonic hedgehog (Shh) [4] and vascular endothelial growth factor (VEGF) [5] towards the floor plate. While it isn't understood why multiple guidance cues are needed to guide axons to the same targets, it is clear they are non-redundant, as interfering with each of these pathways individually results in guidance errors [4–8].
Both Netrin-1 [9,10] and Shh [11,12] diffuse from the floor plate cells which secrete them and establish gradients which guide commissural axons [4,10]. Shh signals through its receptor Boc [8], while Netrin signals through its receptor DCC [7,13]. Shh- and Netrin-1-mediated axon guidance also both require Src-family kinase (SFK) activity [14,15], whose asymmetric activation reflects the direction of the external gradient and is sufficient to cause the growth cone to turn [15,16]. While it is known that both Shh and Netrin-1 form gradients, it is not clear how steep the gradients are in vivo and how this steepness influences axon pathfinding in gradients formed by single or multiple guidance cues. Although theoretical chemotaxis modeling has suggested that two overlapping attractive concentration gradients could increase the probability of a cell making a correct decision about the gradient direction [17], this prediction has not been tested experimentally.
There are several potential mechanisms by which multiple guidance cues could collaborate to improve how well the growth cone estimates the direction of the gradient. In one model, the concentration of individual guidance cues is too low to elicit a robust turning response. When the cues are combined, the response is higher than the sum of responses from the same concentration of either cue individually. We will refer to this as concentration-limited synergy, as the concentration of either guidance cue is limiting for the pathway to be engaged. When a second cue is present, there is some crosstalk or convergence between pathways, which overcomes the activation threshold. In an alternative mechanism, which we will refer to as steepness-limited synergy, the concentration of guidance cue present at the growth cone is not limiting; instead, it is the concentration difference of an individual guidance cue across the growth cone that is too small compared to the ambient guidance cue concentration to be accurately detected by the growth cone. When two guidance cues are present, corroborating directional information is supplied and integrated by the growth cone through crosstalk or convergence between the two guidance cue pathways.
We demonstrate that commissural axon guidance errors occur in vivo when the Shh concentration gradient is relatively shallow. We then use a novel microfluidic guidance assay to show the importance of gradient steepness for commissural axon guidance in vitro. We find that a combined gradient of the attractive guidance cues Shh and Netrin-1 can act in steepness-limited synergy to attract axons when the steepness of a single guidance cue is insufficient to guide axons. Mechanistically, we demonstrate that combined Shh and Netrin-1 gradients polarize SFK phosphorylation in the growth cone at the same gradient steepness when the two cues behaved synergistically to attract axons.
To determine the Shh gradient steepness that growth cones of commissural axons are exposed to in vivo, we examined spinal cord cross sections of embryonic day 9.5 (e9.5) and e10.5 mouse embryos, stages when axons are actively being guided towards the floor plate. We visualized the distribution of Shh protein in paraformaldehyde-fixed spinal cords using immunofluorescence with an anti-Shh antibody [18]. The Shh staining present in the floor plate and the spinal cord were not present in Shh-/- embryos (S1A Fig), demonstrating that the antibody specifically recognized Shh. We then measured the fluorescence intensity profiles of the Shh protein gradient along the dorso-ventral axis at several angles for each image (Fig. 1A) and pooled these measurements from multiple embryos to obtain a prototypical gradient profile (Fig. 1B). Shh fluorescence signal was highest at the floor plate and rapidly decreased for approximately 50 μm from the floor plate, followed by a slower decrease for the remainder of the spinal cord. We observed that the gradient profiles were remarkably consistent between embryos (S1B Fig) and that they did not depend on the concentration of the primary antibody (S1C Fig). We then demonstrated that there is a linear relationship between the fluorescence intensity and the concentration of Shh protein (S1D Fig). Furthermore, the gradient profiles were similar whether the measurements were made medially (as in Fig. 1A) or more laterally, overlapping with Tag-1 positive axons (S2 Fig).
Both the concentration (C) and steepness of the gradient can influence axon guidance responses. Because growth cones must be able to determine the direction of a gradient, it is essential that they can sense a difference in concentration across their width. This can be expressed as the absolute change in concentration across a growth cone (ΔC). The fractional change in concentration (δ = ΔC/C) is a measure of the steepness of the gradient across the growth cone, typically estimated at 10 μm [19]. The fractional change is usually expressed as a percentage and reflects the change in concentration across a growth cone relative to the ambient concentration at the growth cone.
Although it is not possible to accurately quantify absolute protein levels in vivo using immunohistological methods, measuring the fractional change in concentration does not require knowledge of the actual concentration of the cues, only the relative concentration of the cue. Thus we estimated the fractional change in concentration using the Shh fluorescence intensity. Within 50 μm of the floor plate (relative distance of 0–0.1 from the floor plate to the roof plate), there is a rapid decrease in Shh, with a fractional change (δ) of 46%–72% (Fig. 1B). In the region beyond 50 μm of the floor plate, the Shh gradient was shallower.
We then determined where along the spinal cord guidance defects occur for commissural axons from mice genetically deficient for Shh or Netrin-1 signaling. We analyzed images from previously reported guidance cue or guidance receptor mutants [4,6–8] and measured the relative distance from the floor plate at which misguided axons begin to deviate from their normal trajectory (S1 Table). For Shh and Netrin-1 signaling dependent defects, guidance errors occurred at a relative distance of 0.35–0.6, which corresponds to 158–270 μm from the floor plate for a spinal cord ~450 μm in height. In the region where Shh dependent errors occur (relative distance of 0.41–0.56), the Shh gradient at e9.5 was very shallow, with a fractional change of 0.6 < δ < 0.7%. At e10.5, when the majority of the commissural axon growth cones are en route from the roof plate to the floor plate, the fractional change in this region was 1.9 < δ < 2.1%, slightly higher than that measured at e9.5 (Fig. 1B). The Netrin-1 gradient has been previously visualized at mouse e10.5 using alkaline phosphatase immunohistochemistry [10]. Similarly to what we observed for Shh, Netrin-1 signal is highest at the floor plate and decreases rapidly in the first ~50 μm from the floor plate, with a shallow gradient present in the remainder of the spinal cord, which includes the region from the floor plate where Netrin-1 dependent errors occur (relative distance of 0.35–0.6). This gradient shape is reminiscent of the gradient shape for Shh and suggests that the Netrin-1 gradient is also steep close to the floor plate and shallow in the remainder of the spinal cord. However, we were unable to confirm this by more precise quantification using immunofluorescence because the Netrin-1 antibodies that work for immunohistochemistry are no longer available.
Intriguingly, the Shh- and Netrin-1-dependent guidance errors occur in the region of the spinal cord where Shh and most likely Netrin-1 gradients are shallow, not steep (Fig. 1C), indicating that loss of one guidance cue is sufficient to cause guidance defects in shallow gradients. Since guidance defects occur in this shallow gradient region, we hypothesized that having multiple guidance cues may be most important when the fractional change is low, when it is more difficult for a growth cone to obtain an accurate sense of direction from a single gradient.
The guidance of commissural neuron axons towards the floor plate in mice occurs between e9.5 and e11.5 [20,21]. Considering that commissural axons grow at 13–20 μm/h in vivo [21,22] and that the distance from the roof plate to the floor plate is about 500 μm, an individual axon will therefore take ~25–38 h to reach the floor plate. Since neurons vary in when they differentiate and begin their axon outgrowth, we approximate that commissural neurons are exposed to guidance cues en route to the floor plate over 1–2 d.
We thus developed a guidance assay capable of simulating, over 1–2 d, the shallow Shh gradients that we observed in the spinal cord in vivo. Microfluidic mixing networks allow gradients to be controlled in space and time, allowing for long-term gradients to be established, in contrast to passive source-sink diffusion gradients (e.g., pipette assay and Dunn chamber). We used a linear gradient generator because it allowed us to test a range of fractional change (δ) values. We modified a pre-mixer microfluidic gradient generator [23] by increasing both the length and width of the gradient region, thus maximizing the surface area on which neurons could be exposed to the gradient and thus the sample size. By increasing the width of the gradient, we also decreased the range of gradient steepness to physiologically relevant levels, as determined in vivo (Fig. 1B). Our wider gradient chamber required an increase in the number of sequential mixing channels (Fig. 2A), which offered the added benefit of increasing the overall resistance, thus decreasing the flow velocity and resulting shear stress, which can be harmful to axons [24]. With these device improvements, we were thus able to generate stable, long-term gradients.
In our microfludic device (Fig. 2A), gravity-driven flow (Fig. 2B) directs fluid into the mixing network (Fig. 2C), resulting in a linear gradient throughout the chamber (Fig. 2C–E). We used fluorescent dextran to measure the concentration and fractional change of the gradient, and found that as predicted, the gradient is linear and maintained throughout the chamber (Fig. 2F,H), and stable over a 24 h period (Fig. 2J–K). Furthermore, the measured fractional change (δ) values match the predicted values, ranging between 0.3% and 2.2% (Fig. 2G,I). Since the gradient is linear (Fig. 2F,H), the fractional change increases as the concentration decreases across the device (Fig. 2G,I). The device was biocompatible, as dissociated commissural neurons could be cultured in the device and were observed to extend axons (Fig. 2L,M). To test whether the slow flow rate present in the chamber would bias the direction of axon growth, we measured the angle at which the axon emerged from the cell body and the angle at which the tip of the axon was oriented. We found that the presence of fluid flow did not change the random distribution of these angles (S3 Fig), and therefore the shear stress in our device is negligible and does not bias the direction of axon initiation from the cell body nor the direction of axon growth. Therefore, we have developed an assay, which we named le Massif, to challenge commissural neurons with physiologically relevant gradients.
We established gradients of Shh or Netrin-1 after commissural neurons had been cultured for 24 h, when the majority of neurons had already initiated an axon. We calculated the turned angle of an axon as the difference between the base and tip angles (Fig. 3A) and scored the angle as positive if the axon turned towards the gradient and negative if it turned away. By varying the maximal concentration of ligand in a particular chamber, we could test a wide range of concentrations. In a control gradient (Phosphate buffered saline/BSA), we observed a wide range of turned angles towards and away from the gradient, resulting in a net turned angle of 0° (Fig. 3B,E). Neurons exposed to a gradient of Shh (Fig. 3C,E) or Netrin-1 (Fig. 3D,F), however, turned towards the higher concentration of chemoattractant. For either cue, we observed axon turning in response to a wide range of concentrations at the growth cone (Fig. 3E,F). The distribution of turned angles of individual axons confirmed that wide concentrations of Shh and Netrin-1 induced biases towards attraction (Fig. 3G,H). To eliminate the possibility that Shh or Netrin-1 influences the orientation at which the axon exits the cell body (axon base angle), thus confounding our measurement of the angle turned, we performed experiments where gradients were established 4–6 h after the neurons were plated, before the majority of neurons had initiated an axon. We found that Shh and Netrin-1 gradients induced no significant bias in the distribution of axon base angles facing up-gradient (higher concentration) compared to those facing down-gradient (lower concentration) (Fig. 3I,J). Therefore, le Massif generates gradients that can induce axon turning without any effect on axonal initiation.
Since we observed similar turning over a wide range of concentrations (Fig. 3E,F), we then analyzed axon turning as a function of the fractional change in concentration, δ, across a growth cone. The fractional change is a function of the chamber geometry, and not of the maximum concentration used (Fig. 4A). Therefore, the fractional change is independent of the maximum concentration in the gradient chamber, so long as the minimum concentration is 0. We found that axon turning increased as a function of fractional change for both gradients of Shh and Netrin-1 (Fig. 4B,C). This corresponded with an increase in the ratio of axons that turned towards the gradient compared with those turning away (Fig. 4D,E). Thus, there seem to be fewer guidance errors as the fractional change across the growth cone increases. This was also illustrated with the distribution of turned angles of individual axons (Fig. 4F). At low fractional change, the population of axons have variable turned angles, with a slight bias toward attraction. As the fractional change increased, a bias towards attraction became more pronounced, as there were fewer axons that were erroneously repelled. We then compared the turned angles of axons experiencing the same fractional change (δ > 1%) with different concentrations at the growth cone. When δ > 1%, we observed no trend toward increased turning as the concentration at the growth cone increased (Fig. 4G,H). Therefore, for commissural neurons in gradients of Shh or Netrin-1, the turning response is more sensitive to changes in the fractional change than the local concentration at the growth cone.
Since guidance errors occur in the region of the spinal cord where Shh and Netrin-1 gradients are shallow, not steep (Fig. 1), we hypothesized that multiple guidance cues might be most important for guiding axons in shallow gradients. Therefore, we next tested whether combining gradients of two guidance cues might modulate the axon turning response in relation to fractional change. We performed guidance assays with 20 nM Shh and 0.69 nM Netrin in the inlet, generating local concentrations ranging from 2.5 to 18.55 nM for Shh and 0.08 to 0.65 nM for Netrin-1, encompassing the range for which we see axon turning (Fig. 3E–H, Fig. 4G,H). Gradients of Shh or Netrin-1 alone and in combination were established 6 h after neurons were plated and maintained for 45 h. In either Shh or Netrin-1 gradients, the turned angle peaked at the highest fractional change, δ = 2.2% (Fig. 5A). Upon applying both Shh and Netrin-1 simultaneously, the average turned angle increased more quickly as a function of fractional change, such that axons were turning robustly in a region of the double gradient where a single cue was not eliciting much turning (1.4 < δ < 1.8%). Interestingly, at the maximal zone of fractional change (1.8 < δ < 2%), there was no observable difference in the angle turned induced by the individual or combined cues (Fig. 5A).
To assess the relationship between the combined cues compared to the individual cues, we calculated the synergy quotient as the turned angle in the combined gradient divided by the sum of the turned angles to both cues individually (described in the Materials and Methods section). With this measurement, a value below 1 indicates sub-additive effects, 1 is defined as additive, while a value above 1 is synergistic. We observed additive and sub-additive effects for the majority of the fractional change range, apart from fractional change range of 1.44 < δ < 1.82%, in which the effect of the combined cues is much greater than the sum of the individual cues, demonstrating synergy (Fig. 5B). Hence the synergistic effect of the combined gradient is greatest when the fractional change is below the maximum.
At this fractional change (1.44 < δ < 1.82%) in which the combined gradients lead to synergy, axons responded much more robustly to the combined cues than for either cue individually, resulting in a higher average turned angle (Fig. 5C). This effect was remarkably consistent, with every independent gradient chamber with combined cues having strong positive turning in this range of fractional change, whereas the gradient chambers with the single cues had variable turned angles with no consistent bias (Fig. 5D). The synergy occurring when the two cues are present was also demonstrated by the larger proportion of axons which turn up the combined gradient than for either cue individually (Fig. 5E). The influence of combining cues on the proportion of correct versus incorrect guidance decisions was also apparent when we observed the distribution of the turned angles in the different conditions (Fig. 5F). For the control gradient (vehicle) and gradients of Shh and Netrin-1, there was no bias towards either attraction or repulsion. Remarkably, when both cues were presented as a combined gradient, there was a clear bias towards attraction, wherein very few axons failed to reorient their direction. Together, these results indicate that a combination of guidance cues can act in synergy to guide axons when the gradient steepness is sub-optimal for the growth cone to sense the direction of a single cue gradient.
Since SFKs act downstream of Shh [15] and Netrin-1 [14] to guide commissural axons, it has been proposed in a recent review by Dudanova and Klein [25] that Shh and Netrin-1 signaling may converge on SFKs. Furthermore, the active form of SFKs, phosphorylated at Y418 (pSFK), accumulates on the side of the growth cone proximal to the higher concentration of Shh and is sufficient to relay the direction of the gradient [15]. Using le Massif, we challenged commissural growth cones with gradients of either Shh, Netrin-1, or a combination of both for 2 h, and then assessed the distribution of growth cone pSFK along the direction of the gradient (Fig. 6A). Growth cone pSFK distribution was measured by the fractional change in signal intensity across the width of a growth cone (δGC), which represents the difference in the amount of pSFK at the proximal versus the distal side of the growth cone, relative to the overall levels. Growth cones with more pSFK on the proximal side closer to the higher guidance cue concentration had positive δGC values, and growth cones with more pSFK on the distal side closer to the lower guidance cue concentration had negative δGC values. For axons exposed to a fractional change of 1.44 < δ < 1.82% in single cue gradients of Shh (Fig. 6B) or Netrin (Fig. 6C), there was no consistent bias in the direction of pSFK distribution (mean and median δGC ~0%, Fig. 6E,F). In the combined gradient, however, more growth cones had a proximally biased pSFK distribution (Fig. 6D–F).
This shift towards proximally distributed pSFK was also apparent when we calculated the ratio of the number of proximal to distally polarized growth cones in each independent gradient chamber for 1.44 < δ < 1.82%. Independent chambers with single cue gradients of Shh or Netrin-1 vary between having a net proximal or distal pSFK growth cone distribution. In contrast, for the combined Shh and Netrin-1 gradient, there are consistently more chambers with a net proximal pSFK growth cone distribution and not a single chamber in which there is a net distal pSFK growth cone distribution (Fig. 6G). Therefore, single cue gradients of Shh and Netrin-1 that do not elicit axon turning (Fig. 5), also do not elicit a polarized pSFK distribution (Fig. 6). Remarkably, when the Shh and Netrin-1 gradients synergize to elicit turning, this also corresponds to a higher pSFK distribution on the side of the growth cone facing the high concentration of guidance cues. Taken together with our quantification of gradients and guidance defects in vivo, these results indicate that a combination of guidance cues can act in synergy to polarize growth cones in regions where the gradient steepness is sub-optimal for the growth cone to be polarized by a single cue gradient.
In this study, we developed a microfluidic gradient generator that enabled us to directly measure long-term axon turning responses under physiological gradient steepnesses. We demonstrate that commissural axons turn more when the fractional change is high for both Shh and Netrin-1. Additionally, we show that when the steepness of the gradient is limiting, a combined gradient of Shh and Netrin-1 induces axon turning at a gradient steepness at which either cue alone cannot (Fig. 7A). Furthermore, at this same gradient steepness, we observe polarized growth cone activation of SFK only when Shh and Netrin-1 are presented in combined gradients (Fig. 7A). Therefore, we propose that collaboration between Shh and Netrin-1 results in synergy specifically in circumstances in which both gradients are shallow. In this steepness-limited synergy, multiple overlapping signals are necessary for the growth cone to properly interpret the orientation of the gradient when the gradient is shallow. In the developing spinal cord, we propose that this corresponds to a region midway along the commissural axon trajectory (Fig. 7B). Notably, the analysis of the phenotype of four different mouse models (Netrin-1, DCC, Boc, and Smo conditional mutants) shows that when one of these pathways is impaired, guidance errors occur in this shallow gradient region (Fig. 7C). Therefore, our data support a model where guidance cue collaboration is essential to guide axons when the gradient steepness is sub-optimal for them to be guided by a single cue.
An essential component of the current study is the use of microfluidic mixing networks to generate spatially and temporally stable concentration gradients. le Massif guidance assay allows us to assess axon turning over the course of days. Since an image only has to be taken at the final time point, le Massif is compatible with high-content screening microscopes, allowing assays to be performed in a high-throughput manner, such that a large number of axons can be imaged and analyzed (over 200 per chamber).
An additional advantage of le Massif over other axon guidance assays is that it generates gradients with low-to-moderate fractional change, 0.3 < δ < 2.2%, which sits near the lowest fractional change eliciting detectable guidance responses (Fig. 4B,C). This is critical for studying the influence of fractional change on axon turning. This contrasts with techniques such as the pipette assay, which generates gradients with a steep fractional change (5 < δ < 35%) [26]. While printed gradient assays allow precise control over the gradient parameters [19,27–29], the gradient is printed prior to the addition of the neurons, making it difficult to distinguish the effect of the gradient on direct axon turning, rather than differential axon outgrowth or growth rate modulation (a notable exception to this is Mortimer et al. [30], which tested the effect of printing the guidance cue before and after addition of explants). Furthermore, in these assays, axons are either growing along pre-defined corridors or are growing from an explant, making individual axon trajectories often difficult to identify. In contrast, individual axon trajectories can be easily visualized in le Massif because the dissociated neurons are grown at low density, so we can clearly measure directed turning of individual axons. Also, by imposing the gradient after axon outgrowth has commenced, we avoid the gradient influencing the orientation of axon protrusion from the cell body [29]. Thus, owing to the versatile process of microfluidic design, we were able to create a customized gradient generator and generate gradients with physiologically relevant steepness that are stable over days.
In addition to axon turning, axon growth [28] and growth rate modulation [19,31] are also processes important in guiding axons to their correct targets. Compared to direct axon turning, growth-rate modulation occurs when axons growing up-gradient grow faster than those growing down-gradient [31]. Previous studies have found that gradient steepness affects axon growth [28] and growth rate modulation [19,31]. We found that gradient steepness also influences axon turning, with robust turning observed for steepness δ~1%–2%. This contrasts with what has been reported for growth rate modulation by NGF gradients, where steepness as low as 0.1% is sufficient to bias DRG axon trajectories [19,31]. Consistent with our results, these 0.1% NGF gradients had no effect on direct axon turning [31]. Similarly, growth of axons is also modulated by gradients with steepness of ≥0.4% (1% over 25 μm), possibly also by influencing the growth rate [28]. Therefore, our results suggest that steeper gradients of 1%–2% are required to induce direct axon turning rather than growth-rate modulation, as hypothesized by Mortimer and colleagues [31].
The gradient steepness at which robust turning occurs is ~1%–2%, within a similar range to our estimate of Shh gradient steepness in the spinal cord (Fig. 1B). For Netrin-1, the lack of effective antibodies for Netrin-1 for use in immunofluorescent staining hampers our ability to directly measure the Netrin-1 gradient steepness in the spinal cord. Previously published images of Netrin-1 in the spinal cord are not amenable to precise quantification because they use alkaline phosphatase immunohistochemistry combined with darkfield imaging [10]. However, examination of the pattern of Netrin-1 staining in the spinal cord [10] does show that the Netrin-1 gradient is steeper closer to the floor plate and shallower further away from the floor plate, consistent with what we observe with Shh and consistent with our hypothesis that Netrin-1–dependent guidance errors occur in shallow, not steep, regions of the gradient.
While significant evidence indicates that multiple guidance cues act on the same axons, precisely how these cues converge to regulate the behavior of the growth cone is poorly understood. The response to two combined cues may be additive or synergistic, depending on whether the output is equal to or above the combined response of either cue individually. Dudanova and Klein [25] define additivity as resulting from cues that act in parallel pathways, and synergy as resulting from cues that have crosstalk between pathways.
Additive effects of guidance cues have been observed with ephrin-A and glial cell line-derived neurotrophic factor (GDNF) on lateral motor column (LMCL) axons [32], whereas a synergistic attractive response was seen between EphA and GDNF for the same axons [33]. The former demonstrates that ephrin-A and GDNF act in parallel pathways, while the latter demonstrates crosstalk between EphA and GDNF. EphA and GDNF signal through their respective GPI-anchored receptors ephrin-A and GFRα1, and they crosstalk by sharing a common co-receptor, Ret. The co-activation of ephrin-A and GFRα1 through sharing Ret acts as a coincidence detector and generates synergy [33]. The interaction between EphA and GDNF is an example of concentration-limited synergy, as the combination of low concentrations of guidance cues induced turning when neither cue alone was sufficient.
One of our major findings is that synergy occurs between Netrin-1 and Shh to guide commissural axons. In contrast to Bonanomi and colleagues [33], we find that this synergy is steepness-limited rather than concentration-limited. Steepness-limited synergy occurs when the gradient of individual cues is too shallow to guide axons, but a combined gradient of two cues elicits axon turning. We know that in our case the concentration of the individual cues is not limiting because we observe axon turning when the fractional change is high, despite this corresponding to a lower absolute concentration (Fig. 4A). Furthermore, the range of concentrations used in our experiments all elicit axon turning when the steepness is not limiting (Fig. 4G,H). Thus, we demonstrate for the first time that synergy can also be steepness-limited, when the amount of ligand is not limiting but instead the steepness of the gradient is insufficient for the growth cone to estimate the direction of a single cue gradient.
We also identify SFK as a downstream signaling molecule that integrates Shh and Netrin-1 signaling when the two cues synergize. Both Shh and Netrin-1 can activate SFKs, and SFKs are required for Shh and Netrin-1–mediated axon guidance [14,15]. Furthermore, pSFK polarization at the growth cone reflects the direction of the external gradient [15]. Gradients of Shh and Netrin-1 too shallow to guide axons were also insufficient to correctly polarize pSFKs at the growth cone. Only in the presence of Shh and Netrin-1 together was the direction of the gradient correctly reflected by the growth cone pSFK polarization. Hence, activated SFKs appear to be a node where information from the Shh and Netrin-1 gradients are integrated. In addition to synergy resulting from sharing a common co-receptor as for EphA and GDNF, we find that for Shh and Netrin-1, synergy can arise from shared intracellular signaling molecules. Therefore, diverse mechanisms exist through which synergy between two guidance cues can occur, and more mechanisms likely remain to be discovered.
In the developing nervous system, it is likely that many types of synergistic interactions play a role in the correct guidance of axons to their targets. In the developing limb, where guidance cues act at a choice point for motor axons, concentration-limited synergy may be more important than steepness because the gradient is very abrupt. For commissural axons, which climb a shallow gradient of guidance cues over a long distance, steepness-limited synergy may initially be more critical. Later in their journey, when they reach the steep part of the gradient, it appears that one cue alone may be sufficient to guide axons—for example, the axons in Boc mutant mice that cannot respond to Shh but by chance make it close to the floor plate do eventually reach the floor plate [8], possibly because of the effect of the steep Netrin-1 gradient in the ventral spinal cord. Thus, it appears that single steep gradients can guide axons over short distances and allow for more precise guidance near the floor plate, whereas midway along the commissural axon trajectory, synergy between shallow gradients of Shh and Netrin-1 allows these gradients to guide axons that are far from the floor plate, thus extending the distance that guidance cues can act in the spinal cord.
All animal work was performed in accordance with the Canadian Council on Animal Care Guidelines. Wild type C57Bl6 or Shh-/- mouse embryos were sacrificed at e9.5 or e10.5 and fixed in 4% paraformaldehyde (PFA) in phosphate buffered saline for 1–1.5 h at 4°C and cryoprotected in 30% sucrose. 12–20 μm thick serial sections were cut with a cryostat. Sections were rinsed several times in buffered saline, and then treated for 1 h with a blocking solution containing 0.1% Triton X-100 and 10% heat-inactivated goat serum (HiGS). Spinal cord sections were stained with anti-Shh antibody (kindly provided by S. Scales, Genentech) to detect Shh protein [18]. This antibody is specific, as no signal is detected in Shh mutant embryos (S1A Fig). The primary antibody was then replaced with a buffered solution containing 1% HiGS and Alexa Fluor 546-coupled secondary antibody (Molecular Probes; 1:1,000) or Cy3 conjugated secondary antibody (Jackson Immunoresearch, 1:1,000) for 1 h. After staining, slides were mounted with Mowiol (Sigma) and allowed to dry for at least 24 h before imaging. Dot blot immunochemistry was performed by pipetting serial dilutions of recombinant human NShh C24II (R&D) onto a glass microscopy slide, followed by the same procedure and reagent concentrations as above.
For pSFK asymmetry assays, guidance cues were added, then the microfluidic devices were returned to the incubator for 2 h, after which they were fixed with 4% PFA at room temperature for 15 min. Phosphorylated Src-family kinase was detected using a phosphospecific (pY418) primary antibody (Invitrogen, 1:1,000), followed by Alexa Fluor 488-coupled secondary antibody (Molecular Probes; 1:1,000). Chambers were imaged with an IXM high-content screening microscope (Molecular Devices) using a 40X Nikon objective.
Spinal cord cross-sections were imaged on a Leica upright microscope with 10X and 20X objectives at multiple exposure times to ensure that the images contained the entire dynamic range of the gradients that had been revealed by immunohistochemistry. Images were then analyzed with a custom ImageJ macro, which measured the intensity profile along the dorso-ventral axis at five discrete angles ranging from 95° to 105°, emanating from a region just outside the floor plate. This was performed for both sides of the spinal cord for each image (Fig. 1A). The data was then pooled and visualized using a custom MATLAB script to calculate the mean intensity of the Shh gradient. The background fluorescent signal contribution from both the primary and secondary antibodies was determined by measuring the staining intensity in the neural tube of Shh-/- littermates, which were processed simultaneously and imaged identically. The background signal was then subtracted from each quantified Shh gradient profile before the fractional change was calculated. To calculate the fractional change of the measured gradient, the mean intensity profile of the regions of interest were fit to a straight line using Open Office Calc (Maryland), and the fractional change calculated from the fit line.
A microfluidic gradient generator [23] was modified to increase the surface area over which the gradient can be applied. Positive relief master molds were fabricated from a 17.78 cm (7 in) chrome photomask (FineLine Imaging, Colorado) by the McGill Nanotools Microfabrication Facility by spin coating SU-8 2050 (Microchem) to a height of 50 μm onto a 15.24 cm (6 in) silicon wafer. The silicon master wafer with positive relief features was exposed to CHF3 plasma for 1 min, then treated with 3,3,3 trifluoroperfluoro-octylsilane in a vacuum desiccator for 30 min to ensure that the polydimethylsiloxane (Silgard 184—PDMS) would not stick to the SU-8 features. PDMS was then mixed thoroughly as per manufacturer's recommendations (10:1 base polymer: curing agent) before being degassed for >15 min in a vacuum and poured onto the silicon master wafer. The PDMS was cured for >3 d at 60°C. The PDMS was then peeled off from the master and cut to individual chips. Through-holes at the two inlets and outlet were made using a biopsy punch. Glass slides (Schott Glass D) were soaked in concentrated nitric acid for 24–36 h, before being rinsed in milliQ water 12 times over 2 h and sterilized by baking at 225°C for 4–6 h. On the day prior to beginning the experiment, both glass slides and PDMS chips were exposed to an oxygen plasma (Plasmaline 415 Plasma Asher, Tegal Corporation, 0.2 mbar for 30 s at 75 W) before bringing the surfaces into contact to form an irreversible bond. Within 20 min following bonding, devices were filled with 0.1 μg/ml poly-D-lysine (PDL; Sigma) to generate an adhesive substrate onto which neurons could attach. After coating for 1 h, the PDL was removed and the microfluidic chamber rinsed twice by adding sterile milliQ water to the outlet. Fluid reservoirs were crafted by cutting the bottoms from 200 μl PCR tubes and positioning the tubes into the punched holes such that both tubes were an equal height. The tubes were both filled with 200 μl of Neurobasal media containing serum, generating a forward gravity-driven flow, which was left to further rinse the PDL coated channels overnight.
The range of ligand concentration imposed on the axons in the gradient chamber depended on the guidance cue concentration added to the reservoir at inlet 1 (Fig. 2A). The reservoir at inlet 2 was filled with culture media without guidance cue. To visualize and quantify the gradient, we used 40 kDa tetramethylrhodamine-dextran. Hydrostatic pressure was created by filling the inlet reservoirs higher than the outlets (Fig. 2B), which drove fluid flow uni-directionally from left to right throughout the device. When fluid from the two inlets converge, the concentrations at inlet 1 and inlet 2 are mixed and subsequently divided to three discrete concentrations (Fig. 2C). This mixing and splitting occurs a total of 18 times, generating 20 discrete concentrations that are spaced at linear gradations between the concentration at inlet 1 and inlet 2 (no cue). The 20 discrete concentrations then flow from the premixer channels into the gradient chamber, where they meet and diffuse to establish a linear concentration gradient (Fig. 2C). Because the fluid volume is on the microliter scale and the Reynold's number is low (Re < 1), the flow is laminar and there is no convective mixing [34]. Because diffusion is slow over long distances, the diffusion of the guidance cue is slow relative to the flow velocity and the gradient remains linear for the entire 9 mm length of the gradient chamber (Fig. 2E) as long as there is a continuous flow driving the mixing. Consequently, long-term gradients can be maintained without actively controlling the flow rate, so long as the reservoir at the outlet is emptied periodically (approximately every 24 h). The upstream and downstream regions of the gradient chamber were imaged using a 2.5X objective on an upright fluorescence microscope (Leica).
Commissural neurons were prepared from the dorsal fifth of E13 rat neural tubes as described previously [15,35]. Cells were re-suspended in plating media composed of Neurobasal (Gibco) supplemented with 10% heat-inactivated FBS and 2 mM GlutaMAX (Life Tech). 50 μl of plating media was added to both inlet reservoirs and 50 μl of cell suspension (3,160,000–5,630,000 cells/ml) was added to the outlet. One of the inlet reservoirs was removed and a reverse flow induced by connecting a syringe to the inlet hole via a short rubber hose and pulling on the plunger. While observing with an inverted microscope, neurons were drawn into the gradient chamber, after which the flow was stopped by releasing the plunger, disconnecting the syringe, and then returning the reservoir to the hole. After 4–6 h, inlet reservoirs were filled with 200 μl of plating media, again inducing a forward flow. Approximately 15 h later, the plating media was replaced with serum-free growth media composed of Neurobasal (Gibco) supplemented with 2% B27 (Gibco), 2mM GlutaMAX (Gibco) and penicillin/streptomycin (Gibco).
Shh guidance experiments were performed using the recombinant human NShh C24II (R&D). Netrin-1 guidance experiments were performed using the VI,V peptide [36], which was a kind gift from Dr. Tim Kennedy. The guidance assay was started within 24 h of plating, when most of the neurons had initiated a neurite. Culture media (200 μl) was added to one of the inlet reservoirs and guidance cue or vehicle control (0.1% BSA; Sigma) diluted in culture media (200 μl) to the other. Gradient devices were then returned to the incubator until the following morning (~20 h following gradient application), at which point a Pasteur pipette was used to remove any fluid which had accumulated in the outlet. The devices were then returned to the incubator for a further 4 h, for the remainder of the 24 h assay. Guidance assays over 45 h were performed as described above, except the gradient was established 4–6 h after the neurons were loaded into the device.
The assay was ended by quickly removing all culture media from the inlets and outlets by aspiration and adding 4% PFA (100 μl) to the outlet reservoir. After 15 min, the PFA was removed and replaced with a staining buffer consisting of DAPI (Sigma, 1:10,000) to stain cell nuclei, TRITC-phalloidin (Molecular Probes, 1:250) to stain F-actin, and Triton (Sigma, 1:400) to permeabilize the cells. Neurons were left to stain overnight (~12 h) and the staining buffer was replaced with buffered saline for 1–2 h before imaging.
Fixed specimens were imaged using an IXM high content screening automated microscope (Molecular Devices) with a laser-based auto-focus and a 20X objective (Nikon). To include the entire area of the gradient chamber, 275–300 images were obtained for each device using MetaExpress imaging software (Molecular Devices). All analyses were performed by an observer naive to the gradient conditions for each device. For each image, we traced all isolated axons in each field of view using a custom ImageJ macro. So that the observer would be blind to the direction of the gradient, every image had a 50% chance of being flipped vertically when opened. Image files were analyzed by Flatworld Solutions (Bangalore).
All calculations of neuron position, concentration, fractional change, axon length, and turned angle were performed using a custom MATLAB script. We defined the axon base angle as the angle between the proximal 20 μm of the axon and the direction of flow, and the axon tip angle as the angle between the distal 20 μm of the axon and the direction of flow (parallel to the arrow in Fig. 2C). We defined the turned angle as the difference between the base and tip angles of the axon, where the sign of the difference was positive if the axon turned toward the gradient and negative if the axon turned away from the gradient (Fig. 3A). We considered only axons that faced against the direction of the flow, and we excluded those facing directly towards or against the gradient (within 20° of the gradient direction). We excluded from further analysis axons that were shorter than 20 μm. Because of the local flattening of the gradient near the boundaries caused by the no-slip condition, we excluded any neurons positioned within 450 μm of either boundary (red boxes Fig. 2K).
To estimate the concentration of guidance cue at each growth cone position, we calculated each neuron’s position relative to the gradient chamber, and thus relative to the gradient itself. We assumed a growth cone width of 10 μm for fractional change calculations, which we calculated using the difference in concentration between a point 5 μm above and 5 μm below the neuron, divided by the concentration at the neuron's position. All included growth cones experience fractional change within the range 0.3 ≤ δ < 2.2%.
To calculate the synergy quotient, we first calculated a central moving average (CMA) of the turned angle of all axons within a window of 0.5% fractional change for each Shh, Netrin, and the combined gradient Shh+Netrin (Fig. 5A). We then calculated the synergy quotient (SQ) as:
SQ = CMAShh+Netrin/(CMAShh+ CMANetrin).
After being exposed to the gradient(s) for 2 h, neurons were fixed and stained for pSFK. Chambers were imaged with an IXM high-content screening microscope (Molecular Devices) using a 40X Nikon objective. Each growth cone was outlined. Then a line was placed spanning the width of the growth cone outline, parallel to the direction of the concentration gradient. The intensity profile was measured across five parallel lines spaced 1 pixel apart. The average intensity profile of the five lines was then processed using a custom MATLAB script. The fractional change in staining intensity across each growth cone (δGC) was then calculated as the difference in mean intensity between the proximal and distal thirds, divided by the mean intensity of the entire area of the growth cone (Fig. 6A). This value was scored as positive if the higher staining intensity was on the side of the growth cone proximal to the gradient and negative if the higher staining intensity was on the side of the growth cone distal to the gradient.
All analysis of variance, Chi-square and Wilcoxon signed-rank tests were performed using Graphpad Prism 5 (La Jolla, CA). All Rayleigh tests for unimodal deviation from uniformity were performed using the circStat toolbox for MATLAB.
The majority of graphs were generated using GraphPad Prism or Open Office Calc (The Apache Software Foundation), unless otherwise mentioned. Tricolor radial scatter plots (Figs. 3G,H, 4F, 5F) and radial frequency histograms (Figs. 3I,J, S3) were scripted manually with Processing, an open-source sketchpad software (www.processing.org). Random samples of axons were generated with a Processing script using a uniform probability distribution, wherein each axon was equally likely to be selected as the next data point was plotted, and the same data point could not be plotted twice.
|
10.1371/journal.ppat.1007452 | Different functional states of fusion protein gB revealed on human cytomegalovirus by cryo electron tomography with Volta phase plate | Human cytomegalovirus (HCMV) enters host by glycoprotein B (gB)-mediated membrane fusion upon receptor-binding to gH/gL-related complexes, causing devastating diseases such as birth defects. Although an X-ray crystal structure of the recombinant gB ectodomain at postfusion conformation is available, the structures of prefusion gB and its complex with gH/gL on the viral envelope remain elusive. Here, we demonstrate the utility of cryo electron tomography (cryoET) with energy filtering and the cutting-edge technologies of Volta phase plate (VPP) and direct electron-counting detection to capture metastable prefusion viral fusion proteins and report the structures of glycoproteins in the native environment of HCMV virions. We established the validity of our approach by obtaining cryoET in situ structures of the vesicular stomatitis virus (VSV) glycoprotein G trimer (171 kD) in prefusion and postfusion conformations, which agree with the known crystal structures of purified G trimers in both conformations. The excellent contrast afforded by these technologies has enabled us to identify gB trimers (303kD) in two distinct conformations in HCMV tomograms and obtain their in situ structures at up to 21 Å resolution through subtomographic averaging. The predominant conformation (79%), which we designate as gB prefusion conformation, fashions a globular endodomain and a Christmas tree-shaped ectodomain, while the minority conformation (21%) has a columnar tree-shaped ectodomain that matches the crystal structure of the “postfusion” gB ectodomain. We also observed prefusion gB in complex with an “L”-shaped density attributed to the gH/gL complex. Integration of these structures of HCMV glycoproteins in multiple functional states and oligomeric forms with existing biochemical data and domain organization of other class III viral fusion proteins suggests that gH/gL receptor-binding triggers conformational changes of gB endodomain, which in turn triggers two essential steps to actuate virus-cell membrane fusion: exposure of gB fusion loops and unfurling of gB ectodomain.
| Infection by herpesviruses leads to many human diseases, ranging from mild cold sores to devastating cancers. Human cytomegalovirus (HCMV) is among the most medically significant herpesviruses and causes birth defects and life-threatening complications in immuno-suppressed individuals. HCMV infection begins with cellular membrane fusion, a dynamic process involving receptor-binding to gH/gL complexes and drastic transformation of fusion protein gB trimer from the metastable prefusion conformation to the stable postfusion conformation. We have used cryo electron tomography incorporating cutting-edge technologies to observe the three-dimensional structures of gB and gH/gL in their native environments of HCMV particles. Visualizations of gB in both prefusion and postfusion conformations, together with structures of other class III viral fusion proteins and molecular dynamics flexible fitting (MDFF), facilitate a prediction for the structure of HCMV gB in its prefusion conformation. With the further observation of the contact of prefusion gB with gH/gL complex, the conformational changes of gB from pre- to postfusion state lead to a better understanding of herpesvirus fusion mechanism.
| Human cytomegalovirus (HCMV), a member of the Betaherpesvirinae subfamily of the Herpesviridae family, is a leading viral cause of birth defects [1, 2] and a major contributor to life-threatening complications in immunocompromised individuals. As one of the largest membrane-containing viruses, HCMV shares a common multilayered organization with all other herpesviruses, composed of an icosahedrally ordered nucleocapsid enclosing a double-stranded DNA genome, a poorly defined tegument protein layer, and a pleomorphic, glycoprotein-embedded envelope [3]. During infection, herpesviruses fuse their envelopes with cell membranes, resulting in the delivery of nucleocapsid into the cytoplasm of the host cells. This complex process requires a number of viral glycoproteins and host receptors functioning in a coordinated manner. Glycoproteins gB and gH/gL are conserved across all herpesviruses and are essential for virus entry into cells [4]. Receptor-binding to gH/gL-containing complexes—the composition of which differs among clinical and laboratory-adapted HCMV strains and across different herpesviruses [5]—triggers conformational changes of fusion protein gB, leading to fusion of the viral envelope with cell membrane [6]. This use of both a fusion protein and a receptor-binding complex for herpesvirus entry differs from many other enveloped viruses, which use a single protein for both receptor binding and membrane fusion.
Averaging up to tens of thousands of particle images by single-particle cryoEM method has resolved in situ structures of capsid proteins [7–9] and the capsid-associated tegument protein pp150 [10], up to atomic resolution [11]. However, such method is not applicable to the studies of herpesvirus gB and other glycoproteins due to their disorganized distribution on the pleomorphic viral envelope. Instead, the structures of gB ectodomains and various forms of gH/gL from herpes simplex virus (HSV) [12, 13], Epstein-Barr virus (EBV) [14, 15] and HCMV [16, 17] have been solved by x-ray crystallography. The gB ectodomain structures from these studies share structural similarities to other class III viral fusion proteins in their postfusion conformation [18–20]. Among these proteins, vesicular stomatitis virus (VSV) G is the only one whose ectodomain structure has been solved for both prefusion [21] and postfusion [20] conformations, thanks to its pH-reversibility between the two conformations and amenability to crystallization at both high and low pH conditions. At pH 6.3–6.9 conditions, VSV G has also been observed to exist in monomeric forms both in solution and on virion envelope, possibly representing fusion intermediates [22, 23]. By contrast, the prefusion conformation of herpesvirus gB is metastable and its structure has been elusive (even the recent crystal structure of the full-length HSV-1 gB is also in the postfusion conformation [24]). While cryo electron tomography (cryoET) of HSV-1 virions has revealed glycoproteins on the native viral envelope, poor contrast of cryoET reconstructions makes it difficult to distinguish different glycoprotein structures and conformations [25]. Recent efforts resorted to the use of purified HSV-1 gB-decorated vesicles to visualize the prefusion gB, but its domain assignments have been controversial due to difficulties in interpreting cryoET structures with poor contrast and signal/noise ratio (SNR) [26, 27]. As a result, the mechanism underlying the complex process of receptor-triggered membrane fusion remains poorly understood for not only HCMV, but also for other herpesviruses.
Recently, the technologies of electron-counting [28, 29], energy filter and Volta phase plate (VPP) [30] have significantly improved contrast and SNR of cryoEM images and their combined use in cryoET has led to resolution of two functional states of 26S proteasome in neurons [31]. In this study, we first demonstrated the ability to distinguish prefusion and postfusion conformations of the VSV G trimer (171 kD) in situ by employing a combination of VPP, direct electron-counting, energy filtering and subtomographic averaging. Application of the same approach to HCMV virions has allowed us to identify different conformational states of HCMV gB (303 kD) in their native virion environments and to determine the in situ structure of prefusion gB at a resolution of ~21 Å. Moreover, we also observed prefusion gB forming a complex with gH/gL in situ for the first time. Integration of these structures and knowledge of class III viral fusion proteins has led to a working model of how conformational changes drive membrane fusion during HCMV entry into host cells.
We first established the validity of our cryoET method of combining VPP, direct electron detection, energy filtering, and subtomographic averaging by obtaining in situ structures of class III viral fusion proteins with known structures. Towards this end, we took advantage of the relative simplicity of VSV in having a single 57kD glycoprotein, G, on the viral envelope, with its trimeric structures known for both prefusion and postfusion conformations; and used VSV as a gold standard to validate our method. For VSV at pH = 7.5, tomograms reconstructed from tilt series obtained by 300kV Titan Krios equipped with VPP, energy filter and direct electron detection show excellent contrast, enabling the visualizations of G projecting from viral envelope, the helical nucleocapsid, as well as the internal densities corresponding to polymerases L (Fig 1A and 1B). Two conformations of G are readily differentiable based on the height and shape of the ectodomain: the majority is long (12.5nm) and slim, while the minority is short (8.7nm) and fat (Fig 1B). Subtomographic averages of 330 long-form particles and 65 short-form particles from five tomograms both contain a prominent ectodomain, with the long one (~28 Å resolution) fit perfectly with the crystal structure of G ectodomain trimer in the postfusion conformation (Fig 1C–1E) and the short one with that in the prefusion conformation (Fig 1G–1I). Similar structures were observed in a previous electron tomography study performed on negatively stained sample [32].
Both crystal structures of G contain five domains, DI through DV, despite drastic domain arrangements (Fig 1E, 1F, 1I and 1J). The dramatically different appearances between the two conformations are primarily due to the refolding of the short loop (residue 273 to 275) in DIII, resulting in the elongation of the central helix and a taller postfusion trimer. DIII form the trimeric core in both conformations, buried in the center of the cryoET density map (Fig 1E and 1I). The other domains (DI, DII and DIV) undergo a rigid-body type rearrangement—only changing the relative orientations and locations while retaining their domain structures [21] (Fig 1F and 1J). This analysis demonstrates that our cryoET approach incorporating the three cutting-edge technologies can distinguish the two forms of in situ structures of glycoprotein G and allows fitting existing domain structures of individual fusion protein into the density maps for functional interpretation.
Next, we applied the same strategy established above to obtain in situ structures of gB and its interaction with gH/gL complex. We imaged virions of the highly passaged laboratory HCMV strain AD169, taking advantage of its simplicity, as it has lost some glycoprotein genes and does not contain gH/gL/UL128/UL130/UL131A pentamers on its envelope [33]. We recorded cryoET tilt series of HCMV virions with and without VPP in a Titan Krios instrument equipped with an energy filter and a direct electron detector in electron-counting mode. Both the raw images in the tilt series and the reconstructed tomograms show significantly better contrast when VPP was used (S1 Fig, S1–S4 Movies). Typical in virions obtained by high-speed centrifugation, the viral envelopes are pleomorphic and often exhibit membrane blebs likely due to mechanical stress during purification (Fig 2A). [As discussed below, such mechanical stress might also be responsible for triggering some of the “spring-loaded”/higher-energy (prefusion) gB to its lower energy (“postfusion”) form, which were used as an internal control to validate our cryoET subtomographic averaging method.] In the tomograms reconstructed from the tilt series obtained with VPP (referred to as VPP tomograms) (Fig 2), three types of enveloped viral particles are readily recognized: virions with C-capsid containing densely-packed dsDNA genome (Fig 2B), non-infectious enveloped particles (NIEPs) with B-capsid containing a protein scaffold (red arrows in Fig 2A) or with empty A-capsid (cyan arrow in Fig 2A). Inside C-capsids, the dsDNA molecule occupies evenly throughout the entire interior of the capsid with the 20 Å-diameter dsDNA duplex resolved (Fig 2B)—the first time such detailed features ever observed directly by cryoET. In B-capsids, the scaffolding protein (pUL80, up to 1000 copies/capsid [34, 35]) is organized into a density sphere with an outer and inner diameter of ~700 and ~400Å, respectively. In capsids devoid of genome DNA, a portal complex for DNA translocation is visible at one of the 12 vertices of the capsid (Fig 2C). The viral envelope is pleomorphic (Fig 2A) and its membrane resolved into two leaflets 40Å apart (Fig 2D), sporting sparsely and randomly located, and clearly identifiable glycoprotein spikes on the outer leaflet (Fig 2A and Fig 3A).
We used the following three pieces of evidence to establish the identifications of gB trimers on the viral envelope. First, among HCMV glycoproteins, gB is known to only exist as homotrimer with a combined mass of ~300 kD [36] and is the most abundant complex over 100 kD [37]. This mass is expected to occupy an estimated extracellular volume of ~300 nm3. Among the density spikes decorating the outer leaflet of the viral membrane, only two differently shaped spikes with such volume were identified, suggesting that they might be gB trimer at different conformational states (Fig 3B). Second, the two distinctive side-view shapes—one triangular, Christmas-tree like (71%) and the other rectangular, columnar-tree like (29%) (Fig 3B)—are similar to the side-views of the cryoET reconstructions of HSV-1 gB trimers on purified vesicles in their putative prefusion and postfusion conformations, respectively [27]. Third, we performed subtomographic averaging to these two types of spikes, respectively, in order to examine them with a higher SNR. Both of the averaged models exhibit apparent three-fold symmetry with the symmetric axis perpendicular to the plane of viral membrane, despite slight distortion arising from the inherent “missing wedge problem” of electron tomography (S2 Fig). These three pieces of evidence all point to our tentative assignment of the Christmas tree-shaped and the columnar tree-shaped densities on the HCMV envelope as gB trimers in the prefusion and “postfusion” (quotation marks are used here since the conformation is not really caused by fusion but likely triggered by mechanical stress during virion purification with high-speed centrifugation) conformations, respectively. Indeed, as shown below, the available crystal structure of gB in the postfusion conformation matches perfectly with our final subtomographic average of the columnar tree-shaped density, further validating our assignments.
As mentioned above, we performed subtomographic averaging to characterize the two putative gB conformations at a higher resolution. The significantly enhanced contrast afforded by imaging with VPP at a near-focus condition allowed the clear visualizations of different structures in the reconstructed tomograms. For direct comparison, we also obtained tilt series without using a VPP (referred to as non-VPP tomograms). For the latter data, we had to use a significantly larger defocus value (-4μm) to improve image contrast and record much more tilt series (28 total) in order to obtain a similar number of gB particles for subtomographic averaging due to greater difficulties in distinguishing different glycoprotein morphologies in the tomograms (S1 Fig). In addition, the use of large defocus has necessitated correction for contrast transfer function (CTF): the structure obtained without CTF correction contains phase-inverted, incorrect structure information beyond 25 Å (S4E Fig), as reflected by the broken connections between the ectodomain and the viral membrane in the absence of CTF correction (S3B, S4C and S4D Figs).
In total, 350 particles of the columnar tree-shaped and 1509 particles of the Christmas tree-shaped densities were included for subtomographic averaging. For the columnar tree-shaped structure, all particles were extracted from the VPP tomograms due to ambiguities in distinguishing its slender shape from background noise in the non-VPP tomograms. For the Christmas tree-shaped structure, 874 particles, which came from VPP tomograms, were first used and 635 particles from non-VPP tomograms eventually were also included to further improve resolution. Three-fold symmetry was imposed subsequently to improve SNR and the resolution of the averaged structures. Fourier shell correlation (FSC) analyses indicate that the resolutions for the symmetrized 3D subtomographic average of the columnar tree-shaped and Christmas tree-shaped spikes are 26 Å and 21 Å, respectively, based on the gold-standard criterion (S3A Fig).
The subtomographic average of the columnar tree-shaped spike resolves the two leaflets of the bilayer viral envelope and a prominent (161Å in height) ectodomain (Fig 3C–3E, S5 Movie). The ectodomain density matches well with the crystal structure of the HCMV gB ectodomain trimer [16] (Fig 3F–3H), validating our initial assignment of the columnar tree-shaped density as gB structure in its “postfusion” conformation and re-establishing the validity of our approach. The subtomographic average of our putative prefusion gB densities reveals the two leaflets of the bilayer viral envelope with prominent gB densities attached to both: a prominent ectodomain attached to the outer leaflet (130Å in height) and a globular (about 35Å in height and 26Å in width) endodomain to the inner leaflet (Fig 3I–3K). The ectodomain in the putative prefusion gB is shorter than that in the gB “postfusion” conformation and anchors to the membrane with three well-separated densities, forming a tripod (Fig 3I–3K, S6 Movie). Although no crystal structure of prefusion gB is available to fit into our subtomographic average to directly confirm or refute this prefusion gB assignment, it is believed that herpesvirus gB bears structural and mechanistic similarities to other class III viral fusion proteins, which can be used to aid our assignment. Indeed, the postfusion conformation of HCMV gB ectodomain is similar to the postfusion conformations of all other class III viral fusion proteins [18], including the postfusion VSV G (Fig 1C–1F). The lower portion of the prefusion conformation of the VSV G trimer (Fig 1G–1I) has a tripod shape similar to the lower portion of the Christmas tree-shaped density (Fig 3I). The prefusion VSV G trimer is shorter than—and undergoes drastic domain rearrangements towards—its postfusion conformation [20, 21] (Fig 1); likewise, the Christmas tree-shaped density is shorter than the columnar tree-shaped density. Taken together, these characteristic similarities to the prefusion structure of VSV G corroborate our initial assignment of the Christmas tree-shaped density as the in situ prefusion structure of HCMV gB trimer.
Structure-guided sequence analysis (Fig 4A) indicates that each full-length gB protomer contains an N-terminal ectodomain (residues 87–705), a membrane proximal region (MPR, residues 706–750), a single transmembrane helix (residues 751–771) and a C-terminal endodomain (residues 772–906). For the “postfusion” gB trimer, the ectodomain in the subtomographic average can be divided into a base in contact with the membrane, and two lobes—middle and crown—connected by a neck (Fig 3F). The crystal structure of the ectodomain trimer shows that each protomer consists of five domains: DI, DII, DIII, DIV and DV (Fig 3G and 3H) [16]. Except for DV, these domains can be located in our subtomographic average of the “postfusion” gB (Fig 3F). DI, each containing two fusion loops, is located at the base of the trimer; DII and DIV reside, respectively, in the middle and crown lobes, which are connected by DIII in the neck. DV contains a long loop connected by two short helices and is buried, thus is not resolved in our subtomographic average gB trimer due to the limited resolution.
As detailed in the Method, we employed a combination of manual rigid-body fitting of known domain structures from the existing HCMV gB postfusion structure [16], comparative modeling of DIII based on the homologous DIII from VSV G prefusion conformation [21], followed by optimization by the molecular dynamics flexible fitting (MDFF) method [38], to put forward a provisional domain arrangement model of the prefusion gB (Fig 5). DV was not considered in our domain modeling of HCMV gB prefusion conformation due to the lack of a template structure, since DV was truncated in the crystal structure of postfusion VSV G. MDFF not only optimized the chemical interactions among the fitted domains, but also improved overall model to map correlation coefficient from 0.83 to 0.94 (Fig 5H and 5I, S7 Movie).
The model from MDFF does not include the MPR (residues 706–750), which is proposed to lie between the ectodomain and the transmembrane helix (Fig 4A) and “mask” the fusion loops to prevent their premature (non-productive) association with lipid [39]. Helical wheel projection of the first 15 amino acids of the MPR shows an amphipathic helix (Fig 4B) whose hydrophobic side could interact with the fusion loops. This notion is consistent with our interpretation of DI in the subtomographic averages of both prefusion and “postfusion” conformations, with the fusion loops pointing to and in close proximity to the membrane.
Among herpesviruses, gB and gH/gL are highly conserved and known to form a fusion machinery for virus entry [40]. Previous biochemical studies have indicated that gH/gL regulated fusion activity of gB [41] and might form a complex with gB in virions on the basis of co-immunoprecipitation experiments [42]. Besides gB trimer densities mentioned above, “L”-shaped spikes were also observed protruding outwards from the viral envelope, which we interpret as gH/gL complexes on the basis of size and shape similarities to the gH/gL crystal structure [13, 17]. Moreover, among such “L”-shaped spikes, ~7% were observed to be in contact with the Christmas tree-shaped, prefusion gB trimer, forming a gB-gH/gL complex (Fig 6B and 6C), while others were unbound. No “postfusion” gB trimer have been observed involving in gB-gH/gL complex.
A subtomographic average was obtained by aligning and averaging 49 such gB-gH/gL complexes to investigate the contact sites between prefusion gB and gH/gL (Fig 6C–6F), with a resolution around 30Å reported by calcFSC in PEET. The HCMV gH/gL crystal structure [17] fits well in the “L”-shaped density in the subtomographic average (0.75 of the cross-correlation coefficient between the cryoET map and the model filtered to 30Å, Fig 6D and 6E). This fitting, together with the predicted domain arrangement in the prefusion gB structure (Fig 5H and 5I), reveals that DI of gB may contact the gH subunit of gH/gL (Fig 6D). The contact sites on gB and gH are consistent with the gH-binding site on HSV-1 gB suggested by blocking gH binding to gB with SS55 and SS56 antibodies (epitopes mapped to residues 153–363 of gB) (Fig 5H) [43] and the gB-binding sites on gH/gL suggested by anti-gH/gL antibody LP11 for HSV [13], respectively. Mutagenesis of gH cytotail has led to its proposed role of acting as a “wedge” to split the gB endodomain “clamp” to trigger gB ectodomain refolding [44]. Though the details of their interactions in the endodomain are yet to be resolved, this first observation of gH/gL complex making contact with prefusion gB in situ (Fig 6) supports the notion that receptor binding to gH/gL triggers transformation of gB from prefusion to postfusion conformation.
Since the postfusion conformation of gB is energetically favorable and structurally more stable, it is not surprising that purified recombinant gB so far have all adopted the “postfusion” conformation [12, 16,45]. Therefore, imaging gB in its native, virion environment by cryoET seems to be the necessary approach to obtain the in situ structure in its metastable, prefusion conformation. However, a major challenge in interpreting in situ cryoET structures is the intrinsic poor contrast of tomographic reconstructions due to the use of low electron dose in order to avoid radiation damage to specimen. Poor contrast makes it difficult to identify different molecules or structures for subtomographic averaging. Normally for cellular tomography without phase plate, one could image with a large defocus value to achieve better contrast, aiding in distinguishing densities with different characteristics for subtomographic averaging. However, such approach only offers limited improvements in contrast (S1 Fig), and difficulties still exist in identifying the slender gB in postfusion conformation in our tomograms. This experience is consistent with two previous cryoET studies on HSV-1 gB structures, in which large defocus values were used to increase contrast to facilitate subsequent subtomographic averaging, yet the resulting structure either is at much lower resolution [26] than reported here or has led to controversial interpretations [27]. The greatly improved contrast afforded by VPP technology allowed the differentiation of various glycoprotein structures based on their characteristic appearances on the virion membrane (Fig 7A and 7B; S1 Fig). Therefore, cryoET with VPP offers a clear advantage in resolving structures of proteins in the native environments, enabling their identifications and subtomographic averaging to obtain structures of multi-functional states, as also demonstrated by the existence of two states of 26S proteasome inside neurons [31].
A vital step of herpesvirus infections is the fusion of viral and cell membranes, a complicated process involving at least three conserved proteins—gB, gH and gL. The in situ structures of gB at both prefusion and “postfusion” conformations reported here can shed lights on conformational changes of gB during membrane fusion and inform how herpesvirus entry into cell (Fig 7). Prior to fusion, gB needs to be maintained at its inactive, metastable prefusion conformation (Fig 7A). The maintenance of this metastable conformation possibly involves a properly-folded endodomain of gB, since removal of the endodomain caused gB ectodomain to adopt the postfusion conformation [46]. In addition, the direct observation in our cryoET structure of gB-gH/gL complex (Fig 6) and its isolation by co-immunoprecipitation [42] both suggest that the metastable ectodomain of gB might also be stabilized through the interaction with the ectodomain of gH subunit (Fig 7C). Host receptor-binding to gH/gL complex would trigger a conformational change in gH/gL cytotail and its dissociation from, and the destabilization of, the endodomain of gB, which in turn triggers the massive conformational changes of gB ectodomain to expose its fusion loops (step 1). Subsequently, DIII central helix extends, unfurling other domains and swinging the fusion loops to engage with the host membrane (step 2). Facilitated by the intrinsic fluidity in the plane of the membrane, the refolding of gB domains to the lower-energy, postfusion conformation, in which its ectodomain C-terminal end and the fusion loops must come together, leads to fusion of the two membranes and the release of viral DNA-containing capsid into cytoplasm (step 3). In the absence of receptor binding as in the situation of this study, mechanical stress to the membrane caused by such means as high-speed centrifugation could also destabilize the membrane-associated endodomain, triggering metastable prefusion gB to undergo the cascade of transformation events, possibly accompanied by the exposure of the fusion loops (step 1). Lacking host cell membrane, these events, with exposed fusion loops eventually encountering and inserting its hydrophobic moieties into the viral membrane, will be followed by refolding of other domains into the stable, “postfusion” conformation (step 3). Notably, the topology of the conformational change during step 2 to step 3 would preclude transiting from prefusion to postfusion conformation without breaking the three-fold symmetry. Indeed, monomeric intermediates of VSV G have been observed both in solution and on the surface of virions at intermediate pH conditions [22, 23].
In our model, the fusion loops of prefusion gB point to and are in close proximity to the viral membrane, possibly buried within a hydrophobic “mask” of MPR, which is attached to the C-terminal end of the gB ectodomain crystal structure. This membrane-proximal location of the gB fusion loops is the same as that based on the cryoET structure of the HSV-1 gB/anti-fusion loop 2-antibody at 5nm resolution [26] and is consistent with the fusion loop locations in all known atomic structures of classes I and III viral fusion proteins, including influenza HA [47], HIV env trimer [48], VSV G [21] and others [6]. Notably, our model is in stark contrast to the exposed fusion loops assigned to the membrane-distal tips of the “short-form” HSV-1 gB structures [27], which were obtained by cryoET of purified gB-containing vesicles. The ectodomain of the “short-form” vesicular HSV-1 gB structure is 15% shorter in height and 23% wider in diameter than that of our in situ HCMV gB structure, despite both sharing the Christmas tree shape (S5 Fig). Superposition of the domain assignment obtained by the hierarchical fitting approach [27] into the “short-form” HSV-1 gB structure shows that the densities projecting from the lower whorl of the Christmas tree-shaped trimer were unaccounted for (S5B Fig). Moreover, placing the same domain assignment into our in situ HCMV gB prefusion structure reveals that the fusion loops in this assignment are projecting out of the cryoET map, yet the leader density of the map is not accounted for (S5C Fig). When filtering the pseudoatomic model to 25Å, the cross-correlation coefficient is 0.74, as compared to 0.93 of our prefusion structure. We believe that an exposed fusion loop orientation of prefusion gB is unlikely for both chemical and biological reasons—exposed hydrophobic moieties are chemically unfavorable in solution and can lead to unproductive membrane insertion during infection. Indeed, the “short-form” HSV-1 gB structure was cautiously interpreted as an ambiguous “prefusion and/or intermediate” conformation [27], probably to reconcile these contradictory considerations.
Secondary structure prediction indicates that the endodomain is helix-rich (~50%) (Fig 4A). Our results suggest that gB endodomain undergoes significant conformational changes, from prominently visible/stable in the prefusion structure (Fig 3I and 3K), to invisible/flexible in the “postfusion” structure (Fig 3C and 3E). Proteolysis and circular dichroism analyses of the endodomain of the highly homologous HSV-1 gB posit that gB endodomain clamps the viral membrane and stabilizes gB in its prefusion conformation [44,49]. This proposed model is supported by studies on truncation and substitution mutations in endodomain [44,46]. The structured endodomain resolved in the recent crystal structure of full-length gB was thought to be similar to that in prefusion gB [24]. Detergent solubilization of the membrane may be responsible for the postfusion conformation of its ectodomain. Our observation of the endodomain structure of HCMV gB changing from a stable, prefusion conformation (Fig 3I) to a flexible, postfusion conformation (Fig 3E) is consistent with its proposed role in stabilization of gB prefusion conformation on native viral membrane [24].
Human fibroblast MRC-5 cells (ATCC) were cultured in Eagle's Minimum Essential Medium (EMEM, ATCC) with 10% fetal bovine serum (FBS, Omega scientific: FB-11). Cells were grown in T-175 cm2 flasks to 90% confluence and infected with HCMV strain AD169 (ATCC, Rockville, MD) at a multiplicity of infection (MOI) of 0.1–0.5, and incubated for about 7 days. Once the cells showed 100% cytopathic effect, the media were collected and centrifuged at 10,000 g for 15 min to remove cells and large cell debris. The clarified supernatant was collected and centrifuged at 60, 000 g for 1 hour to pellet HCMV virions. Pellets were resuspended in 20mM phosphate buffered saline (PBS, pH 7.4), loaded on a 15%–50% (w/w) sucrose density gradient, and centrifuged at 60,000 g for 1 hr. After the density gradient centrifugation, three light-scattering bands were observed in the density gradient: top, middle and bottom. The middle band contained both HCMV virions and NIEPs (particles with intact viral envelopes as judged by negative-staining EM) and was collected, diluted in PBS and then centrifuged at 60,000 g for 1 hour. The final pellet was resuspended in PBS for further cryoET sample preparation.
VSV virion (Indiana serotype, San Juan strain) samples were produced as previously described [50]. Particularly, the inoculum was passaged multiple times in Hela cells with a very low multiplicity of infection (MOI), 0.001, to suppress the truncated defective-interference particles. The full VSV particles were isolated in a sucrose gradient and the final inoculum was also plaque-purified in Hela cells. We then pelleted the VSV virions at 30,000g for 2 hours and resuspended them in PBS. The stock was subjected to another low speed centrifugation at 12,000g for 5min in a desktop centrifuge to remove large aggregates. After resuspension, the pellets were banded on a 10ml density gradient containing 0–50% potassium tartrate and 30–0% glycerol. The virions-containing band was collected, diluted in PBS, pelleted at 30,000g for 2 hours, resuspended in PBS and kept in 4°C refrigerator for further cryoET sample preparation.
An aliquot of 2.5 μl of the sample mixed with 5-nm diameter gold beads were applied onto freshly glow-discharged Quantifoil Holey Carbon Grids. Grids were blotted and plunge-frozen in liquid ethane cooled by liquid nitrogen using an FEI Mark IV Vitrobot cryo-sample plunger and were stored in liquid nitrogen before subsequent usage. CryoEM imaging and cryoET tilt series acquisition were performed with SerialEM [51] on an FEI Titan Krios 300kV transmission electron microscope equipped with a Gatan imaging filter (GIF), a Gatan K2 Summit direct electron detector, and with or without a Volta phase plate (VPP). Tilt series were recorded by tilting the specimen covering the angular range of -66° to +60° (starting tilt from -48° to +60°, then from -50° to -66°) with 2° or 3° interval, with a nominal magnification of x53,000 (corresponding to a calibrated pixel size of 2.6 Å) and a cumulative electron dose of 100~110 e-/Å2. Exposure time was multiplied by a factor of the square root of 1/cosα (in which α = tilt angle), and the exposure time at 0° was set at 1.2s for the tilt step-size of 2° or 1.6s for the tilt step-size of 3°. Movies were recorded with the frame rate of 0.2 frame/s on a Gatan K2 Summit direct electron detector operated in counting mode with the dose rate of 8–10 e-/pixel/s. An energy filter slit of 20 eV was chosen for the GIF. For imaging with VPP, defocus value was targeted at -0.6μm. Note, one of the benefits of using a phase plate is that the CTF is insensitive to the sign of the defocus value being negative (underfocus) or positive (overfocus) [52]. VPP was advanced to a new position every tilt series, followed by a 2 min waiting for stabilization, and pre-conditioned by electron illumination with a total dose of 12 nC for 60s to achieve a phase shift of ~54° as previously described [53]. For tilt series obtained without VPP, the defocus value was maintained at around -4μm while other imaging parameters were kept the same as those for the tilt series with VPP.
Frames in each movie of the raw tilt series were aligned, drift-corrected and averaged with Motioncorr [54] to produce a single image for each tilt angle. Both sets of tilt series, collected with and without VPP, were reconstructed with IMOD 4.8 software package [55] in the following six steps. All images in a tilt series were coarsely aligned by cross-correlation (step 1) and then finely aligned by tracking selected gold fiducial beads (step 2). The positions of each bead in all images of the tilt series were fitted into a specimen-movements mathematical model, resulting in a series of predicted positions. The mean residual error (mean distance between the actual and predicted positions) was recorded to facilitate bead tracking and poorly-modeled-bead fixing (step 3). With the boundary box reset and the tilt axis readjusted (step 4), images were realigned (step 5). Finally, two tomograms were generated by weighted back projection and simultaneous iterative reconstruction technique (SIRT) method, respectively (step 6). For data collected without VPP, contrast transfer function (CTF) was corrected with the ctfphaseflip program [56] of IMOD in step5. The defocus value for each image in one tilt series was determined by CTFTILT [57], and the estimated defocus value of each image was used as input for ctfphaseflip.
Subtomographic averaging was performed using PEET 1.11 [58, 59]. High contrast SIRT tomograms were 4× binned by the binvol program of IMOD to facilitate particle picking. Particles were picked manually in IMOD as follows. For distinct conformations of VSV G and HCMV gB on viral envelope, two points (head and tail) in one contour were used to define one particle (glycoprotein)—head is the membrane-proximal end of the protrusion density while tail is the membrane-distal end. An initial motive list file, a RotAxes file and three model files containing the coordinates of head, centroid and tail for each particle were generated by stalkInit in PEET. In total, we manually picked 337 long-form particles from 5 VPP tomograms of VSV, and 350 columnar tree-shaped particles and 886 Christmas tree-shaped particles from 11 VPP tomograms of HCMV. Besides, 637 Christmas tree-shaped particles were picked from 28 non-VPP tomograms, averaged either alone or together with those from the VPP tomograms for prefusion gB.
For the reconstruction of the long-form VSV G, subtomographic averaging was performed first with 4× binned SIRT tomograms using the sum of all particles as the initial reference. Through stalkInit, each particle’s tilt orientation (i.e., the axes normal to the membrane) was already coarsely aligned to Y axis, but its twist orientation (i.e., the angle around the axis) was randomized. Therefore, in the first refinement cycle, we set the angular search range 180° max (-180° to 180°) with 9° step in Phi (Y axis), and 5° (-5° to 5°) max with 1° step in both Theta (Z axis) and Psi (X axis), and search distance 3 pixels along all three axes. Due to the known symmetry of postfusion VSV G, the resulting averaged structure was then trimerized and used as the reference of the next refinement cycle. The trimerized structure was the sum of each refined particle and its two symmetrical copies—the two symmetrical copies have the same position and tilt orientation as the refined particle, but twist orientation differed by either 120° or 240°. For subsequent refinement cycles, the newly trimerized structure from the last refinement cycle was used as reference, with both angular and distance search ranges narrowing down gradually. After four refinement cycles, the averaged structure converged based on no further improvement in resolution. The following refinement cycles were performed with 2× binned tomograms reconstructed by weighted back projection, after up-sampling (generations of 2× binned model files and updates of corresponding motive list files from the latest refinement cycle), with small search distance range (4 pixels) and narrow angular search range (-20° to 20°). The reference was updated from the averaged structure of the last refinement cycle (trimerized). For particles with distance of <1 pixel and twist angle difference of <1°, the one representative with lower cross-correlation coefficient was treated as duplicate particle and removed during the refinement. The averaged structure, contributed by 330 particles, converged after eight refinement cycles and was filtered to the final resolution, calculated by calcFSC in PEET based on the 0.143 FSC criterion.
Reconstructions of columnar tree-shaped and Christmas tree-shaped particles on HCMV envelope followed the same refinement procedure as the reconstruction of long-form VSV G, except that trimerization was only applied after three-fold symmetry became apparent in the averaged structures. With the removal of duplicate particles, the final averaged structures of the postfusion (columnar tree-shaped) and prefusion (Christmas tree-shaped) conformations were obtained from 350 particles and 1509 particles, respectively. Furthermore, gold-standard FSC calculations for the structures were performed afterwards by splitting the original dataset of each conformation into two independent groups. The same refinement procedure used above was applied to the two newly-generated groups independently. Upon the convergence of the averaged structures, FSC were calculated by calcUnbiasedFSC in PEET (S3A Fig.).
For the reconstruction of the short-form VSV G, 65 particles were manually picked from five tomograms with single point to define the centroid position. Each particle was manually rotated around X, Y, Z axes to a similar orientation (both the tilt orientation and twist angle) in IMOD slicer window. By slicer2MOTL in PEET, the initial motive list files for subtomographic averaging were generated from the corresponding X, Y, Z rotation degrees. For the Angular Search Range, small search range was set during all seven refinement cycles. The final subtomographic average was Gaussian filtered with width 7 using the “volume filter” tool in UCSF Chimera [60]. Due to the limited number of particles (49 particles), HCMV gB-gH/gL complex was reconstructed with the same strategy above.
We used IMOD [61] to visualize reconstructed tomograms and UCSF Chimera to visualize the subtomographic averages in three dimensions. The crystal structures of prefusion VSV G (PDB: 5I2S) [21], postfusion VSV G (PDB: 5I2M) [20], HCMV postfusion gB (PDB: 5CXF) [16] and gH/gL part from HCMV pentamer (PDB: 5VOB) [17] were fitted into subtomographic averages of prefusion G, postfusion G, postfusion gB and gB-gH/gL complex, respectively, with the tool fit in map in Chimera. Segmentation and surface rendering for the membrane and tegument proteins were done by the tools volume tracer and color zone in Chimera. All membrane glycoproteins were placed back on the viral membrane according to their locations in the original tomogram. A published structure of HCMV capsid with inner tegument protein [11] was filtered to 10 Å and placed back at the same position of the capsid in tomogram.
As outlined below, we employed a combination of initial manual fitting of known domain structures, followed by simulation with MDFF program [38] to generate a gB prefusion model based on our cryoET prefusion gB trimer density map and the existing gB ectodomain postfusion crystal structure (PDB: 5CXF) [16]. First, the ectodomain in the subtomographic averaged density map of prefusion gB trimer was segmented out and its symmetric axis obtained with Chimera’s “volume eraser” tool and “measure symmetry” command, respectively. Second, Chimera’s “fitmap” command with “global search” and 15Å-resolution options was used to refine 1000 initial random DI placements, resulting in 28 refined fitted positions, each with a correlation coefficient (between the fitted model and the density map) and a “clash volume fraction” value (between symmetry-related copies). We chose the fitted position with the largest fitting score, defined as the correlation coefficient subtracted by the “clash volume fraction” penalty value (Fig 5A). Third, we obtained our initial DIII by computationally mutating the DIII model from the existing hypothetic model of EBV prefusion gB [14], as it is known to differ substantially from its postfusion conformation for both herpesvirus gB [14, 62] and homologous VSV G [21]. Compared to that in the postfusion gB, the central helix α4 in DIII in the prefusion gB is bent in order to fit into the top of the Christmas tree-shaped density. This bent varies from only ~30° in our proposed HCMV gB prefusion structure to ~90° in VSV G (Fig 5D) [21] and ~180° in influenza HA [47] and HIV env [48]. This DIII model, and the models of DII and DIV from the gB postfusion crystal structure were manually fitted as rigid bodies into our prefusion gB trimer cryoET density to produce a composite model with the above obtained DI trimer model by referencing the prefusion VSV G crystal structure. Connecting loops were then added to this composite model through the Modloop server [63]. Fourth, the resulting trimer model was used as the initial model for MDFF simulations [38] with grid force scale of 0.3. Secondary structure, cis peptide and chirality restraints were imposed during MDFF simulations. Simulations were performed with NAMD 2.12 [64], using the CHARMM36 force field with CMAP corrections [65].
Secondary structures for residues 707–906 of gB were predicted with Phyre2 [66].
|
10.1371/journal.pbio.2003885 | Neutrophils kill the parasite Trichomonas vaginalis using trogocytosis | T. vaginalis, a human-infective parasite, causes the most common nonviral sexually transmitted infection (STI) worldwide and contributes to adverse inflammatory disorders. The immune response to T. vaginalis is poorly understood. Neutrophils (polymorphonuclear cells [PMNs]) are the major immune cell present at the T. vaginalis–host interface and are thought to clear T. vaginalis. However, the mechanism of PMN clearance of T. vaginalis has not been characterized. We demonstrate that human PMNs rapidly kill T. vaginalis in a dose-dependent, contact-dependent, and neutrophil extracellular trap (NET)-independent manner. In contrast to phagocytosis, we observed that PMN killing of T. vaginalis involves taking “bites” of T. vaginalis prior to parasite death, using trogocytosis to achieve pathogen killing. Both trogocytosis and parasite killing are dependent on the presence of PMN serine proteases and human serum factors. Our analyses provide the first demonstration, to our knowledge, of a mammalian phagocyte using trogocytosis for pathogen clearance and reveal a novel mechanism used by PMNs to kill a large, highly motile target.
| The human parasite Trichomonas vaginalis is a large unicellular, motile eukaryote that causes a highly prevalent sexually transmitted infection in humans: trichomoniasis. While harmful effects of trichomoniasis are associated with inflammation, the immune response to the parasite is sorely under-characterized. Neutrophils are known to be important players in the host response to T. vaginalis, but it was not previously known how effective they are at killing the parasite and the mechanism(s) they use to do this. Here, we show that human neutrophils use trogocytosis, a previously undescribed neutrophil mode of microbial killing, to kill T. vaginalis. Trogocytosis is a process by which a cell takes “bites” of a neighboring cell, a process also referred to as “nibbling.” Using 3D and 4D live imaging, we show that neutrophils rapidly surround and trogocytose T. vaginalis, prior to parasite death. We rule out whole parasite engulfment (phagocytosis) and the employment of neutrophil extracellular traps (NETosis) in this rapid contact-dependent killing. We also show that antibody–fragment crystallizable (Fc) receptor interactions mediate neutrophil trogocytosis and killing and that serine proteases, commonly employed by neutrophils for microbial degradation, additionally play a role in parasite “nibbling”.
| Trichomonas vaginalis is a unicellular, flagellated eukaryote that lives as an obligate extracellular parasite, restricted to humans [1,2,3]. T. vaginalis causes the most common nonviral sexually transmitted infection (STI) worldwide: trichomoniasis [1,4]. The World Health Organization reports approximately 275 million cases each year [1,4]. However, annual reported cases are likely underestimated because the Centers for Disease Control and Prevention estimates that at least 50% of cases are asymptomatic. In the United States, an estimated 1.1 million new infections occur every year [5]. T. vaginalis is routinely treated with microaerophilic-specific 5-nitroimidazoles: metronidazole and tinidazole [1]; however, metronidazole-resistant strains are on the rise [6,7,8,9,10]. Additionally, T. vaginalis infection is associated with increased transmission of and susceptibility to HIV, as well as increased progression of cervical cancer in human papilloma virus (HPV)+ individuals [11,12,13,14,15,16,17]. T. vaginalis infection is also associated with bacterial vaginosis, suggesting that infection may disrupt the microflora. Notably, T. vaginalis was recently classified as a neglected parasitic infection [5] because it disproportionately affects underserved communities [5,18] and contributes to reproductive health disparities; trichomoniasis is linked to pelvic inflammatory disorder, premature and underweight infant birth, infertility, and endometriosis [5,11].
T. vaginalis complications are thought to have an inflammatory origin. Clinical manifestation of trichomoniasis is associated with an influx of neutrophils—also known as polymorphonuclear cells (PMNs)—to the vaginal mucosa [17]. Studies in preliminary mouse models have shown robust vaginal PMN recruitment after inoculation [19], and human PMNs in vitro were shown to swarm and attack T. vaginalis [20]. Interleukin 8 (IL-8), a PMN recruitment chemokine [21], is secreted from human vaginal ectocervical cells (Ect-1) [22], human prostate epithelial cells [23], and human monocytes [24,25,26,27] after T. vaginalis encounter and is found in vaginal secretions of infected patients [25]. PMN involvement during trichomoniasis is likely a major contributor to inflammation-associated pathology because PMNs are potent destructive cells and implicated in other mucosal inflammatory pathologies [21].
PMNs have a diverse arsenal of potent antimicrobial activities and have 3 main modes of killing pathogens: phagocytosis, extracellular degranulation, and casting of Neutrophil Extracellular Traps (NETosis) [21,28]. PMN phagocytosis is aided by pathogen opsonization via antibody and complement [29]. Several types of toxic granules containing numerous effectors, including reactive oxygen species (ROS), pore-forming toxins, elastases, and proteases [30], may be released into the phagosome. When the extracellular degranulation mode of killing is employed, the toxic granules are released extracellularly. Upon activation, PMN granule components work in concert with the NADPH oxidase to achieve H2O2-mediated oxidative bursts of ROS [31,32]. The use of NETosis results in the release of nuclear contents to form a NET, in which a scaffold of DNA studded with granular components ensnares nearby pathogens. The pathogen is then subsequently engulfed or subjected to toxic granule exocytosis [21,28,33]. The 3 known killing mechanisms used by PMNs are not mutually exclusive, and multiple mechanisms can be used in the attack of a single target.
Recently, PMNs have been shown to perform trogocytosis (trogo = to nibble), a mechanism by which fragments, or “bites,” are taken from one cell by a neighboring cell [34,35,36]. Trogocytosis has been described as a way for immune cells to exchange membrane proteins [34,37,38,39,40,41,42,43] and as a process recently shown to be exploited by intracellular bacteria for cell–cell transfer [44,45]. Parasitic amoebae have also been shown to injure or kill host cells via trogocytosis [46,47,48]. It is notable that immune cell trogocytosis to date has not been demonstrated to be lethal to a pathogen, although there is a recent report linking monocyte trogocytosis to cancer cell death [49].
The mechanisms that PMNs use to control T. vaginalis have not been examined. Here, we show that PMN killing of T. vaginalis is mediated in a contact-dependent, NET-independent engulfment process, during which bites of T. vaginalis are taken by PMNs prior to parasite death, demonstrating a previously undefined role for PMN trogocytosis in pathogen clearance. These studies also reveal a novel mechanism that PMNs can use to kill large, highly motile targets.
PMNs have been reported to be abundant during T. vaginalis infection [1,20], but the mechanism(s) used to kill T. vaginalis has not been characterized. We therefore employed an in vitro cytoxicity assay to determine whether PMNs can kill T. vaginalis. In this assay, T. vaginalis were labeled with Cell Tracker (CT), and PMNs were labeled with Carboxyfluorescein succinimidyl ester (CFSE). Both were then cocultured for 2 hours (Fig 1A). Afterwards, numbers of live T. vaginalis remaining (CFSE-CT+ cells) were quantified to determine whether PMNs are capable of killing T. vaginalis (Fig 1B, S1 and S2 Figs, and S1–S3 FCS [fluorescence correlation spectroscopy] files). Cell numbers in each condition were normalized to counting beads to ensure that equivalent volumes were analyzed from sample to sample. By comparing the numbers of CFSE-CT+ cells in T. vaginalis–alone conditions to those in which PMNs were present at 1:1 ratio of T. vaginalis:PMN, we observed an intermediate level of killing (approximately 30%) (Fig 1C and S1 Data). Decreasing the multiplicity of infection (MOI; adding more PMNs) resulted in more efficient killing of T. vaginalis (Fig 1C), with more than 90% of T. vaginalis killed at MOI 0.125 (Fig 1C). We found T. vaginalis killing by PMNs to be reproducible across many different blood donors and the efficiency to be relatively consistent because PMNs from 19 different donors showed similar levels of T. vaginalis cytotoxicity at MOIs within the dynamic range of our titration (Fig 1D and S1 Data).
Because T. vaginalis is highly motile and slightly larger (10–15 um in diameter) [50] than PMNs (average of 8.85-um diameter in suspension) [51], we next asked whether PMN killing of T. vaginalis required contact or whether it could be mediated by modes of PMN killing that do not necessarily require contact: extracellular release of toxic granules and ROS or NETosis, a mechanism involving the casting of DNA nets to capture and kill pathogens [52]. To test whether contact-independent degranulation plays a role in PMN killing of T. vaginalis, we first adapted our T. vaginalis cytotoxicity assay to a trans-well apparatus in which top and bottom chambers share continuous media but are separated by a membrane containing 0.4-um pores (Fig 2A). We found that PMNs were not able to kill T. vaginalis if they were in separate chambers (Fig 2B and S1 Data). We next asked whether this was due to failure to activate PMNs because T. vaginalis–PMN contact may be necessary to trigger degranulation. To test this, we compared the killing of labeled T. vaginalis in the top chamber when PMNs in the bottom chamber were stimulated by either phorbol-myristate acetate (PMA), a standard strong positive stimulus for PMN activation, or were co-incubated with an equivalent amount of unlabeled T. vaginalis in the bottom chamber (Fig 2A, bottom panel). Both T. vaginalis and PMA were found to activate PMNs, as assessed by H2O2 secretion into the media (S3A Fig and S1 Data); however, killing of labeled T. vaginalis in the top wells was not detected (Fig 2B and S1 Data). To eliminate the possibility that PMN viability was affected in these assays, we assessed viability and found that neither T. vaginalis nor PMA stimulation affected the viability of PMNs (S3B Fig and S1 Data). To assess the role of H2O2 in PMN killing of T. vaginalis, we performed cis-well cytolysis assays, coculturing labeled PMNs and T. vaginalis in the presence of catalase, which inactivates H2O2 (S4A Fig and S1 Data). We found no abrogation of PMN ability to kill T. vaginalis (Fig 2C and S1 Data) under conditions in which catalase does not affect PMNs or T. vaginalis viability (S4B Fig and S1 Data). These data indicate that H2O2 -mediated oxidative bursts do not play a critical role in T. vaginalis killing by PMNs.
We next examined whether NETosis contributes to PMN killing of T. vaginalis by performing our cytolysis assay in the presence of DNase I, which has been shown to degrade NETs and allow pathogen escape [53]. We found that the addition of DNase I, under conditions in which control experiments showed that NETs induced by PMA stimulation of PMNs were degraded (S4C Fig and S1 Data), did not affect the ability of PMNs to kill T. vaginalis in our cis-well assays (Fig 2D and S1 Data). Additionally, DNase I did not affect viability of either cell type in the assay (S4D Fig and S1 Data). These data strongly indicate that NETosis does not contribute to PMN killing of T. vaginalis.
Taken together, the data shown in Fig 2B demonstrating that PMNs kill T. vaginalis by a contact-dependent mechanism, and the significant shift of CFSE+ cells into the CFSE+CT+ quadrant suggesting uptake of CT+ T. vaginalis by CFSE+ PMNs (Fig 1B), indicate that PMNs may kill T. vaginalis via engulfment. To test this, we performed our cytotoxicity assay in the presence of inhibitors of actin polymerization and phosphoinositide 3-kinase (PI3K) signaling, both of which have been shown to be necessary for phagocytosis and trogocytosis [36,47,54]. We found that while neither reagent adversely affected the viability of either cell type (S5A Fig and S1 Data), both actin polymerization inhibitor Cytochalasin D and PI3K signaling inhibitor wortmannin significantly decreased PMN killing of T. vaginalis (Fig 3A and 3B and S1 Data). Treatment with Cytochalasin D and wortmannin also dramatically decreased the occurrence of CFSE+CT+ double positives (S5B and S5C Fig and S1 Data), consistent with PMN killing T. vaginalis using engulfment.
Although T. vaginalis is slightly larger than PMNs, the CFSE+CT+ double positive population displayed a smear gradient rather than a clear population shift (Fig 1B), inconsistent with PMNs phagocytosing whole parasites. In contrast, these data are consistent with a mechanism called trogocytosis (trogo = to nibble), which PMNs have been recently described to use to capture membranes of neighboring cells [34,35,36]. A related phenomenon, amoebic trogocytosis, has also been shown to be utilized by amoeba to kill host cells [47]. As a first step towards testing whether T. vaginalis killing proceeds by trogocytosis, a previously undescribed PMN killing mechanism, we visualized CFSE+CT+ events using Imaging Flow Cytometry. We found that CFSE+CT+ cells are PMNs that appear to contain fragments of T. vaginalis (Fig 3C). Using the spot count analysis function of IDEAS software (Seattle, WA), a majority (>90%) of PMNs were found to colocalize with “bites” of T. vaginalis material at the end of the cytotoxicity assay, and the majority of double positives had between 1 and 4 CT+ spots (Fig 3D). Furthermore, the Internalization Erode algorithm of IDEAS software indicated that a vast majority of these spots are internal to PMNs (values >0) (Fig 3E) and not clinging to the circumference of the cell (values <0).
Moreover, to examine whether PMN interaction with T. vaginalis differs depending on whether the parasite is alive or dead, we compared living and heat-inactivated (dead) T. vaginalis cocultured with PMNs and found that dead parasites were engulfed intact, i.e., phagocytosed (Fig 3F). We did not observe small spots of T. vaginalis material in PMNs cocultured with heat-inactivated parasites (Fig 3F and S6A Fig), and cocultures of PMNs with heat-inactivated parasites contained a clear population of CT+ CFSE+ cells that had equal intensity to parasites alone (CT+CFSE−) (S6B Fig, S1 Fig, and S4 FCSfile). These data argue against the interpretation that spots of live T. vaginalis material observed in PMNs result from degradation of parasites that were engulfed whole and instead indicates that bites are taken prior to T. vaginalis death.
To determine whether “bites” of T. vaginalis material are found in PMNs before the death of the parasite, we devised a live imaging strategy to monitor cell interactions and cell death. The T. vaginalis surface was stably labeled with an amine-reactive Biotin and streptavidin Alexa-488 (green), and PMNs were left unlabeled and visualized with bright field imaging. Propidium Iodide (PI; red), a cell-impermeant nucleic acid stain, was added to the media as an indicator of cell death. We found that free-swimming parasites were sought after and surrounded by groups of PMNs (Fig 4A first 3 time points, and S1 & S2 Videos). We then observed transfer and accumulation of green signal into the PMNs (Fig 4A middle 3 time points, and S1 & S2 Videos). At least several minutes later, we observed membrane breach and death of the parasite, as indicated in red (Fig 4A last 2 time points, and S1 & S2 Videos). PI-positive (dead) parasites were clearly discernable from PI-positive (dead) PMNs, the latter having characteristic multilobed nuclei. To rule out the possibility that T. vaginalis were actively shedding fragments of their surface and PMNs were merely passively endocytosing them, we performed the same experiments with Jurkat cells and never observed any shedding of green material from T. vaginalis nor appearance of green signal inside Jurkat cells (S7 Fig).
We observed 96 total killings using 11 different human donors’ PMNs and found that generally 3 to 6 PMNs were involved in the swarm before parasite death was achieved (Fig 4B and S1 Data). We found that an average of 3 to 8 “bites” were usually taken from a parasite before its death (Fig 4C and S1 Data), and in some cases, over 20 bites were observed. We also assessed the time that elapsed between the first bite taken and parasite death (becoming PI positive) and determined that in a majority of cases, 5 to 12 minutes passed between the first bite and T. vaginalis death, with an average of 8.3 minutes elapsed (Fig 4D and S1 Data). These data are inconsistent with “bites” being derived via PMN phagocytosis of dead T. vaginalis fragments generated through contact-dependent degranulation (CDD). We also observed a small percentage (9%) of parasite deaths occurring before any visible “bites,” which may represent death induced by CDD. Alternatively, it is possible that “bites” were not detected due to limitations of 2D imaging.
Next, to increase our confidence that “bites” of T. vaginalis material are indeed engulfed by PMNs, we visualized the early stages of PMN–T. vaginalis interactions using live HyVolution 3D Confocal microscopy (Leica, Wetzlar, Germany). For these experiments, we labeled PMNs with CT, labeled the T. vaginalis surface with Alexa 488, and added PI to ensure that all parasites were alive at the time of imaging. Only early stages of PMN–T. vaginalis interactions were visualized because larger PMN aggregates confounded individual PMN boundary discernment. Using 3D reconstruction and Z-clipping to virtually slice through the inside of cells, we found that T. vaginalis “bites” are internalized by PMNs rapidly after encounter (Fig 4E & 4F and S3 Video).
To assess the mechanism of PMN killing by trogocytosis, we examined the role of PMN proteases in parasite degradation and killing. The most abundant and best-characterized PMN proteases are serine proteases (cathepsin G, neutrophil elastase, proteinase 3, and neutrophil serine protease 4), which mediate major functions in tissue destruction and microbial degradation [55]. Therefore, we tested whether pre-incubation of PMNs with the serine protease inhibitor AEBSF would inhibit PMN trogocytosis of T. vaginalis. Using live imaging, labeling parasites and PMNs as described above, we observed that AEBSF almost completely inhibited “bites” taken from T. vaginalis (Fig 5A and 5B and S1 Data). While we did observe reduced swarming motility of PMNs in the presence of AEBSF, consistent with the involvement of serine proteases in chemokine processing [56], PMNs were still observed to bind to and restrain T. vaginalis (Fig 5A). To prevent artifacts due to reduced swarming, we only imaged parasites that were in direct contact with PMNs and we normalized the number of bites observed to the number of PMNs present in the swarm (Fig 5B). A flow cytometry–based cytolysis assay was also used to assess PMN killing of T. vaginalis after 2 hours, in the presence of AEBSF. This method confirmed the live imaging data showing that parasite killing was reduced by more than 85% in the presence of AEBSF (Fig 5C and S1 Data). In contrast, phagocytosis of dead (heat-inactivated) T. vaginalis was only moderately—and not statistically significantly—affected by AEBSF (S8A and S8B Fig and S1 Data), indicating that PMNs are still able to surround and engulf T. vaginalis in the presence of the inhibitor. Together, these data demonstrate a critical role for PMN serine proteases in the trogocytosis and killing of T. vaginalis.
The killing activity of amoebic trogocytosis requires surface receptor engagement [47]. Trogocytosis by mammalian leukocytes has also been shown to be receptor mediated by antigen receptor-binding interactions in lymphocytes [43] and by antibody–fragment crystallizable (Fc) receptor interactions in phagocytes [36]. To determine whether PMN trogocytosis of T. vaginalis is mediated by serum factors opsonizing the surface of the parasite in our assay, we first devised staining panels to test whether antibody and iC3b complement factor in our human serum bind to the parasite. We found that both are present on the parasite surface after incubation with human serum (Fig 6A and 6B and S1 Fig, and S5–S10 FCSfiles), consistent with prior reports that serum opsonizes T. vaginalis [20,57]. We then tested whether removing the 10% human serum in our cytotoxicity assay would affect the ability of PMNs to kill T. vaginalis. In the absence of human serum, PMNs were unable to kill T. vaginalis (Fig 6C and S1 Data), and CFSE+CT+ double positive cells were reduced (Fig 6E and S1 Data), indicating that parasite opsonization is required for optimal trogocytosis and killing. We also found that 10% human serum was required for PMN phagocytosis of dead (heat-inactivated) T. vaginalis, consistent with a role for human serum factors in T. vaginalis opsonization (S8C Fig and S1 Data). Finally, we performed our T. vaginalis cytotoxicity assay in the presence of Fc receptor–blocking agent and found that T. vaginalis killing was decreased moderately overall and by around 50% in the dynamic range of the assay (MOI 0.25) (Fig 6D and S1 Data), consistent with antibody opsonization mediating PMN cytotoxicity towards T. vaginalis. We similarly observed a decrease in CT+CFSE+ cells in cocultures treated with Fc receptor–blocking agent (Fig 6E and S1 Data), demonstrating a role for Fc receptors in trogocytosis of T. vaginalis. These effects were not due to adverse effects of the Fc-blocking agent on T. vaginalis or PMNs alone because neither affected viability (S9 Fig and S1 Data).
Despite the high prevalence and adverse inflammatory outcomes of T. vaginalis infection, the immune response to this parasite has been poorly characterized. PMNs are known to be major mediators of inflammation and are abundant at the sight of T. vaginalis infection [1,17], but neither their effectiveness nor the mechanism(s) used to kill T. vaginalis have been previously examined. Here we show that the killing of T. vaginalis by human PMNs involves cooperative swarming of PMNs around the parasite, followed by attachment and trogocytosis, which precedes parasite death. This novel mechanism for killing a pathogen is mediated by antibody–Fc receptor engagement and PMN serine protease degradation of T. vaginalis into “bites”.
Trogocytosis is a process by which cells take “bites” of neighboring cells. It has been described to be used by immune cells [37,38,39,40,41,42,43], embryonic cells during development in Caenorhabditis elegans [58,59], and by amoebae [46,47,48]. Trogocytosis by Entamoeba histolytica has been shown to result in death of mammalian target cells [47,60,61,62], and trogocytosis between immune cells has been shown to be exploited as a conduit for transfer of the intracellular bacterium Francisella tularensis from cell to cell while evading the immune system [44,45]. More recently, PMNs have been shown to trogocytose tumor cells coated with monoclonal antibody (mAb) [35], and macrophages have been shown to kill tumor cells using trogocytosis [49]. The data presented here represent, to our knowledge, the first demonstration of mammalian phagocyte trogocytosis being used to kill a pathogen.
Humans are the only hosts infected by T. vaginalis. To date, attempts to establish robust mouse models with reproducible ability to infect and that sustain adequate T. vaginalis titers to allow reliable analysis of immune function have failed. The most promising model required a combination of estrogen and dexamethasone treatment to allow infection [19], both of which are functionally suppressive to PMNs, and generally immunosuppressive [63,64,65], placing severe limitations on the use of mouse models to study the immune response to T. vaginalis. In lieu of an animal model, we have used PMNs derived from human donors to characterize the interaction and killing of T. vaginalis by PMNs. We found that PMN killing of T. vaginalis is most efficient at higher PMN:T. vaginalis ratios (Fig 1C), consistent with our observations using live imaging that PMNs swarm around the parasite (Fig 4A), and that more than one PMN was needed to achieve killing, with rare exceptions (Fig 4B). In vaginal discharge from trichomoniasis patients, the PMN:T. vaginalis ratio was measured at greater than 100:1 [20], indicating that PMNs likely outnumber T. vaginalis in vivo, increasing confidence that these in vitro findings are physiologically relevant. Furthermore, prior studies with lower-resolution imaging techniques and light microscopy of vaginal smears from infected patients have observed “membrane fusion” between T. vaginalis and PMNs from infected patients [66]. Likewise, ex vivo studies of parasites cultured with human PMNs have described “killing by fragmentation” [20]. Both of these observations are consistent with killing using trogocytosis.
We found no evidence of PMN killing of T. vaginalis by NETosis by comparing parasite death in the absence and presence of DNase (Fig 2C). These data were corroborated by live imaging data that reveal that killing of T. vaginalis by PMNs was typically completed within approximately 15 minutes, whereas NETosis is generally a later-stage PMN killing mechanism, optimally assessed after 4 hours [67]. Furthermore, although our live imaging experiments contained an extracellular nucleic acid stain, no NETs were detected during imaging.
PMN killing of T. vaginalis was found to be contact dependent (Fig 2B), indicating that extracellular degranulation of toxic granules is not sufficient for T. vaginalis killing. It is possible that T. vaginalis can inactivate granular components or that the parasite is robust enough to withstand assaults from the granules because T. vaginalis secretes myriad proteases thought to play a role in virulence [3,68,69,70,71,72,73]. However, soluble granule components may not be toxic to T. vaginalis at dilute concentrations but may need to be secreted directly onto the parasite in a synapse, similar to antibody-dependent cell-mediated cytotoxicity (ADCC) or other receptor-mediated CDD [29]. In this regard, it is notable that PMN serine proteases were found to be critical for both trogocytosis and killing of T. vaginalis (Fig 5). Although PMN serine proteases are found at highest concentrations in the granules, they also localize to the plasma membrane upon PMN activation and can act in sealed compartments [55]. Therefore, a possible mechanism for PMN-mediated trogocytosis and death of T. vaginalis may involve recruitment of granules to the parasite surface, allowing proteases within the granules to nibble, taking “bites” of the parasite. Alternatively, serine proteases may be relocalized to the plasma membrane, where they are able to nibble within the tight microenvironment of the PMN–T. vaginalis junction. Such a killing mechanism would employ CDD in strict conjunction with trogocytosis. Future studies should elucidate whether this is a general mechanism used for PMN killing of larger pathogens and targets that otherwise escape phagocytosis and are resistant to contact-independent extracellular degranulation.
Our observation that bites of T. vaginalis material usually occur and accumulate in PMNs at least several minutes before parasite death (average of 8.3 minutes) indicates that trogocytosis proceeds PMN killing of T. vaginalis (Fig 4D). The accumulation of “bites” inside PMNs before parasite death argues against the interpretation that parasites are being killed by serine proteases and then PMNs are merely “cleaning up” the fragments through phagocytosis. The number of bites observed prior to T. vaginalis death was variable, averaging 3 to 8, but as many as 25 bites were observed (Fig 4C). Induction of a wound-repair mechanism during early stages of trogocytosis may allow parasite survival but may become overwhelmed eventually. The observation that death of some parasites occurs after only a few bites whereas others can survive up to 25 bites indicates PMN killing of T. vaginalis is multifactorial—consistent with a role for CDD, dependent on the number of PMNs in the swarm surrounding the parasite and other variables. Number of bites required for death may also be dictated by whether bites are initially taken from the flagellum or the plasma membrane in closer proximity to the cytosol.
Several parallels exist between the lethal trogocytosis mechanism of PMN killing of T. vaginalis described here and amoebic trogocytosis described for the killing of host cells by the parasite E. histolytica [47]. Both mechanisms result in punctate “bites” of the target cell appearing in the effector cell, which precede the death of the target cell. Both also require live bait for trogocytosis because feeding dead cells to effectors results in whole-cell engulfment of the target (Fig 3F & S6 Fig) and both are receptor mediated (Fig 6) [47,48]. Interestingly, the surface molecules that mediate contact during trogocytosis also mediate contact between effector and target cells for other purposes. In the case of T. vaginalis, human serum factors are also necessary for phagocytosis of dead parasites (S8 Fig). We therefore conclude that receptor-mediated surface contact is necessary, but not sufficient, for trogocytosis. The similarity in killing mechanisms used by these very divergent eukaryotic cell types is remarkable and suggests that the ability of cells to trogocytose neighboring cells evolved early in eukaryotic evolution and may be more widespread and common than currently acknowledged.
Anti–T. vaginalis antibodies are present in sera of infected patients [74,75], and cysteine proteases secreted by T. vaginalis have been shown to degrade human immunoglobulin [71]. We have also shown T. vaginalis to be preferentially cytotoxic to B cells using contact-dependent and contact-independent mechanisms [27]. T. vaginalis may therefore have evolved strategies to evade humoral immunity in order to avoid rapid clearance by PMNs. This is consistent with partner reinfection of trichomoniasis being common [76], indicating ineffective adaptive immunity. It will be important to determine whether trogocytosis and PMN killing of T. vaginalis strictly requires the presence of cognate antibody or whether innate factors in serum such as complement can also mediate killing and trogocytosis. iC3b was observed to bind to the surface of T. vaginalis, and Fc-blocking treatment only partially inhibited PMN cytolysis of T. vaginalis (Fig 6), indicating that antibody-independent mechanisms may also mediate trogocytosis and killing.
PMNs are also important in mediating therapeutic effects of directed mAb therapy against cancers. PMN surface Fc receptors can bind opsonized tumor cells, which can then be eliminated by ADCC or phagocytosis [77]. However, recent reports have also attributed failings of some mAb treatments to PMN trogocytosis [78]. Using trogocytosis, phagocytes containing Fc receptors have been shown to clip off exogenous mAbs from the surface of tumor cells, effectively enabling therapeutic escape and cancer cell survival [36,78]. The reason or rationale for this is unclear. A host defense role for PMN trogocytosis in pathogen clearance, demonstrated here, could help explain Fc receptor–mediated trogocytosis by phagocytes.
These data indicate that human PMNs swarm, trogocytose, and kill T. vaginalis, a process that is enhanced by adaptive immunity (antibody) and is dependent on PMN serine proteases. Further characterization of the specific molecular determinants of this process should better inform vaccine strategies and immunotherapies to mitigate adverse inflammatory complications resulting from T. vaginalis infection. Future studies are also likely to illuminate whether this novel mechanism of lethal PMN trogocytosis is broadly used to combat large, motile pathogens.
All human blood material was obtained from the University of California, Los Angeles (UCLA) Center for AIDS Research (CFAR) Virology Core Facility. Use of this material was approved by UCLA Institutional Review Board (IRB) committee Medical IRM 2, protocol 11–000443. All participants were 18 years of age or above and gave written consent.
Blood was obtained periodically at random from among approximately 50 de-identified healthy donors from the UCLA Virology Core. Peripheral Blood Mononuclear cells were removed after Ficoll gradient, and the remaining pellets were subject to a 3% Dextran (Pharmacosmos, Denmark) in 0.9% NaCl gradient for 20 minutes. The top layer was isolated and spun at 250 g for 10 minutes. The pellet was then treated with 20 mL ACK lysing buffer (Invitrogen; Thermo Fisher Scientific, Carlsbad, CA) to remove residual erythrocytes, and resulting PMNs were washed in PBS and stored on ice. PMNs were confirmed to be more than 98% viable and pure as assessed by flow cytometry scatter and dead cell exclusion using Zombie Red viability dye (Biolegend, San Diego, CA), CD11b antibody staining, and microscopy for PMN nuclear morphology using DAPI. Experiments with PMNs were always commenced immediately or within 1 hour of isolation.
T. vaginalis strain G3 (Beckenham, UK 1973, ATCC-PRA-98) was grown as described [27]. Briefly, parasites were maintained in supplemented TYM medium [79] at 37 °C and maintained at approximately 1 x 105 to 2 x 106 cells/mL. To ensure that all strains used in this study were free of the common symbiont Mycoplasma hominis [80], media was supplemented with 50 μg/ml chloramphenicol and 5 μg/ml tetracycline (Sigma-Aldrich, St. Louis, MO), supplemented for at least 5 days, as described previously [27]. Dead, intact controls were generated as described [27]. Briefly, T. vaginalis were reconstituted in complete RPMI media and heated at 65 °C for 1 hour. Parasites were confirmed by live microscopy to be immobile but intact as described [27], and loss of viability was confirmed with PI (Biolegend).
Coculture experiments were performed as described [27]. Briefly, PMNs and T. vaginalis were combined in RPMI 1640 media and incubated at 37 °C with 5% CO2. Ten percent human AB serum from the same batch (Lot 35060105, Corning, Corning NY) was thawed on ice and supplemented into media at the time of coculture for all experiments, except the human serum-deficient condition in Fig 6.
Flow cytometry–based cytotoxicity assays were utilized, as described previously [27]. Briefly, T. vaginalis were labeled with either CT Blue or Red (Molecular Probes) and recovered in complete TYM media for 45 minutes to 2 hours. T. vaginalis was then resuspended in RPMI, and 1.5 x 104 parasites were added to each well of a 96-well u-bottom plate. CFSE-labeled PMNs were added to the plate at the indicated MOI and analyzed by flow cytometry. Counting beads (Life Technologies, Carlsbad, CA) were added to each sample before analysis. Counts of parasites and CFSE+CT+ double positives were determined as in Fig 1B. Percent T. vaginalis killing was calculated as ([number of T. vaginalis in T. vaginalis–alone condition − number of T. vaginalis in coculture condition] / number of T. vaginalis in T. vaginalis–alone condition) x 100. Zombie Red dead-cell exclusion dye (Biolegend) was added to preliminary experiments to ensure that live cell gates did not include dead cells, and then found to be redundant because dead cells completely disappear from the live cell scatter plots (S2 Fig) and was no longer included, for simplicity. Trans-well cytotoxicity experiments were also conducted as described [27]. Briefly, a trans-well plate with 0.4-μm polycarbonate membrane was utilized, and PMNs were placed in the bottom chamber. After the cocultures, both top and bottom chambers were harvested. Where noted, inhibitors/activators were added to T. vaginalis cytotoxicity assays either at the time of coculture (DNase-1 [Worthington], Catalase [Sigma], PMA [Cayman Chemical]), to PMNs 10 minutes before addition of T. vaginalis (Human IgG F(c) fragment [Rockland Chemical]), or to PMNs 20 minutes before the addition of T. vaginalis (Cytochalasin D [Sigma], wortmannin [Tocris], AEBSF [Tocris]).
For quality control experiments, T. vaginalis viability was determined by labelling with CT and counting intact cells on flow cytometry as described above. Because PMNs adhere to plates after activation, the use of counting by flow cytometry was ineffective for determining viability, and instead viability was assessed using cytosolic mammalian-specific lactate dehydrogenase (LDH) detection in supernatants using the CytoTox-One homogeneous membrane integrity assay (Promega, Madison, WI) as described [27]. Briefly, 5 x 105 PMNs were plated in 100 ul in a 96-well u-bottom plate, and after treatment, supernatants were assayed for LDH. Percent death was calculated as described [27]. Inhibitors were added at the same concentration and timeframe as in T. vaginalis cytotoxicity assays.
The Amplex Red extracellular H2O2 detection kit (Invitrogen) was used according to the manufacturer’s instructions with 2.5 x 105 PMN/well in a 96-well plate, with final volume 100 ul. After Amplex Red and Horseradish peroxidase were added, the plate was incubated for 2 hours at 37 °C, 5% CO2 and then read on a Victor3 1420 plate reader (Perkin-Elmer, Waltham, MA). To determine Catalase inhibition, Catalase was added at the beginning of the assay.
NET release from PMNs was detected using the picogreen extracellular DNA quantitation kit (Life Technologies) according to the manufacturer’s instructions. A total of 2.5 x 105 PMNs were plated in 96-well flat-bottom plates and stimulated with 100-nM PMA to serve as a positive control for NET induction. After 2 hours, 50 ul of supernatant was harvested and read on a Victor3 1420 plate reader (Perkin-Elmer), after 2-minute incubation with picogreen.
Samples were run on either an Amnis FlowSight or ImageStream Mark II. A total of 5,000 events were collected for each sample. Single color controls were used for auto compensation by IDEAS software. Analysis was done only on cells in focus, and doublets and debris were eliminated based on area versus aspect ratio. Spot Count, Internalization Erode, Intensity, and circularity algorithms were done in using standard wizards in IDEAS software. Internalization positive scores were generated using the Internalization Erode algorithm with CT as internalizing probe and CFSE as cell boundary. For phagocytosis analyses, percent of T. vaginalis phagocytosed was calculated as follows: ([number of CT+CFSE+ cells] / [Total number of CT+ cells]) x (percentage of CT+CFSE+ cells that are internalization positive / 100).
Parasites were washed 2 times in 5% sucrose PBS and incubated for 45 minutes with cell-impermeant Sulfo-NHS-SS-Biotin (Thermo Fisher Scientific) in 5% PBS sucrose to label amine groups on surface proteins. The reaction was quenched for 5 minutes on ice in 50 uM Tris pH 7.5, and then parasites were washed 3 times with 5% PBS sucrose and incubated on ice for 5 min with 5 ug/ml Streptavidin-Alexa 488 (Biolegend). Parasites were then washed 2x with 5% sucrose PBS, counted, resuspended in RPMI, and stored on ice. For experiments in Fig 4E and 4F, PMNs were labeled with CT Deep Red and stored on ice.
Two-dimensional live imaging videos were taken using a Zeiss confocal microscope outfitted with Yokogawa spinning disc confocal scanner unit and Oko lab 37-degree chamber, as assembled by Intelligent Imaging Innovations (3i, London, UK). SlideBook6 software was used for acquisition and analysis. A total of 1.3 x 106 PMNs were plated in RPMI with 10% human serum in a 35-mm MatTek dish and allowed to settle down to the glass bottom for 15 minutes. Then, 1.3 x 105 T. vaginalis were added slowly in a dropwise fashion around the plate, and video was commenced once parasites reached the PMN layer. Sequential exposures with BF, 488, 561, and 640 nm were taken approximately every 0.5 seconds, and x, y, and z parameters were adjusted by hand during the video to follow individual parasites and trogocytic events. Videos shown are at 0.1 frame/second. For composite data (Fig 4B and 4C), some videos were taken at 40x to visualize more parasites/field, and some were taken at 63x.
Three-dimensional live imaging videos were taken using the same plating strategy as for 2D imaging, but HyVolution super resolution acquisition was done on a Leica Sp8 Scanning Confocal microscope in a 37-degree chamber. Z-sections were taken every 0.3 um over an approximately 20-um section, set by hand depending on parasite and PMN position. Using the white laser and HyD sensors, 488 and 632 wavelengths were excited and collected simultaneously, and 561 was done sequentially, approximately every 2 seconds/time point. Data were then deconvolved using Huygen’s Professional (Scientific Volume Imaging, Hilversum, Netherlands), and processing and 3D reconstruction were done in LASX software (Leica, Wetzlar, Germany). Video shown is 0.25 frames/second.
Human serum Lot 35060105 (Corning, Mediatech) was thawed on ice, and T. vaginalis were immediately incubated with serum for 30 minutes at 37 degrees. T. vaginalis were then washed 2x with FACS buffer (1x PBS, 5% FBS, 0.1 NaN3), and the following staining steps were done for 30-minute incubations on ice with FACS buffer washes in between. For detection of human antibody bound to T. vaginalis, parasites were incubated on ice for 30 minutes with either PE antihuman immunoglobulin light chain kappa or isotype control (Biolegend). For detection of iC3b on T. vaginalis, parasites were incubated with purified mouse antihuman iC3b or isotype control, followed by PeCy7 antimouse IgG1 (Biolegend). Parasites were then stored on ice and analyzed on LSR Fortessa within 2 hours.
All statistical analyses were done using a two-tailed, unpaired t test in Microsoft Excel (Redmond, WA).
|
10.1371/journal.pgen.1002722 | TBC-8, a Putative RAB-2 GAP, Regulates Dense Core Vesicle Maturation in Caenorhabditis elegans | Dense core vesicles (DCVs) are thought to be generated at the late Golgi apparatus as immature DCVs, which subsequently undergo a maturation process through clathrin-mediated membrane remodeling events. This maturation process is required for efficient processing of neuropeptides within DCVs and for removal of factors that would otherwise interfere with DCV release. Previously, we have shown that the GTPase, RAB-2, and its effector, RIC-19, are involved in DCV maturation in Caenorhabditis elegans motoneurons. In rab-2 mutants, specific cargo is lost from maturing DCVs and missorted into the endosomal/lysosomal degradation route. Cargo loss could be prevented by blocking endosomal delivery. This suggests that RAB-2 is involved in retention of DCV components during the sorting process at the Golgi-endosomal interface. To understand how RAB-2 activity is regulated at the Golgi, we screened for RAB-2–specific GTPase activating proteins (GAPs). We identified a potential RAB-2 GAP, TBC-8, which is exclusively expressed in neurons and which, when depleted, shows similar DCV maturation defects as rab-2 mutants. We could demonstrate that RAB-2 binds to its putative GAP, TBC-8. Interestingly, TBC-8 also binds to the RAB-2 effector, RIC-19. This interaction appears to be conserved as TBC-8 also interacted with the human ortholog of RIC-19, ICA69. Therefore, we propose that a dynamic ON/OFF cycling of RAB-2 at the Golgi induced by the GAP/effector complex is required for proper DCV maturation.
| Synaptic transmission is mainly mediated by the triggered release of neurotransmitters from synaptic vesicles (SVs). To regulate synaptic transmission and neuronal activity, neurons also release neuropeptides and hormones from dense core vesicles (DCVs). While SVs can be recycled locally at the synapse, DCVs have to be newly synthesized in the cell body after release. The formation of new DCVs requires a multi-step maturation process. During this maturation, the neuropeptides are processed into their active form and factors that would disturb DCV release are removed. Only properly matured DCVs are able to undergo efficient release after stimulation. Since DCV biogenesis mainly uses the normal secretory pathway, an elaborate machinery must exist that guarantees efficient sorting and retention of DCV cargo. Previously, we identified the small GTPase RAB-2 and its effector RIC-19/ICA69 to be involved in the retention of soluble cargo in DCVs. In a screen for molecules that regulate RAB-2 activity during DCV maturation, we identified the evolutionarily conserved TBC domain-containing protein, TBC-8. We demonstrate that TBC-8 is a putative RAB-2 GAP, which also binds to RIC-19/ICA69. Thus, RAB-2 might recruit its own GAP via its effector RIC-19, which suggests that a highly dynamic cycling of RAB-2 is required for DCV maturation.
| Most neurons secrete both neurotransmitter filled synaptic vesicles (SVs) as well as dense core vesicles (DCVs) that contain neuropeptides, hormones and trophic factors [1], [2], [3], [4], [5]. The stimulated release of neurotransmitters from SVs mediates fast synaptic transmission, while the release of neuropeptides from DCVs modulates neurotransmission and neuronal activity [2], [3]. Classical neurotransmitters are thought to act locally in the millisecond timescale at their site of release. In contrast, neuropeptides and hormones secreted by DCVs act more slowly and over longer distances [3]. Contrary to SVs that can be recycled locally at the site of release after exocytosis, DCVs have to be synthesized de novo in the cell body after release [6]. Despite their importance for the modulation of neurotransmission, neuronal DCV biogenesis is not well understood.
DCVs are generated at the trans-Golgi network (TGN) where neuropeptide precursors and prohormones along with their processing enzymes are sorted and packaged into transport carriers that bud off the Golgi. These immature DCVs (iDCVs) subsequently undergo a maturation process by which processing enzymes, SNAREs (soluble N-ethylmaleimide-sensitive factor-attachment protein receptor) that are required for homotypic fusion of iDCVs and lysosomal proteins that have been accidentally packaged into DCV transport carriers are removed by vesicular transport to endosomes [6], [7]. This remodelling of iDCVs is achieved by clathrin mediated sorting at the Golgi-endosomal interface [6], [7], [8], [9]. During their maturation process, iDCVs are continuously acidified by the action of vacuolar ATPases (v-ATPases). This acidification supports the sorting and retention of cargo in iDCVs [6]. It also activates prohormone convertases that proteolytically process prohormones and proneuropeptides into their bio-active forms [10]. The fully processed neuropeptides and hormones subsequently aggregate to a crystalline matrix forming the dense core of DCVs [6], [7], [8], [9]. Mature DCVs (mDCVs) are then transported along microtubules from the cell body to their distal release sites where they are secreted in a regulated manner [11]. Only mDCVs and not iDCVs have been shown to undergo efficient stimulus dependent exocytosis [12], [13].
Two main models have been proposed by which sorting and retention of cargo in forming DCVs could occur: sorting by entry and sorting by retention. The sorting by entry hypothesis predicts the existence of sorting signals and receptors that would actively sort cargo into nascent DCVs [6], [7], [8], [9]. Considering the number of different post-Golgi transport carriers that are generated at the TGN, it is likely that such active sorting mechanisms exist. Furthermore, short N-terminal sequence motifs have been identified in DCV cargos such as provasopressin, pro-oxytocin, pro-opiomelanocortin, and chromogranin A (CgA) and chromogranin B (CgB), which are sufficient for DCV targeting [14], [15], [16], [17]. In contrast, the sorting by retention hypothesis suggests that DCV cargo could passively enter forming DCVs and then be retained during DCV maturation either by active retention in lipid domains or by its aggregation within iDCVs. This model is based on the observations that several DCV factors such as CgA and CgB as well as secretogranin II (SgII) aggregate at pH below 6.5 and high Ca2+ concentrations, which occur during iDCV formation at the TGN [17], [18]. Furthermore, it has been shown in the case of secretogranin III (SgIII) that there are direct interactions between aggregated DCV cargos and cholesterol-rich membrane domains of DCVs [18]. These interactions between the different aggregating molecules are likely to generate some sort of higher order retention matrix within iDCVs. In addition, proneuropeptides and prohormones, which enter nascent DCVs in a soluble form, are rendered insoluble once processed by their prohormone convertases. Thus, it is conceivable that after an initial active sorting step to enter nascent DCVs, several DCV core components and processed neuropeptides and hormones are subsequently retained in iDCVs by aggregation during the DCV maturation process.
During sorting of cargo into DCVs, it has been shown that a parallel mechanism exists to remove processing enzymes, lysosomal proteins and mannose-6-phosphate receptors from iDCVs by clathrin dependent sorting processes [19]. In a similar manner, the SNARE proteins syntaxin 6 and VAMP4 as well as synaptotagmin IV are also sorted away from iDCVs and are absent on mDCVs [6], [19], [20]. The fact that during DCV maturation proteins are actively removed suggests that mechanisms must exist to ensure retention of factors that are supposed to stay in mDCVs but are not part of the aggregating core. Therefore, the sorting processes on iDCVs have to be tightly regulated to prevent loss of DCV factors to the endosomal system.
In Caenorhabditis elegans, it has recently been shown that RAB-2 is required to retain soluble cargo in maturing DCVs [21], [22]. RAB-2 belongs to the Ras superfamily of small GTPases that act as molecular switches and cycle between an active GTP-bound and an inactive GDP-bound state. Rab GTPases are involved in many steps of vesicular transport including vesicle formation, transport, tethering, docking and fusion of vesicles by binding to effector molecules [23], [24], [25], [26]. In order to regulate the active state of Rab GTPases, two additional enzyme families are required: i) GTPase activating proteins (GAPs) that inactivate Rab proteins by enhancing their low intrinsic GTPase activity, and ii) guanine nucleotide exchange factors (GEFs) that activate Rab GTPases by facilitating the exchange of GDP with GTP [27]. Most of the known Rab GAPs possess a conserved Tre2/Bub2/Cdc16 (TBC) domain (except for Rab3A-GAP [28]) consisting of approximately 200 amino acids. This TBC-domain is sufficient for GAP activity in vitro by employing an arginine finger-based catalytic mechanism [29]. Interestingly, some TBC proteins are modular and contain additional domains that are important mainly for the regulation of protein-protein interactions or for targeting these proteins to membranes, suggesting additional role(s) and/or regulation of these proteins [30].
In order to understand the spatio-temporal regulation of RAB-2 during DCV maturation, we aimed to identify a GAP for RAB-2 in C. elegans. By using a DCV trafficking assay for neuronal TBC domain-containing proteins in C. elegans, we were able to identify the evolutionarily conserved protein, TBC-8, as a putative, neuron specific RAB-2 GAP (GenBank: CAA84706.3). As in rab-2 mutants, loss of TBC-8 showed a similar DCV trafficking phenotype, whereas the transport of SVs is not affected. Furthermore, we could demonstrate that TBC-8 binds to the active form of RAB-2. Lastly, overexpression of TBC-8 in neurons redistributed RAB-2 from Golgi membranes to the cytosol. Taken together our in vivo data suggest that TBC-8 is a putative RAB-2 GAP. Unfortunately, we were unable to obtain soluble full-length or fragments of TBC-8 to show GAP activity towards RAB-2 or other RABs in vitro. Interestingly, we could show that TBC-8 also binds to the RAB-2 effector, RIC-19/ICA69, indicating that RIC-19 might recruit the GAP to RAB-2 positive membranes to inactivate this small GTPase. These results suggest that a dynamic cycling of RAB-2 at the Golgi is necessary for proper DCV maturation.
To identify the Rab GAP that regulates RAB-2 during neuronal DCV maturation, we systematically tested mutants of TBC domain-containing GAPs in C. elegans. These strains were analyzed for trafficking defects of the DCV marker NLP-21-VENUS (Figure S1). NLP-21-VENUS is a fusion protein of the proneuropeptide NLP-21 with the yellow fluorescent protein VENUS. In this assay, we used an integrated strain stably expressing fluorescently-labeled neuropeptide, NLP-21-VENUS, specifically in DA and DB cholinergic motoneurons in C. elegans. It was shown that NLP-21-VENUS is packaged into DCVs, transported to axons of the dorsal nerve cord (DNC), and released into the body cavity [31]. Once secreted, NLP-21-derived VENUS is subsequently endocytosed by six macrophage-like scavenger cells (called coelomocytes), which constantly filter the body fluid by bulk endocytosis [32] (Figure 1A).
In this screen, only tbc-8(tm3802) deletion mutants displayed decreased NLP-21-derived VENUS levels in the DNC axons, similar to unc-108/rab-2 mutants [21], [22] (Figure S1). Inactivation of TBC-8 led to a 74.95±3.51% decrease of NLP-21-derived VENUS fluorescence in axons compared to wild type (Figure 1B). We observed comparable results when tbc-8 expression was downregulated by RNAi (Figure S2). Similar to unc-108/rab-2 mutants, the accumulation of secreted VENUS in coelomocytes was also decreased by 68.17±6.94% (Figure 1C) suggesting that less VENUS was secreted. To determine whether the observed DCV phenotype was caused by TBC-8 malfunction in the nervous system, we expressed tbc-8 cDNA pan-neuronally under the control of the rab-3 promoter. Pan-neuronal expression of tbc-8 as well as expression of tbc-8 under the DA and DB cholinergic motoneuron specific unc-129 promoter (2641 bp promoter fragment upstream of the start ATG of unc-129) was able to rescue the NLP-21-derived VENUS fluorescence levels in DNC and in coelomocytes (Figure 1B, 1C). This showed that TBC-8 is cell-autonomously required in cholinergic motoneurons for proper DCV function. The catalytic GAP activity of TBC-8 was also shown to be indispensible for DCV trafficking as a catalytically inactive TBC-8 (R697A) was unable to rescue the loss of NLP-21-derived VENUS fluorescence (Figure 1B, 1C). This result demonstrates that TBC-8 acts as a RAB GAP required for effective DCV function.
Previously, it was shown that in unc-108/rab-2 mutants less NLP-21-VENUS-derived cargo is present in mature DCVs, because the cargo was partially lost in the endosomal-lysosomal degradation route during DCV maturation [21], [22]. These trafficking defects resulted in aberrant NLP-21-derived VENUS accumulation in the cell bodies of unc-108(n501) mutants. In order to test whether tbc-8(tm3802) mutants displayed the same trafficking defects, we analyzed the size distribution of VENUS positive vesicular structures in the cell bodies of these mutants. Similar to unc-108(n501) mutants, tbc-8(tm3802) mutant worms contained more of these large VENUS positive vesicular structures in the neuronal cell bodies, whereas smaller vesicular structures filled with NLP-21-derived VENUS are less present compared to wild type animals (Figure 1D, 1E). Furthermore, the overall fluorescence of NLP-21-derived VENUS in the cell bodies of motoneurons is decreased in tbc-8(tm3802) mutants (58.23±11.71%) like in unc-108(n501) worms (45.16±18.98%) (Figure 1F). These results indicate that tbc-8(tm3802) mutants have comparable DCV trafficking defects similar to unc-108/rab-2 mutants.
To determine whether neuropeptides other than NLP-21 are also affected in tbc-8(tm3802), we tested the fluorescence levels of FMRFamide related peptide FLP-3-derived VENUS and the insulin-like neuropeptide INS-22-VENUS in the DNC (Figure 1G, 1H). FLP-3-derived VENUS levels in the DNC of tbc-8 mutants were decreased by 87.90±1.19%, suggesting that VENUS derived from other neuropeptides was also lost. However, INS-22-derived VENUS fluorescence levels were unchanged in tbc-8(tm3802), like in unc-108/rab-2 mutants [21], [22]. In addition to different neuropeptides, we also tested a transmembrane protein of DCVs, IDA-1-GFP, in the DNC. IDA-1-GFP levels were unchanged in the tbc-8 background when compared to wild type animals, suggesting no changes in the numbers of DCVs in tbc-8 mutants (Figure 1I). This observation was supported by a high-pressure freeze electron microscopy (HPF-EM) analysis of the DCV numbers and distribution at axonal release sites in tbc-8 mutants, which were unaltered compared to wild type (Table 1). Therefore, in tbc-8 mutants, DCVs are loaded with less soluble NLP-21-derived VENUS, as it has also been shown for unc-108/rab-2 mutants [21], [22]. Thus, DCV maturation is disrupted in tbc-8 mutants in a manner similar to unc-108/rab-2 mutants.
In order to see whether TBC-8 acts in the same genetic pathway as RAB-2, we compared tbc-8(tm3802) with different unc-108/rab-2 mutants. We used two different alleles of RAB-2, (n501) D122N, which is dominant active and constitutively bound to GTP [22], and the deletion (nu415) that serves as a molecular null allele since the protein product could not be detected on Western blots [33]. In unc-108/rab-2(n501) NLP-21-derived VENUS fluorescence levels in the DNC are more strongly decreased as compared to (nu415) by 91.13±1.38% and 43.89±5.59%, respectively (Figure 2A). If TBC-8 is the GAP for RAB-2, then in tbc-8 mutants, RAB-2 should predominantly be in the constitutively GTP-bound form and therefore, tbc-8 mutants should resemble the dominant unc-108/rab-2(n501) mutant. Consistent with this idea, the tbc-8(tm3802) deletion allele indeed showed a decrease in NLP-21-derived VENUS fluorescence (74.00±2.51%), similar to unc-108(n501) mutants (91.13±1.38%) (Figure 2A). Comparatively, the decrease in NLP-21-derived VENUS fluorescence was not as pronounced in the unc-108(nu415) null allele (43.89±5.59%) (Figure 2A). This result suggests that RAB-2 is indeed in the active GTP-bound form in tbc-8 mutants. Therefore, inactivation of RAB-2 in a tbc-8 mutant background should lead to a weaker NLP-21-VENUS phenotype, identical to the unc-108/rab-2(nu415) null allele. In agreement with this hypothesis, unc-108(nu415); tbc-8(tm3802), a combination of both deletion alleles, showed an identical phenotype (46.88±5.20%) as the single unc-108(nu415) deletion allele (43.89±5.59%) (Figure 2A). Furthermore, unc-108(n501); tbc-8(tm3802), a double mutant with unc-108/rab-2 gain of function allele, (91.84±1.37%) also resembled the single unc-108(n501) allele (91.13±1.38%) (Figure 2A). Both double mutants followed the phenotype of the respective unc-108 single mutant indicating that both proteins are involved in the same genetic pathway.
Previously, we have shown that the expression of a constitutively GTP-bound form of RAB-5 (Q78L) could rescue the DCV phenotype in the unc-108(n501) mutant by inhibiting loss of DCV cargos to early endosomes and returned NLP-21-derived VENUS fluorescence levels back to wild type levels [22]. Since TBC-8 and RAB-2 are in the same pathway, RAB-5 (Q78L) should also rescue the NLP-21-VENUS phenotype in the tbc-8(tm3802) deletion allele. Consistent with this hypothesis, expression of RAB-5 (Q78L) could rescue the NLP-21-derived VENUS fluorescence levels at the synapses up to 86.47±10.90%, as seen in Figure 2B. This result indicates that TBC-8, like RAB-2, is required for retaining specific cargo in maturing DCVs by preventing its entry to the late endosomal system.
In order to test whether TBC-8 is involved in neuropeptide processing, we generated a double mutant of the prohormone convertase PC2, egl-3(gk238), and tbc-8(tm3802), and tested it for defects in DCV maturation (Figure 2C). The data was compared with the respective single mutants to identify possible genetic interactions (Figure 2C). Previously, it has been shown that the loss of NLP-21-derived VENUS from maturing DCVs in unc-108/rab-2 mutants can be rescued by inactivation of EGL-3 [22]. Thus, the lack of proneuropeptide processing retains NLP-21-VENUS into the insoluble dense core of DCVs of egl-3(gk238); unc-108/rab-2 mutants. Similarly, in egl-3(gk238); tbc-8(tm3802) double mutants, NLP-21-derived VENUS levels were restored to wild type levels (72.50±9.90%) (Figure 2C). This finding indicates that primarily soluble NLP-21-derived VENUS is lost in unc-108/rab-2 and tbc-8(tm3802) mutants. Although unc-108/rab-2; egl-3(gk238) double mutants had restored levels of NLP-21-derived VENUS in the DNC, analysis of the movement phenotype in the double mutants revealed that they were more severely uncoordinated than either single mutant [22]. This result suggests that, despite rescuing the NLP-21-derived VENUS marker in the DNC, the combined loss of soluble cargo and functional neuropeptides in DCVs leads to additive locomotion defects in unc-108/rab-2; egl-3(gk238) double mutants. Interestingly, tbc-8(tm3802) mutants do not display a movement defect (Figure S3) and double mutants of tbc-8(tm3802); egl-3(gk238) do not yield animals that are similarly uncoordinated to unc-108/rab-2; egl-3(gk238) mutants (data not shown; [22]). These data suggest that RAB-2's role in regulating movement may predominantly occur in a subset of neurons that do not express tbc-8, but perhaps express another GAP.
All these findings highlight that TBC-8 is indeed in the same pathway as RAB-2 and thus, TBC-8 is a likely candidate to be a RAB-2 GAP. However, differences in movement defects between tbc-8 and unc-108/rab-2 mutants suggest the existence of additional GAPs, which may regulate RAB-2 activity in the absence of TBC-8.
For unc-108/rab-2 mutants, it has been shown that DCV maturation is specifically affected while SV trafficking and function are unaltered [21], [22]. To exclude that the lack of TBC-8 would lead to general membrane trafficking defects in neurons, we analyzed SV localization, distribution and morphology. We first examined synaptic morphology by analyzing the localization of the synaptic marker proteins, RAB-3 and Synaptobrevin-1 (SNB-1), using integrated strains expressing YFP-RAB-3 and GFP-SNB-1 in cholinergic motoneurons. The fluorescence of YFP-RAB-3 (90.12±11.96%) (Figure 3A) and GFP-SNB-1 (90.82±9.32%) (Figure 3B) was unchanged in the DNC of tbc-8(tm3802) when compared to wild type. Second, we used HPF-EM to show that there were no obvious changes in the morphology of synapses, neuronal cell bodies and neuronal Golgi complexes in tbc-8(tm3802) mutants compared to wild type worms (Figure 3C). Here, we did not detect any abnormal accumulation of vesicles or degeneration of the Golgi stacks in tbc-8(tm3802) mutants compared to wild type animals.
To exclude that there are more subtle changes in SV distribution at presynaptic active zones in tbc-8(tm3802) mutants, we analyzed SV distributions surrounding the presynaptic dense projection of cholinergic motoneurons by HPF-EM. There were no changes in SV distribution, SV numbers or SV diameter as compared to wild type (Figure 3D, Table 1). This analysis demonstrates that SV function is unaffected in tbc-8(tm3802) mutants, similar to what has been reported for unc-108/rab-2 mutants [21], [22]. For unc-108/rab-2 gain of function mutants, it has been shown that DCV diameters are more variable [21], [22] (Table 1). However, this was not the case for tbc-8(tm3802), where DCV diameters are not significantly different from wild type DCVs (Table 1).
tbc-8 encodes a 903 amino acid (aa) protein. The tbc-8(tm3802) deletion allele leads to a stop codon after the 8th exon, truncating the protein after aa 482 just before the TBC-domain (Figure 4A). The TBC-domain of TBC-8 is located at the C-terminus between aa 621 and 862. In addition to the TBC-domain, TBC-8 also contains a conserved RUN [after RPIP8 (RaP2 interacting protein 8)/UNC-14/NESCA(new molecule containing SH3 at the carboxyl-terminus)] domain (Figure 4B), which has been shown to bind to small GTPases of the Rab and Rap family [34], [35]. TBC-domain proteins with the same domain structure can be found in Drosophila melanogaster and the mammalian system [30] (Figure 4B). The orthologs in D. melanogaster (CG32506-PC, FlyBase ID: FBpp0300194) and in H. sapiens (SGSM1 [36], accession number: NP_001035037) are about 250 aa longer than C. elegans TBC-8. The RUN domains of these proteins show a 64% (SGSM1) and a 63% (CG32506-PC) similarity to the RUN domain of TBC-8, whereas both TBC-domains are about 58% (SGSM1) and 57% (CG32506-PC) similar to the TBC-domain of C. elegans TBC-8. For full protein sequences of TBC-8 and its orthologs see Figure S4.
In order to determine the expression pattern of tbc-8, a construct containing a 2873 bp genomic fragment of the tbc-8 promoter region was fused to gfp and injected into wild type worms (Figure 4C, upper panel). The expression pattern using a transcriptional reporter revealed that tbc-8 is exclusively expressed in neurons including the head and tail neurons as well as neurons in the ventral nerve cord (VNC) (Figure 4C, lower panel). Interestingly, analysis of the expression pattern of unc-108/rab-2 showed a similar high expression in neurons [21], [22].
Based on our expression analysis, we could show that tbc-8 is expressed in the nervous system (Figure 4C). To exclude the possibility that our transcriptional reporter missed important regulatory regions and elements, we tested TBC-8 function in other tissues in which RAB-2 activity was shown to be required [33], [37], [38]. We analyzed the possible role of TBC-8 in postendocytic trafficking in coelomocytes (Figure S5) and in the degradation of apoptotic cell corpses in the germ line (Figure S6).
We did not detect any defects in coelomocytes during the steady-state endocytosis of ssGFP from muscle cells (arIs37[pmyo-3::ssGFP]) [39] in tbc-8(tm3802) mutants (Figure S5A).
In order to test the kinetics of ssGFP uptake and the degradation of endocytosed GFP in coelomocytes, we used a strain that expresses ssGFP under a heat-shock promoter (arIs36[phsp::ssGFP] [39]) which allowed us to perform an in-vivo pulse-chase experiment in C. elegans. While following the fate of secreted GFP within coelomocytes after a short heat-shock, no defects were observed in tbc-8(tm3802) mutants compared to wild type worms for all measured time points (Figure S5B).
In addition, we examined postendocytic trafficking of the fluid-phase marker Texas red-BSA (TR-BSA) in coelomocytes of tbc-8(tm3802) mutants. TR-BSA was microinjected into the pseudocoelom of young adult worms. Using a marker for endosomes, RME-8-GFP, we analyzed the migration of injected TR-BSA through RME-8-GFP positive compartments into later postendocytic compartments (Figure S5C). We did not see obvious defects at any time point in tbc-8(tm3802) mutants compared to wild type animals.
Previously, it was shown that RAB-2 is also required for the degradation of apoptotic cell corpses in C. elegans [37], [38]. To identify whether TBC-8 plays a role together with RAB-2 in the germ line, we assayed the presence of apoptotic cell corpses in tbc-8(tm3802) mutants (Figure S6). Unlike unc-108/rab-2 deletion mutants, tbc-8(tm3802) mutants showed similar numbers of apoptotic cell corpses as wild type worms (Figure S6).
These data together with the expression analysis indicate that TBC-8 likely regulates RAB-2 function only in the nervous system of C. elegans.
In order to examine where TBC-8 functions, we expressed fluorescently labeled TBC-8 under the rab-3 pan-neuronal promoter and determined its sub-cellular localization in motoneurons. Fluorescently labeled TBC-8 showed a cytosolic localization with some punctate membrane staining (Figure 5A) reminiscent of RAB-2. In order to identify which subcellular compartments these cytoplasmic puncta correspond to, we co-localized fluorescently-labeled TBC-8 with different subcellular markers in neurons (Figure 5A). We observed partial localization of TBC-8 with medial (mannosidase II) and trans-Golgi (APT-9) markers and an almost complete co-localization with the early endosomal marker RAB-5. However, we could not observe any localization of TBC-8 to RAB-7-positive late endosomes. This suggests that TBC-8 acts at the Golgi-endosomal interface, similar to RAB-2, where DCV maturation is also believed to take place [7].
RAB-2 was shown to localize to the Golgi [22]. Attempts to co-localize TBC-8 and RAB-2 were difficult, because when YFP-TBC-8 fusion proteins were co-expressed with mCherry-RAB-2 in motoneurons, we observed that RAB-2 was mainly cytosolic (Figure 5B, right cell body). However, in neurons where YFP-TBC-8 was not expressed due to the mosaic nature of extra-chromosomal arrays in C. elegans, a distinct Golgi staining of mCherry-RAB-2 was visible in these cell bodies (Figure 5B, left cell body). This is a strong indication that TBC-8, when overexpressed, inactivates RAB-2, which in turn would be redistributed from the Golgi to the cytosol. This result reinforces the idea that TBC-8 is a RAB-2 GAP as evident from the genetic data.
It was previously shown that TBC domain-containing Rab GAPs possess an essential arginine finger located within the TBC-domain, which is crucial for its catalytic GAP activity [29] (Figure 6A). When mutated to alanine, the TBC-domain could still bind to its Rab partner but was unable to activate the intrinsic GAP activity of the Rab, which would lead to GTP hydrolysis [29]. TBC-8 also contains a conserved arginine finger residue (R697) (Figure 6A) within the catalytic motif that is conserved among its orthologs in humans (SGSM1) and D. melanogaster (CG32506-PC) (Figure 6B). If TBC-8 is the GAP specific for RAB-2, the interaction of both proteins should be detectable in the yeast two-hybrid system (Y2H) if the arginine finger of TBC-8 is mutated [29]. In order to identify the Rab/GAP pair, we screened TBC-8 and the catalytically inactive TBC-8 (R697A) with all constitutively active, GTP bound C. elegans Rab proteins in a Y2H experiment. Interaction of TBC-8 with Rab proteins was captured by growth on histidine selection plates. As expected, RAB-2 (Q65L) interacted with the catalytically inactive TBC-8 (R697A) but not with wild type TBC-8 (Figure 6C, 6D). This suggests that TBC-8 is a GAP for RAB-2 and the interaction between TBC-8 and RAB-2 is confined to the TBC-domain. In this screen, we also detected a specific binding of the GTP bound form of RAB-19 (Q69L) to TBC-8. However, in this case RAB-19 binds to both wild type TBC-8 and the catalytically inactive TBC-8 (R697A) (Figure 6C, 6D). This indicates that TBC-8 might be an effector and not a GAP of RAB-19. Furthermore, we could not detect NLP-21-derived VENUS trafficking defects in rab-19 deletion mutants (Figure S7) indicating that TBC-8 together with RAB-19 has another role besides DCV trafficking. The co-localization and interaction studies strongly suggest that TBC-8 is the GAP for RAB-2 in motoneurons. However, we were unable to test the GAP activity biochemically since full-length TBC-8 as well as TBC-domain fragments were insoluble when purified from bacteria or insect cells.
Previously, it has been shown that activated, GTP-bound RAB-2 recruits the BAR-domain (Bin/amphiphysin/Rvs) containing effector, RIC-19/ICA69, to Golgi membranes [22], [40]. We specifically showed that in the dominant active unc-108/rab-2(n501) allele, more RIC-19-YFP was membrane-bound [22]. A similar phenotype would be expected in tbc-8 mutants where the RAB-2 GAP is missing. Thus, in tbc-8 mutants RAB-2 should be mostly GTP-bound. As shown in Figure 7A, in tbc-8(tm3802) deletion mutants, RIC-19-YFP was indeed more punctate as compared to wild type. The extent of RIC-19-YFP membrane recruitment in tbc-8 mutants is comparable to the dominant active allele of unc-108(n501) [22], [40]
To test genetically whether TBC-8 and RIC-19 are in the same pathway, we constructed a tbc-8(tm3802); ric-19(ok833) double mutant, which displayed the same DCV defects as tbc-8 single mutants (Figure 7B). NLP-21-derived VENUS fluorescence was decreased in tbc-8 mutants by 71.61±3.53%, in tbc-8(tm3802); ric-19(ok833) by 78.25±1.90%, whereas ric-19(ok833) single mutants showed a weaker decrease in fluorescence by 57.44±4.55%. These results suggest that TBC-8 and RIC-19 act in the same pathway.
We reasoned that if inactivation of TBC-8 leads to a constitutive activation of RAB-2 as indicated by RIC-19 recruitment, then overexpression of TBC-8 GAP should cause RIC-19 to be completely cytosolic. Unexpectedly, when tagRFP-TBC-8 was co-expressed with RIC-19-YFP, both proteins co-localized into huge cytosolic puncta in the cell bodies of motoneurons (Figure 7C, upper panel). This result suggested that TBC-8 and RIC-19 might form a complex that localizes to intracellular membranes. The extent of co-localization and the size of the puncta indicate that TBC-8 and RIC-19 strongly drive the membrane localization of the other. Interestingly, this cooperative membrane localization of TBC-8 and RIC-19 forced by overexpression is independent of RAB-2, since it is also seen in unc-108/rab-2(nu415) null mutant backgrounds (Figure 7C, lower panel). This result strongly suggested that RIC-19 and TBC-8 might directly interact. We tested this hypothesis by Y2H analysis as well as by co-immunoprecipitation (co-IP) of RIC-19 and TBC-8 in HEK293 cells (Figure 7D–7E). Both experiments revealed that there is a direct interaction between the RAB-2 effector, RIC-19, and the putative RAB-2 GAP, TBC-8. By Y2H analysis and co-IP, we showed that a fragment of TBC-8 (1–597 aa) containing the RUN domain was sufficient to bind to RIC-19 as well as to the human RIC-19 ortholog, ICA69 (Figure 7D–7E). Thus, the interaction between TBC-8 and RIC-19 seems to be evolutionarily conserved. Any attempt to shorten the TBC-8 RUN domain construct resulted in a loss of the interaction (data not shown). Therefore, our data suggest that active GTP-bound RAB-2 might recruit its own GAP via its effector RIC-19. However, this hypothesis has to be confirmed in further studies. It is also possible that another RAB-2 effector could also recruit TBC-8 in the absence of RIC-19, explaining the less severe DCV phenotype of ric-19(ok833) mutants.
Previously, we have shown that RAB-2 is involved in neuronal DCV maturation in C. elegans and that a functional RAB-2 cycle is important to retain specific cargo in maturing DCVs [21], [22]. To gain insights into the spatio-temporal regulation of RAB-2 during DCV maturation, we screened all TBC domain-containing RAB GAPs in C. elegans for their involvement in DCV maturation. In this screen, we identified TBC-8, an evolutionarily conserved RAB GAP. We demonstrated that TBC-8 is an active RAB GAP since mutations in its catalytic motif “IxxDxxR” prevent the function of TBC-8 for DCV maturation. It was previously shown that mutation of the arginine within the catalytic motif to alanine renders the GAP into a catalytically inactive protein [29], [41]. TBC-8 was specifically expressed in neurons using a transcriptional reporter construct. The strong neuronal expression plus the similarity in DCV maturation defects suggested that TBC-8 is a neuron specific RAB-2 GAP. Furthermore, analysis of TBC-8 activity in non-neuronal tissues revealed that it does not regulate RAB-2 activity in postendocytic trafficking and in the removal of apoptotic cell corpses in the germ line. These results indicate that in non-neuronal tissues other GAPs exist that regulate RAB-2 activity. In addition, we cannot exclude that in different subtypes of neurons, RAB-2 function may be regulated by GAPs other than TBC-8.
Unfortunately, due to the inability of obtaining soluble TBC-8 when expressed in bacteria or insect cells, we were unable to demonstrate GAP activity towards RAB-2 in vitro. Thus, it is still possible that TBC-8 might not be a RAB-2 specific GAP.
However, three findings indicate that TBC-8 acts as a RAB-2 specific GAP in vivo.
First, the analysis of the two double mutants of tbc-8 and unc-108/rab-2 revealed that both proteins are involved in the same pathway. Second, in the yeast two-hybrid system, the GTP-bound form of RAB-2 specifically interacts with the TBC-domain of TBC-8 but only if the TBC-domain was rendered catalytically inactive. Previously, it was shown that an exchange of the catalytic arginine to alanine within the TBC-domain could be used to detect interaction of a GAP with its cognate Rab [41]. RAB-19 was observed to interact with both wild type and catalytically inactive forms of TBC-8 suggesting that it is unlikely that TBC-8 is the GAP for RAB-19. Third, TBC-8 influences the membrane localization of its binding partner RAB-2 when over-expressed in neurons. It was shown that active GTP-bound RAB-2 localizes to discrete puncta at the Golgi apparatus while inactive GDP-bound RAB-2 is mainly cytosolic [22]. However, when TBC-8 was over-expressed, RAB-2 was redistributed to the cytosol. Such redistribution of a Rab upon expression of its GAP has been previously used as an indication to screen for functional Rab/GAP pairs [42]. All these results strongly suggest that TBC-8 is likely to be a neuronal RAB-2 GAP. Irrespective of its specificity, the catalytic activity of the TBC-domain of TBC-8 is absolutely required for proper DCV maturation (Figure 1B, 1C). This strongly suggests that TBC-8 functions as an active Rab GAP during RAB-2 dependent DCV maturation in C. elegans motoneurons.
TBC-8 is conserved throughout evolution. There are two orthologs in mammals: SGSM1 (small G protein signaling modulator 1) also called RUTBC2 (RUN and TBC1 domain containing 2) [36] and SGSM2/RUTBC1 [36], [43]. Although TBC-8 shares domain structure with SGSM1 and SGSM2, it is more similar to SGSM1 based on sequence alignments. The TBC-domain of SGSM2 is about 162 aa longer than the TBC-domain of TBC-8, leading to a higher gap penalty when both proteins were aligned (data not shown). However, multiple splice variants can be found for both SGSM1 and SGSM2. Thus, we cannot exclude the possibility that one of those variants is even more closely related to TBC-8. SGSM1 is predominantly expressed in the adult brain [36]. Furthermore, it has been postulated to be involved in intracellular transport in neurons, where it localizes to the TGN [36]. It was shown that SGSM1 interacts with different Rab proteins; however, interactions with specific domains of SGSM1 were not shown [36]. In addition, SGSM1 binds to all members of the Rap GTPase family [36]. Recently, SGSM2/RUTBC1 has been found as a Rab9 effector [43]. Furthermore, functional GAP activity of SGSM2 towards Rab32 and Rab33 has been demonstrated in vitro [43]. The C. elegans genome does not encode a Rab9 ortholog, nor could we detect any interaction between TBC-8 and RAB-33 or GLO-1, the C. elegans Rab32 ortholog. Based on the fact that Drosophila contains two SGSM1/2 orthologs, it is likely that C. elegans has lost the SGSM2 ortholog along with Rab9 during evolution. Thus, we reason that SGSM1 is more closely related to TBC-8, consistent with their strong neuronal expression.
TBC-8 and its orthologs SGSM1/2 also contain an N-terminal RUN (RPIP8/UNC-14/NESCA) domain, which consists of α-helices [35], [44]. It is proposed that RUN domains are required to facilitate protein-protein interactions, specifically with small GTPases of the Rab and Rap family [34], [36], [45]. A fragment containing the RUN domain of SGSM1 was shown to interact with both Rap1 and Rap2 in a co-IP experiment [36]. We tested all C. elegans Rabs with TBC-8 and showed that RAB-19 binds to TBC-8, whereas no members of the C. elegans RAP family interacted with TBC-8 within the yeast two-hybrid system, suggesting that the interaction between the RUN domain and Rap GTPases might not be conserved (Figure S8).
In this study, we demonstrated that tbc-8(tm3802) mutants have similar defects in DCV maturation as unc-108/rab-2 mutants. Neuropeptides other than NLP-21 are also affected by tbc-8 deletion. The fluorescence of VENUS-derived FMRF-like neuropeptide FLP-3 containing vesicles at the dorsal nerve cord was also decreased in tbc-8 mutants. However, the fluorescence derived from the insulin-like neuropeptide, INS-22-VENUS, was unchanged in tbc-8(tm3802), similar to unc-108/rab-2 [21]. Unlike NLP-21 and FLP-3, the insulin-like neuropeptide INS-22 lacks PC2 (proprotein convertase 2) cleavage sites. Therefore, VENUS is not cleaved off during the maturation process, and the VENUS-tag is aggregated together with neuropeptides within the insoluble core of DCVs [31], [46]. These data further suggest that it is mainly the soluble cargo that is lost in tbc-8 mutants, as observed also in the unc-108/rab-2 mutants.
Here, we could not observe any differences in the axonal fluorescence levels of the DCV integral membrane protein, IDA-1-GFP, indicating that transmembrane cargos are not affected by tbc-8 deletion. However, in unc-108/rab-2 mutants, the axonal fluorescence of the transmembrane cargo IDA-1-GFP is decreased [21]. This result suggests that both soluble and transmembrane cargos are possibly lost from DCVs in unc-108/rab-2 mutants [21]. This discrepancy in transmembrane cargos being affected in unc-108 mutants but not in tbc-8 mutants suggests that in neurons, another GAP may exist at the site where RAB-2 functions to retain transmembrane cargos. Despite this difference, the axonal DCV numbers are not affected by rab-2/unc-108 or tbc-8 mutations as determined by EM. This finding demonstrates that TBC-8 participates with RAB-2 in the retention of specific (soluble) cargo in DCVs during the maturation process. Since a blockade of the endosomal delivery by over-expression of constitutively active GTP-bound RAB-5 (Q78L) could rescue the loss of soluble NLP-21-derived VENUS, TBC-8 and RAB-2 are likely involved in cargo sorting processes during maturation at the Golgi-endosomal interface.
Collectively, our results indicate that tbc-8(tm3802) and unc-108/rab-2(n501) dominant active mutants share similar defects in DCV maturation, reiterating the idea that TBC-8 is the GAP for RAB-2. Despite the similarities in DCV maturation phenotypes, there are major differences in behavioral phenotypes in both tbc-8(tm3802) and unc-108/rab-2(n501) mutants. Unlike unc-108/rab-2 mutants, tbc-8(tm3802) mutants display no defects in locomotion behavior (Figure S3) [21], [22]. There are several possible explanations why both mutants show different locomotion phenotypes. First, tbc-8 may not be expressed in all subtypes of neurons that are necessary to coordinate locomotion, whereas unc-108/rab-2 is more broadly expressed in (all) neurons and other tissues [22], [33]. Interestingly, it was shown that the locomotory defect in unc-108/rab-2 mutants could be rescued by expressing unc-108 under the pan-neuronal synaptobrevin-1 (snb-1) promoter, but not when unc-108 was expressed under the unc-17 promoter, which is expressed in a subtype of motoneurons, the cholinergic motoneurons [33]. Secondly, in the absence of tbc-8 in some neurons, another GAP might substitute for TBC-8. This hypothesis is further supported by the fact that the transmembrane protein IDA-1 is unaffected by tbc-8 deletion, suggesting that another GAP must exist in neurons. Thirdly, in the absence of TBC-8, it is possible that the low intrinsic activity of RAB-2 might partly rescue some of the phenotypes in tbc-8 mutants that are observed in unc-108/rab-2 mutants. Fourthly, an unknown transmembrane factor, which is normally present on DCVs could be lost in unc-108/rab-2 mutants during DCV maturation and thus be responsible for the uncoordinated phenotype. This transmembrane factor could still be present in tbc-8 mutants since no defects in IDA-1 trafficking were observed in tbc-8 mutants.
How does the possible RAB-2 GAP, TBC-8, regulate RAB-2 activity during DCV maturation? We have previously shown that RAB-2 dependent retention is achieved by a dynamic ON/OFF cycle of RAB-2 at the Golgi-endosomal interface [21], [22]. Thus, RAB-2 most likely goes through several rounds of activation and deactivation during DCV maturation. The fact that the RAB-2 effector, RIC-19, interacts with the RAB-2 GAP, TBC-8, supports this view. Thus, once RAB-2 is locally activated at the Golgi-endosomal interface, it recruits the RIC-19/TBC-8 complex, which in turn will lead to its deactivation and release from its membrane localization. This will also then lead to a cytosolic redistribution of RIC-19 and TBC-8. The fact that overexpressed RIC-19 and TBC-8 show a dramatic membrane localization even in the absence of RAB-2 suggests that RIC-19 and TBC-8 might also bind to a yet unidentified factor that is membrane-localized, independent of RAB-2. The observation that this massive membrane localization of RIC-19 and TBC-8 is only seen by overexpression of both proteins suggests that the interactions might be of low affinity. Thus, in a normal situation, the effective concentrations would be insufficient for a productive interaction and thus recruitment. In this case, the interaction would require activated RAB-2 for efficient recruitment of RIC-19 and TBC-8 to membranes. This would postulate that active GTP-bound RAB-2 would interact with this unknown factor as well as with the RIC-19/TBC-8 complex. Accordingly, an inactivation of RAB-2 by its GAP, TBC-8, would lead to a rapid disassembly of the complex and a release of its components for a new round of recruitment. However, from our current set of data, we cannot fully exclude the possibility that RIC-19 and TBC-8 are recruited separately from each other to the Golgi-endosomal interface and start to interact at RAB-2 positive vesicles. More efforts have to be made to solve the function of the TBC-8/RIC-19 interaction.
It is currently unclear whether and how TBC-8 GAP activity is regulated. On the one hand, TBC-8 can be inactive when recruited by active RAB-2 via RIC-19 and then activated after recruitment. On the other hand, TBC-8 can be constitutively active while bound to RIC-19, which then would lead to an immediate deactivation of RAB-2 after recruitment of the effector complex. RIC-19 binds outside of the TBC-domain, to the N-terminal part of TBC-8 containing the RUN domain. Thus, it is currently unlikely that RIC-19 would mask the TBC-domain and inhibit GAP activity. The interaction between RIC-19 and TBC-8 seems to be conserved because the human homolog of RIC-19, the diabetes autoantigen ICA69, could also bind to TBC-8 in the yeast two-hybrid system. This result suggests that TBC-8 and its interaction to ICA69 may also play a role in insulin signaling in humans.
TBC-8 activity might also be regulated by its interaction with active GTP-bound RAB-19. However, we did not detect any defects in DCV maturation in rab-19(ok1845) deletion mutant animals. This result suggests that RAB-19 is not involved in DCV trafficking and that TBC-8 together with RAB-19 might act in another DCV-independent pathway.
So far it has been described that in Rab GAP cascades, the downstream Rab would recruit the Rab GAP for the upstream Rab via its effector complex [25]. For example, in the Ypt1p/Ypt32p cascade Ypt1p associates with the Golgi and regulates ER to Golgi trafficking [47]. The GAP for Ypt1, Gyp1p, is recruited by interaction with activated Ypt32p, the downstream Rab [48]. It has been shown that this counter-current GAP cascade limits the overlap between activated Ypt1p and Ypt32p within the yeast secretory pathway [49], [50]. Therefore, this GAP cascade sharpens the boundaries during the Rab conversion cascades supporting sorting of cargo into the downstream Rab domain and thus, enhancing the directionality of the transport process [51].
However, the situation might be different in the case of RAB-2 and TBC-8 where the active Rab GTPase would recruit its own GAP through its effector complex. This would suggest that RAB-2 might not participate in a traditional Rab conversion cascade. Further support for this view comes from the fact that we were unable to identify a second Rab GTPase in C. elegans showing the same DCV maturation phenotype despite screening all Rab GTPases in C. elegans (our unpublished data). It is therefore likely that RAB-2 participates in a membrane trafficking mechanism distinct from a Rab cascade. However, since we were unable to test RAB-2 specific GTPase activity of TBC-8 due to protein insolubility, we can currently not exclude that RAB-2 might be part of a Rab cascade. RAB-2 might participate in a highly dynamic membrane sorting event by helping to transiently orchestrate the assembly of an acceptor complex required for DCV maturation. This complex would then subsequently be rapidly disassembled to allow sorting to proceed further. It is conceivable that active RAB-2 would help to accept retrograde trafficking carriers within the Golgi-endosomal interface to be delivered back to the maturing DCV compartment. This would explain why in rab-2 mutants, cargo from maturing DCV is lost to the endosomal-lysosomal pathway [21], [22].
The fact that TBC-8 also binds to human ICA69 might suggest that the role of TBC-8 is evolutionarily conserved and that its human ortholog may play an important role in regulating insulin secretion from pancreatic ß-cells since SGSM1 is also expressed in the pancreas [36].
C. elegans strains were maintained at 20°C on nematode growth medium (NGM) seeded with Escherichia coli OP50 as described previously [52]. The following strains were used in this study: wild type N2 Bristol strain, tbc-8(tm3802), unc-108(n501), unc-108(nu415), egl-3(gk238), ric-19(ok833), rab-19(ok1845), eri-1(mg366), tbc-1(tm2282), tbc-2(qx20), tbc-4(tm3255), tbc-11(ok2576), tbc-12(gk362), tbc-13(ok1812) and tbc-18(ok2374). The following integrated transgenic lines were used in this study: nuIs183[punc-129::nlp-21–venus], nuIs195[punc-129::ins-22–venus], ceIs61[punc-129::flp-3–venus], ceIs72[punc-129::ida-1–gfp], nuIs152[punc-129::gfp-snb-1], nuIs168[punc-129::venus-rab-3], bIs34[prme-8::rme-8-gfp], arIs36[phsp::ssgfp], arIs37[pmyo-3::ssgfp] and bcIs39[plim-7::ced-1-gfp]. All these integrated strains were crossed into the strain tbc-8(tm3802). Crosses were carried out using classical genetic approaches, and the progeny was genotyped by PCR. For more detailed information about strains used see Table S1. Transgenic worm lines were generated by microinjections into N2 Bristol strain or into tbc-8(tm3802) [53]. Table S2 lists all transgenic worm lines including the plasmids and co-injection marker concentrations used in this study.
The cDNA of tbc-8 was amplified by PCR using a cDNA library (ProQuest, Invitrogen) with the primers oGQ1622 (tt ccc ggg tta ccg gtt atg tgg agg gcg aag aag cca aca) and oGQ1623 (ggg gcg gcc gcc ctc gag cta ctt gag gtg ttg cac aag gtt) and TA cloned into the vector pGEMT (Promega). Positive clones were verified by sequencing reaction (Qiagen). tbc-8 was then subcloned into different vectors: see Table S3 for details. For expression of genes of interest in the nervous system of C. elegans, the backbone of the vector pPD115.62 was used. The myo-3 promoter in pPD115.62 was exchanged for the rab-3 promoter using PstI and KpnI restriction sites creating prab-3::gfp. The same procedure was used to generate punc-129::gfp. For the tbc-8 promoter construct, a fragment of 2873 bp upstream of tbc-8 start codon together with the first five tbc-8 codons was PCR amplified using N2 genomic DNA with the following primer pair: oGQ1793 (ctt aag ctt ctg cag gaa ctt ttc cat ctg) and oGQ2055 (agt aac cgg tgc cct cca cat atc tgc cga tga atg ccg).
The catalytic arginine finger of TBC-8 was mutated to alanine using a site-directed mutagenesis approach. The following primers were used: oGQ1622, oGQ1623, oGQ1698 (gac gtg gag gca tgc gat aga aat ttg atg ttc) and oGQ1699 (tct atc gca tgc ctc cac gtc ctt gtc aat tct).
For yeast two-hybrid analysis, all dominant active forms of rab genes were cloned into the bait vector pGBKT7 whereas all tbc-8 variants were cloned into the prey vector pGADT7 (Clontech) (Table S3).
For confocal microscopy, live worms were paralyzed with 50 mM NaN3 (Sigma) on 2% agarose (Invitrogen) pads. An inverted Confocal Laser Scanning Microscope (SP2, Leica) with a 100× oil objective (NA = 1.4) (co-localization studies) or a 63× oil objective (NA = 1.32) (DCV assay, expression pattern studies) was used. GFP was excited with a laser at 488 nm, YFP at 514 nm and tagRFP as well as mCherry at 561 nm. The scan was performed with a resolution of 1024×1024 pixels, and the pinhole was set to 1 airy unit. To study co-localization of TBC-8 with different subcellular markers tagged with fluorescent proteins, images of neuronal cell bodies from the ventral nerve cord were taken. Image stacks were captured and average intensity projections were obtained using the Leica software. These images were then edited using ImageJ software (National Institutes of Health).
Furthermore, images of expression pattern of tbc-8 were taken using a Perkin Elmer Spinning Disc Confocal Microscope. These images were edited using Adobe Photoshop software.
For quantification studies from DNC, neuronal cell bodies and coelomocytes (DCV assay, imaging of endocytosed ssGFP in coelomocytes secreted from muscle cells), young adult worms were imaged as described previously [31]. For cell body and DNC imaging, the neuronal cell bodies and DNC were oriented toward the objective, whereas for coelomocytes imaging, the posterior coelomocytes were oriented laterally. Image stacks of the regions of interest were captured and maximum intensity projections were obtained using the Leica software. For all obtained images, the same settings were used. These projections were thresholded and quantified using the ImageJ software (National Institutes of Health). All these data were normalized to wild type.
For quantification of vesicle sizes in neuronal cell bodies in nuIs183 background strains, obtained images of cell bodies (explained above) were analyzed using the ImageJ software (National Institutes of Health).
In order to image apoptotic cell corpses, image stacks of the distal gonad arms were captured and projections were obtained using the Perkin Elmer Spinning Disc Confocal Microscope. From these pictures the number of cell corpses in the distal gonad arm and in the gonad loop was counted manually for each strain.
The Matchmaker yeast two-hybrid assay was performed according to the manufacturer's protocol (Clontech). The appropriate plasmid combinations were transformed into the yeast strain AH109 (Clontech) and spread onto selective growth media lacking leucine and tryptophan for plasmid selection. Protein interactions were tested as follows: several clones of transformants were mixed and diluted to OD600 of 0.4 (RAB interaction studies) or 0.2 (RAP interaction studies). Five microliters of this yeast dilution was spotted onto selective plates lacking leucine, tryptophan and histidine. Interactions were identified by growth after 3–4 days. All interacting proteins were tested for self-activation as described above using the appropriate empty vector pGBKT7 or pGADT7, respectively.
HEK293 cells were grown in high glucose (4.5 g/l) DMEM supplemented with 10% FBS, 110 mg/l sodium pyruvate, 2 mM glutamine, 100 U/ml penicillin, and 10 µg/ml streptomycin in a 5% CO2 incubator at 37°C.
For co-immunoprecipitation, 4×106 HEK293 cells were plated onto two 10 cm petri dishes 24 hours before transfection, which was performed using TurboFect in vitro Transfection Reagent according to the manufacturer's protocol (Fermentas). After 24 hours, cells were washed with PBS and harvested in lysis buffer (50 mM Tris pH 7.5, 150 mM NaCl, 1% Triton X 100, 0.5 mM EDTA, 10% glycerol, Complete Mini Protease inhibitor (Roche)) for 30 min at 4°C. Lysates were pre-cleared by centrifugation at 4°C before supernatant was incubated with 2 µg monoclonal anti-GFP antibody (clone 3E6, Invitrogen) for 3 hours at 4°C. Protein G Plus-sepharose (Pierce) beads were added. After another incubation time of 2 hours, the beads were washed three times with washing buffer (50 mM Tris pH 7.5, 500 mM NaCl, 0.1% Triton X 100, 0.5 mM EDTA, 10% glycerol, Complete Mini Protease inhibitor (Roche)) and resuspended in Laemmli loading buffer. Samples were resolved on 10% SDS-polyacrylamide gels and blotted onto a nitrocellulose membrane. The detection of co-precipitated proteins were performed by applying a mixture of two monoclonal mouse anti-GFP antibody (1∶1000) (clones 7.1 and 13.1, Roche) and monoclonal anti-V5 antibody (1∶5000) (Invitrogen) followed by goat anti-mouse horseradish peroxidase-conjugated secondary antibody (1∶10,000) (Jacksons Laboratory). A FujiFilm LAS 3000 processor was used to develop images, which were edited using the ImageJ software (National Institutes of Health).
A 100 µm deep aluminum platelet (Microscopy Services, Flintbek) was filled with E. coli OP50 suspension. About 20 young adult worms were transferred into the chamber and immediately frozen using a BalTec HPM 10. Freeze substitution was carried out in a Leica AFS2. Incubations were performed at −90°C for 100 h in 0.1% tannic acid, 7 h in 2% OsO4, and at −20°C for 16 h in 2% OsO4, followed by embedding in EPON at room temperature [54] (all solutions w/v in dry acetone). Fifty nanometer sections were mounted on copper slot grids and placed for 10 min on drops of 4% (w/v) uranyl acetate in 75% methanol and then washed in distilled water. After air drying, the grids were placed on lead citrate [55] for 2 min in a CO2-free chamber, and rinsed in distilled water. Micrographs were taken with a 1024×1024 CCD detector (Proscan CCD HSS 512/1024; Proscan Electronic Systems, Scheuring, Germany) in a Zeiss EM 902A, operated in the bright field mode. The SV and DCV diameter and distribution at the synapse of motoneurons were analyzed by a semi-automated analysis software XtraCount (manuscript in preparation).
For the analysis of presynaptic terminals and SV and DCV distributions, cross sections of young adult animals were used to image cholinergic neuro-muscular junction (NMJ) synapses in the ventral nerve cord, posterior to the nerve ring. These cholinergic NMJ synapses were defined as polyadic synapses projecting onto muscle arms as well as other neurons according to the standard convention in the C. elegans EM field. Axons showing a clearly visible presynaptic density and synaptic vesicles were defined as synapse. The mean area of presynaptic terminal was measured in cross sections using the surrounding axonal membrane as border.
For the morphological analysis of neurons, ten motoneuronal cell bodies localized in the ventral nerve cord in wild type worms were compared with seven cell bodies in tbc-8(tm3802) worms.
For the analysis of postendocytic trafficking within coelomocytes of the fluid-phase endocytosis marker TR-BSA, the integrated strain bIs34[prme-8::rme-8-GFP] was crossed into tbc-8(tm3802) to label RME-8 positive endosomes. TR-BSA (1 mg/ml) was injected into the body cavity in the pharyngeal region of young adult worms as described previously [56]. Uptake and postendocytosis was analyzed after 10, 30 and 50 min after injection by confocal microscopy. At least five animals were injected for each time point. Single fluorescence images at the middle plane of each coelomocyte were taken and line-averaged. Images were edited using ImageJ software (National Institutes of Health).
The strain arIs36[phsp::ssGFP] [39] was crossed into tbc-8(tm3802). Young adult worms were grown at 20°C before a heat-shock at 33°C for 30 min was performed. Worms were put back to 20°C to recover until they were used for imaging. The uptake of ssGFP into coelomocytes and degradation of endocytosed GFP was monitored after 3.5, 6 and 28 hours after heat-shock. All fluorescence pictures were taken with the same settings. At the time points where fluorescence was hard to detect during imaging, DIC (differential interference contrast) images of the respective specimen were taken. Later, these images were used to outline the cell boundaries of coelomocytes. Images were edited using the ImageJ software (National Institutes of Health).
RNAi by feeding was performed as described previously [57]. The plasmids L4440 and L4440-tbc-8 were transformed into the E. coli strain HT115, respectively. Overnight cultures of these bacteria were seeded onto NGM plates that contained 100 mg/ml ampicillin and 1 mM IPTG. Ten L4 worms of the strain eri-1(mg366);nuIs183 were placed onto these plates to allow egg laying and were transferred onto new RNAi plates every 12 hours. After the third round of transferring worms, the parents were removed and the progeny was imaged when they reached the young adult worm stage. Fluorescence of NLP-21-derived VENUS in the dorsal nerve cord was normalized to the VENUS fluorescence of mock RNAi (L4440) worms.
In order to assay locomotion, young adult worms of each strain were transferred to non-seeded NGM plates. After an initial adjustment time of 30 min, the number of body bends (one sine wave) was counted over a period of 3 min.
|
10.1371/journal.pntd.0005648 | Congenital toxoplasmosis in Austria: Prenatal screening for prevention is cost-saving | Primary infection of Toxoplasma gondii during pregnancy can be transmitted to the unborn child and may have serious consequences, including retinochoroiditis, hydrocephaly, cerebral calcifications, encephalitis, splenomegaly, hearing loss, blindness, and death. Austria, a country with moderate seroprevalence, instituted mandatory prenatal screening for toxoplasma infection to minimize the effects of congenital transmission. This work compares the societal costs of congenital toxoplasmosis under the Austrian national prenatal screening program with the societal costs that would have occurred in a No-Screening scenario.
We retrospectively investigated data from the Austrian Toxoplasmosis Register for birth cohorts from 1992 to 2008, including pediatric long-term follow-up until May 2013. We constructed a decision-analytic model to compare lifetime societal costs of prenatal screening with lifetime societal costs estimated in a No-Screening scenario. We included costs of treatment, lifetime care, accommodation of injuries, loss of life, and lost earnings that would have occurred in a No-Screening scenario and compared them with the actual costs of screening, treatment, lifetime care, accommodation, loss of life, and lost earnings. We replicated that analysis excluding loss of life and lost earnings to estimate the budgetary impact alone.
Our model calculated total lifetime costs of €103 per birth under prenatal screening as carried out in Austria, saving €323 per birth compared with No-Screening. Without screening and treatment, lifetime societal costs for all affected children would have been €35 million per year; the implementation costs of the Austrian program are less than €2 million per year. Calculating only the budgetary impact, the national program was still cost-saving by more than €15 million per year and saved €258 million in 17 years.
Cost savings under a national program of prenatal screening for toxoplasma infection and treatment are outstanding. Our results are of relevance for health care providers by supplying economic data based on a unique national dataset including long-term follow-up of affected infants.
| Toxoplasma gondii is a widespread parasitic disease. In the event of primary infection during pregnancy, this parasite can be transmitted from mother to unborn child. Clinical presentation of congenital toxoplasmosis varies from asymptomatic to life-threatening risk for the fetus and infant and in later life. Prevention programs and screening strategies of health care providers vary in different countries. Austria has implemented mandatory prenatal screening for toxoplasmosis for four decades. The screening is free of charge for families and costs are covered by national health care providers. Compliance with the national program is good and outcomes for infected pregnant women and their infants since 1992 are well documented. We compared lifetime costs of screening, treatment, and follow-up with costs in a No-Screening scenario in an economic decision-analytic model. Prenatal screening resulted in substantial cost savings due to reduction in congenital toxoplasmosis and consequent injuries in affected children.
| Toxoplasma gondii (T. gondii) is a protozoal parasite that infects up to 30% of humans globally, although prevalence of infection varies widely, from 10% to 80%, among world regions and within regions [1]. While the definitive host is the cat, sources of infection for humans include food, the water supply, and organ transplants as well as direct contact with cat feces in the soil and domestic litter [1–5]. Additionally and of particular concern is maternofetal transmission during pregnancy after primary infection.
Prevalence is high in South America and tropical Africa (>50%) [6], moderate in parts of western, central, and southern Europe (30% to 50%), and relatively low (10% to 30%) in northern Europe, North America, Southeast Asia, and the Sahara [7,8]. Prevention entails adequate cooking of meat and washing of fruits and vegetables as well as drinking water free of contamination with oocysts. Educational programs for prevention, however, can only reduce infection rates, not eliminate new infections, because most people, even those who are aware of the infection routes, do not know the source of their infection [3,4,6]. Most people infected postnatally have no recognized symptoms, but immune suppression due to medical conditions or treatments can lead to serious damage to the brain and eyes as a consequence of T. gondii infection. Infection with T. gondii that occurs during pregnancy can be transmitted to the unborn child and may have serious consequences, before or after birth, even in apparently asymptomatic infected newborns [9–11]. Three European countries—Austria, France, and Slovenia—have instituted mandatory prenatal screening for primary infections of T. gondii to minimize the harmful effects of infection on infants. This is the first systematic study of the cost of a European national prenatal screening program to reduce congenital toxoplasmosis (CT) and its sequelae [12].
Women with primary infection with T. gondii during pregnancy may exhibit no symptoms, but there is about a 50% risk of transmission to the fetus and the possibility of mild to profound injury to the unborn child in untreated women [1]. The risk of maternofetal transmission increases over the course of the pregnancy, from very low risk in the first trimester to nearly 100% in the final weeks of pregnancy. In the event of transmission, risk of injury to the fetus varies inversely with gestational age, with the risk of profound injury greatest in the first trimester and the possibility of mild disease or no recognized symptoms in later stages of gestation [1,6,13,14].
Consequences of CT can include retinochoroiditis, hydrocephaly, cerebral calcifications, splenomegaly, hearing loss, blindness, and death [1,6,15,16]. In countries with prenatal screening for primary infections and consequent pre- and postnatal treatment, rates of CT and severity of symptoms in infants are lower than in countries without screening programs or compared to historical data before screening was initiated [7,10,17,18]. In comparison, a recent study of children in the United States with CT who had no pre- or postnatal treatment found that 91% of the children referred had visual and/or mental impairment by age 12 [9].
The risk of CT is complicated, however, by the diversity of genotypes of T. gondii. Type II predominates in Europe and was thought to be the predominant genotype in North America [6,19–21]. Recent research has identified greater diversity in US wild and domestic animals than was previously thought [22–24]. Types I and III and atypical genotypes are more common in Central and South America [25–27]. These latter strains are more virulent and are associated with ocular disease even when acquired postnatally by immune-competent persons [28]. South American genotypes are also associated with more serious injuries in CT [19,20,28–30].
Prevalence of infection with T. gondii varies considerably in Europe, from 7% in Norway [31], 10% in the United Kingdom [32], 19% in Italy [33], 32% in Spain [34], 33% in Austria [31,35,36], and 34% in Slovenia [37], to 37 to 44% in France [7,38] (all reported since 2000). Over the past 20 years, prevalence has fallen rather dramatically in most of the high prevalence countries coincident with national education campaigns, which have perhaps led to changes in food preparation [7,31]. Systematic screening of pregnant women also plays an educational role in highlighting the importance of food safety and hygiene for the health of the unborn.
Countries with high prevalence in the past similarly had high rates of primary infection in women during pregnancy. This may seem paradoxical since the higher the prevalence among women of child-bearing age, the higher will be the proportion of women entering pregnancy who are immune. Since prevalence, however, increases with age, the majority of young women are not immune and continue to be at risk, presumably with the same food preparation habits as before.
The substantial drop in prevalence from the 1990s to the present was accompanied by a substantial drop in maternal incidence after an initial rise [7,17]. Austria in 1974, France in 1992, and Slovenia in 1995 initiated mandatory prenatal screening programs aimed at reducing maternofetal transmission as well as the severity of injury from CT. Numerous studies have reported that systematic prenatal screening and treatment were coincident with substantial reductions in maternofetal transmission and sequelae of CT [7,10,12,13,17,18,36,37,39–45].
No systematic economic evaluation of those programs, however, has been published. The countries with systematic prenatal screening and treatment programs face the paradox of successful prevention. Now there are so few children with serious, disabling symptoms of CT that it can appear that the risk of maternal infection does not warrant the expenditure for universal prenatal screening programs. Health budgets are under continual scrutiny. In some countries political currents have changed and the assumption of state responsibility for health is questioned. Moreover, there are diverse stakeholders in the decision to allocate funds to prenatal screening or to other national health needs: the Ministry of Health, insurance funds, the Ministry of Education, social security administrations, and families of affected children.
The purpose of the current work is to compare the societal costs of CT under the Austrian national program of prenatal screening for primary toxoplasmosis with the societal costs that would have occurred in the absence of the screening program.
In 1961, Thalhammer revealed a rate of CT of 78 per 10,000 live births for the Austrian population [46]. In response, mandatory prenatal screening for toxoplasma infection for all pregnant women was implemented in 1974 under the auspices of the national health care system [46,47]. This prenatal screening is part of a national prevention program called “Mother-Child-Booklet-Program” for all pregnant women and their infants through early childhood. The costs are covered by the government and the local health insurance funds; the program is free of charge for families.
The Austrian national program is described in detail in previous works [12,31,48]. Serological prenatal screening is performed ideally on a bimonthly schedule, at 8, 16, 24, and 32 weeks of gestation as well as a maternal or neonatal test for women seronegative up to the time of birth and women who have not been tested during pregnancy. In women with proven seropositivity before the current pregnancy, no further toxoplasma testing is necessary. Women who are tested and found to have been seropositive before conception require only one test. Those with suspected primary infection during pregnancy are tested twice. In Austria during this screening program, the local laboratories used 9 different test methods for the detection of IgM Toxo antibodies, each performed according to manufacturer recommendations. In the case of primary infection in a pregnant woman or to clarify suspicious test results, blood samples were retested in the reference laboratory. The Toxoplasmosis Laboratory at the Medical University of Vienna routinely uses the in-house Sabin Feldman dye test, immunosorbent agglutination assay (ISAGA)-IgM (bioMérieux, France), VIDAS Toxo IgG Avidity (bioMérieux, Frankreich), and PCR diagnostics for the detection of toxoplasma infections in pregnant women and their children.
In women with primary infection, amniocentesis and polymerase chain reaction of the amniotic fluid is recommended, but costs for those additional tests are not covered by the program. A positive result from amniocentesis identifies an affected fetus prenatally and influences the treatment during pregnancy. The routine PCR analysis used for the B1 gene after amniocentesis showed a sensitivity and specificity of 87.2% and 99.7%. Furthermore, the results revealed a positive predictive value and negative predictive value of 94.4% and 99.3% [48]. More recently, using the 529-bp PCR protocol improved sensitivity up to 100.0% [49].
Pregnant women are treated after the diagnosis of primary infection until birth, and infants with proven or suspected congenital infection are treated during the first year of life. In cases of CT, additional investigation, including cranial ultrasound, funduscopy, and complete blood count, is part of the program. The screening program reached 93% of pregnant women over the period covered by this analysis, although the ideal schedule was not achieved for most women [31].
The Austrian Toxoplasmosis Register records the serology history and birth outcomes for 1,387,680 pregnant women from 1992 to 2008 [12]. All cases of CT are recorded in the Register and thus it provides the basis for evaluating the costs of the program and pediatric outcomes over the 17-year period. In 10% of women no toxoplasma testing was necessary due to proven seropositivity before pregnancy. Screening confirmed additional infected women, resulting in seroprevalence of 34.4% used in the model [31]. The Register reported 70 women with primary infection of T. gondii and 8 cases of CT per year. The management of women and infants was stable, as was the rate of toxoplasma infection, during the observation period. Pediatric long-term follow-up revealed that 81% of infants with T. gondii infection did not show any clinical signs as of May 2013. All clinical variables for infection, transmission, and outcomes in infants are shown in Table 1.
We retrospectively analyze data from the Austrian Toxoplasmosis Register for birth cohorts from 1992 to 2008 and clinical data from pediatric long-term follow-up to May 2013 [12]. Data were recorded at the Medical University of Vienna, Austria, in coordination with local nurses, physicians, specialists, and medical care centers. Average annual number of births was 81,628 over the 17-year period [12] and 76,547 over the last decade (www.statistik.at). We compared societal costs of illness over the lifetimes of affected children of the Austrian national program as it was carried out with the lifetime societal costs estimated in the hypothetical scenario of Austria if it had not implemented prenatal screening in those years.
We use TreeAge Pro Suite 2015 software (TreeAge Software, Inc., Williamstown, MA, USA) to construct a decision-analytic model. Using a societal perspective, we include the costs of treatment, care, and accommodation of injuries projected over the lifetimes of affected children, and lost productivity that would have occurred in a No-Screening scenario with the actual costs of screening, treatment, projected lifetime care and accommodation, and lost productivity in Austria for all of the children in the Register from 1992 to 2013.
The current work follows a template established in a decision-analytic model for a hypothetical prenatal serologic screening program for the United States [51]. The current work is the first to use clinical data on specific child outcomes with local costs to calculate the lifetime costs and benefits of a mandatory national prenatal screening program as it has been carried out over time compared to the costs that would have occurred if there had been no screening program.
The model (decision tree) contains two kinds of variables: probabilities at chance nodes (circles) and costs of outcomes at terminal nodes (triangles). Clinical variables are listed in Table 1 and represent the chance of primary infection during pregnancy, fetal infection, and pediatric clinical long-term outcomes. For the No-Screening branch, probabilities are based on international experience reported in peer-reviewed literature and synthesized in the US model [51]. Because this is a retrospective study dating back to 1992, the use of historical data for the counterfactual No-Screening scenario is appropriate. The risk of fetal infection in the No-Screening scenario is taken from the actual rate of transmission among untreated women recorded in the Austrian Toxoplasmosis Register [12]. In the Screening branch, probabilities for results at the 8-week screening are also derived from the literature in [51] because the small number of cases in the Austrian Register makes a comparison unreliable. For all other branches of the Screening arm, the probabilities are calculated from the Austrian Toxoplasmosis Register and thus represent actual Austrian experience recorded by the Toxoplasmosis Laboratory at the Medical University of Vienna, Vienna, Austria [12].
Costs of serology, treatment, and lifetime costs of special care and lost productivity for affected infants and their parents are listed in Table 2. Costs of serology, other tests, and medications are derived from recorded expenses of the Austrian program from the years 1999–2013 and adjusted to 2012 prices. For test and medication costs we use the average of costs reported by health insurance funds. Lifetime costs of injuries of CT include treatment, accommodation, special schooling, loss of earnings for affected infants, and loss of parental earnings. Earnings are used as a proxy for the lifetime productivity that is lost by the family and the society for infants affected by CT and their parents. Estimates of costs are derived from the literature for Austria (adjusted to costs for 2012) and, when necessary, for neighboring countries (adjusted to Austrian costs for 2012). Direct costs and productivity losses are discounted annually at 3% for as long into the future as each cost occurs. Direct costs for medication represent average maternal and infant treatment costs. Costs of special treatment are assigned to the actual child outcomes in the Austrian Register [12]. Detailed explanation of cost derivation can be found in the methodological supplement, S1 Methods. For the cost assigned to death, in utero or neonatal, we derive a Value of Statistical Life (VSL) for Austria in 2012 using the recommendation of the OECD (Organization of European Cooperation and Development), which is based on a meta-analysis of more than 800 studies of VSL [58]. The background on the use of VSL and the derivation of our estimate for Austria are explained in the methodological supplement, S1 Methods.
In addition to the costs that are assigned to each outcome as terminal nodes in the tree, we include the costs of amniocentesis with PCR, which is assigned to the group of women with primary infection. It is unnecessary to assign the costs to specific women because it does not change the overall costs. In Austria over the period, 60% of women with gestational toxoplasma infection underwent amniocentesis. The cost of PCR, which was €363.45, was absorbed by the local prenatal care centers. The total cost was almost €256,000. Since the decision tree calculates the cost per birth in the country, we assign the cost of PCR as overhead on all 1,387,680 births over the period. It is expected that the cost of PCR will drop significantly in the near future, to an estimated €100 when the testing is done routinely, reducing costs overall. Although the women and their insurers did not bear the cost, the expense does represent a societal cost and so we include it in the analysis.
The decision tree shows the probabilities of all possible outcomes and the costs associated with each outcome. In Fig 1, each outcome has a conditional probability that is the product of the probabilities along each branch. The formulas at the terminal nodes for each outcome list the direct and indirect costs that are explained in Table 2.
The method outlined above is the conventional way to calculate the lower-cost option, including societal costs that are borne by affected infants, their families, and the economy as a whole, regardless of who pays. There could be times, however, that a Ministry of Health or other institution would like to know just the impact of a program on the government budget, not societal cost. For that reason, we also calculate the cost-saving option considering only those costs paid by government and public insurance funds, that is, omitting lost productivity of affected children and their parents and VSL.
To test the robustness of our results to variations in costs, we perform a sensitivity analysis using an Incremental Tornado diagram varying all costs –10% and +10%, except for test costs, which have a lower bound of €4, and VSL, which is given a range of €800,000 to €6,700,000. The former amount represents only the discounted valuation of productivity loss over the lifetime, and the latter amount is the upper bound of the OECD estimate of VSL. (See S1 Methods for explanation of VSL derivation.)
The maternal screening study was approved by the local ethics committee at the Medical University of Vienna, Vienna, Austria (824/2009). All adult subjects and parents of any child participants gave their informed consent orally in person or by telephone at the time of inclusion. The individuals were included in the nationwide toxoplasmosis register performed 1992‒2008 and their oral consent was documented in the register data file. Written consent could not be obtained, due to the fact that this was a nationwide study. The data were processed anonymously. The current economic study utilized anonymous data from the national screening program.
In Austria, a country with a moderate seroprevalence of T. gondii during childbearing years, we recorded a total of 1,387,680 women giving birth between 1992 and 2008 (www.statistik.at). Fig 2 shows the decision tree after calculation of the lower-cost option, based on the probability of each outcome and the costs associated with each. As shown in Fig 2, and summarized in Table 3, lifetime societal costs of CT sequelae in the No-Screening scenario would have been €426 per birth, or about €35 million for all Austrian births in one year. Total societal costs in Austria that would have occurred without prenatal screening for nearly 1.4 million births over the 17 years would have been about €591 million, including costs for lifelong treatment and accommodation, as well as loss of earnings for affected children and their parents.
In contrast, prenatal screening for toxoplasma infections according to the Austrian national program including costs of screening, maternal treatment, infant treatment, and lifetime costs for those infants with CT sequelae amounted to €103 per birth. The total cost of the Screening scenario, including lifetime costs of CT sequelae, was €8.4 million for all births in one year for Austria, or €143 million for 1.4 million births in the 17-year period.
As shown in Table 3, the prenatal screening option resulted in savings of €323 per birth, or about €26 million per year compared to No-Screening. For all births, screening saved about €448 million in 17 years.
Adding the cost of amniocentesis with PCR for 60% of the women with primary toxoplasma infection during pregnancy increased the cost per Austrian birth in the period by about €0.18, changing the difference in cost per birth of the entire screening and treatment program by a trivial amount.
The TreeAge program calculates all of the costs that occur in each scenario—the counterfactual (No screening) compared to all actual lifetime costs in Austria resulting under the screening scenario. Thus the TreeAge program attributes costs to the Screening scenario that result from treating infants who are infected despite the program, including those whose mothers were not screened or were screened inadequately, with the lifetime costs of follow-up, accommodation, and parental work time lost. In Austria, if there were no screening program, one must assume that the state would provide health care for a child born with, or who later develops, CT symptoms. So the costs of diagnosing and caring for a symptomatic infected child are not really costs of the screening program itself. They would occur (and in much larger numbers) without the national screening. The €8.4 million a year under the Screening scenario represents the costs of the screening program plus the lifetime societal cost for the affected children born during the 17-year period.
The screening program itself entails very little cost. It includes only testing all pregnant women (except those already known to be seropositive) and treating women with primary infection. It also would include the cost of treating the very few asymptomatic infected infants because without screening they would be missed, but with screening, they would be treated from birth. Under the screening program, there have been 70 incident infections in mothers per year. Without treatment, there would be a fetal infection rate of 0.508 [12] and a probability of asymptomatic CT of 0.06 [1]. Thus, there would be two asymptomatic infected newborns treated per year because of the screening program who would have been overlooked without screening (70 x 0.508 x 0.06 = 2.10). Costs for each of those children would be: 5 infant IgG test, 5 infant IgM test, pediatric treatment, CBC, ECG, cranial ultrasound, and 17 funduscopies, which amount to €1,372.
The costs of the screening program, shown in Table 4, total approximately €1.9 million per year for all pregnancies, or €25 per pregnancy. A new diagnostic appears likely with a test cost of about €4. Recalculating with a test cost of €4 would reduce the total cost of prenatal screening and maternal treatment from about €1.9 to about €1.2 million (calculation not shown).
The costs of the screening program can be compared to the cost of caring for a child whose mother is not treated. The costs for individual services and productivity losses are listed in Table 2, but each symptomatic child generates multiple kinds of costs. In the tree before rollback (calculation), Fig 1, all the costs for an individual child for each outcome are listed at the terminal node. For example, in the No-Screening scenario, a child with severe visual, cognitive, and hearing impairment (Terminal node #14 in Fig 1) will incur the following costs (assuming symptoms at birth that lead to testing, treatment, and follow-up care): 5 infant IgG tests, 5 infant IgM tests, pediatric treatment, CBC, ECG, cranial ultrasound, and 17 funduscopies, as well as the direct costs and productivity losses for child and parents associated with severe visual, cognitive, and hearing impairment, and special education costs.
Fig 2 (Terminal node #14) shows the sum of those costs. The lifetime cost for one child with severe visual, cognitive, and hearing impairment is €1.8 million (€1,778,210). Thus the costs of the entire screening program for one year are nearly the same as the potential costs for a single severely affected child whose mother was not treated. A child with only severe visual impairment generates costs of €482,811 (at terminal node #9). The costs for four such children exceed the annual cost of the screening program. Without prenatal treatment, more than 90% of infected children have been found to have some form of serious impairment [1,9,52,53]. Prenatal screening with pre- and postnatal treatment as needed prevents or mitigates most injuries.
Austria has 70 primary infections per year [12]. If we assume 50% maternofetal transmission without prenatal treatment, as seen in Austrian women who were not treated [12], that would be 35 cases of CT each year, rather than the 8 cases per year under the treatment program, with symptoms ranging from mild visual impairment to fetal death. Because the model calculates costs on a population basis, the cost of €426 in the tree is a cost per Austrian birth, which is multiplied by the number of births, resulting in potential costs of €35 million for the 35 children who would be infected under the No-Screening scenario. The screening program costs €1.9 million per year while the societal costs of the No-Screening scenario are €35 million per year.
It is useful to see these costs in relation to overall Austrian government spending and Gross Domestic Product (GDP). The annual cost of the screening and treatment program is 0.007% of total Austrian public spending on health and 0.003% of overall Austrian government spending. The annual cost of the program is 0.0006% of Austrian GDP (Derived from data at www.focus-economics.com/country-indicator/austria/gdp-eur-bn; World Development Indicators, www.wdi.worldbank.org).
Calculating just the impact on the Austrian public budget—that is, omitting the lifetime costs of lost earnings that fall on affected children, their families, and society, and VSL for fetal and infant deaths, we find that the maternal screening program is still cost-saving. As seen in Fig 3, and summarized in Table 3, expenditures by government and government-sponsored insurers, based on Austrian experience over the period 1992 to 2008, cost €33 per birth compared to an estimated €219 per birth if the prenatal screening program had not been implemented in Austria. (As explained above, this overstates the budgetary cost of the screening program itself because it includes diagnosis and care of children who would be cared for under the Austrian health care system even without a screening program.) Even from the extremely narrow budgetary perspective, the Austrian national program has more than paid for itself in reducing the costs to the state and state-sponsored institutions of treating and educating children injured by CT by €186 per birth for 1.4 million births over the period. That amounts to a total budgetary saving of more than €258 million, or more than €15 million per year.
Results of the sensitivity analysis show that the savings both to society and to the government budget are robust to variations in all costs. Varying costs by ±10% had a trivial effect on cost per birth in the No-Screening and Screening scenarios and consequently on the savings that result from screening, for both the full societal cost and for the public budget. Fig 4 shows an Incremental Tornado Analysis from the societal perspective. The x axis shows the difference in costs per birth between the No-Screening and Screening scenarios with an Expected Value (EV) of €323. The horizontal bars show the full variation in the Expected Value (savings per birth) resulting from the range of values for each cost parameter. Both Fig 4 and Table 5 demonstrate the trivial impact on the large savings that result from screening. The variation in VSL had the greatest effect on costs, but even then the difference between low and high values for savings was only €56 and the savings from screening never fell below €275 per birth. Fig 5 shows the one-way sensitivity analysis on VSL in the societal model, which again demonstrates that whether one includes only the loss of earnings (€800,000) or the upper bound of the OECD estimate for VSL (€6.7 million), there is little impact on the savings derived from the screening program, showing the same minimum savings of €275 per birth seen in Fig 4 and Table 5.
Fig 6 shows the Incremental Tornado Analysis for the Budget impact. The Expected Value, that is savings per birth, is €186. The variation in savings per birth never exceeds €17 and the minimum savings from the screening program for the budget is never less than €178 per birth, as seen also in Table 6.
In this retrospective study we compare the costs for a national program of prenatal screening for T. gondii with a No-Screening scenario for Austria, a country with moderate seroprevalence in women of childbearing-age and 1,387,680 births over the years 1992 to 2008. There have been few economic analyses of CT-prevention programs [51,65]. To our knowledge this is the first report of an economic decision-analytic model incorporating surveillance data from pregnancy through long-term pediatric follow-up for an entire nation over nearly two decades of observation. Thus our data are of special interest for physicians, health care providers, and policy makers in considering the implementation of a prevention program for CT.
The substantial reductions in primary infection, maternofetal transmission, and fetal and child injuries resulting from T. gondii infection during the implementation of the Austrian prenatal screening program from 1992 to 2008 have been reported elsewhere [12]. In the current work, our major finding demonstrates that a national program of prenatal screening and treatment to prevent congenital toxoplasmosis or reduce clinical symptoms in affected infants is cost-saving for governmental health care providers and for Austrian society. Under the Austrian national prenatal screening program, total societal savings are €323 per birth. Consequently, the screening program saved about €448 million in costs to Austrian society for the birth cohorts from 1992 to 2008. Even in narrowly budgetary terms, the prenatal screening program has saved the Ministry of Health, the Ministry of Education, and government-sponsored insurance funds €186 per birth, or more than €258 million over the period, averaging more than €15 million a year, because of injuries prevented in children of women with primary toxoplasma infection. Even large variations in all costs make little difference in results. This is not surprising given the profound injuries that can occur without treatment and the low cost of the intervention.
Even in a country where prevalence is falling due to greater awareness and success of primary prevention, prenatal screening is needed. Lower prevalence means that more women enter pregnancy susceptible to infection. Since seroprevalence increases with age, women in their childbearing years are among the vulnerable population that has grown over the past decades as prevalence has declined.
Under the Austrian national program of prenatal screening, there has been a dramatic reduction in maternofetal transmission of T. gondii and in the degree of injury in affected children compared to historical data before implementing the prenatal screening [46,47]. Interestingly, the Austrian Toxoplasmosis Register shows even greater success in child outcomes than observed in France even though the French protocol mandates monthly testing, compared to the Austrian program of bimonthly testing [12]. It seems, however, that in Austria, while women are attending prenatal checkups, most are not receiving the recommended number of blood tests for primary toxoplasma infection [12,31,35]. Education of women and obstetric staff should be a relatively inexpensive solution that would improve even further the success of the Austrian CT-prevention program and increase the cost saving beyond what we have measured based on actual experience.
Further examination of the Austrian data demonstrates that 49% of amniocentesis testing was unnecessary and was not based on a primary infection during pregnancy [48]. Such testing is expensive and brings unnecessary risk to the unborn and anxiety to parents. Ongoing education for gynecologists should help to eliminate this unnecessary cost and risk.
In sum, while the Austrian prenatal screening protocol to minimize the effects of primary infections of T. gondii during pregnancy is cost-saving, additional cost saving could be achieved by enhancing the education of obstetric staff. There is a need to increase the number of susceptible women who receive the recommended number of screening blood tests at the recommended intervals. There is also a need to use amniocentesis only when indicated by proven primary infection during pregnancy.
Successful screening and treatment programs, such as Austria’s, face two challenges, both of which derive from their success. As with other public health programs, the European prenatal screening programs and education campaigns confront the paradox of success. People do not see or hear about infants affected by CT as they did in the past when infant deaths or profound brain injuries and visual impairment of varying degrees were more common, due to high rates of CT. Prevention programs only seem expensive in the absence of disease. In the face of budget pressure, the absence of infants with injuries of CT can be misunderstood to mean there is no longer a risk. On the contrary, it has taken two decades of successful prenatal screening and treatment to make the risk invisible. Moreover, the success of education programs in reducing prevalence in the population, while it may protect women by making them more aware of the risk of eating undercooked meat and unwashed fruits and vegetables, actually creates a larger population of women still at risk of infection, and particularly so since even the water supply is a source of infection in some regions.
The second challenge to the prenatal screening programs comes from the methodological debate over the validity of observational studies versus randomized controlled trials as the evidence base for interventions. Numerous authors have suggested that the question of efficacy of prenatal screening and treatment can only be adequately answered with randomized controlled trials (RCTs) [13,39,66,67]. RCTs, however, pose an insurmountable ethical problem in countries where prenatal screening has been associated with significant improvement in outcomes for infants whose mothers were treated prenatally. An RCT requires equipoise, which is lacking in countries with successful screening programs (Austria, France, and Slovenia, for example) and in countries with similar epidemiology and access to care. Without equipoise, it is doubtful that one could construct an ethical trial that would require random assignment of some pregnant women to denial of a treatment with demonstrated efficacy [6,7]. Blinding could be incompatible with informed consent. It is also unlikely that such trials would have sufficient power because, with informed consent, few parents would be likely to choose not to medicate. The resulting selection bias would also invalidate the results of the trial. This ethical question is not unique to prenatal screening programs for CT. Interventions to reduce smoking, for example, were implemented based on observational data. Any RCT assigning participants to smoking would not have passed ethical review. It has been impossible to construct valid RCTs for treating sexually transmitted diseases to reduce HIV incidence because observational studies and an earlier trial demonstrated that such treatment is beneficial [68]. Similarly, any other effective treatments for cofactor infections cannot ethically be withheld from controls [68,69]. Observational and historical data from Austria, France, and Slovenia, and perhaps even comparative data from the United States, have eliminated the equipoise necessary for an ethical RCT of prenatal screening and treatment for primary infection of T. gondii. The European screening programs for CT have had noteworthy success, reducing the number of deaths and profound injuries in affected infants. That success itself in reducing preventable suffering and death commends the programs for continuation. The cost savings for national health care systems and society at large reinforce the argument for continuation.
CT is a health problem worldwide and it is not possible to eliminate all sources of infection for pregnant women, nor is a vaccine likely to be developed in the near future. There are, however, successful CT-prevention programs that are reducing clinical effects of CT and saving money for national health administrations and cost to society.
Our results understate the benefits of following the Austrian national program because the costs associated with injuries to infants whose mothers were not tested in accordance with the protocol are attributed to the screening scenario [31]. If those mothers had been tested on schedule, the injuries in the infants would most likely have been fewer and less severe, as was the case for the infants tested on schedule. Another source of overstatement of costs of actual Austrian practice is that we show the direct costs of ideal compliance with the protocol in obstetric visits, including the cost for all susceptible women having five tests, whereas, in practice, 97% of women had fewer than three tests. With fewer tests, that also means shorter treatment and lower treatment costs than the ideal. The average time between tests was 14 weeks, rather than the prescribed eight weeks. For two women whose infants were profoundly affected, the time between tests was 19 weeks [31]. If Austrian practice were in full compliance with the protocol, actual direct costs of screening and prenatal treatment would have been slightly higher, but the costs of treatment and accommodation of infants injured by CT and the loss of their productivity and that of their parents would have been substantially lower because fewer infants would have slipped through the screening process. The costs of screening and preventive treatment are negligible compared to the costs of treatment and accommodation for infants whose injuries are not prevented. Net benefits strongly favor screening.
As demonstrated by the Austrian national program, prenatal screening and treatment result in substantial cost saving, both from the conventional societal perspective and even from the narrow perspective of budgetary impact. Results in both cases are robust to wide variations in parameter values. Our data show the positive economic value of such a prevention measure. In summary, our findings of this economic analytic-decision model represent an important base for the discussion regarding implementation or continuation of prenatal screening for toxoplasma infection.
|
10.1371/journal.pgen.1006719 | Discovery and fine-mapping of adiposity loci using high density imputation of genome-wide association studies in individuals of African ancestry: African Ancestry Anthropometry Genetics Consortium | Genome-wide association studies (GWAS) have identified >300 loci associated with measures of adiposity including body mass index (BMI) and waist-to-hip ratio (adjusted for BMI, WHRadjBMI), but few have been identified through screening of the African ancestry genomes. We performed large scale meta-analyses and replications in up to 52,895 individuals for BMI and up to 23,095 individuals for WHRadjBMI from the African Ancestry Anthropometry Genetics Consortium (AAAGC) using 1000 Genomes phase 1 imputed GWAS to improve coverage of both common and low frequency variants in the low linkage disequilibrium African ancestry genomes. In the sex-combined analyses, we identified one novel locus (TCF7L2/HABP2) for WHRadjBMI and eight previously established loci at P < 5×10−8: seven for BMI, and one for WHRadjBMI in African ancestry individuals. An additional novel locus (SPRYD7/DLEU2) was identified for WHRadjBMI when combined with European GWAS. In the sex-stratified analyses, we identified three novel loci for BMI (INTS10/LPL and MLC1 in men, IRX4/IRX2 in women) and four for WHRadjBMI (SSX2IP, CASC8, PDE3B and ZDHHC1/HSD11B2 in women) in individuals of African ancestry or both African and European ancestry. For four of the novel variants, the minor allele frequency was low (<5%). In the trans-ethnic fine mapping of 47 BMI loci and 27 WHRadjBMI loci that were locus-wide significant (P < 0.05 adjusted for effective number of variants per locus) from the African ancestry sex-combined and sex-stratified analyses, 26 BMI loci and 17 WHRadjBMI loci contained ≤ 20 variants in the credible sets that jointly account for 99% posterior probability of driving the associations. The lead variants in 13 of these loci had a high probability of being causal. As compared to our previous HapMap imputed GWAS for BMI and WHRadjBMI including up to 71,412 and 27,350 African ancestry individuals, respectively, our results suggest that 1000 Genomes imputation showed modest improvement in identifying GWAS loci including low frequency variants. Trans-ethnic meta-analyses further improved fine mapping of putative causal variants in loci shared between the African and European ancestry populations.
| Genome-wide association studies (GWAS) have identified >300 genetic regions that influence body size and shape as measured by body mass index (BMI) and waist-to-hip ratio (WHR), respectively, but few have been identified in populations of African ancestry. We conducted large scale high coverage GWAS and replication of these traits in 52,895 and 23,095 individuals of African ancestry, respectively, followed by additional replication in European populations. We identified 10 genome-wide significant loci in all individuals, and an additional seven loci by analyzing men and women separately. We combined African and European ancestry GWAS and were able to narrow down 43 out of 74 African ancestry associated genetic regions to contain small number of putative causal variants. Our results highlight the improvement of applying high density genome coverage and combining multiple ancestries in the identification and refinement of location of genetic regions associated with adiposity traits.
| Obesity is a worldwide public health epidemic, with current US estimates of 37.9% obese and 7.7% morbidly obese adults [1]. Disparities in obesity rates, as well as rates of comorbidities and mortality, are evident across sex and racial/ethnic groups. Estimates from NHANES for 2013–2014 [1] show that obesity is more prevalent among African Americans (48.5%) than among non-Hispanic Whites (37.1%). In addition, obesity rates are higher among African American women (57.2%) than among African American men (38.2%). For comparison, the obesity rates in non-Hispanic Whites were 38.7% and 35.4%, respectively, for women and men.
Genome-wide association studies (GWAS) in diverse populations have identified > 300 loci associated with measures of adiposity including body mass index (BMI) and waist-to-hip ratio (adjusted for BMI, WHRadjBMI) in populations of European [2–9], African [10–12], and East Asian ancestry [13–15]. The majority of associated variants are common (MAF >5%) with small effect size, and jointly explain only a fraction of the phenotypic variances [7–8]. It has long been hypothesized that low frequency (MAF = 0.5–5%) and rare (MAF < 0.5%) variants may also contribute to variability in complex traits. However, these variants are not well captured in previous GWAS imputed to the HapMap reference panel [16–17]. The availability of higher density reference panels such as the 1000 Genomes Project (38M variants in 1092 individuals from phase 1) [18] has demonstrated improved imputation quality in European populations particularly for low frequency variants (aggregate R2 ~0.6 for MAF = 0.5%). However its impact is less clear for non-European populations [19]. We took this opportunity to use higher density imputation to reevaluate our previous GWAS for associations with anthropometric traits in individuals of African ancestry (AA) including African Americans and Africans.
The African Ancestry Anthropometry Genetics Consortium (AAAGC) previously identified seven genome-wide significant loci for BMI in up to 71,412 AA individuals, and an additional locus when combined with European ancestry (EA) data from the Genetic Investigation of ANthropometric Traits (GIANT) consortium using GWAS imputed to the HapMap Phase 2 reference panel [11]. No genome-wide significant loci were identified for WHRadjBMI in a GWAS of up to 27,350 AA individuals [12]. The low yield of discovery in AA studies is likely due to their relatively smaller sample sizes in comparison to EA studies [7–8], as well as their lower degree of linkage disequilibrium (LD) and thus poorer imputation quality. Here, we extended our previous work in the AAAGC to perform meta-analyses and replication of GWAS imputed to the 1000 Genomes reference panel in up to 52,895 AA individuals for BMI and up to 23,095 AA individuals for WHRadjBMI. We aimed to 1) discover novel variants, 2) fine map established loci, and 3) evaluate the coverage and contribution of low frequency variants in genetic associations in AA populations.
We conducted sex-combined and sex-stratified meta-analyses of GWAS summary statistics across 17 studies for BMI (N = 42,752) and 10 studies for WHRadjBMI (N = 20,384) in AA individuals in stage 1 discovery (S1 and S2 Tables, S1 Fig). Missing genotypes in individual studies were imputed to the 1000 Genomes Project cosmopolitan reference panel (Phase I Integrated Release Version 3, March 2012) [18] using MaCH/minimac [20] or SHAPEIT2/IMPUTEv2 [21–22] (S3 Table). Among all variants with MAF ≥ 0.1% in the largest Women’s Health Initiative (WHI) study, the average info score was 0.81 and 90.5% had imputation info score ≥ 0.3 (S4 Table). Genomic control corrections were applied to each study and after meta-analysis (λ = 1.07 for BMI, 1.01 for WHRadjBMI) (S3 Table, S2–S5 Figs). Association results for ~18M variants for BMI and ~21M variants for WHRadjBMI were subsequently interrogated further.
From stage 1 meta-analyses, variants associated with BMI (3,241 in all, 1,498 in men, 2,922 in women) and WHRadjBMI (2,496 in all, 1,408 in men, 2,827 in women) at P < 1×10−4 were carried forward for replication in AA and EA. Stage 2 included 10,143 AA (2,458 men and 7,685 women) for BMI and 2,711 AA (981 men and 1,730 women) for WHRadjBMI analyses. Stage 3 included 322,154 EA (152,893 men and 171,977 women) for BMI and 210,086 EA (104,079 men and 116,742 women) for WHRadjBMI analyses by imputing HapMap summary statistics results [7–8] to 1000 Genomes [23] (S1 Fig). Meta-analyses were performed to combine either sex-combined or sex-specific results from AA (stages 1+2, N ≤ 57,895 for BMI, ≤ 23,095 for WHRadjBMI in sex-combined analyses) and both AA and EA (stages 1+2+3, N ≤ 380,049 for BMI, ≤ 233,181 for WHRadjBMI in sex-combined analyses, S6–S9 Figs). Variants that reached genome-wide statistical significance (P < 5×10−8) were assessed for generalization of associations with BMI to children in two additional AA cohorts (N = 7,222).
Among the locus-wide significant established loci (44 for BMI given two of 45 lead regional variants were identical in two loci, and 21 for WHRadjBMI), and novel loci (three for BMI and six for WHRadjBMI) derived from the sex-combined and sex-stratified analyses, we performed fine mapping to localize putative causal variants. We constructed 99% credible sets containing variants that jointly account for 99% posterior probability of driving the association in a locus using the corresponding sex-combined or sex-stratified meta-analysis results from AA, EA and combined ancestry (S13 Table). A smaller number of variants in a credible set represent a higher resolution of fine mapping and we considered a credible set containing ≤ 20 variants as “tractable’ for follow up. The credible sets in the EA analyses were generally smaller than those in the AA given their larger sample size. As compared to the EA analyses, the number of tractable loci in the meta-analyses of AA and EA increased from 23 to 26 for BMI, and from 14 to 17 for WHRadjBMI.
Among these 43 tractable loci, the lead variants in the combined ancestry analyses had posterior probability ≥ 0.95 in six BMI loci (SEC16B, TLR4, STXBP6, NLRC3, FTO and MC4R) and seven WHRadjBMI loci (DCST2, PPARG, ADAMTS9, SNX10, KLF13, CMIP and PEMT) (S13 Table). Functional characterization of variants within the tractable credible sets revealed two loci contain nonsynonymous variants (ADCY3: rs11676272 S107P; SH2B1: rs7498665 T484A from the ATP2A1 locus), but they had low posterior probability to drive the respective associations (0.02 and 0.15, respectively) (S14 Table). On the other hand, the ADCY3 non-coding variants rs10182181 and rs6752378 had higher posterior probability (0.26–0.72) and are cis-eQTLs of ADCY3 and nearby genes. Several BMI loci including MTCH2, MAP2K5, NLRC3 and ATP2A1, and WHRadjBMI loci including TBX15-WARS2 and FAM13A, also contained cis-eQTL variants regulating nearby gene expression in subcutaneous and/or visceral adipose tissue (S14 Table).
In our large-scale meta-analyses of GWAS in up to 52,895 and 23,095 individuals of African ancestry for BMI and WHRadjBMI, respectively, we identified three novel (IRX4/IRX2, INTS10/LPL and MLC1) and seven established (SEC16B, TMEM18, GNPDA2, GALNT10, KLHL32, FTO and MC4R) BMI loci, as well as three novel (TCF7L2/HABP2, SSX2IP and PDE3B) and one established (ADAMTS9-AS2) WHRadjBMI loci in either sex-combined or sex-stratified analyses. By employing a recently developed method [23] to impute European GWAS summary statistics to the denser 1000 Genomes reference panel, followed by meta-analyses of both African and European ancestry individuals, we also identified three additional novel loci (SPRYD7/DLEU2, CASC8 and ZDHHC1/ HSD11B2) for WHRadjBMI. While all lead variants from established loci are common (MAF ≥ 5%), four of the nine lead variants from novel loci were low frequency (0.5% ≤ MAF < 5%). In addition, the lead variants from established loci including TMEM18 and ADAMTS9-AS2 were absent in HapMap. Overall, these results suggest the deeper genome coverage and/or improved imputation quality using 1000 Genomes, and complemented with additional sex-stratified analyses, facilitate the discovery of novel loci and identification of variants with stronger effects in established loci.
Among the novel sex-specific BMI loci (IRX4/IRX2, INTS10/LPL and MLC1), we did not identify any putative coding variants or regulatory regions underlying our association signals. Additionally, no associations have been reported with other metabolic traits in these novel BMI-associated signals. The first lead variant rs112778462 is located between the IRX4 and IRX2 genes which are members of the Iroquois homeobox gene family. IRX2 expression has been associated with deposition of fat in the subcutaneous abdominal adipose tissue but no sex difference was observed [29–30]. Irx4 knock out mice demonstrated cardiomyopathy with compensated increased Irx2 expression [31]. The second lead variant rs149352150 is located between the INTS10 and LPL genes. LPL encoded lipoprotein lipase is expressed in several tissues including adipose to mediate triglyceride hydrolysis and lipoprotein uptake. The serum LPL mass [32] and LPL activity and fat cell size of adipose tissues at gluteus and thigh [33] have been reported to be higher in women than in men. Previous GWAS demonstrated association of LPL with triglycerides and HDL cholesterol [34–35]. However, the reported lead variant rs12678919 was not in strong LD with rs149352150 (r2 = 0.005 in AFR and 0.006 in EUR). The third lead variant rs56330886 is located in a gene-rich region on chromosome 22q13 including MLC1. No biological candidates are identified in this region, therefore further analyses may be needed to explain the causative mechanism for this association signal.
Among the novel WHRadjBMI loci, rs116718588 is located between TCF7L2 and HABP2. TCF7L2 is the most significant type 2 diabetes locus in African Americans [36] and other populations [37]. However, rs116718588 was not in LD (r2 < 0.01 in AFR) with the reported type 2 diabetes associated variants. The second lead variant rs2472591 is located near SPRYD7, DLEU2 and TRIM13. This locus was associated with height in previous GWAS [6], but rs2472591 was not associated with height in our study (P > 0.05), suggesting different variants in this locus regulate different measures of body size. In addition, a surrogate of rs2472591, rs790943, is a cis-eQTL for TRIM13 [26] suggesting it may be the target gene. TRIM13 encodes an E3 ubiquitin-protein ligase involved in endoplasmic reticulum-associated degradation. The third lead variant rs140858719 is located between SSX2IP and LPAR3. LPAR3 is a plausible candidate as it encodes a receptor for lysophosphatidic acid (LPA). The autotaxin/LPA pathway mediates diverse biological actions including activation of preadipocyte proliferation [38], suppression of brown adipose differentiation [39], and promotion of systematic inflammation [40] which lead to increased risk for cardiometabolic diseases including obesity and insulin resistance [41–42]. LPA receptor 1 which is highly expressed in adipocytes and the gut primarily mediates these effects [43]. It has also been reported that LPA, via LPA1 and LPA3 receptors, mediated leukocytes recruitment and pro-inflammatory chemokine secretion during inflammation [44]. The fourth lead variant rs185693786 is located at intron 2 of PDE3B. The association signal spanned a large genomic region and harbors GWAS loci for adiponectin and height. Phosphodiesterase 3B is critical for mediating insulin/IGF-1 inhibition of cAMP signaling in adipocytes, liver, hypothalamus and pancreatic β cells [45]. Pde3b-knockout mice exhibited multiple alterations in regulation of lipolysis, lipogenesis, and insulin secretion, as well as signs of peripheral insulin resistance [46]. PDE3B expression has been reported to be higher in microvascular endothelial cell culture derived from skeletal muscles from male rats than in female rats [47]. The fifth lead variant rs6499129 is located intergenic between ZDHHC1 and HSD11B2. HSD11B2 encodes 11β-hydroxysteroid dehydrogenase type 2 which converts the active glucocorticoids to inactive metabolites. HSD2 activity was elevated in severe obesity and negatively associated with insulin sensitivity [48]. HSD2 expression is higher in omental than abdominal subcutaneous adipose tissue which may contribute to adipocyte hypertrophy and visceral obesity [49]. The sixth lead variant rs378854 is located at the long non-coding RNA CASC8. Associations of variants at CASC8 have been reported for various cancers [50–52] but no association was reported for cardiometabolic traits.
In our SNP and locus transferability analyses, a moderate number of EA-derived BMI and WHRadjBMI associated variants shared the same trait-raising alleles and displayed nominally significant associations in AA individuals, similar to previous findings [11–12]. While the BMI variants were similar in terms of their effect sizes and frequencies of trait-raising alleles between EA and AA populations, there were more discrepancies for WHRadjBMI variants. In addition, a substantial proportion of lead regional variants in AA were not in strong LD with EA lead variants, suggesting AA populations either have different association signals or the results may be spurious. Taken together, only <30% of EA loci were associated with BMI and WHRadjBMI in AA.
Trans-ethnic fine mapping improved resolution to refine putative causal variant(s) in some loci as compared to using EA studies alone. In the meta-analyses of AA and EA GWAS, four BMI loci (SEC16B, STXBP6, FTO and MC4R) and six WHRadjBMI loci (PPARG, ADAMTS9, SNX10, KLF13, CMIP and PEMT) only contained one variant in the 99% credible sets. Among 16 BMI and 3 WHRadjBMI loci that were examined in both the previous trans-ethnic meta-analysis studies using HapMap imputation [7–8] and the present study, the number of variants and the interval of credible sets were either the same or lower in the present study for 13 and 15 loci, respectively. The majority of credible variants are non-coding in those sets containing ≤ 20 variants. Several of them located at the MTCH2, MAP2K5, NLRC3, ATP2A1, TBX15-WARS2 and FAM13A loci are cis-eQTL variants regulating nearby gene expression in subcutaneous and/or visceral adipose tissue, suggesting the putative causal variants may have a regulatory role instead of directly altering protein structure and function. Despite the low posterior probabilities, the coding changes of credible variants at ADCY3 and SH2B1 suggest that they may be the causal genes in the respective loci modulating BMI. Further studies are warranted to delineate putative causal variants including functional annotation in trans-ethnic fine mapping efforts [53].
Our large-scale GWAS meta-analyses in African ancestry individuals imputed to the 1000 Genomes reference panel, complemented by imputation of European GWAS using summary statistics and additional sex-stratified analyses, boosts the study power and improves resolution, leading to the identification of nine novel loci and fine mapping 37 loci with tractable credible sets. We observed significant associations for variants with MAF ≥ 0.5%, but rare variants were unlikely to be detected due to limited power and poor imputation quality. Large scale sequencing studies are needed to evaluate the contribution of rare variants in modulating complex traits such as BMI and WHR. Given the substantially larger sample size in European than in African ancestry samples, the trans-ethnic fine mapping results are largely driven by variants showing strong associations in Europeans. Future trans-ethnic studies including additional non-European populations will further improve the fine mapping effort.
We used a three-stage design to evaluate genetic associations with BMI and WHRadjBMI in sex-combined and sex-stratified samples (S1 Fig). Stage 1 included GWAS meta-analyses in AA individuals and stage 2 included replication of top associations from stage 1. Stage 3 included meta-analysis of top associations from stages 1 and 2 AA studies and EA meta-analysis results. In the discovery stage 1 of AAAGC, 17 GWAS of up to 42,752 AA individuals (16,559 men and 26,193 women; 41,696 African Americans and 1,056 Africans) were included for the BMI analyses. A total of 10 GWAS of up to 20,384 AA individuals (4,783 men and 15,601 women; all African Americans) were included for the WHRadjBMI analyses. For variants with P < 1×10−4 in either the sex-combined or the sex-stratified meta-analyses, stage 2 replication was performed in additional AA individuals from AAAGC (N = 10,143 for BMI, N = 2,711 for WHRadjBMI), followed by meta-analysis with EA individuals from the GIANT consortium (322,154 for BMI, 210,086 for WHRadjBMI). Variants that reached genome-wide significance (P < 5×10−8) were assessed for associations with BMI in two cohorts of children (N = 7,222). All AA participants in these studies provided written informed consent for the research, and approval for the study was obtained from the ethics review boards at all participating institutions. Detailed descriptions of each participating study and measurement and collection of height, weight, waist and hip circumferences are provided in S1 Text, S1 and S2 Tables.
Genotyping in each study was performed with Illumina or Affymetrix genome-wide SNP arrays. Pre-phasing and imputation of missing genotypes in each study was performed using MaCH/ minimac [20] or SHAPEIT2/IMPUTEv2 [21–22] using the 1000 Genomes Project cosmopolitan reference panel (Phase I Integrated Release Version 3, March 2012) [18]. The details of the array, genotyping and imputation quality-control procedures and sample exclusions for each study are listed in S3 Table. In general, samples reflecting duplicates, low call rates, gender mismatch, or population outliers were excluded. Variants were excluded by the following criteria: call rate < 0.95, minor allele count (MAC) ≤ 6, Hardy-Weinberg Equilibrium (HWE) P < 1×10−4, imputation quality score < 0.3 for minimac or < 0.4 for IMPUTE, or absolute allele frequency difference > 0.3 compared with expected allele frequency (calculated as 1000 Genomes frequency of AFR × 0.8 + EUR × 0.2).
We evaluated the performance of 1000 Genomes imputation using the largest study, the Women’s Health Initiative (WHI) (N = 8,054). A total of 25.1 million variants with MAF ≥ 0.1% were imputed to the 1000 Genomes reference panel. Of these, 98.1% (8.8 million) common variants, 95.4% (9.3 million) low frequency variants (0.5% ≤ MAF < 5%), and 72.5% (4.6 million) rare variants (0.1% ≤ MAF < 0.5%) were well imputed with IMPUTE info scores ≥ 0.3 (S4 Table). Notably, these frequencies are slightly lower than those obtained by imputation using 1000 Genomes phase 1 interim reference panel in Europeans [54]. However, 72.6%, 95.5% and 99.5% of the common, low frequency and rare variants, respectively, from the 1000 Genomes reference panel were not present in the HapMap and therefore demonstrate deeper coverage of the genome, particularly for the low frequency and rare variants.
At all stages, genome-wide association analyses were performed by each of the participating studies. BMI was regressed on age, age squared, principal components and study site (if needed) to obtain residuals, separately by sex and case-control status, if needed. WHR was regressed on age, age squared, principal components, BMI and study site to obtain residuals, separately by sex and case-control status. Principal components were included to adjust for admixture proportion and population structure within each study. Residuals were inverse-normally transformed to obtain a standard normal distribution with mean of zero and standard deviation of one. For studies with unrelated subjects, each variant was tested assuming an additive genetic model with each trait by regressing the transformed residuals on the number of copies of the variant effect allele. The analyses were stratified by sex and case-control status (if needed). For studies that included related individuals, family based association tests were conducted that took into consideration the genetic relationships among the individuals. Sex stratified, case-control stratified and combined analyses were performed. Association results with extreme values (absolute beta coefficient or standard error ≥ 10), primarily due to small sample sizes and/or low minor allele count, were excluded for meta-analysis.
The latest summary statistics of sex-combined and sex-stratified meta-analyses of BMI and WHRadjBMI imputed to the HapMap reference panel in EA from the Genetic Investigation of ANthropometric Traits (GIANT) consortium were obtained from http://www.broadinstitute.org/collaboration/giant/index.php/GIANT_consortium_data_files [7–8]. These association summary statistics were used to impute z-scores of unobserved variants at the 1000 Genomes Project EUR reference panel (Phase I Integrated Release Version 3) using the ImpG program [23]. In brief, palindromic variants (AT/CG) and variants with allele mismatch with the reference were removed from the data. Using the ImpG-Summary method, the z-score of an unobserved variant was calculated as a linear combination of observed z-scores weighted by the variance-covariance matrix between variants induced by LD within a 1 Mb window from the reference haplotypes. The sample size of each unobserved variant was also interpolated from the sample sizes of observed variants using the same weighting method for z-score as Ni=∑t=1t=T|wi,t|∑ |wi,t|Nt. Here, t = 1,2,….,T, where T is the number of observed variants, wi,t is the element of the covariance matrix Σi,t for the unobserved variant i and the observed variant t within window. The performance of imputation was assessed by r2pred, with similar characteristics as the standard imputation accuracy metric r2hat [20]. Results of variants with r2pred ≥ 0.6 were used in subsequent analyses.
In the discovery stage 1, association results were combined across studies in sex-combined and sex-stratified samples using inverse-variance weighted fixed-effect meta-analysis implemented in the program METAL [55]. The study-specific λ values of association ranged from 0.97 to 1.05 for BMI, and 0.98 to 1.05 for WHRadjBMI (S3 Table). Genomic control correction [56] was applied to each study before meta-analysis, and to the overall results after meta-analysis (λ = 1.07 for BMI, 1.01 for WHRadjBMI). Variants with results generated from < 50% of the total sample size for each trait were excluded. After filtering, the numbers of variants reported in the meta-analyses were 17,972,087 for BMI, and 20,502,658 for WHRadjBMI.
Variants with P < 1×10−4 in stage 1 sex-combined or sex-stratified meta-analyses were carried forward for replication in additional AA individuals (stage 2) and EA individuals (stage 3). For each of the replication AA studies, trait transformation and association were performed as in stage 1 and results were meta-analyzed using the inverse-variance method in METAL. For the replication study in EA, HapMap imputed summary statistics of each trait from the GIANT consortium were used to impute z-scores of unobserved variants at the 1000 Genomes.
In stages 1 and 2, meta-analysis results of AA studies were combined using the inverse-variance weighted method. In all stages including both AA and EA studies, meta-analysis results expressed as signed z-scores were combined using the fixed effect sample size weighted method in METAL due to the lack of beta and standard error estimates from the ImpG program [23]. Evidence of heterogeneity of allelic effects between males and females, within and across stages were assessed by the I2 statistic in METAL. Genome-wide significance was declared at P < 5×10−8 from each of the sex-combined and sex-stratified meta-analysis including AA and/or combined AA and EA individuals. Difference in effects between men and women was assessed using Cochran’s Q test and nominal Phet < 0.05 declared as significant. A lead variant in a locus was defined as the most significant variant within a 1 Mb region. A novel locus was defined as a lead variant with distance > 500 kb from any established lead variants reported in previous studies. By convention, a locus was named by the closest gene(s) to the lead variant.
For the genome-wide significant loci identified in sex-combined and sex-stratified analyses in AA (stages 1+2), we used the program GCTA [57–58] to select the top independent associated variants from summary statistics of the meta-analyses. This method uses the LD correlations between variants estimated from a reference sample to perform an approximate conditional association analysis. We used 8,054 unrelated individuals of African ancestry from the WHI cohort with ~15.7M variants available as the reference sample for LD estimation. To select the top independent variants in the discovery and replication meta-analysis results, we first selected all variants that had P < 5×10−8 and conducted analysis conditioning on the selected variants to search for the top variants iteratively via a stepwise model to select the independent variants from this list. Then we proceeded to condition the rest of the variants that had P > 5×10−8 on the list of independent variants in the same fashion until no variant had conditional P that passed the significance level P < 5×10−8. Finally, all the selected variants were fitted jointly in the model for effect size estimation.
We also tested if the genome-wide significant variants identified from sex-combined GWAS in AA and the locus-wide significant variants identified from sex-combined and sex-specific locus transferability studies in AA were independent from nearby established loci identified from EA studies [7–8]. First, the published lead variants from EA studies were used to search for all surrogate variants that were in high LD (r2>0.8 in 1000 Genomes Project EUR population). Second, these variants were pruned to select only variants in low LD in AA (r2<0.3 in the 1000 Genomes Project AFR population) to avoid collinearity in conditional analysis. Third, association analysis was conducted on the AA significant variants conditioned on the selected EA lead and surrogate variants, using the program GCTA and estimated LD correlation from the WHI cohort. For genome-wide significant loci, an AA derived association signal is considered as independent from the established EA signals when the difference in–logP <3 and difference in effect size < 1 standard error after conditional analysis. For locus-wide significant loci, given the lower level of significance, independence is only considered as difference in effect size < 1 standard error after conditional analysis.
We investigated the transferability of EA BMI and WHR associated variants and loci in AA individuals from stage 1 sex-combined and sex-stratified meta-analyses. First, we tested for replication of lead variants previously reported to be associated with BMI (176 variants from 170 loci) and WHRadjBMI (84 variants from 65 loci) at genome-wide significance in sex-combined and sex-stratified analyses from the GIANT consortium studies [7–9]. We defined SNP transferability as an EA lead variant sharing the same trait-raising allele at nominal P < 0.05 in AA individuals. To account for differences in local LD structure across populations, we also interrogated the flanking 0.1cM regions of the lead variants to search for the best variants with the smallest association P in AA individuals. Locus-wide significance was declared as Plocus < 0.05 by Bonferroni correction for the effective number of tests within a locus, estimated using the Li and Ji approach [59].
We compared the credible set intervals of established loci that showed locus-wide significance (Plocus < 0.05) in the sex-combined or sex-specific analyses from this study in summary statistics datasets including the 1000 Genomes imputed results from GIANT, AAAGC and meta-analysis of GIANT and AAAGC. In each dataset, a candidate region is defined as the flanking 0.1cM region of the lead variant reported by the GIANT consortium. Under the assumption of one causal variant in a region of M variants, the posterior probability of a variant j with association statistics Z driving the association, P(Cj|Z), was calculated using the formula P(Cj|Z)=exp(12zj2)∑j=1Mexp(12zj2). A 99% credible set was constructed by ranking all variants by their posterior probability, followed by adding variants until the credible set has a cumulative posterior probability > 0.99 [53].
Given our sample sizes in the discovery and replication stages in our African ancestry populations, we have >80% power to detect variants explaining 0.08% variance for BMI that corresponds to effect sizes of 0.09 and 0.20 SD units for MAF of 0.05 and 0.01, respectively. For WHRadjBMI, we have >80% power to detect variants explaining 0.18% variance that corresponds to effect sizes of 0.14 and 0.30 SD units for MAF of 0.05 and 0.01, respectively.
|
10.1371/journal.pgen.1002735 | Allelic Variation and Differential Expression of the mSIN3A Histone Deacetylase Complex Gene Arid4b Promote Mammary Tumor Growth and Metastasis | Accumulating evidence suggests that breast cancer metastatic progression is modified by germline polymorphism, although specific modifier genes have remained largely undefined. In the current study, we employ the MMTV-PyMT transgenic mouse model and the AKXD panel of recombinant inbred mice to identify AT–rich interactive domain 4B (Arid4b; NM_194262) as a breast cancer progression modifier gene. Ectopic expression of Arid4b promoted primary tumor growth in vivo as well as increased migration and invasion in vitro, and the phenotype was associated with polymorphisms identified between the AKR/J and DBA/2J alleles as predicted by our genetic analyses. Stable shRNA–mediated knockdown of Arid4b caused a significant reduction in pulmonary metastases, validating a role for Arid4b as a metastasis modifier gene. ARID4B physically interacts with the breast cancer metastasis suppressor BRMS1, and we detected differential binding of the Arid4b alleles to histone deacetylase complex members mSIN3A and mSDS3, suggesting that the mechanism of Arid4b action likely involves interactions with chromatin modifying complexes. Downregulation of the conserved Tpx2 gene network, which is comprised of many factors regulating cell cycle and mitotic spindle biology, was observed concomitant with loss of metastatic efficiency in Arid4b knockdown cells. Consistent with our genetic analysis and in vivo experiments in our mouse model system, ARID4B expression was also an independent predictor of distant metastasis-free survival in breast cancer patients with ER+ tumors. These studies support a causative role of ARID4B in metastatic progression of breast cancer.
| A person's individual genetic background influences not only the likelihood of developing breast cancer, but also the likelihood of that cancer becoming metastatic. The identification of metastasis susceptibility genes using human samples is rendered impractical by the high degree of genetic diversity among people. Our laboratory's strategy is to cross genetically defined inbred mouse strains to recapitulate a degree of genetic diversity that is more readily studied. By breeding these panels of inbred mouse crosses to a mouse model of breast cancer, we can identify regions of the genome that correlate with observed phenotypic variation including metastatic density and then identify individual candidate genes. This manuscript describes the identification of Arid4b as a candidate gene of interest and the experiments we performed to validate its role in metastasis. High expression of Arid4b enhances cell migration and invasion and, conversely, knockdown of Arid4b inhibits metastasis of breast tumor cells to the lungs. The mouse gene and human ARID4B are highly conserved, and among women with ER+ tumors ARID4B expression level is predictive of which patients will progress to develop metastatic disease.
| Breast cancer remains the most commonly diagnosed malignancy among women in the United States [1]. Because the vast majority of breast cancer related mortality is attributable to disseminated metastatic disease, a clear need exists to identify factors that modulate breast cancer metastatic progression. In addition to acquired somatic mutations, there is accumulating evidence that the genetic background on which a tumor arises can influence disease progression [2]. Identifying and characterizing metastasis susceptibility genes would provide additional insights into the mechanisms associated with tumor dissemination and growth, leading not only to better understanding of this complex process but also ultimately to new targets and strategies for clinical intervention.
Due to the complex interactions between inherited factors and somatic mutations in metastatic progression, as well as the genetic complexity of human populations, identification of inherited susceptibility genes directly in human populations is difficult. To circumvent this our laboratory has chosen to apply a systems genetics approach on a mouse model of metastatic luminal breast cancer, the FVB/N-TgN(MMTV-PyMT)634Mul (MMTV-PyMT) transgenic model. The MMTV-PyMT transgenic mouse model, which expresses the polyoma virus middle T antigen under the control of the mouse mammary tumor virus promoter, rapidly develops tumors in approximately 100% of female mammary glands and >85% of these animals develop pulmonary metastases by 14 weeks of age. When the MMTV-PyMT model is bred onto a variety of different mouse strains, the F1 progeny display broad and strain-dependent heterogeneity in primary tumor latency, primary tumor growth rate and lung metastatic density [2].
Two strains, the highly metastatic AKR/J and poorly metastatic DBA/2J, were found to have a 20-fold difference in their metastatic capacity but no significant difference in any other measured tumor phenotype. These strains were also the progenitor strains for the AKXD recombinant inbred panel of mice, which consists of more than 20 substrains that are composites of the original parental strains AKR/J and DBA/2J. The MMTV-PyMT model was therefore bred to 18 different AKXD strains, the F1 mice were phenotyped with respect to primary tumor latency and burden and lung metastatic density, and the phenotypes were compared to haplotype maps of the AKXD strains to determine quantitative trait loci (QTLs) associated with mammary tumor progression [3]. Subsequently, RNA was also harvested from F1 tumors and gene expression analysis was performed to define individual genes whose expression correlated with progression [4].
In this study we have utilized these resources to identify Arid4b as a novel candidate metastasis susceptibility gene. Although the precise molecular functions of ARID4B are unknown, it has been shown to associate with the SIN3A histone deacetylase (HDAC) complex [5]. As predicted by the genetic linkage and gene expression data, higher expression of Arid4b is associated with more rapid tumor growth in animal models, as well as increased tumor cell motility and invasion. These effects are associated with differential binding of the AKR and DBA alleles of ARID4B to HDAC complex members mSIN3A and mSDS3. ARID4B was also found to bind the mSIN3A-associated breast cancer metastasis suppressor protein BRMS1. Stable shRNA-mediated knockdown of Arid4b significantly inhibited the pulmonary metastatic efficiency of orthotopic mammary tumors without inhibiting primary tumor growth. Consistent with impaired metastasis in the Arid4b knockdown lines was decreased expression of a recently described metastasis-predictive gene network [6]. High expression of ARID4B was associated with an approximately 2-fold increased risk of metastatic progression in human breast cancer patients who were lymph node negative at diagnosis. Taken together these results demonstrate a causal role for Arid4b in tumor growth and metastatic progression and suggest that mechanisms of action involve modification of epigenetic state via the mSIN3A complex and regulation of the conserved Tpx2 gene network.
Previously a cross between the highly metastatic PyMT model and the AKXD recombinant inbred (RI) panel was performed to map QTLs associated with inherited predisposition to developing pulmonary metastasis [3]. In addition to metastasis susceptibility loci on chromosomes 6 and 19, linkage analysis revealed a potential peak on proximal chromosome 13 (Figure S1). In a subsequent study, gene expression analysis was also carried out on these samples to examine the effect of varying metastatic genotypes on tumor transcriptional patterns [7]. To discover potential candidate genes that may affect metastatic predisposition, correlation analysis was performed using GeneNetwork [8] to identify genes whose differential expression was highly associated with metastasis. Upon integrating the data from these two studies we found that of the top ten genes most significantly associated with metastasis in our expression correlation analysis, two also mapped to potential QTLs: Ttc9c and Arid4b. The potential role of Ttc9c was investigated and no significant differences were detected with respect to orthotopic tumor growth or metastasis of 6DT1 mouse mammary carcinoma cells stably expressing Ttc9c compared to vector control cells (data not shown). Similarly, we detected no significant effects on tumor growth or metastasis when Ttc9c BAC transgenic mice were bred to the MMTV-PyMT model (data not shown). The most likely explanation for why Ttc9c did not pass our validation experiments is that its initial identification in our screens was a false positive owing to its close physical proximity on chromosome 19 to the metastasis modifier gene Sipa1 [9]. Our current studies have therefore focused on Arid4b, which maps within the chromosome 13 locus and whose mRNA expression was positively associated with metastatic disease and tumor growth (Figure 1), suggesting a possible causative role as a progression modifier.
To validate the potential differences in Arid4b expression between strains, microarray data from AKR and DBA normal tissues were examined [4]. Consistent with the AKXD RI results, Arid4b expression was 2.3-fold higher in thymus (p = 9.32×10−5, FDR = 0.0004) and 2.5-fold higher in bone marrow (p = 1.28×10−5, FDR = 0.0005) of DBA mice compared to AKR, suggesting that constitutional polymorphisms can influence Arid4b expression levels in normal tissues. Sequence analysis was also performed to both validate SNPs in the public database as well as identify potential new variants between the AKR and DBA alleles of Arid4b. Complete exon sequencing revealed that the DBA allele matched the consensus C57BL/6 sequence. Analysis of the AKR allele revealed numerous silent SNPs as well as polymorphisms encoding eleven amino acid substitutions, as shown in Figure 2. Interestingly, eight of these eleven polymorphisms are located in exon 22 and their encoded substitutions are densely clustered towards the C-terminal end between amino acids 1171 and 1198. These results are consistent with the possibility that inherited variation of Arid4b may contribute to tumor progression.
Analysis of the data revealed that increased Arid4b expression and increased metastatic susceptibility were associated with the DBA rather than the AKR genotype at the chromosome 13 QTL. This result suggests that the DBA allele at this locus promoted metastatic progression relative to the AKR allele, and a series of in vitro and in vivo assays were performed to test this hypothesis. V5-tagged AKR and DBA alleles of Arid4b were ectopically expressed in the mouse mammary carcinoma cell line Met-1, which was originally derived from tumors arising in the MMTV-PyMT transgenic model [10]. Because the Met-1 line was derived from an FVB strain background, we also sequenced the FVB allele of Arid4b and found it to be identical to the DBA and C57BL/6 alleles. Cell lines were then identified that expressed the epitope tagged constructs at levels that were only two to three-fold higher than endogenous levels as measured by QRT-PCR (Figure S2A), consistent with the approximately two-fold range of Arid4b mRNA between high and low metastatic AKXD strains. Furthermore, the ectopically expressed AKR and DBA alleles were detected at approximately equal levels in our stable lines as assessed by western blots (Figure S2B).
Orthotopic implantation assays were then performed to examine the role of Arid4b expression in vivo (Figure 3A). By four weeks post-implantation, cells expressing the DBA allele formed tumors with a 2.6-fold larger mass compared to control cells (741 mg versus 284 mg; p = 6.08×10∧−7). The AKR allele expressing cells formed tumors with a median mass of 480 mg, which was significantly larger than control tumors (p = 0.010) but significantly smaller than the DBA cohort (p = 7.73×10∧−3), consistent with our previous genetic analysis and our in vitro studies.
In vitro assays were performed to address the potential affect of the Arid4b polymorphisms on tumor cell behavior. In vitro growth assays demonstrated no significant difference in proliferation between cells expressing the DBA or AKR alleles or control cells (data not shown). In contrast, ectopic expression of either allele significantly increased the abilities of Met-1 cells to migrate through a porous membrane and to invade through Matrigel, compared to control cells expressing lacZ (Figure 3B). Notably, Met-1 cells stably expressing the DBA allele were significantly more migratory and invasive than those expressing the AKR allele. Since both cell lines express the epitope-tagged construct at approximately the same level, these results suggest a potential functional consequence for the amino acid substitutions present between the two variants in addition to the effects associated with differential expression.
Because Met-1 cells are poorly metastatic in our laboratory, and because we were unable to stably overexpress Arid4b in several more aggressive mouse breast cancer cell lines, we adopted a knockdown strategy to examine the role of Arid4b in lung metastasis in vivo. To this end, the highly metastatic 6DT1 cell line [11] was transduced with five lentiviral shRNAs targeting Arid4b, or a scrambled control, and knockdown of ARID4B protein was evaluated using western blots (Figure 4A) and densitometry (Figure 4B) to select stable shRNA lines for in vivo studies. No significant knockdown was observed using the scrambled control shRNA. Cells stably transduced with Arid4b shRNAs designated H3 and H4 expressed 81% and 85% less ARID4B protein, respectively, compared to controls, and were therefore selected for further in vivo study.
Following orthotopic implantation of 10∧5 cells into the mammary fat pad we observed only slight differences in median primary tumor mass between the scrambled control, H3, and H4 cohorts, and these data did not achieve statistical significance (p = .070, Kruskal-Wallis; Figure 4C). In contrast, we observed a 2-fold decrease in the median number of macroscopic lung metastases in the H3 cohort (10 vs. 22; p = .013) and a 7-fold decrease in the H4 cohort (3 vs. 22; p = 9.72×10∧−5) compared to controls (Figure 4D). Differences in lung metastasis between the two Arid4b knockdown cohorts were not statistically significant following post hoc testing (p = .066, Conover-Inman). These data demonstrate that ARID4B protein levels are a critical determinant of pulmonary metastatic efficiency in this model system.
Previous studies demonstrated that ARID4B is a member of the mSIN3A HDAC complex and that binding to mSIN3A involves the C-terminal domain of ARID4B [5], where the majority of the amino acid substitutions were found between the AKR and DBA variants (Figure 2). Co-IP analysis was therefore performed to examine a potential effect of the observed amino acid substitutions on ARID4B-mSIN3A binding. For these experiments V5-tagged ARID4B was transiently transfected into HEK293 cells and immunoprecipitated using an anti-V5 antibody. Binding to endogenous mSIN3A and another component of the mSIN3A complex, mSDS3 [12], was evaluated by western blots (Figure 5). Input controls for ARID4B, mSIN3A, and mSDS3 were approximately equal as were the amounts of the two Arid4b variants immunoprecipitated; however, a marked decrease in binding to mSIN3A was observed along with diminished mSDS3 association for the DBA variant (Figure 5A). Densitometry analysis revealed that binding of the DBA variant was reduced by 51% and 37% for mSIN3A and mSDS3, respectively, compared to AKR (Figure 5B). These results demonstrate a functional consequence of Arid4b polymorphisms and provide insight into one potential molecular mechanism whereby Arid4b may modulate breast cancer progression.
Breast cancer metastasis suppressor 1 (BRMS1) belongs to the same family of proteins as mSDS3 and is known to associate with the mSIN3A complex as well as ARID4A [13]. Because ARID4B is also known to bind mSIN3A, mSDS3, and ARID4A [14], we postulated that ARID4B might physically bind BRMS1. Proteomics screens to identify BRMS1 interacting proteins also support this association: in a yeast two-hybrid screen for proteins binding full-length BRMS1, ARID4B was the number one hit identified, and ARID4A and mSDS3 were also detected (unpublished data). In a separate screen, mass spectrometry was performed to identify mSIN3A binding proteins in MCF10A human breast epithelial cells. Peptides representing endogenous ARID4B and BRMS1 were detected, providing further evidence for this interaction and demonstrating that it is not simply an artifact of supraphysiologic expression in transfected cells (Douglas Hurst; personal communication). To validate this interaction we performed co-IPs using lysates from 293 cells transiently transfected with the FLAG-tagged AKR or DBA variants of ARID4B along with either HA- or myc-tagged BRMS1. HA-BRMS1 was readily detected following pull-down of ARID4B using an anti-FLAG antibody (Figure 6A). Likewise, ARID4B was efficiently co-precipitated with myc-BRMS1 (Figure 6B). Unlike the associations with mSIN3A and mSDS3 however, the AKR and DBA variants of ARID4B did not exhibit differential binding to BRMS1. One possible explanation for this observation is that BRMS1 binds to a different region of ARID4B than the polymorphic C-terminal domain that mediates binding to mSIN3A.
Because little is known about the specific cellular processes regulated by Arid4b that might influence the metastatic phenotype, we performed expression microarray analysis on the 6DT1 cell lines stably expressing Arid4b shRNAs to identify genes that are differentially expressed as a function of Arid4b levels. Based on the western blot densitometry shown in Figure 4A–4B cell lines expressing hairpins H3 and H4 were chosen to represent the Arid4b knockdown cohort, and the control cohort consisted of untreated 6DT1 cells and lines expressing the scrambled control shRNA or hairpin H5. We detected 2,048 unique genes whose expression was significantly different (p<0.05, ANOVA) between the two groups and those with the greatest fold change are summarized in Table 1. While the most highly upregulated genes function in pathways with diverse biological roles, it was noted that among the most downregulated genes were multiple factors associated with centromeres (Cenpi, Cenpq), microtubule and spindle dynamics (Kif2c, Kif4a, Sass6), and cell cycle regulation (Ccne, Cdc25c). Consistent with this observation were the results of pathway analysis conducted to identify biological functions impacted as a consequence of Arid4b knockdown (Table 2). The most differentially regulated processes based on gene ontology were checkpoint control and DNA repair, and processes related to centrosome, centriole, and chromosome dynamics.
In examining the microarray data we noticed a striking overlap between genes downregulated in the Arid4b knockdown lines and components of the TPX2 gene network. This transcriptional network was recently identified based on expression profiling of three mouse data sets and two human breast cancer data sets [6]. The TPX2 network is tumor cell-autonomous and conserved across species, its activation is predictive of reduced distant metastasis-free survival (DMFS) in ER-positive patients, and the nine common hub genes in the TPX2 signature (TPX2, BUB1, UBE2C, CDC20, CCNB2, KIF2C, BUB1B, CEP55, CENPA) that were conserved across all five data sets consist primarily of genes involved in microtubule and mitotic spindle function. To determine how Arid4b levels influence the activation state of the TPX2 network, the fold changes of the 311 TPX2 network genes were examined in the Arid4b knockdown lines. Compared to control cell lines, 119 network genes were significantly downregulated (p<0.05) including Tpx2 itself and the other eight common hub genes, versus only 5 network genes upregulated (Figure 7; high resolution available as Figure S3). The downregulation of this gene network concomitant with the inhibition of metastasis observed in the Arid4b knockdown lines provides further support for the role of the TPX2 network in metastatic susceptibility and suggests that a significant portion of this network may be regulated by Arid4b.
Because Arid4b was identified as a candidate gene in part based on differential expression between high and low metastatic strains of mice in the AKXD panel, and because Arid4b expression levels were associated with tumor growth and metastasis in mice as well as the activity of the metastasis-associated TPX2 network, we tested whether ARID4B expression alone correlated with human patient outcomes. A search of publically available breast cancer microarray data sets using Oncomine (Compendia Bioscience, Ann Arbor, MI) revealed that ARID4B expression was 2.3-fold higher in 40 ductal breast carcinoma samples compared to 7 normal breast tissue samples in the Richardson study [15], confirming that high ARID4B expression is clinically associated with breast cancer (Figure S4). Analysis of a pooled breast cancer dataset using GOBO (http://co.bmc.lu.se/gobo/) [16] showed that among the subgroup of patients with ER-positive tumors, the cohort with high expression of ARID4B had significantly reduced DMFS compared to the low or median ARID4B cohorts (Figure 8). Because this association was significant among patients with ER-positive tumors who were lymph node negative at the time of diagnosis (Figure 8A), this finding indicated that ARID4B expression level is predictive of patient progression to metastatic disease. As determined by multivariate analysis (Figure 8B), the hazard ratio compared to the high ARID4B tercile was 0.54 for middle ARID4B (95% C.I. = 0.33–0.89; p = .015) and 0.42 for the low ARID4B tercile (95% C.I. = 0.26–0.70; p = 7.51×10∧−4), indicating that patients with tumors expressing high levels of ARID4B are approximately twice as likely to develop metastatic disease. The association of ARID4B with reduced DMFS was also highly significant among ER-positive patients not receiving adjuvant therapy (Figure 8C–8D), indicating that ARID4B expression level plays a significant role in the natural metastatic progression of ER-positive breast cancer in human patients and its relevance is not confined solely to our mouse model systems.
Arid4b was identified as a candidate gene of interest through linkage and expression correlation analyses, and the in vitro and in vivo data presented here provide the first direct evidence of a causal role of Arid4b in mammary tumor progression and metastasis. The initial QTL analysis revealed association with Arid4b on proximal chromosome 13, and Arid4b was among the most highly correlated genes with the most significant p values in the subsequent eQTL analysis in the AKXD recombinant inbred panel. It was noted that although AKR/J is the more highly metastatic of the two parental strains, progression was associated with the DBA/2J allele, suggesting that the metastasis promoting influence of Arid4b is likely masked by other suppressive factors in a pure DBA/2J background. Although the AKXD recombinant inbred panel lacks the power to detect these epistatic interactions, ongoing experiments using the latest generation of recombinant inbred mice including the Collaborative Cross [17], [18] will enable higher resolution QTL mapping and more robust systems genetics analyses going forward.
Mouse Arid4b encodes a protein of 1314 amino acids that shares 89% identity and 95% similarity to the 1312 amino acid human protein. Alternate nomenclature includes breast cancer-associated antigen 1 (BRCAA1), retinoblastoma-binding protein-1-like protein-1 (RBP1L1), and mSIN3A-associated protein of 180 kDa (SAP180). Indeed, there are multiple lines of evidence implicating Arid4b in breast cancer. A ten amino acid peptide was found to represent an antigen epitope expressed in 65% of breast cancer specimens and was significantly upregulated in the sera of breast cancer patients compared to healthy donors [19]. ARID4B was also found to associate with the mSIN3A HDAC complex [5], which is in turn known to be bound by the breast cancer associated tumor suppressor ING1 [20], [21], the well-characterized breast cancer metastasis suppressor BRMS1 [13], and the ARID family homolog ARID4A/RBP1 [22], which has also been identified as a breast cancer associated antigen [23].
Ectopic expression of Arid4b at a physiologically relevant two- to three-fold increased level resulted in a 3-fold increase in orthotopic tumor mass relative to controls for Met-1 cells expressing the DBA allele, while the AKR allele induced 1.9-fold larger tumors versus control cells. To our knowledge, this is the first direct evidence that Arid4b upregulation promotes tumor growth. Although Met-1 orthotopic tumors do not readily metastasize in our experience, transwell assays in vitro demonstrated that upregulation of either allele of Arid4b increased tumor cell migration and invasion, consistent with a role of Arid4b in metastatic progression, and cells expressing the DBA allele were significantly more migratory and invasive than cells expressing the AKR allele. While stable upregulation of Arid4b did not induce Met-1 cells to metastasize with any greater frequency, stable knockdown of Arid4b in the highly metastatic 6DT1 cell line did cause a dramatic reduction in pulmonary metastases, raising the possibility that ARID4B may represent a novel therapeutic target. Taken together, the results of the orthotopic implantation and transwell assays are broadly consistent with our genetic linkage and expression correlation analyses that showed an association of the DBA haplotype on chromosome 13 with metastatic progression in the MMTV-PyMT×AKXD mice, and validate a functional role of Arid4b polymorphism in modulating the breast cancer phenotype.
While the molecular mechanisms of Arid4b are incompletely understood, an examination of its sequence and conserved domains provides further insight into its potential functions. Arid4b contains a nuclear localization signal (NLS) towards the C-terminus as well as conserved Tudor, RBB1NT, ARID/BRIGHT, and Chromo domains in the N-terminal half of the protein. The ARID domain mediates binding to DNA, although the affinity for AT-rich sequences varies among members of the Arid superfamily [24], [25]. The RBB1NT domain is present in many Rb binding proteins including ARID4A, although it is noteworthy that unlike ARID4A, ARID4B does not contain the LCXCE motif necessary for RB binding [26], and no interaction was observed when we attempted to co-IP ARID4B with RB (data not shown); therefore, the function of the RBB1NT domain of ARID4B remains uncertain. Tudor domains are present in many RNA binding proteins [27] and also bind methylated lysine residues on histone tails [28]. Chromo domain-containing proteins have also been shown to bind methylated lysines and mediate the recruitment of chromatin modifying complexes [29]. Because mSIN3A itself lacks intrinsic DNA binding capability, targeting of mSIN3A-associated HDAC activity depends on interactions with other transcription factors including Mad1 and KLF repressors among others [30]. The presence of putative DNA and histone binding domains in the N-terminal half of ARID4B suggest that its influence on mammary tumor progression involves directing the HDAC activity of mSIN3A complexes to chromatin. This is supported by our observations that the high and low metastatic alleles of ARID4B have a dense cluster of amino acid polymorphisms in the C-terminal domain and bind with different affinities to mSIN3A and mSDS3, though the biochemical significance of this observation remains to be determined. Diminished expression levels or binding affinity of ARID4B may allow mSIN3A to be bound by other proteins with different DNA sequence specificity, perhaps not resulting in a global change in the abundance of any one particular histone mark but rather altering the expression of different subsets of genes.
It is noteworthy that the pro-metastatic ARID4B and the metastasis suppressive BRMS1 bind each other and also to the mSIN3A complex in vitro. This observation reinforces the significance of the mSIN3A complex in metastatic progression, and it is tempting to speculate that an HDAC complex may be caught in a molecular tug-of-war between these two metastasis modifier genes. However, the mSIN3A complex is modular in nature and interacts with a great variety of transcriptional regulators [30], [31]. Many different complexes exist, and their precise composition and function within the context of breast cancer are not well understood. Further studies will be necessary to define a role, if any, for the ARID4B-BRMS1 interaction in human disease.
While ARID4B expression was not a significant predictor of DMFS across all patients in a meta-analysis of an 1,881 sample data set, statistical significance emerged when patients were stratified based on ER status. The observation that ARID4B is predictive of metastatic progression only in ER+ patients is consistent with the identification of Arid4b as a candidate gene in the context of the MMTV-PyMT mouse model system, in which tumors arise from a predominantly ER+ luminal epithelial cell population [32]. Loss of ER and PR is detected during progression to late carcinomas, however in a systematic analysis of gene expression profiles these late PyMT tumors clustered most closely with human luminal tumors [33], which are ER+. Also consistent with ARID4B promoting metastatic progression of ER+ tumors is our observation that Arid4b knockdown caused a significant downregulation of the core components of the Tpx2 gene network. The TPX2 signature was tumor cell autonomous and predictive of DMFS only in those patients who were ER+ at diagnosis, and was distinct from a CD53 network that was associated with ER-negative stromal components [6]. Polymorphisms in several other tumor cell autonomous metastasis susceptibility genes identified in our laboratory including Sipa1, Rrp1b, and Brd4 are prognostic only in ER+ patients [9], [34]–[37], and stable expression of Brd4 can also differentially regulate the Tpx2 network [6]. The association of multiple metastasis susceptibility genes with a transcriptional network comprising many cell cycle and mitotic spindle checkpoint regulatory genes highlights the possibility that these cellular functions are critical determinants of metastatic efficiency. Further experiments are underway in our laboratory to determine whether upregulation of the TPX2 network is causative in promoting metastasis.
Using genomic DNA from AKR/J and DBA/2J mice as templates, PCR was performed to amplify the protein coding region, exons 2 through 24 (Table S1). PCR Products were subjected to agarose gel electrophoresis, bands isolated using the QIAquick Gel Extraction kit (Qiagen) according to manufacturer's recommendations, and used as templates for sequencing. All sequencing runs were performed by the DNA Sequencing and Gene Expression Core, NCI, Bethesda, MD. Genomic sequences for the AKR and DBA alleles were aligned using pairwise BLAST [38] and non-synonymous polymorphisms verified by manual comparison of chromatograms using Chromas software (Technelysium).
V5-tagged AKR and DBA alleles of Arid4b were generated using long range PCR with forward primer 5′-AACAAAGGTGCAGGTGAAGC-3′ and reverse primer 5′-CCTGCACTCAACTGACATTCCATTC-3′ to amplify Arid4b, and PCR products were cloned into pcDNA3.1/V5-His-TOPO (Invitrogen). FLAG-tagged Arid4b vectors were constructed by the Protein Expression Laboratory, SAIC-Frederick, Inc. using Gateway technology (Invitrogen). Briefly, the AKR or DBA allele of Arid4b was PCR amplified and cloned into entry vector pDonr-253, then subcloned by Gateway LR recombination into pDest-737 to generate an expression construct with CMV promoter and N-terminal 3xFLAG tag. Full-length BRMS1 was epitope tagged at the N-terminus by PCR with the HA or myc tag sequence incorporated into the forward primer and cloned into pCMV or pcDNA3-hygro (Invitrogen), respectively. Correct sequences of all vectors were confirmed prior to use.
Met-1 cells [10] were a gift from Dr. Robert Cardiff (University of California, Davis, CA). 6DT1 cells [11] were a gift from Dr. Lalage Wakefield (NCI, NIH, Bethesda, MD). HEK293 cells were purchased from ATCC (Manassas, VA). Cell lines were maintained in DMEM supplemented with 10% FBS, 2 mM L-glutamine, penicillin and streptomycin. Cells were confirmed to be free of mycoplasma contamination using the MycoAlert detection kit (Lonza).
Met-1 cells seeded onto 10 cm tissue culture plates were co-transfected with 6 µg of the appropriate V5-tagged Arid4b construct described above, or pcDNA3.1/V5-His-TOPO/lacZ (Invitrogen) as a control vector, plus 600 ng of pSuper.Retro.Puro (Oligoengine) as a selectable marker, using FuGENE 6 transfection reagent. Cells were selected using 1 mg/ml G418 plus 4 µg/ml puromycin and clones derived by limiting dilution. Stable upregulation of Arid4b was verified by performing QRT-PCR using forward primer 5′- GGTGAGTGGGAGCTGGTCTA-3′ and reverse primer 5′- ATAAAGGGCCCACTGAAGGT-3′, and western blotting for endogenous and ectopically expressed ARID4B as described below. 6DT1 cells were transduced with one of five different lentiviral shRNAs targeting Arid4b (RMM4534-NM_194262, Open Biosystems) or a scrambled control shRNA in the same pLKO.1 vector. Stable cells were selected using 10 µg/ml puromycin and pooled clones were analyzed for Arid4b knockdown by western blot.
Met-1 or 6DT1 stable lines were orthotopically implanted into the fourth mammary fat pad of six week old female NU/J or FVB/NJ mice using 105 cells suspended in 100 µl of PBS per animal. Primary tumors and lungs were harvested 28 days later. All experiments were performed according to the National Cancer Institute Animal Care and Use Committee guidelines.
Met-1 cells stably expressing Arid4b or lacZ were seeded at 75,000 cells per well into invasion chambers coated with Matrigel basement membrane matrix (534480, BD Biosciences) or control chambers lacking Matrigel (354578, BD Biosciences). After 24 hours, cells were fixed in 100% methanol, stained with crystal violet, and mounted onto glass slides using mineral oil. Cells were visualized at 400× magnification and five fields were counted for each of three experiments.
HEK293 cells were transfected with the V5- or FLAG-tagged AKR or DBA allele of Arid4b, with or without HA-BRMS1 or myc-BRMS1 where appropriate, using FuGENE 6 transfection reagent. After 30 hours, cells were harvested in mild IP lysis buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% NP-40, 5% glycerol) supplemented with protease inhibitors (11836170001, Roche) and phosphatase inhibitors (P-5726, Sigma). Protein samples were quantitated using Bradford assays. Gammabind G Sepharose beads (17088501, GE Healthcare) were washed twice in NET buffer (50 mM Tris pH 8.0, 150 mM NaCl, 5 mM EDTA, 1% NP-40, 0.5% BSA, 0.04% sodium azide) supplemented with protease and phosphatase inhibitors, and resuspended to form a 50% bead slurry. Lysates were precleared by adding 40 µl of bead slurry and rotating for 30 minutes at 4°C. Samples were centrifuged at 10,000 rpm for 1 minute at 4°C and pre-cleared supernatant transferred to a fresh tube. Anti-V5, anti-FLAG, or anti-myc tag antibodies was added to a final concentration of 1.0 µg/ml and samples rotated for 1 hour at 4°C, then 50 µl of bead slurry was added and co-IPs performed overnight at 4°C. Beads were then washed four times with NET buffer and resuspended in SDS-PAGE sample buffer.
NuPAGE precast gels and buffers (Invitrogen) were used according to manufacturers recommendations and gels were transferred onto Immobilon-P (Millipore). Membranes were blocked in TBS with 0.5% Tween-20 (TBST) plus 5% nonfat dry milk for 1 hour at room temperature then incubated with the primary antibody diluted in blocking buffer overnight at 4°C. The following primary antibodies and concentrations were used: anti-Arid4b (1∶3,000; A302-233A, Bethyl Laboratories), anti-V5 (1∶5,000; 37–7500, Invitrogen), anti-mSIN3A (1∶1,000; sc-994, Santa Cruz), anti-mSDS3 (1∶2,000; A300-235A, Bethyl Labs), anti-β-actin (1∶10,000; ab6276, Abcam), anti-FLAG (1∶3,000; F-3165, Sigma), anti-HA (1∶5,000; 11867423001, Roche), anti-myc tag (1∶2,000; 2276, Cell Signaling). After three washes in TBST, membranes were incubated with one of the following horseradish peroxidase conjugated secondary antibodies diluted in TBST plus 0.5% milk for 1 hour at room temperature: anti-mouse (1∶5,000; NA931V, GE Healthcare), anti-rabbit (1∶10,000; sc-2004, Santa Cruz), anti-goat (1∶10,000; sc-2304, Santa Cruz). Membranes were washed an additional three times in TBST and proteins detected using the Amersham ECL Plus system (RPN2132, GE Healthcare) and Amersham Hyperfilm ECL (28906837, GE Healthcare) according to manufacturer's recommendations. Densitometry data were collected and analyzed using a ChemiDoc-It Imaging System and VisionWorksLS software (UVP).
Total RNA was isolated from pooled clones of 6DT1 Arid4b knockdown cell lines using RNeasy kits (Qiagen) and then arrayed on Affymetrix GeneChip Mouse Gene 1.0 ST arrays by the Microarray Core in the NCI Laboratory of Molecular Technology. Expression data were normalized using Partek Genomics Suite to identify genes whose expression was significantly different (p<.05) between the Arid4b normal cohort (untreated, scrambled control, and H5 lines) and the Arid4b knockdown cohort (lines H3 and H4). The gene list and expression values were then analyzed using Ingenuity Pathways Analysis (Ingenuity Systems, www.ingenuity.com) to identify differentially regulated signaling pathways and biological functions. Expression of the Tpx2 transcriptional network was visualized and figure generated using Cytoscape software [39]. Microarray data are available through the Gene Expression Omnibus under accession number GSE35731.
|
10.1371/journal.pgen.1001153 | Alternative Splicing at a NAGNAG Acceptor Site as a Novel Phenotype Modifier | Approximately 30% of alleles causing genetic disorders generate premature termination codons (PTCs), which are usually associated with severe phenotypes. However, bypassing the deleterious stop codon can lead to a mild disease outcome. Splicing at NAGNAG tandem splice sites has been reported to result in insertion or deletion (indel) of three nucleotides. We identified such a mechanism as the origin of the mild to asymptomatic phenotype observed in cystic fibrosis patients homozygous for the E831X mutation (2623G>T) in the CFTR gene. Analyses performed on nasal epithelial cell mRNA detected three distinct isoforms, a considerably more complex situation than expected for a single nucleotide substitution. Structure-function studies and in silico analyses provided the first experimental evidence of an indel of a stop codon by alternative splicing at a NAGNAG acceptor site. In addition to contributing to proteome plasticity, alternative splicing at a NAGNAG tandem site can thus remove a disease-causing UAG stop codon. This molecular study reveals a naturally occurring mechanism where the effect of either modifier genes or epigenetic factors could be suspected. This finding is of importance for genetic counseling as well as for deciding appropriate therapeutic strategies.
| Mild disease outcomes associated with premature termination codons can result from at least three different mechanisms, but none of these mechanisms explain the mild phenotype observed in some patients. Subtle differences in alternative transcripts have recently been reported at NAGNAG tandem acceptor motifs, which can be detected in 30% of human genes. We provide the first experimental evidence of premature termination codon removal by alternative splicing at a NAGNAG acceptor splice site. Our study emphasizes the biological significance of such alternative splicing in the context of disease-causing mutations and defines a new phenotype-modifying mechanism that buffers nonsense mutations.
| Premature termination codons (PTCs) are usually associated with severe phenotypes. However, mild disease outcomes can occur by at least three different mechanisms [1]. First, translation can be initiated at an internal start codon located downstream from the PTC [2]. Second, PTCs can trigger nonsense-mediated mRNA decay (NMD), a pathway that protects the cell from aberrant transcripts [3]. Third, nonsense-associated alternative splicing (NAS) [4] can remove the exon harboring the PTC.
Subtle changes in alternative splicing events have recently been reported at particular tandem acceptor splice sites, NAGNAG sites (where N represents any nucleotide) [5]–[7]. The use of the intron proximal or distal splice site results in the production of two distinct isoforms distinguished by three nucleotides (NAG). This alternative splicing could result in the creation or deletion of a stop codon [5]. The latter event would thus constitute another mechanism of PTC removal, but has never been described in human pathophysiology.
Splice events at short-distance tandem sites are widespread and contribute to transcriptome and proteome complexity [5]. NAGNAG acceptor motifs are present in 30% of human genes and several studies based on computational analysis using expressed sequence tag (EST) databases showed that at least 5% of human genes contain an experimentally confirmed NAGNAG tandem site [5]. A subtle splice event associated with Stargardt's disease 1, in which a mutation in the ABCA4 gene produced an indel of one amino acid in 50% of the transcripts from one patient, has been described [8]. Nonetheless, the involvement of mutations that alter tandem sites have not been extensively studied in disease [9]. Therefore, to explore further the effect of mutations located within a NAGNAG acceptor motif, we studied the cystic fibrosis transmembrane conductance regulator gene (CFTR). Querying the Tandem Splice Site DataBase (TassDB), a comprehensive online database dedicated to recognizing tandem acceptor sites, identified two NAGNAG motifs in CFTR. Mutations in the CFTR gene, which encodes a cAMP-regulated Cl− channel located at the apical membrane of epithelial cells, cause cystic fibrosis (CF). CF is the most common severe autosomal recessive genetic disorder in Caucasians [10] and affects the physiology of the lung, gastrointestinal tract, reproductive organs, and sweat glands. Mutations in the CFTR gene induce a continuum of phenotypes ranging from mild manifestations with isolated features such as congenital bilateral absence of the vas deferens (CBAVD), or nasal polyposis, to severe disease symptoms. Therefore, CF is a good model to identify novel modifier mechanisms [11]. Here, we show that alternative splicing at a tandem acceptor site removes a premature UAG stop codon and leads to synthesis of a functional protein. This novel PTC removal mechanism explains the mild phenotype detected in several patients.
In this study, we focused on the rare E831X mutation (2623G>T) which affects the first nucleotide of exon 14a which forms part of one of the two NAGNAG acceptor sites detected in the CFTR gene. We had the opportunity to study a consanguineous family of Turkish origin which included three patients (Figure 1A) homozygous for the E831X mutation (Figure 1B). To evaluate the functional consequences of this mutation in a tandem splice motif, we performed transcript analysis on epithelial cells obtained from patient III10 (Figure 1A) by nasal brushing, which is the least invasive technique and provides the most reliable cells to study CFTR mRNA.
Mutations that generate PTCs can reduce the steady-state level of mRNA via nonsense-mediated decay (NMD) [3]. Total RNA from nasal epithelial cells of patient III10 was first quantified to evaluate the amount of CFTR transcripts. Quantitative analysis by real-time PCR was normalized to that of keratin 18 (KRT18), a marker of ciliated and secretory epithelial cells [12], [13]. The results clearly showed that the level of CFTR mRNA was reduced by half in the sample from patient III10 compared to three control samples (48%±9% compared to WT, Figure 2A), showing that the CFTR mRNAs were subject to NMD.
To evaluate the mRNA pattern associated with the 2623G>T mutation, a semi-quantitative RT-PCR analysis was performed and the products were analyzed by capillary electrophoresis. In control samples, two peaks could be detected: a major peak corresponding to the full-length mRNA (97%±0.5%) and a minor peak corresponding to exon 14a skipping (3%±0.5%) (Figure 2B), a feature of CFTR splicing that has been described previously [14]. In the patient III10 sample, a more complex pattern appeared, with the presence of three distinct peaks (Figure 2B) that were subsequently identified by sequencing: a major peak corresponding to exon 14a skipping (76%±2%) (Figure 2B and 2C) and two additional peaks corresponding to a full-length mRNA containing the stop codon (16%±2%) and an mRNA lacking these three nucleotides (8%±0.8%) (Figure 2B and 2D).
Hence, the 2623G>T substitution induced two effects: a reduced level of full-length mRNA containing the stop codon by NMD and the generation of additional mRNA isoforms. These mRNAs were identified as an mRNA lacking exon 14a, a full-length mRNA containing the premature UAG stop codon, and a third mRNA lacking the three nucleotides encoding the UAG stop codon.
To investigate whether the alternative splicing was due to the 2623G>T mutation, we constructed a hybrid minigene containing CFTR exon 14a and approximately 400 nucleotides of its flanking intronic regions. After transfection in the bronchial epithelial cell line BEAS-2B, RT-PCR samples were separated by capillary electrophoresis analysis and isoforms were identified by sequencing. As in the direct transcript analysis, the wild-type construct revealed a low level of exon 14a skipping (7%±0.8%). The 2623G>T substitution increased exon 14a skipping up to 92%±2%. In addition, two peaks could be detected corresponding to full-length mRNA (7%±2%) and to the isoform lacking three nucleotides (0.8%.±0.5%) (Figure 3A and 3C). To focus on these two lower abundance isoforms, a reverse primer within exon 14a was used to amplify mRNAs containing exon 14a specifically. A single peak was detected in mRNA obtained from cells transfected with the wild-type construct, whereas two peaks were detected in mRNA from cells transfected with the 2623G>T construct. The major peak corresponded to full-length mRNA containing the UAG stop codon (88%±0.5%) and the mRNA in the minor peak lacked these three nucleotides (12%±0.5%) (Figure 3B and 3C). The ratio between these two isoforms was comparable using either reverse primer. The relative amounts of the three isoforms differed from the amounts measured for endogenous CFTR, but the minigene construct containing the mutated exon 14a sequence reproduced the in vivo splicing pattern. Therefore, the single nucleotide substitution (2623G>T) is the main determinant of alternative splicing at this site.
Recognition of the 3′ splice site upstream of exon 14a generates full-length mRNA leading to synthesis of the entire CFTR protein. The CFTR channel is composed of two transmembrane spanning domains, two nucleotide binding folds, and a regulatory domain (R domain). The regulatory domain is encoded by exon 13 (orange) and exon 14a encodes the region linking the regulatory domain to the seventh transmembrane segment (blue) (Figure 4A). The 2623G>T mutation which occurs at the first base of exon 14a induced multiple splicing defects, including exon skipping. An mRNA lacking exon 14a encodes a protein missing the linker domain between the R domain and the seventh transmembrane segment (CFTR-del831-873, Figure 4B). In addition to the full-length mRNA containing the UAG stop codon, alternative splicing at the NAGNAG site generated an mRNA lacking this stop codon. The resulting proteins would be either truncated after the regulatory domain (CFTR-E831X, Figure 4C) or missing one amino acid (CFTR-ΔE831, Figure 4D), respectively.
The processing of CFTR can be assessed by examining its glycosylation state [15]. Western blot analysis of wild-type CFTR protein, transiently (Figure 5A, left panels) or stably (Figure 5A, right panel) expressed in HEK293 cells, revealed two bands. The diffuse band of 170 kDa (band C) corresponds to the mature, fully glycosylated protein and the thin band of approximately 140 kDa (band B) represents the core-glycosylated immature CFTR. Similar analysis of transiently expressed CFTR-del831-873 revealed a unique thin band indicating a maturation defect (Figure 5A, higher left panel) that was further investigated using an N-Glycosidase F assay (Figure S1). After enzymatic treatment of CFTR-WT, both bands B and C were converted to the lower apparent molecular weight non-glycosylated band A. CFTR-del831-873 was not affected by the enzymatic treatment, indicating an absence of glycosylation of this protein, thus identifying the first naturally occurring non-glycosylated CFTR mutant.
Stably expressed CFTR-E831X protein was detected as a single band at the expected size (Figure 5A, right panel), while transiently expressed CFTR-ΔE831 showed a normal maturation profile with the presence of both bands B and C (Figure 5A, lower left panel). Immunostaining of transiently transfected HeLa cells confirmed a processing defect in both the CFTR-del831-873 and CFTR-E831X mutant proteins as they could only be detected in intracellular compartments close to the nucleus, whereas both CFTR-WT and CFTR-ΔE831 showed clear cell surface staining (Figure 5B). Finally, functional assays showed the absence of CFTR-dependent anion transport in cells expressing either CFTR-del831-873 or CFTR-E831X. In contrast, a high level of anion transport was observed with CFTR-WT and CFTR-ΔE831, in accordance with the biochemical and immunocytochemical results above (Figure 5C and 5D). Therefore, we conclude that CFTR-ΔE831 represents the functional form accounting for the mild phenotype observed within this family.
CF was suspected in the first year of life in patient III10 (Figure 1A) due to recurrent bronchitis, and this was confirmed by a positive sweat test (70 mmol/L). The patient was tested and found to be homozygous for the nonsense mutation E831X. After familial study, CF was diagnosed in his 18-year-old sister (III5), who also had a positive sweat test and the same genotype. However, the two siblings, now aged 13 and 30, are in good health with no evidence of pancreatic insufficiency, and have normal lung function tests. The female patient married her first cousin who was found to be an E831X carrier. Their son (IV1) is homozygous for the E831X mutation. Now 5 years of age, he presents with a positive sweat test (74 mmol/L), but yearly clinical assessments are normal (Table 1).
Interestingly, this mutation was first described in a female CF patient carrying the severe missense substitution G551D on the other allele. E831X was considered as a severe mutation because she presented with meconium ileus at birth, a neonate pulmonary infection and an elevated sweat test [16]. Given our results, we requested yearly clinical assessments, which indicated no pulmonary exacerbation, an almost normal chest radiograph, moderate pancreatic insufficiency and normal abdominal ultrasound. Thus, her clinical outcome was better than expected at birth. This mutation was also reported in two 13-year-old male twins of Turkish origin carrying an in-frame deletion on the other allele (591del18) [17]. These twins had persistent nasal polyps and elevated sweat tests, but no pancreas or lung involvement [17]. Subsequently, E831X was reported to be the allele present in cohorts of men with CBAVD and mild to severe mutations, such as ΔF508, for the other allele [18]–[20]. These clinical observations and our data bring the deleterious nature of the E831X mutation into question; but how can the minor functional CFTR-ΔE831 isoform lead to such mild phenotypes? The minimal level of CFTR mRNA required to maintain normal function differs between organs, with the vas deferens being the most sensitive tissue [21]. CFTR-ΔE831 mRNAs did not appear to reach this threshold in all tissues, leading to mild phenotypes such as nasal polyposis or CBAVD in compound heterozygotes. In the nasal epithelial cells of patient III10, mRNAs encoding functional CFTR-ΔE831 were estimated at 8%. As total CFTR transcripts were reduced to 48% compared to WT levels, this amount can be corrected down to 4%, a level comparable to previous studies associated with mild lung disease [22], [23]. However, the relative level of each transcript may differ between patients, as the same PTC has been shown to elicit NMD with variable efficiency in CF nasal epithelial cells, thus reducing the amount of full-length mRNA containing the stop codon [13].
The intron 13/exon 14a boundary, illustrated in Figure 6, shows the presence of a proximal acceptor site in the intron and a distal acceptor site in the exon, typical of a NAGNAG motif. The use of the proximal or distal acceptor site is regulated by multiple factors. First, the strength of the acceptor site depends on the site itself (CAG>TAG>AAG), the polypyrimidine tract, and the branch point sequences. In addition, the branch point sequence-to-NAGNAG region in the 3′ tandem splice site was shown to participate in 3′ splice-site selection [24]. This selection could also be modulated by RNA-binding proteins because exonic or intronic splicing enhancer (ESE, ISE) and silencer (ESS, ISS) sites have been described as overabundant in the vicinity of tandem 3′ splice sites compared to constitutively spliced exons [25]. Lastly, the nucleotides composing the NAGNAG acceptor site are also tightly implicated in the recognition of either the proximal or distal splice site. NAGNAG acceptor sites have been classified with respect to their splicing plausibility. Plausible sites allow the use of both acceptor sites, whereas implausible ones allow the use of a single site [26]. The 2623G>T substitution converts the NAGNAG acceptor motif from an implausible CAGGAG (use of the proximal site) into a plausible CAGTAG motif (use of either proximal or distal splice sites). Therefore, the mutation would favor the use of the distal splice site, resulting in bypass of the PTC and explaining the mechanism leading to an mRNA lacking three nucleotides (Figure 6).
In silico analysis using software dedicated to analyzing NAGNAG motifs (BayNAGNAG) [27] predicted a shift in the probability of using both the proximal and distal acceptor sites from 0.2% to 81.2%, consistent with our in vivo results (Table 2). The first description of a disease-causing mutation in a NAGNAG splice site was reported in the ABCA4 gene [8], before the description of the NAGNAG motif by Hiller et al. in 2004 [5]. This mutation (2588G>C) was shown to produce two isoforms leading to either an indel of Gly863 or to a missense defect, Gly863Ala, by shifting the NAGNAG motif from an implausible sequence (TAGGAG) into a plausible sequence (TAGCAG). Similarly, in silico analysis predicted an increase in the probability of using either distal or proximal acceptor sites from 0.1% to 57.7% (Table 2). Our study demonstrated that disease-causing mutations in NAGNAG are predictive of alternative splicing, as previously proposed for single nucleotide polymorphisms in NAGNAG acceptors [26]. Therefore, in the absence of mRNA samples, such splicing events could be anticipated by in silico analysis.
The removal of a deleterious PTC could apply to many genes, as the configuration we have described is frequent in the human genome. Indeed, NAGGAG represents the most frequent NAGNAG motif recognizable upstream of a coding exon [5]. Querying the TassDB retrieved 4882 occurrences of the NAGGAG motif with GAG in a coding exon, representing 4597 genes. Among these NAGGAG motifs, 52.2% (n = 2551) are in intron phase 0, a configuration that leads to the first amino acid of the exon being a Glu, as shown in this study. We can therefore hypothesize that bypass of a deleterious PTC could occur when similar GAG to TAG mutations affect one of the 2551 NAGGAG motifs within the human genome.
In addition to contributing to proteome plasticity [9], alternative splicing at tandem 3′ acceptor sites can also result in in vivo removal of a premature UAG stop codon. This feature differentiates the UAG stop codon from UAA or UGA and may help explain phenotype/genotype discrepancies. The novel PTC removal mechanism would be expected to lead to mild phenotypes, with the deletion of a single amino acid being potentially less deleterious than a truncating mutation. This study also emphasizes the biological significance of alternative splicing at tandem acceptor sites in the context of disease-causing mutations. Subtle splicing events could be considered simply as noise tolerated by the cell [28]; however, considering point mutations at NAGNAG acceptor sites clearly provides evidence of their functional relevance. Alternative splicing at these acceptor sites could define a novel phenotype-modifying mechanism buffering deleterious nonsense mutations. Overall, this study highlights the importance of thoroughly characterizing the molecular defects in patients with milder than expected phenotypes. Indeed, such molecular studies can reveal naturally occurring mechanisms where modifier genes or epigenetic factors have been suspected to have an effect.
Informed consent was obtained from all subjects and the local ethics committee approved the study.
Cells obtained from nasal brushings were immediately transferred into RNA-Later buffer (Qiagen) and total RNA was purified as recommended by the manufacturer using QiaQuick Spin columns (Qiagen). Two independent RT-PCR assays were performed with 400 ng of total RNA for each sample (Applied Biosystems).
CFTR and KRT18 primers and probes (TaqMan FAM/NFQ-MGB probe format) were designed by Applied Biosystems. PCR reactions contained the TaqMan Gene Expression Assay Mix, TaqMan Universal PCR Master Mix, no AmpErase UNG, and 1 µl cDNA (or 1 µl of DNase/RNase free water for the No Template Control) in a final volume of 20 µl. Samples were placed in 96-well plates and amplified in an ABI 7900HT Sequence Detection System (Applied Biosystems). Amplification conditions were 10 min at 95.0°C, followed by 40 cycles of 15 s at 95.0°C and 1 min at 60.0°C. All reactions were run in triplicate and each sample was run in two QPCR assays. To correct for variations in the amount of input RNA and efficiency of the reverse transcription, KRT18 (which is specifically expressed in ciliated and secretory epithelial cells) was quantified and results were normalized to these values. Relative amounts of CFTR mRNA were measured using the 2 −ΔΔCT method [29]. A control sample was chosen as a calibrator, i.e., as the baseline for the comparative results.
Semi-quantitative RT-PCRs were performed using 1 µl cDNA templates with a sense FAM-labeled primer in exon 13 (5′-AGTGTCACTGGCCCCTCAG-3′) and a reverse primer in exon 17 (5′-GTGTCGGCTACTCCCACGTA-3′) or in exon 14a (5′-CATGTAGTCACTGCTGGTATGCT-3′) in a 20 µl total volume. PCR conditions were 94°C for 5 min followed by linear phase amplification of 29 cycles or 30 cycles at 94°C for 20 sec, 60°C for 20 sec, and 72°C for 20 sec. All samples were then extended at 72°C for 1 min and, finally, cooled to 4°C in a 9700 thermocycler (Applied Biosystems). Capillary electrophoresis analysis (GeneScan) was performed using 1 µl of the diluted PCR mixture (1/40) added to 18.5 µl of formamide and 0.1 µl of ROX 400 HD fluorescent size standards (Applied Biosystems). The mixture was then denatured at 95°C for 5 min and cooled to 4°C. Amplified products were separated on an ABI 3130 XL DNA analyzer using 3130 POP6 and analyzed with the GeneMapper 4.0 software (Applied Biosystems). Ratios of splicing isoforms were determined as the peak area for the CFTR isoform divided by the total peak areas for the three isoforms. Data represent the mean±SE of at least two independent measurements performed in triplicate.
For sequencing of the different splice variants, PCR was performed using 1 µl of cDNA templates with a sense primer in exon 13 (5′-AGTGTCACTGGCCCCTCAG-3′) and a reverse primer in exon 17 (5′-GTGTCGGCTACTCCCACGTA-3′) or in exon 14a (5′-CATGTAGTCACTGCTGGTATGCT-3′) using the same conditions as described previously for 40 cycles. Following RT-PCR, products were resolved on a 2% agarose gel and bands of interest were excised and purified using a gel extraction kit (Promega). Samples were subsequently sequenced using both forward and reverse primers.
HEK293 and HeLa cells were grown at 37°C, 5% CO2 in DMEM medium supplemented with 10% SVF and 1% PS. BEAS-2B cells (CRL-9609) cells were grown in LHC-8 medium supplemented with 10% fetal calf serum and 1% Penicillin Streptomycin. Cells were transfected with Lipofectamine 2000 according to the manufacturer's instructions. HEK293 cells stably expressing CFTR-WT or CFTR-E831X were selected using Zeocin (50 µg/mL).
CFTR-WT cDNA was subcloned in pTracer [30] and CFTR-E831X, CFTR-ΔE831 and CFTR-del-831-873 were generated by site-directed mutagenesis (Stratagene) and each construct was sequenced. A fragment comprising exon14a, 387 bp of the upstream intron 13 and 371 bp of the downstream intron 14a was PCR amplified from genomic DNA obtained from a healthy volunteer or from patient III10 using the following primers: 5′-GGACCCCTGAAGAAACAGGT-3′ and 5′-GCCTTCTACTTTGAGCTTTCG-3′. PCR products were subcloned into pCR II (Invitrogen) and inserts sequenced before subcloning into the pET01 vector (Mobitec) at the BamHI and XbaI restriction sites.
Minigenes containing either the exon 14a-WT or the 2623G>T mutation and flanking introns (387 bp upstream and 371 bp downstream of exon 14a) were transfected in BEAS-2B cells seeded in 6-well plates. Total RNA was purified as recommended by the manufacturer using QiaQuick Spin columns (Qiagen). The mRNA concentrations were measured using a NanoDrop spectrophotometer. The mRNA (1.5 µg in a final volume of 20 µl) was DNAse treated for 30 min at 37°C before heat denaturation of the enzyme (Quiagen). Treated mRNA (400 ng) was used to perform RT-PCR using the High capacity cDNA Reverse Transcription kit (Applied Biosystems). Ten percent of the RT-PCR product was PCR amplified using primers specific to the splice donor and splice acceptor exons of the pET01 plasmid (5′-FAM-GTGACAGCTGCCAGGATCG-3′ and 5′-CAGTGCCAAGGTCTGAAGGT-3′). PCR was conducted at 94°C for 5 min followed by linear phase amplification of 20 cycles or 21 cycles at 94°C for 20 sec, 60°C for 20 sec, and 72°C for 20 sec. All samples were then extended at 72°C for 1 min and, finally, cooled to 4°C in a 9700 thermocycler (Applied Biosystems). Capillary electrophoresis analysis of the PCR products was performed as previously described. Ratios of splicing isoforms were determined as the peak area of the considered isoform divided by the total peak areas. Sequencing was performed on the 21 cycle PCR products generated with the internal primer using an unlabeled forward primer located within the splice donor exon of the pET01 plasmid. Data represent the mean±SE of at least two independent measurements performed in duplicate.
HEK293 cells were plated on 100 mm dishes and transiently transfected with CFTR-WT, CFTR-del831-873 or CFTR-ΔE831. Immunoprecipitation was performed as previously described [31] with a C-terminal directed anti-CFTR antibody (24-1, R&D systems). Cells stably expressing CFTR-E831X or CFTR-WT were lysed in 1X RIPA buffer and samples containing 30 µg total protein were analyzed by Western blot. After gel electrophoresis and transfer, membranes were probed using the MM13-4 antibody (Millipore) directed against the N-terminal region of CFTR. Direct recording of the chemiluminescence was performed using the CCD camera of the GeneGnome analyzer and quantification using the GeneTools software (Syngene BioImaging Systems, Synoptics Ltd).
HEK293 cells were plated on 60 mm dishes and transiently transfected with CFTR-WT or CFTR-del831-873. Cells were washed twice with ice cold PBS and lysed in N-Glycosidase F buffer containing 20 mM sodium phosphate, pH 7.5, 0.1% SDS, 50 mM β-mercaptoethanol, and 1% Igepal, supplemented with protease inhibitors. N-Glycosidase F (3 Units, Roche) was added to lysates and incubated overnight at 37°C and one tenth of the volume of each sample was analyzed by Western blot.
HeLa cells were seeded on glass coverslips and transfected with the appropriate construct. The following day, cells were fixed using ice-cold methanol. Cells were permeabilized (1% BSA, 0.1% TritonX-100 in PBS) and, then, incubated with the primary antibody, MM13-4 anti-CFTR, diluted 1/200, for 2 h at room temperature. Secondary antibody (1/500 dilution, AlexaFluor 488 conjugated, Invitrogen) was then added and incubated 1 h. Coverslips were mounted using Vectashield mounting medium containing DAPI (4, 6-Diamino-2-phenylindol) and analyzed using a Leica DMR epifluorescence microscope.
CFTR activity was determined in transiently transfected HEK293 cells using the halide-sensitive yellow fluorescent protein YFP-H148Q/I152L as described previously [32]. Cells were plated in 96-well microplates (2.5×104 cells/well) and co-transfected with CFTR and halide-sensitive YFP. The CFTR functional assay was carried out 48 h after transfection. Cells were incubated for 30 minutes with PBS containing forskolin (20 µM) before being transferred to an Olympus IX 50 fluorescence microscope (Chroma; excitation: HQ500/20X; emission: HQ535/30M; dichroic: 515 nm), equipped with a photomultiplier tube (Hamamatsu) for detection of fluorescence. Cell fluorescence was continuously measured before and after addition of NaI (final NaI concentration: 100 mM). The signal was digitized using a PowerLab 2/25 acquisition system (ADInstruments). Cell fluorescence recordings were normalized to the initial average value measured before addition of NaI. Signal decay was fitted to a double exponential function to derive the maximal slope corresponding to initial influx of I− into the cells. Maximal slopes were converted to rates of change in intracellular I− concentration (in mM/s) using the equation: d[I−]/dt = KI[d(F/F0)/dt] where KI is the affinity constant of YFP for I− and F/F0 is the ratio of the cell fluorescence at a given time vs. initial fluorescence [33].
Bayesian network predictions of splicing outcomes at NAGNAG tandem acceptors in wild-type and mutant conditions were calculated using the BayNAGNAG software, (http://www.tassdb.info/baynagnag/form.html).
Query for the number of genes containing a NAGGAG motif:
SELECT count(distinct GeneID) FROM Gene G, Transcript T, SS2Transcript2SED SS2Tr2SED, SpliceSite SS, SpliceEventData SED WHERE G.ID = T.GeneID AND T.ID = SS2Tr2SED.TranscriptID AND SS2Tr2SED.SpliceSiteID = SS.ID AND SS2Tr2SED.SpliceEventDataID = SED.ID AND SS.Type = ‘acceptor’ AND SED.NumEESTs> = 0 AND SED.NumIESTs> = 0 AND ((SS.pattern = ‘CAGGAG’) OR (SS.pattern = ‘TAGGAG’) OR (SS.pattern = ‘AAGGAG’) OR (SS.pattern = ‘GAGGAG’)) AND G.species = ‘Homo sapiens’. Query for the number of genes having a NAGGAG motif with the GAG within the coding exon:
SELECT count(distinct GeneID) FROM Gene G, Transcript T, SS2Transcript2SED SS2Tr2SED, SpliceSite SS, SpliceEventData SED WHERE G.ID = T.GeneID AND T.ID = SS2Tr2SED.TranscriptID AND SS2Tr2SED.SpliceSiteID = SS.ID AND SS2Tr2SED.SpliceEventDataID = SED.ID AND SS.Type = ‘acceptor’ AND SED.NumEESTs> = 1 AND SED.NumIESTs = 0 AND ((SS.pattern = ‘CAGGAG’) OR (SS.pattern = ‘TAGGAG’) OR (SS.pattern = ‘AAGGAG’) OR (SS.pattern = ‘GAGGAG’)) AND G.species = ‘Homo sapiens’;
Query for the number of genes having a NAGGAG motif with the GAG within the coding exon and with the intron phase = 0:
SELECT count(distinct GeneID) FROM Gene G, Transcript T, SS2Transcript2SED SS2Tr2SED, SpliceSite SS, SpliceEventData SED WHERE G.ID = T.GeneID AND T.ID = SS2Tr2SED.TranscriptID AND SS2Tr2SED.SpliceSiteID = SS.ID AND SS2Tr2SED.SpliceEventDataID = SED.ID AND SS.Type = ‘acceptor’ AND SED.NumEESTs> = 1 AND SED.NumIESTs = 0 AND ((SS.pattern = ‘CAGGAG’) OR (SS.pattern = ‘TAGGAG’) OR (SS.pattern = ‘AAGGAG’) OR (SS.pattern = ‘GAGGAG’)) AND SED.phaseUTR like ‘intron phase 0’ AND G.species = ‘Homo sapiens’.
|
10.1371/journal.ppat.1005911 | Identification of Novel Rosavirus Species That Infects Diverse Rodent Species and Causes Multisystemic Dissemination in Mouse Model | While novel picornaviruses are being discovered in rodents, their host range and pathogenicity are largely unknown. We identified two novel picornaviruses, rosavirus B from the street rat, Norway rat, and rosavirus C from five different wild rat species (chestnut spiny rat, greater bandicoot rat, Indochinese forest rat, roof rat and Coxing's white-bellied rat) in China. Analysis of 13 complete genome sequences showed that “Rosavirus B” and “Rosavirus C” represent two potentially novel picornavirus species infecting different rodents. Though being most closely related to rosavirus A, rosavirus B and C possessed distinct protease cleavage sites and variations in Yn-Xm-AUG sequence in 5’UTR and myristylation site in VP4. Anti-rosavirus B VP1 antibodies were detected in Norway rats, whereas anti-rosavirus C VP1 and neutralizing antibodies were detected in Indochinese forest rats and Coxing's white-bellied rats. While the highest prevalence was observed in Coxing's white-bellied rats by RT-PCR, the detection of rosavirus C from different rat species suggests potential interspecies transmission. Rosavirus C isolated from 3T3 cells causes multisystemic diseases in a mouse model, with high viral loads and positive viral antigen expression in organs of infected mice after oral or intracerebral inoculation. Histological examination revealed alveolar fluid exudation, interstitial infiltration, alveolar fluid exudate and wall thickening in lungs, and hepatocyte degeneration and lymphocytic/monocytic inflammatory infiltrates with giant cell formation in liver sections of sacrificed mice. Since rosavirus A2 has been detected in fecal samples of children, further studies should elucidate the pathogenicity and emergence potential of different rosaviruses.
| We identified two novel picornaviruses, rosavirus B and C, infecting street and wild rats respectively in China. While rosavirus B was detected from Norway rats, rosavirus C was detected from five different wild rat species (chestnut spiny rat, greater bandicoot rat, Indochinese forest rat, roof rat and Coxing's white-bellied rat) by RT-PCR. Anti-rosavirus B antibodies were detected in Norway rats, whereas anti-rosavirus C antibodies were detected in Indochinese forest rats and Coxing's white-bellied rats, supporting potential interspecies transmission of rosavirus C. Genome analysis supported the classification of rosavirus B and C as two novel picornavirus species, with genome features distinct from rosavirus A. Rosavirus C isolated from 3T3 cells causes multisystemic diseases in a mouse model, with viruses and pathologies detected in various organs of infected mice after oral or intracerebral inoculation. Our results extend our knowledge on the host range and pathogenicity of rodent picornaviruses.
| Picornaviruses are positive-sense, single-stranded RNA viruses with icosahedral capsids. They infect various animals and human, causing various respiratory, cardiac, hepatic, neurological, mucocutaneous and systemic diseases [1, 2]. Based on genotypic and serological characterization, the family Picornaviridae is currently divided into 29 genera with at least 50 species. Among the various picornaviruses belonging to nine genera that are able to infect humans, poliovirus and human enterovirus A71 are best known for their neurotropism and ability to cause mass epidemics with high morbidities and mortalities [3, 4]. Picornaviruses are also known for their potential for mutations and recombination, which may allow the generation of new variants to emerge [5–10].
Emerging infectious diseases like avian influenza and coronaviruses have highlighted the impact of animal viruses after overcoming the inter-species barrier [11–15]. As a result, there has been growing interest to understand the diversity and evolution of animal and zoonotic viruses. For picornaviruses, numerous novel human and animal picornaviruses have been discovered in the past decade [1, 16–27]. We have also discovered a novel picornavirus, canine picodicistrovirus (CPDV), with two internal ribosome entry site (IRES) elements, which represents a unique feature among Picornaviridae [28]. Moreover, novel picronaviruses were identified in previously unknown animal hosts such as cats, bats and camels [29–31], reflecting our slim knowledge on the diversity and host range of picornaviruses. The discovery and characterization of novel picornaviruses is important for better understanding of their evolution, pathogenicity and emergence potential.
Although rodents can be infected by several picornaviruses, the picornaviral diversity is probably underestimated, given the enormous species diversity of rodents. Moreover, little is known about the pathogenicity of the recently discovered rodent pricornaviruses, such as rodent stool-associated picornavirus (rosavirus) A1, mouse stool-associated picornavirus (mosavirus) A1, Norway rat hunnivirus and rat-borne virus (rabovirus A) [32, 33]. In this report, we explored the diversity of picornaviruses among rodents in China and discovered two potentially novel picornaviruses, “Rosavirus B” and “Rosavirus C”. While rosavirus B was detected in the street rat, Norway rats, rosavirus C was detected in five different wild rat species, suggesting potential interspecies transmission. Their complete genome sequences were determined, which showed that “Rosavirus B” and “Rosavirus C” represent two novel picornavirus species distinct from Rosavirus A. Rosavirus C isolated from cell culture causes multisystemic diseases in a mouse model, with histopathological changes and positive viral antigen expression in lungs and liver of infected mice.
A total of 2450 respiratory and alimentary samples from 1232 rodents of six different species were obtained (Table 1). Initial screening by RT-PCR for a 159-bp fragment of 3Dpol gene of picornaviruses was positive in 18 respiratory and 24 alimentary samples from 37 rodents of six different species. The sequences from these positive samples had <84% aa identities to the corresponding segments of 3Dpol genes of known picornaviruses, suggesting the presence of potential novel picornaviruses. Subsequent RT-PCR using specific primers targeting these potential novel picornaviruses on the 2450 samples was positive in 62 respiratory and 70 alimentary samples from 92 rodents. The sequences from these 132 samples had <86% aa identities to the corresponding segments of 3Dpol genes of known picornaviruses, being most closely related to Rosavirus A. Moreover, these sequences fell into two distinct clusters, one formed by sequences from wild rodents of five different species from Hong Kong (chestnut spiny rat, greater bandicoot rat, Indochinese forest rats and roof rat) and Hunan and Guangxi (Coxing's white-bellied rat), and the other formed by sequences from street rodents from Hong Kong (Norway rat) (Table 1 and Fig 1). This suggested the presence of two potentially novel picornavirus species, “Rosavirus B” and “Rosavirus C.”
Viral sequences belonging to “Rosavirus B” were only detected from the street rodent species, Norway rat (Rattus norvegicus), whereas sequences belonging to “Rosavirus C” were detected from five different wild rodent species, greater bandicoot rat (Bandicota indica), chestnut spiny rat (Niviventer fulvescens), roof rat (Rattus rattus) and Indochinese forest rat (Rattus andamanensis) from Hong Kong, and Coxing’s white-bellied rat (Niviventer coxingi) from Hunan and Guangxi. Among the five wild rodent species, rosavirus C showed the highest detection rate in Coxing’s white-bellied rats. qRT-PCR showed that the viral load of rosavirus B and C in the positive samples ranged from 1.7 ×104 to 1.6×108 copies/ml.
A total of 13 complete genomes from samples of four different wild rodent species (chestnut spiny rat, Coxing’s white-bellied rat, roof rat and Indochinese forest rat) positive for “Rosavirus C” and one street rodent species (Norway rat) positive for “Rosavirus B” were sequenced directly from the positive respiratory or alimentary samples and characterized. These 13 strains were selected because they were detected from different rodent species or geographical locations (Hong Kong, Hunan and Guangxi) to allow comparison between host species- or geographically distinct strains. The G + C contents of the three rosavirus B and 10 rosavirus C genomes range from 50 to 53%, with genome size 8639 to 9094 bases, after excluding the polyadenylated tract (Table 2). However, the genome sizes of some strains may be larger, as further sequencing of the ends may have been hampered by secondary structures. They share similar genome organization typical of picornaviruses, with UTR at both 5' (470–752 bases) and 3' (693–948 bases) ends, and a large open reading frame of 7449–7476 bases, which encodes potential polyprotein precursors of 2482–2491 aa known to be cleaved by virus-encoded proteases. The predicted protease cleavage sites at P1 (encoding capsid proteins) P2 and P3 (both encoding non-structural proteins) are shown in Table 3. Notably, “Rosavirus B” differed from Rosavirus A in VP2/VP3 (P1’), VP3/VP1 (P1’), VP1/2A (P1’) and 2C/3A (P1’) cleavage sites, whereas “Rosavirus C” differed from Rosavirus A in VP4/VP2 (P1’), VP2/VP3 (P1’), VP3/VP1 (P1 and P1’), VP1/2A (P1’), 2A/2B (P1’), 2B/2C (P1’) and 2C/3A (P1’) cleavage sites.
Phylogenetic trees constructed using the aa sequences of P1, P2 (excluding 2A) and P3 (excluding 3A) of rosavirus B and C are shown in Fig 2 and the corresponding pairwise aa identities are shown in Table 2. 2A and 3A regions were excluded to avoid bias due to poor sequence alignment. In all three trees, the sequences from the present rodent picornaviruses formed two distinct clusters among known picornaviruses, being most closely related to Rosavirus A of the genus Rosavirus. However, their genomes shared only 56.3–59.8% nt identities with that of Rosavirus A. In the polyprotein, P1 and 2C/3C/3D regions, rosavirus B/C possessed only 57.3–57.4%/53.2–54%, 54.3–55%/51.1–52.5% and 65.9–66.1%/62.1–62.8% aa identities respectively to Rosavirus A. Moreover, rosavirus B and C shared only 58.9–60%, 57.7–59.4% and 67.6–68.3% aa identities respectively to each other in these regions. The predicted 2A protein of rosavirus B and C also showed low aa identities to that of rosavirus A, suggesting that they represent two novel picornavirus species, proposed to be named “Rosavirus B” and “Rosavirus C.”
Comparison of genome features of rosavirus B and C to those of rosavirus A is summarized in Table 4. The conserved sequence Yn-Xm-AUG is present in the 5'UTR of rosavirus B and C. While Y7-X19-AUG is found in rosavirus A, the number of Y (6 or 7) and X (19–21) varies among rosavirus B and C. The putative translation initiation sites were contained by an optimal Kozak context (RNNAUGG), with in frame AUG at position 471 to 753. The 5'UTR of many picornaviruses possesses an internal ribosomal entry site/segment (IRES) which is responsible for directing the initiation of translation in a cap-independent manner, and requires both canonical translation initiation and IRES trans-acting factors [3, 34]. Similar to rosavirus A [32], rosavirus B and C also contained a type II-like IRES with stem loops, major domains [19, 27, 35–40] and conserved motifs (Fig 3). However, domain E was only present in rosavirus C strains, RASK8F, RATLC11A, NFSM6F and RASM14A, and rosavirus B strain RNYL1109081R, but not other strains. The pyrimidine-rich region was located near 3' end of 5'UTR. One to three stem-loop structures were present upstream of the start codon and/or between the pyrimidine-rich region and start codon of the polyprotein [41, 42].
The predicted “VP0” of rosavirus B and C are probably cleaved into VP4 and VP2 based on sequence alignment [32]. In contrast to rosavirus A of which the VP4 possessed the myristylation site, GXXX[ST], involved in capsid assembly or virus entry [43], such myristylation site is absent in rosavirus C and variably present in rosavirus B (present in strains RNCW0602091R and RNCW1002091R but absent in strain RNYL1109081R). The predicted 2A of rosavirus B and C exhibited ≤45.1% aa identities to that of rosavirus A, possessed the conserved H-box/NC involved in cell proliferation control, but not Asn-Pro-Gly-Pro (NPGP) motifs [44, 45]. Their predicted 2C possessed GXXGXGKS motif for NTP-binding [46] and DDLXQ motif for putative helicase activity [47]. Their predicted 3Cpro contained the catalytic triad H-D-C [48], conserved GXCG motif in the protease active site and GXH motif [49, 50]. Their predicted 3Dpol contained conserved KDE[LI]R, GG[LMN]PSG, YGDD and FLKR motifs [51], although the second Gly was replaced by Ala in GG[LMN]PSG.
Although rosavirus C were detected in five different rodent species from Hong Kong, Hunan and Guangxi, no major distinct genome features were identified between strains from different rodent species or geographical locations. Yet, the two strains from Coxing’s white-bellied rat from Hunan (NCHN06IO) and Guangxi (NCGX12IN), were always clustered together in the P1, P2 and P3 trees, suggesting that geographically distinct strains may be genetically closely related (Fig 2). These two strains from mainland China possessed a total of 432 unique nucleotide substitutions over the entire genomes compared to the other eight rosavirus C strains from Hong Kong.
Viral sequences belonging to “Rosavirus B” were only detected from the street rodent species, Norway rat (Rattus norvegicus), whereas sequences belonging to “Rosavirus C” were detected from five different wild rodent species, greater bandicoot rat (Bandicota indica), chestnut spiny rat (Niviventer fulvescens), roof rat (Rattus rattus) and Indochinese forest rat (Rattus andamanensis) from Hong Kong, and
Using available rosavirus B and C genome sequences for analysis, the Ka/Ks ratios for various coding regions were estimated (Table 5). The Ka/Ks ratios for most coding regions were low, supporting purifying selection.
Of the various cell lines inoculated with the 11 rodent samples positive for rosavirus B (three samples) or rosavirus C (eight samples), viral replication was detected by RT-PCR in the lysates of 3T3 cells infected by rosavirus C strain RASM14A, with viral load of 4.5×108 copies/ml (3.2 × 103 TCID50) at day 7. Cytopathic effect (CPE), mainly in the form of rounded and refractile cells rapidly detaching from the monolayer, was also observed in infected 3T3 cells five days after inoculation, which showed viral VP1 expression by immunofluorescence in 40% of cells (Fig 4). Electron microscopy of ultracentrifuged cell culture extracts from infected 3T3 cells showed the presence of non-enveloped viral particles of around 25–30 nm in diameter compatible with those described for members of the family of Picornaviridae (Fig 4).
To determine the seroprevalence of rosavirus B and C among different rodent species, western blot analysis was performed on available rodent serum samples to test for specific antibodies against rosavirus B or C recombinant VP1 protein. The purity of the recombinant VP1 proteins was confirmed by the dominant band observed at the predicted size of 40 kDa upon SDS polyacrylamide gel electrophoresis. Anti-rosavirus B antibodies were detected in two (3.3%) of 61 Norway rats from Hong Kong whereas anti-rosavirus C antibodies were detected in three (4.3%) of 70 Indochinese forest rats from Hong Kong and three (7.5%) of 40 Coxing's white-bellied rats from Hunan Province. However, the antibodies from Norway rats against rosavirus B were likely of low levels, as reflected by the relatively weak band observed (Table 1 and Fig 5). Using sera with anti-rosavirus B antibodies against rosavirus C recombinant VP1 protein and sera with anti-rosavirus C antibodies against rosavirus B recombinant VP1 protein, no cross reactivities were observed between the two proteins. Neutralization assays showed that five of the six rats with anti-rosavirus C antibodies by western blot analyses were positive for neutralizing antibodies against rosavirus C RASM14A with titer 1:10 to 1:40.
We attempted to study the pathogenicity in mice challenged with rosavirus C RASM14A isolated from infected 3T3 cells. To mimick the fecal-oral route of transmission typical of many picornavirus infections, oral inoculation of rosavirus C RASM14A was performed on 21 four-day-old suckling mice. One of the suckling mice was eaten by its mother on day one post-challenge. All the remaining 20 suckling mice survived after viral challenge till sacrifice, but some mice exhibit transient roughening of hair two to three days after challenge. Among the nine mice sacrificed on day 3 post-challenge, rosavirus C RASM14A was detected in the intestine and lung of all nine mice, kidney of one mouse, and spleen and liver of three and four mice respectively by RT-PCR. Among the five mice sacrificed on day 7 post-challenge, rosavirus C RASM14A was detected in the intestine of all five mice, liver and lung of four mice, and spleen and kidney of two and one mice respectively by RT-PCR. Among the three mice sacrificed on day 14 post-challenge, rosavirus C RASM14A was detected in the lung of one mouse by RT-PCR. Among the three mice sacrificed on day 21 post-challenge, rosavirus C RASM14A was detected in the lung of one mouse by RT-PCR. Anti-rosavirus C VP1 antibody was detected in none of the mice sacrificed on day 3 and 14, two of the five mice sacrificed on day 7, and one of the three mice sacrificed on day 21 by Western blot assay (Table 6). qRT-PCR of tissues positive by RT-PCR showed high levels of mean viral RNA copies in lung (2.1 ×106 copies/g) and intestine (2.8 ×105 copies/g) tissues of mice sacrificed on day 3 (Fig 6).
Since some picornaviruses have been associated with neurovirulence, intracerebral inoculation was also performed on another group of 21 one-day old suckling mice. One of the suckling mice was eaten by its mother on day one post-challenge. All the remaining 20 suckling mice survived after viral challenge till sacrifice. Among the nine mice sacrificed on day 3 post-challenge, rosavirus C RASM14A was detected in the lung, liver, brain and spleen of eight mice, and intestine and kidney of four mice by RT-PCR. Among the five mice sacrificed on day 7 post-challenge, rosavirus C RASM14A was detected in the lung of three mice, brain and liver of five mice, intestine of two mice, and spleen and kidney of four mice by RT-PCR. Among the three mice sacrificed on day 14 post-challenge, rosavirus C RASM14A was detected in the brain, intestine, liver and lung of one mouse by RT-PCR. Among the three mice sacrificed on day 21 post-challenge, rosavirus C RASM14A was detected in the lung of one mouse by RT-PCR. Anti-rosavirus C VP1 antibody was detected in none of mice sacrificed on day 3, four of the five mice sacrificed on day 7, and two of three mice sacrificed on day 14 and 21 by Western blot assay (Table 6). qRT-PCR of tissues positive by RT-PCR showed high levels of mean viral RNA copies in various tissues (0.8 to 9.6 ×105 copies/g) of mice sacrificed on day 3 (Fig 6).
Histological examination of various organs revealed alveolar fluid exudation, interstitial infiltration, alveolar fluid exudate and wall thickening in lung sections of mice sacrificed on day 3 after oral or intracerebral inoculation. Moreover, hepatocyte degeneration and lymphocytic/monocytic inflammatory infiltrates with giant cell formation were observed in liver sections of mice sacrificed on day 3 after oral or intracerebral inoculation (Fig 7). Immunohistochemical staining with guinea pig anti-serum against rosavirus C VP1 protein antibody revealed viral antigen expression in bronchiolar and bronchial epithelial cells in lung sections, and hepatocytes in liver sections (Fig 7).
We report the discovery of two novel rodent picornaviruses, rosavirus B and C, from six rodent species in southern China. Though being phylogenetically most closely related to Rosavirus A, “Rosavirus B” and “Rosavirus C” should represent two novel species distinct from Rosavirus A under the genus Rosavirus, according to the criteria for International Committee on Taxonomy of Viruses species demarcation for different members of Picornaviridae [52]. Rosavirus B and C also exhibited different genome features when compared to rosavirus A. Notably, the absence of myristylation site in VP4 of rosavirus C and its varying presence in rosavirus B is intriguing. The VP4 myristylation site has been shown to play a role in localization of the capsid protein for cellular entry and permeability [43, 53]. Its absence in some rosavirus strains suggests that rosaviruses may utilize alternative strategies for capsid localization on cellular targets. In addition, the variations in Yn-Xm-AUG sequence in 5’UTR may also suggest different translational dynamics among rosaviruses. Besides phylogenetic and genomic evidence, the two viruses infect different host, with “Rosavirus B” infecting Norway rats (a street rat) and “Rosavirus C” infecting various wild rats, as supported by positive specific and/or neutralizing antibodies in the respective animals. Further studies are required to better understand the epidemiology of these novel rosaviruses in other rodent population.
Our findings support the potential of rosaviruses for interspecies transmission, and extend our knowledge on the diversity and host range of picornaviruses in rodents. Picornaviruses belonging to Cardiovirus, Hunnivirus, Kobuvirus, Mosavirus, Parechovirus and Rosavirus are known to infect different rodents [32, 33, 54]. Rosavirus A was first discovered in a wild canyon mouse (Peromyscus crinintus) in California [32]. Subsequently, a variant, rosavirus 2, was detected in the fecal specimens of children in the Gambia [26], which prompted further studies to investigate potential transmission of rosavirus from rodents to humans. In this study, the observed low Ka/Ks ratios in various coding regions supported that Norway rats (street rats) in Hong Kong, and wild rats across Hong Kong and mainland China, are natural reservoirs of rosavirus B and C respectively. In particular, Coxing’s white-bellied rats appeared to be an important host for rosavirus C. The relatively low seroprevalence of rosavirus C, as compared to RT-PCR detection rate, among tested Coxing’s white-bellied rats (7.5%) may be due to delayed antibody response during acute infection when the animals were still shedding viruses. Together with the ability of rosavirus C in infecting house mouse (Mus musculus), our findings provided evidence for the interspecies transmission potential of rosavirus C among different rodent species. This is in line with the ability of bat picornaviruses group 1 to 3 in infecting bats of different genera or species [29]. Encephalomyocarditis virus (EMCV), of which swine is the major reservoir, can also infect different animals including rodents, elephants, boars, macaques and humans [55]. Further investigations are warranted to elucidate the ability of rosaviruses to cross species barrier and emerge in other animals or human.
The present results suggest that rosaviruses can be pathogenic to their hosts. Although rosavirus A has been detected in rodents and human previously [26, 32], no virus isolate was available for pathogenicity studies. A few picornaviruses, such as EMCV (cardiovirus A) and Theilovirus (cardiovirus B) under Cardiovirus and Ljungan virus (LV) (parechovirus B) under Parechovirus, were also known to cause systemic infections in infected rodents. In particular, Theiler's murine encephalomyelitis virus (TMEV), which primarily causes asymptomatic enteric infections in mice, has been intensively studied because of its ability to cause myocarditis, type 1 diabetes and acute or persistent demyelinating infections mimicking multiple sclerosis [35, 56]. On the other hand, rodents experimentally infected with EMCV may develop type 1 diabetes mellitus, encephalomyelitis, myocarditis, orchitis and sialodacryoadenitis [57]. Interestingly, LV, which may cause type 1 diabetes and fetal deaths in infected rodents, has been recently found in human intrauterine fetal death and sudden infant death syndrome [58–61]. However, the pathogenicity of other rodent picornaviruses, such as rosavirus A1, mosavirus A1, murine kobuvirus 1, Norway rat hunnivirus and rabovirus A, was less clear [32, 33]. The ability of rosavirus C in causing multisystemic infection in mice with high viral loads in infected organs suggested that rosaviruses may cause severe infections in their host. Further experimental studies using other rosaviruses and rodent species may help to better understand the pathogenicity of members of the genus Rosavirus in different rodents and human.
Rodents are the largest order of mammals on earth, accounting for 43% of the approximately 4,800 living mammalian species. They are widely distributed, being found in all habitats except the oceans. The order, Rodentia, with around 2050 species under 28 families, is further classified into five suborders: Anomaluromorpha, Castorimorpha, Hystricomorpha, Myomorpha and Sciuromorpha. [62, 63]. Viruses of at least 22 families, including Adenoviridae, Arenaviridae, Arteriviridae, Astroviridae, Bunyaviridae, Caliciviridae, Circoviridae, Coronaviridae, Flaviviridae, Hepadnaviridae, Hepeviridae, Herpesviridae, Papillomaviridae, Paramyxoviridae, Parvoviridae, Picobirnaviridae, Polyomaviridae, Reoviridae, Rhabdoviridae, Togaviridae, Picornaviridae and Poxviridae, are known to infect rodents [64, 65]. Rodent pathogens may infect human either by direct contact such as bites and inhalation of aerosolized animal excreta, or indirectly through vectors such as ticks and fleas. Urban rodents may pose particular risk to human health, as in the case of Hantavirus and lymphocytic choriomeningitis virus infections. More epidemiological studies should be performed to explore the diversity of rodent picornaviruses and their potential risks to human.
A total of 1232 wild and street rodents, belonging to eight different species, were captured from various locations in both rural and urban areas of Hong Kong, Hunan Province and Guangxi Province of China over a five-year period (September 2008 to August 2013) (Table 1). Samples from Hong Kong were provided by the Agriculture, Fisheries and Conservation Department (AFCD) and Food, Environment and Hygiene Department (FEHD), the government of the Hong Kong Special Administrative region (HKSAR), as part of a surveillance program on local rodents. All rodents were individually trapped and samples were collected from each rodent using procedures described previously [11, 66]. To prevent cross contamination, collection of samples were performed using disposable swabs with protective gloves changed for each rodent. Wild rodents in rural areas of Hong Kong were released back to nature after sample collection. Samples from street rodents in urban areas of Hong Kong and rodents from China were collected immediately after euthanasia as routine policies for disposal of captured rodents. All samples were placed in viral transport medium (Earle’s balanced salt solution, 0.09% glucose, 0.03% sodium bicarbonate, 0.45% bovine serum albumin, 50 mg/ml amikacin, 50 mg/ml vancomycin, 40 U/ml nystatin) to inhibit bacterial and fungal overgrowth, and stored at -80°C before RNA extraction.
Viral RNA was directly extracted from the respiratory and alimentary samples in viral transport medium using Viral RNA mini kit (QIAgen, Hilden, Germany). The RNA was eluted in 60 μl of RNase-free water and was used as the template for RT-PCR.
Initial picornavirus screening was performed by amplifying a 159-bp fragment of the 3Dpol gene of picornaviruses by RT-PCR using conserved primers (5’-GGCGGYTNGAYGGYGCSATGCCGT-3’ and 5’-CCGACCARCACRTCRTCRCCRTA-3’) and previously described protocols [6, 24, 29]. The primers were designed by multiple alignment of the nucleotide sequences of the 3Dpol genes of all known picornaviruses, based on the conserved 3Dpol motifs, GG[LMN]PSG and YGDD. All samples positive by RT-PCR were confirmed by sequencing. Briefly, reverse transcription was performed using the SuperScript III kit (Invitrogen, San Diego, CA, USA) and the reaction mixture (10 μl) contained RNA, first-strand buffer (50 mM Tris-HCl pH 8.3, 75 mM KCl, 3 mM MgCl2), 5 mM DTT, 50 ng random hexamers, 500 μM of each dNTPs and 100 U Superscript III reverse transcriptase. The mixtures were incubated at 25°C for 5 min, followed by 50°C for 60 min and 70°C for 15 min. The PCR mixture (25 μl) contained cDNA, PCR buffer (10 mM Tris-HCl pH 8.3, 50 mM KCl, 2 mM MgCl2 and 0.01% gelatin), 200 μM of each dNTPs and 1.0 U Taq polymerase (Applied Biosystem, Foster City, CA, USA). The mixtures were amplified in 40 cycles of 94°C for 1 min, 50°C for 1 min and 72°C for 1 min and a final extension at 72°C for 10 min in an automated thermal cycler (Applied Biosystem, Foster City, CA, USA). Standard precautions were taken to avoid PCR contamination and no false-positive was observed in negative controls.
All PCR products were gel-purified using the QIAquick gel extraction kit (QIAgen, Hilden, Germany). Both strands of the PCR products were sequenced twice with an ABI Prism 3130xl DNA Analyzer (Applied Biosystems, Foster City, CA, USA), using the two PCR primers. The sequences of the PCR products were compared with known sequences of the 3Dpol genes of picornaviruses in the GenBank database.
As initial RT-PCR of the 3Dpol gene revealed at least two potential novel picornavirus species in 18 respiratory and 24 alimentary tract samples, all the 2450 respiratory and alimentary tract samples were re-tested using specific RT-PCR assays to enhance the sensitivities for detection of these novel picornaviruses. Primers were designed by multiple alignment of the 3Dpol gene sequences obtained during genome sequencing from the initial positive samples. The PCR assays were targeted to a 253 bp (5’- ATGCTCCTGTTCTCATGCTTTT -3’ and 5’- GAAAATCTGGGTCAGGGGTGAA -3’) fragment and a 243-bp (5’- TGTTCTCTTGYTTYTCCCAGAT -3’ and 5’- AAYTGCGGGTCYGGDGTGAA -3’) fragment of the 3Dpol gene of the potential novel picornaviruses. The components of the PCR mixtures and the cycling conditions were the same as those described above. Purification of the PCR products and DNA sequencing were performed as described above, using the corresponding PCR primers. The sequences of the PCR products were compared with known sequences of the 3Dpol genes of picornaviruses in the GenBank database.
Real-time RT-qPCR was performed on samples positive for the novel picornaviruses by RT-PCR as described previously [1, 29]. Briefly, specific primers targeting a 144-bp (5’-TGTCAGATGGTGTCAACAGTCAAA-3’ and 5’-TCATGGCGCACTTTCACATT-3’), a 137-bp (5’-ACAAATCTACAGCCAAATTCCAAA-3’ and 5’-GTAGGGTATGCCTTTCTGGTCAA-3’) and a 112-bp (5’-CAGCCAAATTCCAAATTCAGAT-3’ and 5’-CCAGATCAGCCATGTTTGGAA-3’) fragment of the 2C genes were used for RT-qPCR by Thermal Cyler FastStart DNA Master SYBR Green I Mix reagent kit (Roche). cDNA was amplified by Thermal Cycler 7900HT (Applied Biosystems) with 20-μl reaction mixtures containing FastStart DNA Master SYBR Green I Mix reagent kit (Roche). A plasmid containing the target sequence was used for generating the standard curves.
Thirteen genomes of the two novel picornavirus species were amplified and sequenced, with RNA directly extracted from respiratory or alimentary samples as templates [6, 24, 29]. RNA was converted to cDNA by a combined sequence-specific-priming, random-priming and oligo (dT) priming strategy. As initial results showed that the two novel picornaviruses are distantly related to known picornaviruses, the cDNAs of three initial strains were amplified by 5’-rapid amplification of cDNA ends (RACE) using the SMARTer RACE cDNA Amplification Kit (Clontech, USA). The first strand cDNA for the 5’ sequence of the genome was constructed with specific primers designed according to results of the first and subsequent rounds of sequencing and SMARTer II A Oligonucleotide by SMARTScribe Reverse Transcriptase. The 3’ sequence of the genome is completed by specific primers designed for the 3’ end from the results of the first and subsequent rounds of sequencing and oligo (dT) primer. Sequences were assembled to produce final sequences of the viral genomes. The genomes of the remaining strains were amplified and sequenced by the specific primers designed from the initial three genomes and the 5’ ends of the viral genomes were confirmed by RACE using the SMARTer RACE cDNA Amplification Kit (Clontech, USA).
The nucleotide (nt) sequences of the genomes and the deduced amino acid (aa) sequences of the open reading frames (ORFs) were compared to those of known picornaviruses. Phylogenetic tree construction was performed using maximum-likelihood methods from PhyML 3.0 program. Secondary structure prediction in the 5’UTR was performed using RNAfold [67] and the IRES elements were determined based on sequence alignment with EMCV as described previously [28, 41].
To estimate the selective pressure in driving viral evolution among different regions of the genomes, the number of synonymous substitutions per synonymous site, Ks, and the number of non-synonymous substitutions per non-synonymous site, Ka, for each coding region between different strains of rosavirus B and C were calculated using the Nei-Gojobori method (Jukes-Cantor) in MEGA 5.0 as described previously [68].
Since the VP1 is the largest and most surface-exposed protein which contains most of the motifs important for interaction with neutralizing antibodies in picornaviruses, (His)6-tagged recombinant VP1 proteins of rosavirus B strain RNCW1002091R from a Norway rat and rosavirus C strains, RASK8F from an Indochinese forest rat and NCGX12IN from a Coxing's white-bellied rat, were cloned as described previously [28, 30]. Briefly, the VP1 gene was amplified and cloned into the NheI site of expression vector pET-28b(+) (Novagen, Madison, WI, USA) in frame and downstream of the series of six histidine residues. The (His)6-tagged recombinant VP1 polypeptide was expressed and purified using the Ni2+-loaded HiTrap Chelating System (GE Healthcare, Buckinghamshire, UK) according to manufacturer’s instructions. Western blot analysis was carried out using available rodent sera using the purified recombinant VP1 protein as described previously [30]. Briefly, the purified (His)6-tagged recombinant VP1 protein was loaded into each well of a sodium dodecyl sulfate (SDS)–10% polyacrylamide gel and subsequently electroblotted onto a nitrocellulose membrane (Bio-Rad, Hercules, CA, USA). The blot was cut into strips and the strips were incubated separately with serial dilutions of sera collected from different rodent species with for IgG detection. Antigen-antibody interaction was detected with horse radish peroxidase-conjugated secondary antibodies and ECL fluorescence system (GE Healthcare, Buckinghamshire, UK).
Eleven samples tested positive for the novel picornaviruses were subject to virus isolation in various cell lines including Vero E6 (African green monkey kidney; ATCC CRL-1586), CrFK (Crandell feline kidney; ATCC CCL-94), in-house HFL (human embryonic lung fibroblast), 3T3 (mouse embryonic fibroblast, ATCC CCL-92) cells, RD (human embryo rhabdomyosarcoma; ATCC CCL-136), RK3E (rat kidney; ATCC CRL-1895) and TCMK1 (mouse kidney; ATCC CCL-139) cells as described previously [28, 69]. Briefly, after centrifugation, samples were diluted five folds with viral transport medium and filtered. Filtrates were inoculated to Minimum Essential Media (MEM) and the mixtures were added to 24-well tissue culture plates by adsorption inoculation. After 1 h of adsorption, excess inoculum was discarded, and the wells were washed twice with phosphate buffered saline and replaced by serum-free MEM. Cultures were inspected daily by inverted microscopy for CPE. Subculturing to fresh cell line was performed from time to time even if there was no CPE and culture lysates were collected for RT-PCR for monitoring viral replication. Immunostaining and electron microscopy were performed on samples that were RT-PCR positive.
3T3 cells successfully infected by rosavirus C RASM14A were subject to negative contrast electron microscopy as described previously [69, 70]. Tissue culture cell extracts infected with rosavirus C RASM14A were centrifuged at 19 000 g at 4°C, after which the pellet was resuspended in phosphate-buffered saline and stained with 2% phosphotungstic acid. Samples were examined with a Philips EM208s electron microscope.
Neutralization assays for rosavirus C RASM14A were carried out as described previously with modifications [11]. Briefly, rodent sera serially diluted from 1:10 to 1:80 were mixed with 100 TCID50 of rosavirus C RASM14A. After incubation for 2 h at 37°C, the mixture was inoculated in duplicates onto 96-well plates of 3T3 cell cultures. Results were recorded after 3 days of incubation at 37°C.
Virus stock used to inoculate mice was obtained from at least the 18th passage of rosavirus C RASM14A in 3T3 cells. Groups of 20 suckling balb/c mice were infected orally (4-day-old) and intracerebrally (1-day-old) as described previously [71]. Approximately 200μl (500 TCID50) of virus suspensions was applied orally and 30μl (100 TCID50) intracerebrally. Two mice challenged with culture media from uninfected cells were included as negative controls in both groups. Mice were monitored daily for signs of disease. Nine/ten, five, three and three mice were sacrificed at day 3, 7, 14 and 21 respectively. After euthanasia, necropsies of mice were performed to obtain the following tissues: intestine, spleen, kidney, liver, lung and brain. Blood was collected for antibody tests by western blot analysis as described above.
To perform immunhistochemical staining on infected cell lines and rodent tissues, guinea pig antiserum against the VP1 protein of rosavirus C was produced by subcutaneously injecting 100 μg purified recombinant rosavirus C VP1 protein to three guinea pigs, using an equal volume of complete Freund’s adjuvant (Sigma) as described previously [28]. Incomplete Freund’s adjuvant (Sigma) was used in subsequent immunizations. Three inoculations at once every two weeks per guinea pig were administered. Two weeks after the last immunization, 1 ml of blood was taken via the lateral saphenous vein of the guinea pigs to obtain the sera.
To examine the histopathology and viral replication of rosavirus C RASM14A in tissues of challenged mice, necropsy organs of the mice were subject to both viral RNA detection by RT-PCR and immunohistological studies as described previously [28]. Tissues for histological examination were fixed in 10% neutral-buffered formalin, embedded in paraffin, and stained with hematoxylin and eosin (H&E). Histopathological changes were observed using Nikon 80i microscope and imaging system. Infected cell lines and tissues from challenged mice that were tested positive for rosavirus C RASM14A by RT-PCR were subject to viral load studies and immunohistochemical staining for viral VP1 protein as described previously [28, 69]. Tissue sections were deparaffinized and rehydrated, followed by blocking endogenous peroxidase with 0.3% H2O2 for 25 min, and then with 1% BSA/PBS at room temperature for 25 min to minimize non-specific staining. The tissue sections were then pre-treated with streptavidin solution and biotin solution at room temperature for 30 min respectively to avoid high background signals due to the endogenous biotin or biotin-binding proteins in the tissues. The sections were incubated at 4°C overnight with 1:2000 dilution of guinea pig anti-VP1 anti-serum, followed by incubation of 30 min at room temperature with 1:2000 dilution of biotin-conjugated rabbit anti-guinea pig IgG, H & L chain (Abcam). Streptavidin/peroxidase complex reagent (Vector Laboratories) was then added and incubated at room temperature for 30 min. Sections were counterstained with hematoxylin. Cells infected or uninfected by rosavirus C RASM14A were included as positive and negative controls respectively in each staining. Cells were fixed in chilled acetone at -20°C for 10 min before incubation with antibodies for staining. Color development was performed using 3,3'-diaminobenzidine and images captured with Nikon 80i imaging system and Spot-advance computer software.
The collection of rodent samples was approved by the Committee on the Use of Live Animals in Teaching and Research (CULATR), The University of Hong Kong (HKU) (CULATR 2284–10) and the Department of Health, the Government of the HKSAR under the Animals (Control of Experiments) Ordinance, Chapter 340 (10-580/581/582 in DH/HA&P/8/2/3 Pt.23 & 13-34/35/36 in DH/HA&P/8/2/3 Pt.47). The production of antiserum against VP1 of rosavirus C in guinea pig experiment was conducted according to Canadian Council on Animal Care (CACC) guidelines (2002) with the protocol approved by CULATR, HKU (CULATR 2489–11) and the Department of Health, the Government of the HKSAR under the Animals (Control of Experiments) Ordinance, Chapter 340 (11-584/585 in DH/HA&P/8/2/3 Pt.31). The mice challenge experiment was approved by CULATR, HKU (CULATR 2545–11) and the Department of Health, the Government of the HKSAR under the Animals (Control of Experiments) Ordinance, Chapter 340 (11-845/846 in DH/HA&P/8/2/3 Pt.33). The mice study was carried out in strict compliance with animal welfare regulations. The mice were anesthetized by fetanyl/fluanisone/diazepam during the whole experiment. Standard guidelines prescribed in Pain and distress in laboratory rodents and lagomorphs, Laboratory Animals 28, 97–112 (1994) were strictly followed and the well-being of animals were monitored daily with a scoring sheet to ensure minimal pain and distress experienced by the mice.
|
10.1371/journal.pgen.1002355 | Signatures of Environmental Genetic Adaptation Pinpoint Pathogens as the Main Selective Pressure through Human Evolution | Previous genome-wide scans of positive natural selection in humans have identified a number of non-neutrally evolving genes that play important roles in skin pigmentation, metabolism, or immune function. Recent studies have also shown that a genome-wide pattern of local adaptation can be detected by identifying correlations between patterns of allele frequencies and environmental variables. Despite these observations, the degree to which natural selection is primarily driven by adaptation to local environments, and the role of pathogens or other ecological factors as selective agents, is still under debate. To address this issue, we correlated the spatial allele frequency distribution of a large sample of SNPs from 55 distinct human populations to a set of environmental factors that describe local geographical features such as climate, diet regimes, and pathogen loads. In concordance with previous studies, we detected a significant enrichment of genic SNPs, and particularly non-synonymous SNPs associated with local adaptation. Furthermore, we show that the diversity of the local pathogenic environment is the predominant driver of local adaptation, and that climate, at least as measured here, only plays a relatively minor role. While background demography by far makes the strongest contribution in explaining the genetic variance among populations, we detected about 100 genes which show an unexpectedly strong correlation between allele frequencies and pathogenic environment, after correcting for demography. Conversely, for diet regimes and climatic conditions, no genes show a similar correlation between the environmental factor and allele frequencies. This result is validated using low-coverage sequencing data for multiple populations. Among the loci targeted by pathogen-driven selection, we found an enrichment of genes associated to autoimmune diseases, such as celiac disease, type 1 diabetes, and multiples sclerosis, which lends credence to the hypothesis that some susceptibility alleles for autoimmune diseases may be maintained in human population due to past selective processes.
| Adaptation to local environments is one of the most important factors shaping human genetic variation among different geographically distributed populations. Here we develop a statistical framework aimed at identifying signals of genetic adaptation. We correlate the spatial distribution of allele frequencies of a large sample of SNPs, genotyped in more than 50 populations distributed worldwide, to a set of environmental factors, describing local geographical features such as climate conditions, diet regimes, and pathogens load. Our results show an excess of putative functional variants for high levels of population differentiation, measured by the degree to which genetic variation correlates with a set of environmental variables. We demonstrate that selection on pathogens is the primary driver of local adaptation and affects the distribution of genetic variation at a large number of genes. Among the selected genes, we also identify an excess of genes associated with autoimmune diseases, such as celiac disease, type 1 diabetes, and multiples sclerosis.
| Anatomically modern humans appeared in East Africa about 200 k years ago, spread out from sub-Saharan Africa approximately 100 k years ago, and subsequently colonized the rest of the world in a series of migratory events [1]. During this period humans encountered a wide range of different environmental conditions, which may have induced a number of genetic adaptations.
Recent evidence suggests that the observed phenotypic diversity among human population groups may to some extent be a product of local adaptive processes (e.g. reviewed in [2]), driven by regional variation in pathogen environment, diet, or climate [3]. For example, both genome-wide scans and studies on candidate loci identify genes under selection associated with skin pigmentation, presumably due to the different needs for skin protection in regions with different UV radiation intensity [4]-[8]. A number of genomic scans for loci under selection have been conducted in humans, using methods based on the distribution of SNP allele frequencies, [8]-[10], haplotype structure [4], [7], [11], [12], strength of population subdivision [13]-[15] or a combination of multiple measures [16]. These scans all attempt to identify the signature of a recent selective sweep (the effect of an advantageous mutation as it increases in frequency in the population). Most of these methods have reasonable power to detect a ‘hard sweep’, i.e. a sweep caused by a single new advantageous mutation affected by strong selection. However, they do not identify the underlying environmental factors (if any) that induced the selection acting on the target gene. Additionally, it has been recently suggested that most selection in humans may not be caused by hard sweeps, but rather by selection acting on standing variation in many genes (‘soft sweeps’) [17], [18]. Much evidence for selection may have been missed by focusing strongly on hard sweeps.
A promising alternative strategy for elucidating signatures of human local adaptation, especially when individual beneficial variants have a weak phenotypic effect, is to identify polymorphisms that strongly correlate in frequency with environmental variables [19]. Indeed, heat adaptation in human populations has been shown to correlate with latitude, precipitation and temperature [20]. Based on this observation Young and colleagues hypothesized that past adaptation to climate may be the main cause for the current widespread susceptibility to hypertension [20]. More recently, a scan for selection in candidate genes involved in metabolic disorders, suggested increases levels of positive selection in these disease pathways due to adaptation to local climatic conditions [21]. Similarly, signatures of adaptations to local dietary specializations have been observed [22].
Infectious diseases are one of the most important causes of mortality in human populations. Polymorphisms associated with response to infectious diseases are, therefore, likely targets of selection. Human genetic adaptation may to a large extent be driven by response to microbial, viral or parasite presence. Indeed, numerous studies have identified immune- and defense-related genes targeted by positive selection in the human genome [23]-[37] (reviewed in [38]).
By correlating population allele frequencies with local pathogen diversity, several studies have argued that pathogen-driven selection have been an important force in local adaptation in the MHC class I loci [39], blood group antigen genes [25], and interleukin genes and their receptors [27]. Also, genome-wide scans of adaptation to pathogens identified gene networks correlated with specific pathogen species such as viruses [40], protozoa [41] and helminthes (parasitic worms) [42].
Despite these observations, there is still great uncertainty about the relative importance of the role of pathogens and other ecological factors as selective agents in local adaptive process. Similarly, the degree to which adaptation to infectious agents or to other environmental factors has shaped the distribution of complex-disease alleles in humans is still under debate [24], [27], [38], [43]-[45].
The objective of this study is to identify signatures of human genetic adaptation to local environments, separating the contributions of different environmental factors such as climate, subsistence, and pathogenic environment. We show that the latter factor is the strongest driver of local adaptation, and identify specific pathways in the immune system, and specific disease susceptibility alleles, affected by selection related to the local pathogenic environment.
We first identify possible signatures of human genetic adaptation to local environments via a statistical framework based on exploring correlations between population allele frequencies and environmental variables. Our major goal is to determine the relative contributions of different environmental predictors in driving local adaptation.
Its is often assumed that natural selection is more likely to act on genic rather than non-genic polymorphisms because the former are more likely to be of functional significance. In line with this assumption, previous studies have found a significant enrichment of genic polymorphisms among SNPs with high levels of population genetic differentiation [14], [46].
We verified these observations by correlating population allele frequencies of nearly 500k SNPs genotyped in 55 distinct human populations to a set of 14 environmental variables describing each geographic location (Table S1, Table S2, Table S3). By applying a Projection to Latent Structure multiple regression with an Uninformative Variable Elimination algorithm (UVE-PLS) we computed the prediction accuracy Q2 for each SNP, and used this as a measure of how well the environmental variables predict distributions of allele frequencies. Q2 serves here as a measure of genetic differentiation among populations, but instead of using geographic distances, or implicitly assuming an equal weight of all populations, population genetic differentiation is measured relative to the defined environmental variables. A high Q2 value indicates that populations which are very different in terms of environmental variables also are very different in terms of allele frequencies.
We examined the relative abundance of genic versus intergenic SNPs in the upper tail of the distribution of Q2 values, as in the study by Coop et al. [46]. Significance and confidence intervals were determined using the Moving Block Bootstrap (MBB) estimates (see Materials and Methods). Notably, we found an enrichment of genic SNPs compared to non-genic SNPs for high values of prediction accuracy (Figure 1A, Figure S1A), suggesting the action of natural selection in driving the differential allele frequency distribution among human populations. Indeed, for the highest examined bin of Q2 (75-87.5%) the median value for the re-sampled distribution of the enrichment statistic (see Materials and Methods) was equal to 1.065, which was found to be significantly larger than 1 (p<0.05).
Genic and non-genic SNPs differ in a number of different ways, most importantly in their average allele frequencies and level of linkage disequilibrium. We therefore examined directly if the observed excess could be explained by a difference in the distribution of allele frequencies, levels of population differentiation (measured as FST) or recombination rates (and therefore linkage disequilibrium) between genic and non-genic SNPs, by directly comparing non-genic and genic SNPs with similar minor allele frequency (MAF), FST and recombination rates. We observed an enrichment of genic SNPs for the highest bins of prediction accuracy in almost every classes of equal MAF, FST, or recombination rate in which we divided our sample of SNPs (Table S4). This suggests that the enrichment of genic SNPs for high values of prediction accuracy is not affected by these confounding factors.
We also observed an even stronger excess of non-synonymous vs. non-genic SNPs at increasing values of prediction accuracy (Figure 1B, Figure S1B). The median value of the re-sampled distribution for the highest interval of prediction accuracy (62.5-75%) was 1.123 which is significantly greater than 1 (p<0.05). There was a similar enrichment of non-synonymous vs. synonymous SNPs for high values of Q2 (the median value of re-sampled distribution was 1.031), but this value was not significantly different from 1, possibly due to the small number of SNPs and linkage between non-synonymous and synonymous SNPs.
We next tested whether different environmental variables differed in their contribution to allele frequency differences among populations. Using the same statistical framework, we repeated the multiple regression analysis separately using only climate, subsistence strategies or pathogen predictors for each SNP. The results show that there is a greater abundance of genic compared to inter-genic SNPs for high levels of Q2 for each of the three variables (median values at last Q2 bin are equal to 1.091, 1.027 and 1.034 for pathogens, subsistence or climate variables, respectively) (Figure 2). Genic enrichment at the highest bin of prediction accuracy (62.5-75%), computed by modeling the relationship between allele frequencies and environments including only pathogens predictors, is the highest among the examined factors and it is significantly greater than 1 (the lower bound of the 95% confidence interval is equal to 1.002), while the enrichment for the two other predictor classes were not significantly larger than 1.
Temperature and precipitation rate ranges have been shown to influence the biological diversity and distribution of pathogen species [47]. Nevertheless, when considering annual temperature range and annual precipitation range rather than mean levels as climate variables, we still did not observe an enrichment of genic SNPs for high values of prediction accuracy (median value at last Q2 bin is 0.974).
Our next goal was to identify the relative fraction of loci for which population genetic variation is significantly correlated with specific environmental factors, and to use these results to further elucidate the role played by different environmental variables in shaping human variation.
An unusually high correlation between allele frequencies and environmental variables may help identify loci involved in local human adaptation. However, these correlations are strongly affected by the non-independence of allele frequencies between closely related populations [48]. One method for circumventing this problem would be to estimate parameters of an explicit demographic model that describes the distribution of allele frequencies among populations. Unfortunately, in our case this is not computationally feasible because of the large number of populations. Instead, we assessed the relationship between the population genetic distances of each gene with at least one genotyped SNP, and a distance matrix of environmental variables via partial Mantel correlations [49], while statistically correcting for the genome-wide allele frequencies differences (Figure S2) (see Materials and Methods for further details).
As expected, most of the genetic distance variance is explained by population demography. On average, the overall population genetic distance explains more than 95% of genetic variation for most of genes. However, the average improvement of explained variance I(R2), a measure of the relative importance of each environmental variable in explaining the distribution of allele frequencies for a particular gene (see Materials and Methods), is about 1.5% for pathogen or subsistence factors, and about 0.5% for climate (Table S5). We observed a non-negligible fraction of genes (outliers in the distribution) showing highly elevated values of I(R2), with some values as high as 15% (Figure 3A, Table S5). Again, such extreme values are more common for pathogens and subsistence factors than for climate, using either temperature or precipitation rate mean levels (Figure 3A) or range levels (Figure S3A).
Other environmental variables, or other quantifications of the environment, could potentially provide different results. These analyses may not have captured the main factors affecting fitness when quantifying the environment. Nevertheless, when testing each variable individually and assigning the maximum I(R2) within each environmental category, we still more frequently observed high values for pathogens and subsistence factors than for climate, using either temperature or precipitation rate mean levels or range levels (Figure 3B, Figure S3B). We used a permutation procedure described in the Materials and Methods section to determine statistical significance of the I(R2) values.
Examining the distribution of p-values, which assesses the strength of the evidence against a model in which the environmental variables do not affect allele frequencies, we observed a strikingly larger number of loci with allele frequencies significantly associated with the pathogen distance matrix (Table S5). Indeed, 103 genes show a significant I(R2) value when considering the pathogen distance matrix (corrected p-value <0.05) while no genes were detected when considering the subsistence distance matrix or the climate distance matrix, using either temperature or precipitation rate mean levels or range levels. Again, when testing each variable individually, we observed a larger number of genes significantly correlated (corrected p-value <0.05) with at least one pathogen variable (229), rather than one subsistence (10) or one climate variable (9).
We also applied the Bayenv software (see Materials and Methods) to our data set and compared the results to the ones obtained using the Mantel test procedure. In general, the results are very consistent (Table S6) and the two statistics (improvement of explained variance and ranked Bayes factors) are highly correlated when testing each environmental category separately (all Spearman's rank correlation coefficients being positive, ranging from 0.1 to 0.3, with p-values lower than 1e-5). More importantly, among 2120 genes having at least one SNP correlated with pathogens using Bayenv, 419 are highly correlated using our method (showing an uncorrected p-value lower than 5e-03) and this overlap is significantly greater than expected by chance (χ2 test p-value <0.01).
When applying the Bayenv software to our data set, the number of genes having at least one SNP showing an extreme Bayes Factor value for at least one climate variable is slightly higher than the number obtained using pathogen variables (Table S6). This apparent contradictory result may be due to the fact that Bayenv tests each variable at each SNP for each locus. Therefore, for each environmental category it uses the maximum of as many different values as there are typed SNPs multiplied by the number of variables within each environmental class. Indeed, genes correlated with subsistence strategies or climate conditions (ranked Bayes Factor >0.995) show a significantly greater number of tested values for each locus than genes correlated with pathogens (one-side Wilcoxon rank sum test p-values of 3.62e-05 and 3.27e-06, respectively), while no difference is observed when comparing subsistence and climate variables (two-side Wilcoxon rank sum test p-value = 0.62). This suggest that the discrepancy between the results using our method and the results using Bayenv is caused by the larger number of tests carried out for subsistence and climate factors in the Bayenv analysis (Table S6).
A correlation between the strength of ascertainment bias and a bias in pathogen reporting may potentially affect our results and lead to an inflation in the importance of pathogen-driven selection. To investigate this possibility, we validated our finding using low-coverage new-generation sequencing data from 1000 Genomes Project [50]. Specifically, we correlated genetic variation of more than 1,500 genes located on chromosome 1 from 9 distinct human populations with our set of environmental variables, controlling for demographic effects (see Materials and Methods). Again, we observed a fraction of genes showing highly elevated values of I(R2) and such extreme values were more common for pathogens factors than for subsistence or climate predictors (Figure S4). Assessing statistical significance, we again observed a larger number of loci with allele frequencies highly correlated (uncorrected p-value <0.05) with the pathogen distance matrix (109), while only 0 and 11 genes were detected when considering the subsistence distance matrix or the climate distance matrix, respectively. Among the 109 genes correlated with pathogens using sequencing data, 53 exhibited a previous p-value, computed using genotype data, lower than 0.05 and this overlap was statistically significant (χ2 test p-value = 1.24e-07). The latter result suggests that, despite the different number of populations and the difference in the genetic data analyzed, the results are qualitative concordant suggesting that a correlation between ascertainment bias and pathogen reporting cannot explain our results.
It is possible that different groups of pathogens have differed in their impact on human local adaptation. To test this hypothesis, we recomputed the partial Mantel correlations for each gene using different environmental matrices relating to different aspects of pathogen diversity by removing one pathogen group (viruses, bacteria, protozoa and helminthes) in the calculation of the environmental distances. By comparing the distribution of I(R2) for the model with all pathogen species (I(R2)FULL) to the ones missing one of the pathogen groups (e.g. virus diversity, I(R2)w/o VIRUS) it is possible to evaluate the impact of the missing pathogen group on the relationship between genetic variation and pathogen diversity. The reduction in I(R2) when not considering a particular pathogen group provides a measure of the relative impact of this particular group in explaining local adaptation. QQ plots showing the difference between including all pathogens and dropping one of them are illustrated in Figure 4. Clearly, removing helminth diversity from the model leads to a drastic decrease in the distribution of I(R2) (Figure 4). Conversely removing virus diversity from the pathogen distance matrix results in an apparent increase in I(R2), presumably because the matrix then is more strongly dominated by the helminth distances.
The above results demonstrate that pathogens diversity is a major factor in human adaptation to local environments. We further examined the genes whose genetic variation is correlated with the pathogen diversity matrix using analysis of Gene Ontology (GO), by testing whether certain ontology terms are enriched for SNPs correlated with pathogen diversity.
By investigating statistically enriched terms in our set of pathogen-associated SNPs, we found an over representation of shared GO terms relating to regulation of immune system, defense and inflammatory response (Table 1, Table S7). In addition to these more general ontology categories, we found enrichment in categories related to mechanisms involving a direct response to external agents or host-pathogen interaction (e.g. response to wounding, JAK-STAT cascade, antigen binding, oxidoreductase activity, endosome).
We further interrogated the KEGG PATHWAY database, focusing our attention on hierarchical categories related to the immune system or immune diseases. We identified 3 KEGG pathways showing a significantly higher than expected number of pathogen-associated genes as determined by a bootstrap procedure (Table 2). Two of the identified KEGG pathways are related to immune-related signaling processes (Leishmaniasis pathway and Toll-like receptor signaling pathway), while the remaining one involves allograft rejection.
It has previously been argued that SNPs associated with susceptibility to complex diseases, or other important phenotypic traits, are more likely to be targets of natural selection than random genes [27], [38]. To test this hypothesis, we extracted a collection of more than 2,000 SNPs which have been associated with specific phenotypic traits and/or diseases and typed in on our panel from the GWAS database (www.genome.gov).
To elucidate the biological impact of the inferred local selection, we compared -log10 of the I(R2) p-values in a set of 770 genes containing at least one significant SNP from a GWAS study. Genes were divided into three categories depending on whether the SNPs were associated with an autoimmune disorder, another disease, or a quantitative trait. We use the p-values as more appropriate measures of the strength of the evidence rather than the I(R2) themselves. Genes associated with autoimmune diseases show a clear increase in the proportion of genes with low I(R2) p-values (large -log10 values; Figure 5).
We then investigated which autoimmune diseases more often have been targeted by natural selection. Similarly to our previous Gene Ontology analysis, we identified celiac disease (susceptibility), ulcerative colitis (susceptibility), multiple sclerosis (susceptibility, severity or age of onset), and type 1 diabetes (susceptibility) (Table 3), as the most common disease categories.
In recent years, great efforts have been made to assess the role played by natural selection during human evolution [2], [3], [17], [19], [46], [48]. Genome-wide scans for recent positive natural selection identified a putative list of non-neutrally evolving genes involved in specific biological pathways including metabolism, immune function, and skin pigmentation [4], [7]-[16]. These findings suggest that selective pressures related to adaptation to local environmental conditions might have contributed in shaping human genetic variation.
Here we developed a statistical framework for identifying signals of adaptation to local environments. We correlated the spatial allele frequency distribution of a large sample of SNPs, genotyped in more than 50 populations distributed worldwide, to a set of environmental factors, describing local geographical features such as climate conditions, diet regimes (measured as subsistence strategies) and pathogen loads.
Previous studies have shown that SNPs with an increased degree of population genetic differentiation (measured using FST or other statistics) are enriched for genic SNPs [14], [46]. Our analyses confirm these observations by finding a significant enrichment of genic SNPs, in particular non-synonymous SNPs, vs. intergenic SNPs for high values of the regression prediction accuracy, Q2 (Figure 1). Q2 provides a measure of genetic differentiation among populations relative to the defined set of environmental variables.
Interestingly, the enrichment of non-synonymous SNPs is quantitatively greater than the enrichment of genic variants, in line with the hypothesis that a larger fraction of non-synonymous SNPs has direct functional effects.
We can exclude the possibility that this enrichment is explained by different distributions of recombination rates, allele frequencies or FST between genic and intergenic SNPs, as the enrichment is also apparent when stratifying according to MAF, recombination rate or FST.
It is worth noting that the SNP data analyzed here suffer from an ascertainment bias, owing to the protocols used for selecting SNPs for the genotyping platforms. One of the main effects of the ascertainment bias is a shift toward common variants with non-negligible consequences on various statistics, such as measures of population structure [51]-[54]. However, as previously mentioned, the results hold up even when stratifying with respect to MAF and FST, suggesting that ascertainment biases, which primarily affect the data through the allele frequency, do not have a strong effect on our results.
Overall these results strongly indicate that the enrichment of genic and nonsynonymous variants among SNPs with a high value of Q2 may truly reflect the action of natural selection.
Importantly, we find a quantitatively higher, and statistically significant, enrichment of putative functional SNPs for high values of Q2 for models comprising pathogens as predictors rather than climate or diet (Figure 2), even we testing for additional climate variables such as temperature and precipitation annual ranges. Although all the environmental factors we have investigated contribute to Q2, our results suggest that pathogens are a more important driver of local adaptation than other factors explored in this paper.
To further investigate this issue, we computed partial Mantel correlation between the locus-specific population genetic distance and three different matrices describing pathogen load, diet regimes or climate conditions. In doing so, we used the average distance of allele frequencies as a covariate to control for background demographic processes. As expected, most (approx. 95%) of the variance in allele frequencies among populations can be explained by non-adaptive processes. Nonetheless, we were able to identify a non-negligible contribution of selection. Several loci showed large values (>15%) in the improvement of explained variance I(R2), when adding a specific environmental matrix (pathogen, diet or climate; see Materials and Methods; Figure 3, Figure S3, Table S5).
Genes with a statistically significant I(R2) are likely targets of local selection because I(R2) measures the increase in explained variance by an environmental factor when taking average distances among populations into account. In particular, there is a strikingly larger number of genes significantly correlated to the distance matrix describing pathogen diversity compared to the ones related to climate conditions or diet regimes. A total of 103 genes are significantly correlated in frequency with pathogen predictors while none correlates with climate or subsistence strategies. This predominant role of pathogen-driven selection in the human genome is confirmed when testing each variable within each environmental category separately (229, 10 and 9 genes significantly correlated in frequency with at least one pathogen, subsistence and climate variable, respectively).
Furthermore, we validated our results using low-coverage sequencing data for a smaller set of SNPs and populations, ruling out the possibility that ascertainment bias coupled with a bias in reporting pathogen diversity may lead to the observed prevalence of pathogen-driven selection. We should add that other factors could affect local adaptation than the factors examined here. The quantitative measures used here may not be the ones that correlate most closely with the components of the environment that affect fitness. Other measures of local climate or subsistence, that include variables not examined here might show a stronger effect on local adaptation. However, among the quantitative measures of environmental factors explored here, it is clear that pathogen load has been the most important factor shaping human genetic diversity.
It is perhaps not surprising that selection related to pathogens appears to be the most dominating driver of local adaptation, given the number of studies reporting pathogen related selection in humans, including selection on proteins used by pathogens to infect cells (such as certain blood group antigens [25], [55]), pathogen receptors (such as the TLR family [30], [31] and glycosylated extracellular membrane proteins [56]) and selection on genes product directly involved in immune/defense response to pathogens (e.g. [26]-[28], [33], [57]-[62]).
Infectious diseases have represented, and still represent, one of the major causes of death for human populations, especially in developing countries [63], [64]. Not surprisingly, genes responsible for heritable variation in the response to pathogens are likely targets of natural selection.
It may be more surprising that the pressure imposed by parasitic worms (helminthes) on human genes has been stronger than the one due to viral, protozoa or bacterial agents (Figure 4). Perhaps this is due to the fact that helminthes evolve slower than unicellular/viral agents and that they often have complex life cycles which results in a relatively stable geographic distribution [65]. Evolutionary changes in the helminthes, therefore, occur at a similar time-scale to that of humans, allowing for a true co-evolutionary interaction between humans and the pathogen. Faster evolving species (e.g., viruses) may perhaps not exert the same selective pressure for long enough time to induce a sufficiently strong change in allele frequencies.
We identified signatures of pathogen-mediated selection in 103 distinct human genes. Overall, genes highly correlated with pathogen diversity show a significant enrichment of immunity related functions, according to Gene Ontology analysis (Table 1). Again these findings strongly suggest that the candidate loci we detected truly are targeted by natural selection due to adaptation to pathogens.
Among 103 genes targeted by pathogen-driven selection, 23 are directly related to immunity processes, according to ImmPort database (www.immport.org). These genes encode signaling molecules involved in the inflammatory response, such as IL6, LRRC19, and PON2, cell surface proteins participating in immune functions, such as ADAM17, ITGAL, and LAG3, and signal transducers of the innate and adaptive immune response such as MYD88 (Table S8). In particular, ADAM17 has been shown to be involved in viral entry and to participate in intestinal inflammation triggered by Toll-like receptors (TLRs). In addition to ADAM17, we have identified 9 other genes with high I(R2) values when using pathogen diversity as covariate that also participate in the Toll-like receptor signaling pathway (Table 2). One of these genes, MYD88, encodes a cytosolic adapter protein central for the transduction of the immune response. This protein is implicated in sensing retroviral infections by endosomes [66]. MYD88 is also implicated in the immune response to Bacteroides fragilis [67], Plasmodium berghei [68] and helminth infections [69]. Several of the 23 immunity-related genes with high I(R2) values have previously been reported to be related with pathogen infection, mainly to bacterial infections and viral infections (Table S8).
Interestingly, the two enriched signaling pathways we identified relate to two very different categories of immune response and they function in the defense against different pathogen groups (Table 2). Toll-like receptors (TLR) are molecules involved in the innate immunity and account for the first-line defense against viruses, bacteria, fungi and protozoa (reviewed in [70]), although previous studies have demonstrated the TLR-mediated signaling pathway is also important for resistance to helminthes in mice (Schistosomal-derived lysophosphatidylcholine is involved in eosinophil activation and recruitment through Toll-like receptor-2-dependent mechanisms).
While different TLRs have previously been shown to be targets of natural selection [30], our data indicate that pathogens have also exerted a pressure on genes that impinge on the cellular pathways associated with these receptors.
The second signaling pathway enriched with 13 genes targeted by pathogen-driven selection genes is Leishmaniasis (Table 2). Leishmania are obligate intracellular parasites (protozoa) that produce diseases in humans and mice. When associated with malnutrition, Leishmania infection can produce extremely serious symptoms, and a recent WHO survey indicates that epidemics of visceral leishmaniasis can lead to massive deaths in affected areas (http://www.who.int/leishmaniasis/). Thus, the parasite is likely to have exerted a strong selective pressure during human evolutionary history.
Dendritic cells (DC), sentinels of the immune system, detect Leishmania in vivo. It has been shown that MyD88-dependent receptors are implicated in the direct recognition of Leishmania by DC [71], [72], pointing again to MyD88 as an important element in host-pathogen recognition.
Genes related to immunity and inflammation regulation are known to be common targets of natural selection [38]. In particular, recent reports have suggested that a portion of susceptibility alleles for autoimmune diseases might be maintained in human population because they confer increased resistance against infection [27], [38], [43]. The identification of several autoimmune disease-related genes as target of natural selection may be consistent with the hygiene hypothesis [73]. This model states that humans have adapted to a pathogen-rich environment that no longer exists in industrialized societies. This change has reduced the exposure of the immune system to antigens, causing an overreacting immune response which favors the development of chronic inflammatory conditions [73].
Indeed, our data indicate that SNPs with allele frequencies that correlate highly with pathogen variables are enriched for GWAS SNPs associated with autoimmune diseases (Figure 5). Specifically, among our candidate genes we identified several loci that have been associated with celiac disease, ulcerative colitis (UC), type 1 diabetes (T1D), Crohn's disease (CD), and multiples sclerosis (MS) (both susceptibility and disease severity) (Table 3). Signatures of natural selection at risk alleles for celiac disease, UC and CD have previously been described [27], [38], although these variants were located in genes different from the ones we describe herein. Conversely, only a minority of genes involved in the susceptibility to T1D and MS have been described as possible selection targets [38], although a certain degree of overlap among genes involved in MS pathogenesis and loci subjected to virus-driven selection has previously been noticed [40]. Therefore, our data further support the notion that natural selection has contributed to shaping the pattern of genetic variability relating to this common disorder.
Hancock and colleagues recently performed a genome-wide scan for selection signals by detecting SNPs strongly correlated in frequency with climate [74]. They investigated genetic variation in a similar set of populations, and a similar data set of genotyped SNPs as this study. They retrieved a number of SNPs putatively subjected to climate-mediated selection, while we found only weak signals for genetic adaptation to climate conditions. There are several possible reasons for this apparent discrepancy. First, Hancock and colleagues' and our method are intrinsically different both in the analyzed elements (SNPs rather than genes, respectively) and in the approach to detecting significant signals (extreme Bayes Factors versus p-values, respectively). Most likely, our criterion for selecting extreme genes is more conservative than the one used by Hancock and colleagues. However, when applying their approach to our data set, we retrieved a significant overlap of genes correlated with different environmental factors (Table S5, Table S6). These observations suggest that the two studies, although examining different climate variables in a different sample of populations, lead to concordant results. Second, they found evidence of selection for SNPs located in immune-related genes or previously associated with autoimmune diseases and inflammatory conditions. As stated by authors themselves, it is likely that the selective pressure imposed on these genes is related to pathogen resistance/susceptibility [74], which is in agreement with our main results.
A major assumption in this study, is that the number of different pathogen species (pathogen richness or diversity) transmitted in a given geographic location is a good estimate of the pathogen-driven selective pressure for populations living in that area [25], [39]. Indeed, there is evidence that pathogen richness is a suitable and more effective measure than standard epidemiological parameters (like prevalence or mortality) for estimating the selective pressure exerted by infection agents, and that it better captures the signatures left by adaptation to specific pathogens throughout recent human evolution [27], [40], [41]. It is worth noting that our measure of pathogen evolutionary is noisy, discrete, possibly affected by report biases and calculated on a country level.
More accurate worldwide epidemiological data, as well as more detailed description of diet regimes for human population, are required to obtain a clearer picture of the effect of genetic adaptation to pathogen load or subsistence strategies, especially when comparing with adaptation to climate.
However, any inadequacies of the statistics we use to measure pathogenic environment will lead us to underestimate the role of the pathogenic environment in human local adaptation. Perhaps pathogen related selection plays an even stronger role in human evolution than what has been evidenced in this study and in previous studies.
We investigated the spatial distribution of allele frequencies using genotype data for 55 distinct human populations, comprising more than 1,500 individuals, by joining data from the Human Genome Diversity Panel (HGDP-CEPH) [75], [76] and from HapMap Phase III [77], not considering admixed populations (Table S1).
A total of more than 500k SNPs were analyzed after removing those not covered in both panels and those located on sex chromosomes (Table S2). We used the folded frequency spectrum of genotyped SNPs to quantify allele frequencies.
Two categories of SNPs were considered: genic and intergenic. SNPs were defined as genic if they were located in transcribed regions or were no further than 500 bp upstream transcription start sites. SNPs were defined as intergenic if they were located in a region larger than 100 kbp containing no annotated gene, according to USCS Genome Browser database of gene predictions based on data from RefSeq, Genbank, CCDS and UniProt (http://genome.ucsc.edu). In case of multiple isoforms, the longest transcript was used. A total of 225,502 and 216,151 polymorphisms were classified as genic and intergenic, respectively. A total of 14,804 genes containing at least one genotyped SNP was retrieved.
Data from the 1000 Genomes Project [50] were retrieved from the dedicated website (http://www.1000genomes.org/). Low-coverage SNP genotypes for each one of the nearly 1,5 K analyzed genes located on chromosome 1 were organized in a MySQL database. Populations from countries not included in the HGDP-CEPH panel were excluded. A total of 727 unrelated individuals belonging to 9 distinct populations located in 8 different countries were analyzed. A set of programs was developed to retrieve genotypes from the database and to analyze them according to selected populations. These programs were developed in C++ using the GeCo++ [78] and the LibSequence [79] libraries.
We defined a set of environmental variables for each country from which SNP data were available. Previous studies examining signatures of adaptation to local climates, selected specific variables to represent the physiological effects of different climates on humans [21]. Similarly, adaptation to diet has been investigated by examining correlations between genetic variation and different subsistence strategies among populations [22]. Finally, previous studies have suggested that pathogen diversity (i.e. the number of the different pathogen species transmitted in a given geographic location) is a reasonable proxy for the selective effects exerted by pathogens in an area [25], [39]. Based on these previous studies, we chose a total of 14 environmental variables as proxies for local environmental conditions (Table S3), including climatic and geographic factors (distance from the sea, mean annual temperature, mean annual precipitation rate, mean annual relative humidity, mean annual short wave radiation flux), subsistence strategies (relative amount of human activity spent in agriculture, animal husbandry, fishing, hunting, gathering) and pathogen diversity (number of different species of viruses, bacteria, protozoa, helminthes). Additionally, we used annual temperature range and annual precipitation rate rather than annual mean levels as supplemental climatic factors.
We obtained climatic data from the NCEP/NCAR Reanalysis Project database (http://www.esrl.noaa.gov/psd/data/reanalysis/reanalysis.shtml) by averaging annual values across the last 50 years. Subsistence strategy data were collected from Murdock's Ethnographic Atlas (1967): for each population we retrieved the percentage of activity spent in each of the examined subsistence activities. For three populations we could not assess an unambiguous set of subsistence values. The number of different pathogen species (pathogen richness) was retrieved from the Gideon database (http://www.gideononline.com/), as in references [25], [27], [40]-[42]. Cases of transmission due to tourism and immigration were not included; thus, only species that are transmitted within the countries were included. However, species that recently have been eradicated, for example, as a result of vaccination campaigns, were recorded as present in the country.
We applied a Projection to Latent Structures (PLS, also known as Partial Least Squares) multiple regression [80] to model the relationship between population allele frequencies of each SNP and a matrix describing environmental factors. This algorithm can handle highly correlated predictors and can effectively separate the weight of each predictor in the multiple correlation even in case of strong collinearity among variables (which is likely to be the case for environmental factors) [80], [81]. For the model including all the environmental factors we applied an Uninformative Variable Elimination algorithm [82] before the regression. In this way we could greatly increase prediction accuracy by not considering predictors with very low regression coefficients.
For each SNP we assessed the relationship between population allele frequencies matrix (F) of dimensions 55x1 and environmental predictors matrix (M) of 55x14 dimensions. F describes minor allele frequency at each population for the examined SNP, whereas M describes all the 14 environmental variables for each population. For each regression we computed the cross-validated prediction accuracy (Q2), estimated by a leave-one-out procedure, as:where PRESS is the Predictive Residual Sum of Squares calculated from the models obtained on the reduced data of the leave-one-out procedure, and SS is the Sum of Squares of F corrected for the mean. Formally, PRESS is the sum of squares of the differences between observed and predicted responses:and thus is a measure of the predictive ability of the model, whereas SS is computed as:where N is the number of observations (55 populations in our case), is the predicted response at the ith population and is the mean value for the response matrix (see [80], [83], [84] for further mathematical aspects of PLS regressions and model parameters estimation).
Q2 provides a measure of how well a model predicts the observed data using a cross-validation procedure, based on a partitioning of the sample into complementary subsets of observations. Iteratively each of the partitions are treated as a training set and the level of fit of the model is computed by using the remaining partition as a validation set. In our case, Q2 measures how well a model with environmental variables as covariates predict the observed distribution of allele frequencies among populations. We use it as a measure of population subdivision along the axes defined by the environmental variables. If populations which differ strongly in terms of environmental variables also differ strongly in terms of allele frequencies, Q2 will be large. If allele frequencies do not covary with the environmental variables, Q2 will be small.
We divided the distribution of Q2 values into bins, and for each bin we calculated the relative enrichment of genic, non-synonymous and intergenic SNPs, as previously proposed [46]. Enrichment is here defined as the proportion of genic/non-synonymous (or intergenic) SNPs in the bin divided by the total proportion of genic/non-synonymous (or intergenic) SNPs. We applied a Moving Block Bootstrap (MBB) re-sampling procedure to correct for the non independence among loci taking into account the possibility of an increased level of linkage disequilibrium (LD) near the positively selected site [85], [86]. The MBB method consists of drawing blocks of fixed length uniformly at random and with replacement and joining them to form a new sequence. Standard deviations are estimated from the distribution of bootstrap values. Critical values used in hypotheses testing were obtained directly from the quantiles of the distribution. We performed this procedure 1,000 times for each chromosome separately by creating (n+b-1) overlapping blocks (where n equals to number of polymorphic sites and b equals to block size) and drawing n/b blocks. Block sizes were set to 40 and 100 contiguous SNPs corresponding to segments of approx. 200 k bp and 500 k bp on average, respectively. Bins with less than 100 SNPs were merged with the immediately lower bin. The results were qualitatively similar using 40 and 100 SNPs, and only the results for 40 SNPs are discussed in the main text. The results for 100 SNPs are also provided in Figures S1, S2, S3, S4.
FST, a measure of population genetic difference, was estimated as previously proposed [87]. Recombination rates were obtained from the UCSC Genome Browser (table ‘recombRate’) as estimates computed in 1 M bp intervals based on the deCODE maps [88].
Nearly 15,000 genes containing at least one genotyped SNP were retrieved, according to UCSC Genome Browser database (http://genome.ucsc.edu). For each retrieved gene we computed population genetic distances, as proposed by Reynolds and colleagues [89]. Distances between the environmental values in different locations were calculated as Euclidean distances, which correspond to the square distances between the two vectors of environmental variables. The latter have been scaled to unit variance to ensure that the dissimilarity matrices are not to be dominated by the variables with the largest variance. The dissimilarity matrices were then obtained by averaging distances over each variable within each environmental category (pathogen diversity, subsistence strategies and climate conditions).
For each gene we assessed the relationship between the locus-specific population genetic distances matrix (Y) and the distance matrix for environmental variables (X) via partial Mantel correlations [49]. Each row or column in X and Y correspond to a population. Y contains locus specific population genetic distance values and X contains environmental distances. Partial Mantel correlations are a non-parametric statistical procedure for quantifying association between two distance matrices, while controlling for the effect of a third distance matrix. The latter independent distance matrix, Z, is here the overall population genetic distance among populations [89] computed over all loci. As shown in Figure S2, the Z matrix reflects the general patterns of human population structure.
Our procedure, therefore, accounts for the non-independence of populations and controls, at least in part, for the correlations caused by standard neutral demographic processes (Figure S2). For each variable we calculated the improvement of explained variance [90], here called I(R2), and used this as a measure of the relative importance of each environmental variable in explained the distribution of allele frequencies for a particular gene.
I(R2) is calculated as:where R2 is the explained variance of the model, estimated as:
rYZ is the correlation coefficient between Y and Z, while rXY/Z is the partial correlation coefficient between X and Y given Z defined as:
Again rXY is the correlation coefficient between X and Y while rXZ is the correlation coefficient between X and Z.
Statistical significance was assessed by permuting rows (populations) for the dependent matrix (Y) and recalculating our statistic for the permuted data. Rows were permuted only within the same stratus (continent). P-values are then calculated as the fraction of permuted values that are greater than the observed value of the statistic. This imposes the need for a large number of permutations when estimating small p-values which is computationally expensive and unfeasible for some problems. To reduce computational time we computed approximate p-values using a previously proposed asymptotic method [91]. Briefly, p-values were computed by approximating the right tail of the distribution of permuted statistics by a Generalized Pareto distribution. Parameters of the Pareto distribution are fitted using either the asymptotic maximum likelihood method or a ‘combined method’ previously proposed [92]. P-values were corrected for multiple testing using a procedure which controls the false discovery rate under dependence assumptions [93]. For partial correlations, we considered only 52 populations to which we could unambiguously assess all environmental variables (Table S3). We performed the same procedure when using low-coverage new-generation sequencing data from the 1000 Genomes project, with the Z matrix, the overall population genetic distance among populations (see above), computed by averaging distance matrices over all loci.
We also used the method by Coop and colleagues [94] to calculate Bayes factors relating to the effect of an environmental variable on the geographic distribution of allele frequencies. For each SNP a Bayes factor is calculated, providing a measure of the increase in fit of a model with a linear relationship between allele frequencies and an ecological variable over a null model. A Gaussian model is assumed with a covariance matrix of allele frequencies among populations estimated from a sub-sample of segregating sites. SNPs which are outliers in the empirical distribution of Bayes factors may possibly be affected by local selection in response to the environmental variable. The covariance matrix was estimated on 1,000 random polymorphisms. SNPs were ranked according to their Bayes factor after dividing them in 10 classes of similar minor allele frequency. A significance threshold for ranked Bayes factors was set to 0.995.
Both algorithms use the empirical distribution of genome-wide genetic distances between populations as a null model. A major difference between the methods, in addition to the fact that one is Bayesian and the other frequentist, is that the partial correlation approach used here combines the information from all SNPs in a region, or in a gene, whereas the approach by Coop et al. [94] performs the analyses SNP by SNP. In addition, our method is non-parametric and does not rely on assumptions of normality. The degree to which either of these approaches are preferable depends perhaps on the degree to which the parametric assumptions are met by the data. Our objective here is not to compare the two methods, which were developed in parallel. However, we have included results based on the methods of Coop et al. [94] to show that our conclusions are not dependent on the specific choice of statistical method.
All computation were performed in the R environment using the following packages: vegan, pls, gPdtest, VGAM. All data and scripts used are available at http://bioinformatics.emedea.it.
Gene ontology (GO) analyses were performed with GONOME [95], an algorithm that identifies GO terms over-represented in a set of genomic position (in our case SNPs) compared to that expected in random positions. GONOME avoids biases toward GO terms linked to larger genes or genes having more genotyped SNPs and also can take into account non-independence between linked sites by setting an appropriate cluster distance value (50 k bp in our case).
Analyses of functional pathways were investigated querying the KEGG database (www.genome.jp/kegg) in the following hierarchical categories: Immune system, Immune system diseases, Neurodegenerative diseases, Infectious diseases. Enriched pathways were retrieved as having a significantly higher than expected number of associated genes using 10,000 bootstrap samples. For GO analyses we considered pathogen-associated genes to be those having an uncorrected I(R2) p-value lower than 5e-3 in order to increase the number of analyzed loci and thus statistical power.
Genome-wide association studies (GWAS) data were retrieved from A Catalog of Published Genome-Wide Association Studies [96] (www.genome.gov), updated on December 1st, 2010.
|
10.1371/journal.pntd.0004997 | Molecular Diagnosis of Chagas Disease in Colombia: Parasitic Loads and Discrete Typing Units in Patients from Acute and Chronic Phases | The diagnosis of Chagas disease is complex due to the dynamics of parasitemia in the clinical phases of the disease. The molecular tests have been considered promissory because they detect the parasite in all clinical phases. Trypanosoma cruzi presents significant genetic variability and is classified into six Discrete Typing Units TcI-TcVI (DTUs) with the emergence of foreseen genotypes within TcI as TcIDom and TcI Sylvatic. The objective of this study was to determine the operating characteristics of molecular tests (conventional and Real Time PCR) for the detection of T. cruzi DNA, parasitic loads and DTUs in a large cohort of Colombian patients from acute and chronic phases.
Samples were obtained from 708 patients in all clinical phases. Standard diagnosis (direct and serological tests) and molecular tests (conventional PCR and quantitative PCR) targeting the nuclear satellite DNA region. The genotyping was performed by PCR using the intergenic region of the mini-exon gene, the 24Sa, 18S and A10 regions. The operating capabilities showed that performance of qPCR was higher compared to cPCR. Likewise, the performance of qPCR was significantly higher in acute phase compared with chronic phase. The median parasitic loads detected were 4.69 and 1.33 parasite equivalents/mL for acute and chronic phases. The main DTU identified was TcI (74.2%). TcIDom genotype was significantly more frequent in chronic phase compared to acute phase (82.1% vs 16.6%). The median parasitic load for TcIDom was significantly higher compared with TcI Sylvatic in chronic phase (2.58 vs.0.75 parasite equivalents/ml).
The molecular tests are a precise tool to complement the standard diagnosis of Chagas disease, specifically in acute phase showing high discriminative power. However, it is necessary to improve the sensitivity of molecular tests in chronic phase. The frequency and parasitemia of TcIDom genotype in chronic patients highlight its possible relationship to the chronicity of the disease.
| Chagas disease is a neglected tropical disease caused by the parasite Trypanosoma cruzi that shows tremendous genetic diversity evinced in at least six Discrete Typing Units and massive genetic diversity within TcI. Two clinical phases exist where acute phase shows high parasitemia and chronic phase shows low and intermittent parasite dynamics. One particularity of the disease is the diagnosis, because the parasitemia is highly variable during the phases of the disease. Molecular tests allow detecting DNA of the parasite in all clinical phases. Herein, we determined the operating characteristics of two molecular tests (cPCR and qPCR) to evaluate the performance of these tests for diagnosis of Chagas disease in 708 Colombian patients. We determined the parasitic loads and DTUs to assess how is the behaviour of these characteristics in relation to the clinical phases. We found that the performance of qPCR was higher compared to cPCR and the molecular tests are a precise tool for diagnostic of Chagas disease, mainly in the acute phase. The parasitemia was higher in the acute phase compared to chronic phase and the DTU predominant in Colombian patients was TcI. The behaviour of TcIDom genotype in the chronic phase patients evidenced possible relationship with the chronicity of the disease.
| Chagas disease is a zoonotic parasitic disease caused by the protozoan Trypanosoma cruzi. It is considered a public health problem in Latin-America, where approximately 6 million people are currently infected [1]. The acute phase of the disease is characterised by usually mild fever that in a small proportion of cases can be accompanied by myocarditis and other lethal complications. Most of the patients continue through the chronic phase that is initially characterised by an asymptomatic clinical course during two or three decades, and about 30% of the infected patients will develop heart or digestive complications afterwards [2].
T. cruzi parasite shows significant genetic variability and classified into at least six Discrete Typing Units TcI-TcVI (DTUs), that present associations with the geographical distribution, epidemiological transmission cycles, insect vectors and clinical manifestations of Chagas disease [3–5]. Recent studies suggest the occurrence of an emerging clade within TcI named TcIDom which is distributed in the Americas and associated with domestic cycles of transmission and human infection [6–10]. Recently, a genotype detected in anthropogenic bats and named as TcBat has been described in Panama, Ecuador, Colombia and Brazil including a case of human infection in Colombia [11–14].
The diagnosis of Chagas disease is complex due to the dynamics of parasitemia in the phases of the disease. During the acute phase the parasitemia is high, therefore the diagnosis is performed by direct parasitological tests [15,16]. Nevertheless, direct parasitological tests are not useful in the chronic phase due to the low and intermittent parasitemias. Therefore, the diagnosis of Chagas disease in the chronic phase is determined by serological tests such as ELISA: enzyme-linked immunosorbent assay, IFA: indirect immunofluorescence assay or HAI: Hemagglutination Inhibition Test [17–19]. Recently, molecular techniques such as cPCR (conventional PCR) and qPCR (quantitative real-time PCR) have been considered as supportive diagnostic tests due to their ability to determine parasitic loads of T. cruzi in all clinical phases of the disease [20–22]. The operating characteristics of molecular tests for diagnosis of T. cruzi infection have varied according to clinical phase and technical specifications. Sensitivity for identifying chronic infection with cPCR has ranged between 22 and 75% [23,24] and in both cases with a specificity of 100%. Contrastingly, for qPCR, sensitivity has ranged between 60 and 80% [22,25,26] in chronic phase and between 88% and 100% for acute phase [25,26], whereas specificity is between 70–100% [26–28]. Sampling methods have not been always clearly stated and the role of these techniques for diagnosis of Chagas disease in the different clinical phases still remains poorly understood.
The objective of this work was to determine the operating capabilities of qPCR and cPCR targeting the satellite nuclear DNA region, compared with standard diagnosis methods for acute and chronic Chagas disease. Additionally, we evaluated the plausible associations between parasitic load and DTUs in Colombian patients from the acute and chronic phases to untangle the natural course of T. cruzi infection in terms of parasite dynamics.
All patients who attended the Colombian National Health Institute (Overall 985 individuals) seeking diagnostic tests for Chagas disease in acute (113 patients) or chronic phase (872 patients) between 2004 and 2015 were considered as potential participants. Inclusion criteria were: i. Clinical or epidemiological suspect of Chagas disease in acute or chronic phase ii. Not having received aetiological treatment for Chagas disease iii. Positive serological tests for Chagas disease (IFA, ELISA and/or HAI) iv. Adequate blood and serum samples available for performing diagnostic tests according to the clinical phase. v. Acceptance to participate and sign the informed consent.
The Technical Research Committee and Ethics Research Board at the National Health Institute in Bogotá, Colombia approved the study protocol CTIN-014-11. Participation was voluntary and patients were asked for informed written consent authorising to take blood and serum samples and access information on their clinical records.
The total sample size (N) was calculated for test binary outcomes and separately for each clinical phase: acute and chronic. Considering, n = Z2 S (1−S) d2, where for a confidence level of 95% (1- α, with α = 0.05) Z is inserted by 1.96, and a maximum marginal error of estimate, d, is a desired value for precision based on researchers judgment, and S is a pre-determined value of sensitivity [29]. Based in previous studies, for the acute phase S was pre-established at 92% and with d at 8% [25,26], whereas for chronic phase S was pre-established at 60% with d at 5% [22–26]. Then, N = n /P, being P the estimated prevalence in this specific population under study. Given this is a selected population, composed of patients with some suspicion of the infection and remitted to a reference centre, P was specified at 60% in suspected cases for both acute and chronic phases. This value was obtained as an approximation based on the laboratory records at the NHI (Bogota, Colombia). The minimum total sample sizes were then calculated as N = 74 and N = 615 for suspected cases in acute and chronic phases respectively. The tests were performed to all subjects without knowing their previous clinical status. Clinical evaluation was conducted simultaneously to all individuals as part of the study to determine health status and then to the confirmed cases to evaluate heart complications. The inclusion of participants was conducted retrospectively for the period 2004 to 2012, and prospectively between 2013 and 2015. At the end, a total of 86 suspected acute patients and 622 suspected chronic patients were included in the study (Table 1).
Operating characteristics of the molecular tests were estimated by comparing against standard diagnosis (described above). Sensitivity, specificity, positive (+LR) and negative likelihood ratio (LR-), predictive values (PV), diagnostic precision (DP), Area under the curve (AUC), and Kappa index (K) were estimated for each phase of the disease (acute and chronic), the clinical stage of chronic patients (determined and undetermined) and according to DTUs and TcI genotypes identified (TcI sylvatic/TcIDom) (S2 Appendix). Results are presented as percentages, with corresponding 95% confidence intervals (95% CI). Additionally, operational capabilities in chronic patients were calculated in two ways: the first including negative patients without risk factors since they are the true negative and the second including all the negative patients (with and without risk factors). Due to over dispersion of parasitic loads, medians and quartiles are presented. Comparisons are based on Mann-Whitney test between clinical phases, chronic clinical stages and the different T. cruzi DTUs and genotype groups identified. A p value at <0.05 was considered as statistically significant. All analysis was performed in Stata: Data Analysis and Statistical Software version 12.
Overall, 985 patients were included, 872 suspected of chronic and 113 of acute infection. General demographic characteristics are shown in Table 1. Out of the initial potential participants, 27 and 129 were excluded for incomplete samples to perform all analysis from the acute and chronic groups, respectively and 121 from the chronic group due to absence of clinical information (Fig 1). The inclusion of patients was prospective, whereas the sample collection was both retrospective (for the period 2004–2011) and prospective (for the period 2012–2015). This means that for the retrospective component the samples were part of the repository. The repository consists of 144 samples, collected between 2004 and 2011, and corresponds to serum samples stored at (-80°C). In these samples, serological tests were repeated and it was found that the results were the same that they had been reported at the time of collection of samples and molecular tests were performed. The prospective component consists of 564 samples, collected in the period between 2012 and 2015, and maintained in guanidine hydrochloride solution until processing.
In patients from the acute phase, the qPCR test was positive in 95.7% of the patients and cPCR in 84.5%. In patients from the undetermined chronic phase, qPCR was positive in 68.0% of the cases and in 55.4% by cPCR. In the cardiac chronic phase, qPCR positivity was 59.1% and 58.6% by cPCR. The positive samples for satellite nuclear PCR (qPCR and cPCR), were confirmed by kPCR. In patients that were negative by serology but with risk factors cPCR (2.6%) and qPCR (3.6%) were positive. In febrile and negative patients without risk factors both tests were negative in all samples. In all samples analyzed we detected the internal amplification control for both cPCR and qPCR, the average Ct value in all samples tested was 21.
The operating characteristics including all negatives patients of chronic phase (Negatives with and without risk factors) are presented in Tables 2 and 3.
Performance of qPCR was higher compared to cPCR in both acute (AUC 0.98 vs 0.92) and chronic phase including only negatives with risk factors (0.82 vs 0.78) (Fig 2). Likewise, the performance was significantly higher in acute compared with chronic phase and in overall a specificity higher than sensitivity particularly in chronic phase (Tables 2, 4 and 5).
Parasitic loads were determined in samples that tested positive by qPCR. Significantly different median values were detected in acute (4.69 parasite equivalents/mL) versus chronic phase (1.33 parasite equivalents/mL). A statistically median difference was also found between determined and undetermined chronic phase (Fig 3).
In samples that tested positive (n = 407) by cPCR, the DTUs TcI-TcVI and TcI (TcI Dom, TcI Sylvatic) were evaluated. The distribution of DTUs was 74.2% for TcI, 17.2% for TcII, 1.48% for TcIII, 0.5% for TcV and 6.7% for mixed infections. For the latter seven different combinations were identified: TcIDom/TcII/TcV, TcIDom/TcII, TcIDom/TcISylv, TcIDom/TcISylv/TcII, TcIDom/TcISylv/TcIII, TcIDom/TcIV, TcISylv/TcII. With respect to TcI, the genotyping was feasible in 290/302 samples. Out of them, 28.7% were classified as TcI Sylvatic and 71.4% as TcIDom. The median load parasitic value for TcII (4.68 parasite equivalents/mL) was significantly different to the one for TcI (2.87 parasite equivalents/mL) and TcIII (1.72 parasite equivalents/mL) (Fig 4).
The genotype distribution according to clinical phase evidenced that TcIDom was significantly more frequent in chronic phase compared with acute phase (Table 6). The operating characteristics of molecular tests for the different genotypes were calculated, observing that the sensitivity for identifying TcII was slightly higher than for TcI, mainly for qPCR (S1 Table). The median parasitic load for TcIDom was significantly higher (2.58 parasite equivalents/ml) compared with TcI Sylvatic (0.76 parasite equivalents/ml) in chronic phase (Fig 5).
The main limitation involved in this study is the fact that there is not a gold standard test for all clinical phases of Chagas disease. Particularly for chronic phase, the best comparators are serological tests but these techniques measure the immune response and not the relative presence of the parasite. This particular situation impacts the evaluation of new diagnostic tests. This is reflected mainly in the kappa index (Tables 2 and 4) that presented very low values in the undetermined and determined chronic phases. Unfortunately, it has not a simple solution and more understanding of the course of the infection is still needed.
The results obtained for the molecular diagnosis in acute phase were optimal in terms of sensitivity for both qPCR (95.7%; 95%CI: 88.3–98.5) and cPCR sensitivity (84.5%; 95%CI: 74.3–91.2), and same specificity. Although the results are showing a potential superior performance of the sensitivity of qPCR compared with cPCR, this difference needs a cautious interpretation. This might be explained due to the fact that detection by qPCR increases the sensitivity and specificity because of the hybridization of the Taqman probe in the amplicon, whereas in the case of the cPCR it requires a considerable amount of amplicon so that it can be observed in agarose gels [25,26] In addition, the confidence intervals were slightly overlapped, meaning that there is some indication of this difference but it is not statistically significant, so not definitive. The performance of the molecular tests in the acute phase is explained because there are large numbers of parasites, for example in cases of reactivation in immunosuppressed patients and in oral outbreaks. The values obtained for LR evinced the high probability that positive results correspond to diseased patients (LR+) and the low probability that the diseased patients present negative results (LR-). In addition, the DP was very optimal specifically for qPCR test confirming that this molecular test is very useful for the diagnosis in the acute phase, considering that the direct diagnosis is complex when the parasitemia is low (As is the case of the acute patients detected more than a month after the infection where the parasitemia normally begins to decrease due to the control of the immune response) and are required many tests for the confirmation of the acute cases (direct tests, serology tests and clinical information). Regarding the predictive power of molecular tests in the acute phase, these tests are very good predictors of the disease presence when positive results are obtained (PPV) but their performance as predictors of absence of the disease are less (NPV). However, it is worth noting that the predictive values depend on disease prevalence in the evaluated population.
The analysis of operational capabilities in the chronic phase was conducted in the first instance including only negative patients without risk factors or true negatives. For the chronic phase, qPCR sensitivity was 64.2% and 56.8% for cPCR and in concordance with previous reports obtained by qPCR that have shown sensitivity ranging from 60–80% and 20–70% for cPCR [22–24,26,28,42]. These sensitivity results may be due to low and intermittent parasitic loads during chronic phase.
The performance of qPCR was better than cPCR in the chronic undetermined phase, while that was very similar between the two tests in the determined chronic phase (Tables 3 and 5). The discriminative power of the two molecular tests was acceptable in the chronic phase. For qPCR, the AUC and DP values obtained (Tables 3 and 5) were better for the undetermined phase than for determined phase. The differences between undetermined and determined phases for qPCR of the chronic phase can be explained by the natural course of the disease, in which the parasitic load decreases while increases the infection time. This is supported by several studies showing that there is no relationship between the evolution of the cardiac form of the disease and parasitemia but it declines with time as observed in this study [43,44]. Also, some studies show that cardiac form is mainly related to different types of strains, increased parasitemia, reinfection or immune system disorders in chronic patients [45,46]. In the cPCR AUC values were the same for both phases, while the value of DP was best for the determined phase. Possibly, this is because the detection limit of the cPCR is lower than qPCR, for this reason the cPCR behaves similarly in the two phases. In the two stages of the chronic phase, there is a high probability that patients with negative results in the molecular tests have the disease (LR-) and these tests are not good predictors of the absence of the disease (NPV) (Table 5). Therefore, the use of molecular methods as diagnostic tests is not appropriate due to the better performance displayed by serology. The probability that the results are positive is high in diseased individuals with respect to healthy individuals (LR +) and the molecular tests are excellent predictors of the presence of disease (PPV). Thus, these tests could be used in situations in which the diagnosis is doubtful, allowing the confirmation of the parasite in diseased patients, which is of great importance for example when monitoring etiological treatment. However, it is necessary to improve the sensitivity, which can be performed by analysing serial samples for each patient as seen in some studies in which such sensitivity improved from 69.2% to 85.2% with the addition of a second sample or conducting DNA extraction from a larger volume of the sample [47,48].
In addition, the operating capabilities of patients in chronic phase were calculated including all negatives by serology with and without risk factors (Table 1, N = 141). It was observed in the group of negative patients with risk factors a positivity of 2.6% (3 patients) by cPCR and 3.6% (4 patients) by qPCR, possibly due to an immunosuppression issue in these patients preventing the detection of antibodies or infection. Three patients are from the department of Casanare, which is an endemic area, and five patients had less than 24 years of age suggesting a recent infection. Also, all patients reported to know the vectors and have lived during his/her childhood in homes with features such as thatched or ‘barheque’, floor or wood and/or tread walls of earth, wood or ‘barheque’. Two of the seven patients that were negative by serology and had risk factors, whose ages were 36 and 51 showed the presence of symptoms at cardiac level. In this group of 7 patients, 4 presented the ELISA absorbance values greater than 0.200 and 4 detectable titles in the IFA (1/8 and 1/16).
As the operating capabilities calculated including all negative patients, a small percentage of decreased specificity in the two platforms was observed (S3 Appendix). The positivity of these serologically negative patients that generated the decrease can probably be explained because cases of recent infection or patients with some form of immunosuppression that has generated the absence of detectable antibodies. In fact, in the group of acute patients, 4 patients whose serology was negative showed positive PCR, in these patients the detection was achieved by direct parasitological methods. Regarding the molecular techniques, given that in all PCR runs were included negative controls including reagents controls, a plausible contamination with parasite DNA is discarded. Significantly, the DP and AUC values showed no obvious changes unlike the values obtained for the NPV and the Kappa index, in which there was a marked increase. However, the changes obtained do not change the interpretation of the usefulness of the test in the clinical setting, but can show that there are few cases where serological tests may have false negatives as noted previously using cPCR by Ramirez et al., 2009 [23]. Even though serological tests are considered the best current option for the diagnosis of Chagas disease, in a meta-analysis of high quality tests their sensitivity has been estimated at 90% [49]. Given this, we believe that an improvement of diagnostic tests for Chagas disease is needed for both serology and PCR techniques. An appropriate use of the comparator as gold standard and the inclusion of different phases of the disease are crucial to understand the utility of different diagnostic tests.
To our knowledge, this is the first study to include statistical calculation of the sample, which allowed the analysis of operating characteristics of the molecular tests in all clinical phases of Chagas disease. In addition, this study is the first in analysing the two PCR platforms (qPCR and PCR) for the same target (stDNA) in patients from all clinical phases of Chagas disease. The conventional technique was included, due to the vast use of this technique in the diagnosis and its ease implementation in laboratories with restricted equipment (a Real Time PCR machine is not available) [23,24,28]. Lastly, acute patients had a less median age than chronic phase patients and in turn the largest number of acute cases are male. This possibly is because economic activity in endemic areas is developed by males that assist to the field and this facilitates direct patient contact with the vector and therefore with the parasite. On the other hand, females ratio and the median age were higher in chronic phase patients that are usually detected by screening blood banks or present cardiac abnormalities in chronic phase, then the detection occurs at a greater age. Additionally, in Colombia most blood donors are women facilitating their diagnosis.
Regarding the parasitemia, it is observed that the median parasitemia was higher in acute patients compared to chronic phase, which is expected given the dynamics of parasitemia in the disease [25,26]. As for the group of chronic patients, the herein reported median of parasitemia is similar to those previously reported for Colombia [22,26]. In addition, the difference in medians between cardiac chronic and undetermined chronic stages was statistically significant, being higher in the undetermined chronic phase unlike the findings described by Ramirez et al, 2015 [26], in which statistically significant difference was not detected. However, our results are in accordance with the natural history of the disease where parasitic loads decrease with the chronicity of the infection and this is probably associated with the type of strain and/or the immune response [2].
The DTU with highest frequency was TcI, both in acute and chronic patients, consistent with findings previously reported in Colombia [8,39,50,51]. Followed by TcII most often detected in chronic than acute patients. These findings are congruent due to the predominance of TcII in domestic cycles of transmission for the case of Colombia [50]. Regarding the parasitic loads of the DTUs detected, we observed that TcII had higher median parasitemia than other DTUs, consistent with the number of copies that has been reported in the DNA nuclear satellite region being higher for TcII than for TcI [52–54]. These findings highlight the importance of using the most representative DTU to generate the standard curves for quantification [22,25,26]. In addition, in murine models TcII shows higher parasitemias than TcI when performing individual and mixed infections [55].
In this study, acute cases are likely caused by vector transmission and possible oral route. In most of the cases TcI (TcI sylvatic), TcII and TcIII infection was observed. These findings are consistent with previously documented reports for acute patients where DTUs associated with the sylvatic cycle of transmission were depicted [4,5,40,51,56–61]. An interesting finding was the detection of TcV in the patients surveyed. This DTU has been already reported in dogs and Rhodnius prolixus from eastern Colombia but this would be the first report of TcV human infection in the country [50]. It is necessary to conduct further studies to understand the host-parasite associations of this foreseen DTU in patients from northern areas of the continent. It is well known that TcV infection is endemic in Bolivia, Brazil and Argentina but in Colombia is a novel case that requires further investigation; in fact high-resolution markers have been applied to the few isolates of Colombian TcV showing a tailored hybrid profile suggesting a Pan-American import from south America [62]. The DTU TcVI, is mainly detected in the South Cone of Latin America. Normally associated with megavisceral syndromes and some cases of congenital heart disease [4]. In Colombia, TcVI has been very rare and almost infrequent. In fact it is limited to a report in which was detected in humans and R. prolixus isolates (4% and 1.4% respectively). In addition, in different studies with a considerable number of patients conducted in Colombia it was not detected, confirming the low prevalence of the DTU in the country [39,51,63].
Recently, it has been highlighted the emergence of a genotype named as TcIDom and associated to human infection and domestic transmission cycles via different molecular markers [5,6,8,64–66]. Other studies have shown the presence of TcI Sylvatic genotype in tissue and TcIDom in bloodstream of patients with Chagas cardiomyopathy [41]. In murine models was observed that TcIDom produced high parasitemia and low tissue invasion, a process that allows an adaptation to the host prolonging its permanence and likely generation of chronicity, opposite process to what happened with the TcI sylvatic strains [67]. In accordance with these previous findings, our results show that in chronic patients the frequency and parasitemia of TcIDom genotype were significantly higher in chronic patients than in acute patients, supporting the hypothesis that this genotype may be related to chronicity in patients with Chagas cardiomyopathy.
In conclusion, the molecular diagnostic tests are becoming a precise tool to complement the standard diagnostic methods for Chagas disease. This study shows that in general qPCR has a better performance than cPCR. Also, the results confirm that PCR is highly specific for both acute and chronic clinical phases, whereas sensitivity is acceptable for acute phase but still very low for chronic patients. This situation could be partially explained by the higher parasitic loads detected in acute phase and the intermittent nature of the parasite release to the bloodstream in chronic phase. We explored for the first time in a large cohort of Chagas disease patients the DTU parasitemia and the natural course of infection. This type of studies is required in Latin-America for a better understanding of disease progression and molecular epidemiology of Chagas disease. This makes PCR a potential tool for its use in acute phase diagnosis in a routine basis, and potentially for determining aetiological treatment failure when tests positive but not substantially useful when tests negative and these results must be interpreted cautiously as in the clinical trials previously published [21,68]. Further research is needed to improve the sensitivity of this test and the mandatory deployment of new diagnostic tests.
|
10.1371/journal.ppat.1002695 | ATG5 Is Essential for ATG8-Dependent Autophagy and Mitochondrial Homeostasis in Leishmania major | Macroautophagy has been shown to be important for the cellular remodelling required for Leishmania differentiation. We now demonstrate that L. major contains a functional ATG12-ATG5 conjugation system, which is required for ATG8-dependent autophagosome formation. Nascent autophagosomes were found commonly associated with the mitochondrion. L. major mutants lacking ATG5 (Δatg5) were viable as promastigotes but were unable to form autophagosomes, had morphological abnormalities including a much reduced flagellum, were less able to differentiate and had greatly reduced virulence to macrophages and mice. Analyses of the lipid metabolome of Δatg5 revealed marked elevation of phosphatidylethanolamines (PE) in comparison to wild type parasites. The Δatg5 mutants also had increased mitochondrial mass but reduced mitochondrial membrane potential and higher levels of reactive oxygen species. These findings indicate that the lack of ATG5 and autophagy leads to perturbation of the phospholipid balance in the mitochondrion, possibly through ablation of membrane use and conjugation of mitochondrial PE to ATG8 for autophagosome biogenesis, resulting in a dysfunctional mitochondrion with impaired oxidative ability and energy generation. The overall result of this is reduced virulence.
| Leishmaniasis is a disease of humans that is of major significance throughout many parts of the world. It is caused by the protozoan parasite Leishmania and mammals are infected through the bite of a sand fly in which the parasite develops. Parasite remodelling crucial for generation of the human-infective forms is aided by the catabolic process known as autophagy in which cell material is packaged within organelles called autophagosomes and subsequently broken down in the digestive lysosomal compartment. Here we show that autophagy in Leishmania requires the coordinated actions of two pathways, one of which involves a protein called ATG5. We have generated parasite mutants lacking this protein and shown that ATG5 is required for both autophagosome formation and also maintenance of a fully functional mitochondrion. The mutants lacking ATG5 have increased mitochondrial mass and phospholipid content, high levels of oxidants and reduced membrane potential, all being hallmarks of a dysfunctional mitochondrion with impaired ability for energy generation. Our results have thus revealed that a functional autophagic pathway is crucial for phospholipid homeostasis and mitochondrial function in the parasite and important for the parasite's differentiation, infectivity and virulence to its mammalian host.
| Leishmania are widespread and important parasites of humans and dogs that produce a spectrum of diseases collectively called the leishmaniases. Differentiation between the three distinctive morphological forms, the procyclic promastigote, metacyclic promastigote and amastigote, is crucial for progression through the parasite's digenetic life cycle and requires extensive remodelling of its cellular constituents, a process in which the macroautophagic pathway is involved [1], [2].
Macroautophagy (hereafter autophagy) is a catabolic system that degrades and recycles organelles and proteins [3]–[5]. In yeast and mammals, two ubiquitin-like conjugation systems, involving ATG8 and ATG12 respectively, are normally required for autophagosome formation although other mechanisms (non-canonical autophagy) have recently been recognised [6]. These two conjugation systems also utilise proteins encoded by six of the thirty-two known autophagy genes (designated ATG) with the conjugation of ATG8 to phosphatidylethanolamine (PE) occurring in one pathway and ATG12 to ATG5 in the other (see Figure 1).
In the ATG5-ATG12 pathway, ATG12 is activated by the E1-like enzyme ATG7, a thioester bond is formed between the carboxyl of its C-terminal glycine residue and the active cysteine of ATG7 [7]. ATG12 is then transferred to the active cysteine residue of the E2-like enzyme ATG10 [8] and subsequently to the ε-amino group of a conserved lysine residue of ATG5; an isopeptide bond with the exposed glycine residue of ATG12 being formed [5]. This process requires ATP and the complex subsequently interacts with the ATG6-Vps34 complex, ATG2, ATG14, ATG16, ATG18 and ATG21, proteins on the pre-autophagosomal membrane known as the phagophore [9]. The ATG5-ATG12 complex is crucial for the curvature of the phagophore in canonical autophagy. Work with mammalian cells and yeast has suggested that the phagophore is initially formed by membrane invagination of the centre of a phosphatidylinositol-3-phosphate (PI3P)-enriched spot, called the omegasome, formed by the action of phosphatidylinositol kinase, Vps34, on PI3P [10]. Cell membranes are now known to be involved in autophagosome initiation and the endoplasmic reticulum (ER), mitochondria, plasma membrane and Golgi apparatus have all been implicated [11]–[13].
It is to this developing phagophore that the attachment of ATG8-PE occurs; a key event for autophagosome formation. This is after the C-terminus of the precursor ATG8 has been cleaved by the ATG4 cysteine peptidase, to expose a C-terminal glycine. It is this glycine that is conjugated to PE through the catalytic actions of the E1-like and E2-like enzymes ATG7 and ATG3, respectively. The ATG5-ATG12 complex also contributes to this lipidation of ATG8 to PE through its E3-like activity which enhances the activity of ATG3 [14]; this reinforces the role of the complex in autophagosome biogenesis. The origin of the PE required for this process has been considered to be the ER in most mammalian cells [11], although it was recently shown to be the mitochondrion in mammalian cells under starvation conditions [12]. ATG8 incorporation onto the phagophore marks the start of cargo recruitment and acquisition. Adaptor proteins such as p62 and NIX attached to protein aggregates and damaged organelles, respectively, bind to ATG8-PE embedded on the nascent autophagosome [15], [16].
In time, the phagophore expands and there is closure of the autophagic membranes (with the cargo contained therein), processes that rely upon the ability of ATG8-PE to oligomerise and form aggregates and hemifusions [17]. The ATG5-ATG12 complex dissociates from the nascent autophagosome just before or after the nascent autophagosome buds off the omegasome with closure via a zippering mechanism [5]. For delivery and degradation of the autophagosome to the lysosome or vacuole and subsequent degradation of the contents, there is the requirement for ATG8 on the outer membrane of the autophagosome to be cleaved by ATG4 from its anchoring PE to facilitate fusion of the autophagosome with the endosomal and lysosome systems.
Some analyses of the genome of L. major suggested that the mechanism of autophagy in Leishmania may differ from that in other mammals and yeast in that genes encoding proteins required for the ATG5-ATG12 conjugation pathway appeared to be absent; this prompted speculations that this conjugation pathway may have evolved relatively recently [18]. These in silico findings also lead to the hypothesis that an alternative process known microautophagy may be especially important in these protozoa, which was supported in a report on glycosome turnover [19]. Nevertheless, in our previous studies we showed that autophagy involving ATG8 lipidation to PE occurs in Leishmania [1], [2] and that L. major does have genes that encode proteins with some apparent similarity to ATG5, ATG7, ATG10 and ATG12 [20]. Thus one objective of this study was to test experimentally the hypothesis that these proteins constitute a canonical ATG5-ATG12 conjugation pathway that is a key component of autophagy in Leishmania and to further characterise the pathway itself.
One of the functions of autophagy is recycling organelles including peroxisomes (pexophagy) and mitochondria (mitophagy). The mitochondrion is required for energy production via β-oxidation and oxidative phosphorylation but is also potentially able to regulate cell signalling pathways, maintain calcium and phospholipid levels, and promote cell death via apoptosis [21]. Thus its homeostasis is vital and autophagy is thought to have some role in this [22]. Evidence for interplay between autophagy and mitochondria has been increasing in recent years [23] with reports of mitochondrial function being compromised in the absence of a functional autophagic pathway [24], [25] and mitochondria regulating autophagy via signalling pathways [26]. However, the full mechanisms mediating this interplay are not understood fully. Mitophagy, which involves engulfment of the damaged mitochondrion into an autophagosome [22], [27], [28], has not been reported in Leishmania and the presence of a single mitochondrion, albeit comprising a large complex network, in the parasite raises questions on whether mitophagy per se can occur and if so how. Thus a second aim of this study was to elucidate the extent to which autophagy plays a role in mitochondrion homeostasis, with the hypothesis that the unitary mitochondrion may well necessitate interactions that differ from those that occur in mammalian cells and yeast.
PE is crucial for the binding of ATG8 in the formation of autophagosomes, but more generally it is a major component of biological membranes, especially mitochondrial membranes, and is involved in a wide range of biological processes from cell signalling, cell division, membrane fusion and trafficking events [29], [30]. There are two main routes known for PE synthesis, the Kennedy pathway and via phosphatidylserine decarboxylase (PSD). The latter occurs in the mitochondria in typical eukaryotes, utilising translocated phosphatidylserine (PS) synthesised in the ER [30]. However, it is thought to be insignificant in PE synthesis in Leishmania [31], [32], although present in the Leishmania mitochondrion [33], because sphingolipid metabolism in Leishmania is, unlike the situation in mammals, intrinsically linked with PE metabolism and provides the Kennedy pathway, which appears to terminate in the mitochondrion in trypanosomatids, with ethanolamine-phosphate [31], [32]. Thus the evidence as far as it stands for Leishmania suggests that PE is synthesised in the single mitochondrion before being distributed to other cell membranes. Therefore we hypothesised that the PE required for autophagosome formation may all be obtained from the mitochondrion directly in Leishmania, unlike the situation in most mammalian cells and yeast under normal conditions.
Thus this study was founded on the concept that the unusual nature of Leishmania in terms of mitochondrial structure and phospholipid biosynthesis distinguishes it from mammalian cells and makes it an interesting organism in which to study the interplay, if any, between autophagy and the mitochondrion. Our experimental approach to test the various hypotheses was to generate mutants lacking ATG5, and analyse the phenotype of the resulting mutant. This has not only allowed analysis of the interplay between autophagy and mitochondrial homeostasis but also the importance of autophagy for parasite viability, differentiation and virulence. The findings show clear correlation between autophagy and mitochondrial homeostasis and suggest that one contribution of autophagy to this is maintenance of appropriate PE composition in the mitochondrion. A consequence of the changes is markedly reduced virulence.
We have previously shown using western blot analysis of Leishmania lysates with an ATG12-specific antibody that ATG12 exists in two forms, one corresponding to the molecular mass of ATG12 and a second of higher molecular mass that was predicted to be an ATG5-ATG12 conjugate [20]. To provide further evidence that Leishmania has an ATG5-ATG12 conjugation system, we expressed and purified ATG5, ATG7, ATG10 and ATG12 recombinant proteins and analysed their ability to catalyse the formation of an ATG5-ATG12 conjugate in a reconstitution assay similar to those described previously [14], [34] The purified recombinant ATG7, ATG10 and a mutant ATG12 terminating at the scissile glycine (and named ATG12g; see ref [20]) were mixed with histidine-tagged ATG5 and ATP. Western blot analysis of the resultant mixture with α-His antibody detected the 50 kDa ATG5 and 70 kDa ATG12g-ATG5 conjugate (Figure 2A, lane 5). Analysis of the 70 kDa protein by mass spectrometry identified peptide fragments of both ATG5 and ATG12. The omission of ATG10 (Figure 2A, lane 1), ATG7 (Figure 2A, lane 2) or ATP (Figure 2A, lane 4) abolished the formation of the ATG5-ATG12 conjugate, suggesting that all the components were required. Further, no ATG5-ATG12 conjugate was formed when ATG10 was replaced with ATG3 (Figure 2A, lane 3), showing that in this assay there is no functional redundancy between the two L. major E2 enzymes ATG10 and ATG3. In total, the data suggest that recombinant ATG5, ATG7, ATG10 and ATG12g comprise the protein components required to form the ATG5-ATG12 conjugate in L. major, and the process is energy-dependent.
To determine if formation of the ATG5-ATG12 conjugate required lys128 of ATG5 and the terminal gly185 of ATG12, we prepared recombinant ATG12 and ATG12g and recombinant ATG5 in two forms - native and a mutant form with the lys128 substituted by ala (designated ATG5 and ATG5K128A, respectively). Western blot analysis confirmed that a constitution assay mix with L. major's ATG12g and native ATG5 formed the ATG5-ATG12 conjugate (Figure 2B, lane 1), whilst the native ATG12 and native ATG5 (Figure 2B, lane 2) and the ATG12g and ATG5K128A (Figure 2B, lane 3) did not. The lack of activity of ATG5K128A in the assay is excellent evidence that it is specific for ATG5 itself and is not promiscuous. In addition, as the L. major ATG12 has a key ATG8-like feature (a C-terminal extension beyond the scissile glycine that requires processing before conjugation), we replaced the truncated ATG12g with a similarly truncated ATG8g in the reconstitution assay. However, no ATG5-ATG8 conjugate could be detected (Figure 2C, lane 1) whereas the control experiment with ATG12g under the same conditions formed the ATG5-ATG12 conjugate (Figure 2C, lane 2). These data confirm the functional difference between the proteins, which we had putatively identified as ATG8 and ATG12. Overall, these results suggest that the ATG5-ATG12 conjugate is formed by a reaction between the exposed glycine residue of ATG12 and the ε-amino group of lys128 of ATG5. The results also indicate ATG7 and ATG10 function as E1 and E2 enzymes, respectively. In addition, they show that the native ATG12 in Leishmania needs processing to enable it to function - a control mechanism that is not present in ATG12 from yeast or higher eukaryotes; the enzyme mediating this cleavage is unknown.
Previously we showed that ATG12 and ATG8 co-localized in L. major, but that most ATG8-containing autophagosomes lacked ATG12 [20]. This was consistent with L. major ATG12 being associated with nascent phagophores, but not fully formed autophagosomes containing cargo. To investigate the occurrence and location of the ATG5-ATG12 conjugate, we have now studied co-localisation of the two proteins in living promastigotes. mCherry-ATG5 (mC-ATG5) and green fluorescent protein-ATG12 (GFP-ATG12) were expressed singly and also co-expressed in L. major promastigotes and the resulting lines were analysed by fluorescence microscopy.
When grown in nutrient-rich medium and at early logarithmic growth phase, most cells expressing mC-ATG5 had the fluorescence evenly distributed throughout the cytoplasm (Figure 3A, left panels) with only 2% having a single mC-ATG5-labelled punctum in the cytosol (Figure 3A, right panels). However, under starvation conditions for an hour or more (known to induce autophagy [1]) 20±3% of the cells had mC-ATG5 puncta and of these 80±2% had just one (Figure 3B). All of the puncta in cells with both mC-ATG5 and GFP-ATG12 contained both labels (Figure 3C). To investigate co-localisation of ATG5 and ATG8, mC-ATG5 and GFP-ATG8 were co-expressed in promastigotes and late logarithmic stage cells analysed for puncta by fluorescence microscopy. Of the puncta in the promastigotes, 60±16% of mC-ATG5-labelled puncta also contained GFP-ATG8 but only 31±6% of GFP-ATG8-labelled puncta also had mC-ATG5 (Figure 3D).
The dynamics of the appearance of the mC-ATG5-labelled puncta was studied by using shorter starvation incubation periods. This showed that there was an early phase of up to 30 min starvation when several puncta were observed before the number declined (Figs. 3B and 3E). With this period of starvation, promastigotes expressing either GFP-ATG8 or GFP-ATG12 alone did not have puncta. These data are consistent with ATG5 being the first of these proteins to become recruited when the biogenesis of autophagosomes is initiated.
Interestingly, the early puncta were distributed around the promastigotes (Figure 3E) in a way consistent with the distribution of the typically reticulate mitochondrion that is present in the cells. This prompted us to look for co-localisation between nascent autophagosomes and the mitochondrion. We used two mitochondrial proteins as markers for the mitochondrion, the ubiquitin-like peptidase MUP (LmjF26.2070) and the serine peptidase rhomboid (LmjF04.0850). MUP fused to GFP was used as a marker for the outer membrane of the mitochondrion, as the mammalian homologue of MUP is located on the surface of mitochondria [35] and we showed that MUP-GFP co-localised with MitoTracker Red in Leishmania (Figure 3F). When MUP-GFP was co-expressed with mC-ATG5, more than half of the puncta after 30 min starvation were associated with the mitochondrion (Figure 3G). Differentiation of the procylic promastigote form to the metacyclic promastigote form in nutrient-rich conditions produced a localization profile for mC-ATG5 similar to that described for cells after 1 h starvation, with 19±3% of the cells having puncta of which 66±16% were in association with the mitochondrion (Figure 3H). The serine peptidase rhomboid is predicted to be located in the inner mitochondrial membrane [36], so we expressed rhomboid-GFP (ROM-GFP) and confirmed it as a second mitochondrial marker by co-localisation with MitoTracker Red (MTR, data not shown). Co-expression of this and mC-ATG5 revealed that 62±25% of the ATG5-labelled puncta were associated with the mitochondrion (Figure 3I).
We also looked for association of ATG8-labelled puncta with the mitochondrion. Fluorescence microscopy of late log phase promastigotes expressing both MUP-GFP and red fluorescent protein-ATG8 (RFP-ATG8) showed that 55±3% had ATG8-labelled puncta of which 60±9% were in apparent association with the mitochondrion (Figure 3J). Equivalent experiments with ROM-GFP rather than MUP-GFP gave similar data with, on average, 53±4% of the ATG8-labelled puncta being associated with the mitochondrion (Figure 3K). However, in no case was MUP-GFP or ROM-GFP fluorescence detectable within RFP-ATG8-labelled puncta, nor were the labelled MUP or ROM detectable with a RFP-ATG8-labelled elongated structure sometimes appearing in stationary phase promastigotes and thought likely to be the MVT-lysosome ([1] and Figure 3L). Relocation of mitochondrial components to the lysosome is one assay for mitophagy in yeast [37]. These data together suggest that the processes that we observed under the conditions of our experiments were not mitophagy.
Our observations are consistent with a large proportion of the phagophore biogenesis being initiated at the mitochondrial membrane with the involvement of ATG5 and ATG12, with subsequent recruitment of ATG8 as the nascent autophagosomes develop; these then lose ATG5 and the autophagosomes become located in the cytosol.
In order to investigate further the involvement of ATG5 in the parasite, promastigote mutants lacking both copies of the ATG5 gene were generated by homologous recombination and verified by Southern blot analysis (Figure S1). An add-back line was generated by integrating ATG5 with an N-terminal 6× histidine tag into the ribosomal locus. These cloned lines were named Δatg5 and Δatg5::ATG5, respectively, and were used to infect mice from which the parasites were re-isolated to provide promastigotes for phenotypic analysis. The growth rate of Δatg5 promastigotes was reduced compared with the wild type (WT) and the add-back lines (Figure 4A). Δatg5 were unable to form GFP-ATG8 labelled autophagosomes in either nutrient-rich media or under starvation conditions (Figs. 4B–C), and there was very little conjugation of GFP-ATG8 to PE to generate GFP-ATG8-PE (designated GFP-ATG8-II, Figure 4D; see [1]); consistent with the cells being incapable of forming autophagosomes. Together, these findings show that ATG5 is crucial for autophagosome biogenesis and autophagy in Leishmania.
Transmission electron microscopy revealed the mitochondrion in Δatg5 to be swollen with an extended membranous structure (Figure 5A). This was suggestive of an increased mitochondrial mass, which was confirmed by MitoTracker Green (MTG) labelling (Figure 5B, solid bars; MTG is a green-fluorescent mitochondrial stain which localizes to mitochondria regardless of mitochondrial membrane potential). The additional membranes that were apparent also suggested an increased lipid content. Analysis of Δatg5 expressing MUP-GFP revealed a variety of mitochondrial morphologies (Figure 5C) that ranged from the reticular network characteristic of WT promastigotes (∼30% of the cells; left panel), through fragmented forms (∼25%; centre panel), to swollen mitochondrion with little apparent structure (∼45%; right panel). To investigate if mitochondrion function was compromised, the cells were stained with MitoTracker Red (MTR; this is a red-fluorescent dye that stains mitochondria in live cells and the accumulation of which is dependent upon membrane potential). Δatg5 were found to have less than half of the MTR fluorescence compared with the WT promastigotes (Figure 5B, open bars), indicating a loss of mitochondrial membrane potential. This was confirmed by co-staining the cells with MTR and MTG, which showed total co-localisation in the WT promastigotes but a lower degree of co-localisation in Δatg5 (Figure 5D). Alamar Blue reduction was also less in Δatg5 than WT (Figure 5E), indicating reduced mitochondrial respiration in the mutant. Assessing the levels of reactive oxygen species (ROS) using 2′,7′-dichlorodihydrofluorescein diacetate (H2DCFDA, intracellular cleavage and oxidation of this to yield the highly fluorescent 2′,7′-dichlorofluorescein [DCF] is a measure of ROS) showed these to be higher in Δatg5 than WT promastigotes (Figure 5F). Together these results suggest mitochondrial dysfunction on several levels.
The mitochondrial changes resulting from deletion of ATG5 were suggestive of effects upon lipid content and thus we compared the lipidome of Δatg5 and WT promastigotes cultured in vitro under standard growth conditions. The total intensity obtained from analysis of the extracted metabolites from 2×106 promastigotes by liquid chromatography mass spectrometry (LC-MS) indicated that overall PE and phosphatidylcholine (PC) levels in Δatg5 were significantly higher (p<0.02) than the levels in the same number of WT promastigotes (PE, 3.4±1.3×107 compared with 1.5±0.8×107; PC, 1.1±0.4×108 compared with 6.5±2.3×107). Phosphatidylinositol (PI) and phosphatidylserine (PS) levels remained unchanged (data not shown). The apparent increase in PE and PC levels could be contributing to the increased membrane content of the mitochondrion in Δatg5. As these data suggested a link between autophagy and phospholipid homeostasis of the cell, we investigated the phospholipid composition of WT and Δatg5 promastigotes in more detail using electrospray mass spectrometry. Survey scans using negative ion mode of the WT L. major between 600–900 m/z showed a wide range of molecular species from the three classes of phospholipid (Figure 6, all of the molecular species identified are detailed in Table S1). The major PE species between 680–745 m/z was the plasmalogen (alkenyl-acyl) at 726.4 and 728.4 m/z (for a-18∶1, 18∶2 and a-18∶1, 18∶1, respectively, where a = (alkylacyl) [38]) but the diacyl PE species was also identified at 714.4 m/z (for C34∶2). Several inositol phosphoceramide (IPC) species were observed between 680–810 m/z, the major species being the previously identified d16∶1, 18∶1-IPC at 778.4 m/z [39]. The third class of phospholipids detected were PIs, with an envelope of species between 800–900 m/z, the major species being at 835.4 and 863.5 m/z (diacyl 34∶1 and 36∶1, respectively). The equivalent negative ion survey scans for Δatg5 cells showed the presence of most molecular species identified in WT, but the majority of PE species increased significantly in Δatg5 cells relative to their WT counterpart (Figure 6, compare A and B). Large increases were apparent for PE species at 698.4, 726.4 and 738.4 m/z (a-34∶3, a-36∶3 and diacyl-36∶4, respectively) and IPC at 780.4 m/z for d16∶1, 18∶0 compared with the d16∶1, 18∶1 species at 778.4 m/z. In contrast to PE, no differences in any of the diacyl or alkenyl-acyl PI species or cardiolipin were obvious. More quantitative analysis of the overall PE levels in WT and Δatg5 promastigotes was facilitated by inclusion of an internal PE standard. Higher levels of the PE species (686–748 m/z) were clearly visible in the Δatg5 lipid extracts; normalisation using the internal standard indicated an approximate 2.5-fold increase in the total PE level compared with the WT promastigote levels (Figure S2). To investigate further how the observed increase in PE species in the Δatg5 cells could be due to the lack of ATG5 and autophagy, both WT and Δatg5 cells were grown in the presence of D3-serine prior to lipid extraction and analysis. As expected, D3-serine was incorporated into the phospholipid pool and manifested itself primarily in the IPC species [31]; the serine being utilised in de novo sphingolipid synthesis (Figure S3). Notably, the serine was not apparently incorporated into PE via decarboxylation of PS in either WT or Δatg5 promastigotes, as there was no detectable evidence of deuterated-PE with the same lipid moiety as the tiny amounts of detectable PS species (770–776 m/z, corresponding to a-36∶3 to a-36∶0) or of any other corresponding observable PE species, i.e. 686, 700, 714, 742 m/z. These data show that the only important route for PE synthesis in L. major is the Kennedy pathway.
We applied several approaches to investigate whether differentiation from procyclic promastigote to metacyclic promastigote and infectivity is impaired in Δatg5. We found firstly that peanut agglutinin-negative metacyclic promastigotes [40] were less abundant in Δatg5 promastigotes than in the WT line (Figure 7A). Secondly, Δatg5 expressed lower levels of the metacyclic marker protein HASPB [41] than WT promastigotes (Figure 7B). Thirdly, Δatg5 promastigotes were taken up into macrophages to a similar extent as WT, but survived poorly intracellularly with most macrophages being cleared of parasites by day 5 (Figure 7C). This reduced virulence of Δatg5 was also evident in vivo, inoculation of Δatg5 promastigotes into mice generated rump lesions that were significantly smaller than those inoculated with WT promastigotes or the re-expressor line at weeks 3 and 4 (p<0.01 and p<0.05, respectively at each time point; Figure 7D). Analysis of parasite morphology by scanning electron microscopy showed that the parasites isolated from an infected mouse were predominantly amastigotes (∼88%) with sizes ranging from ∼2–4 µm with no apparent morphological differences from WT (Figure 7E, compare panels on left). Surprisingly, ∼12% of Δatg5 had a spindle-shaped body that was 6–10 µm in length and 75% of these had no external flagellum (Figure 7E, panels in centre and on right).
Amastigotes of Δatg5 were extracted from a mouse lesion and transformed in vitro to promastigotes which were used for all of the phenotypic characterization of Δatg5 in this study. The cells exhibited unusual morphological features and so we applied scanning electron microscopy to analyse the morphology of the cell population. Whilst many of the Δatg5 population on day 5 of in vitro culture were typical promastigotes, others were ovoid and amastigote-like and others were spindle-shaped, with or without an external flagellum (Figure 8A). Forms with no external flagellum represented ∼20% of the cells in logarithmic growth phase populations and their abundance increased to ∼60% in stationary phase of growth. Morphometric analysis of promastigotes of L. major in in vitro cultures reflected these differences and the mean flagellum and body lengths for stationary phase cells were significantly lower for Δatg5 parasites than for WT promastigotes (Figs. 8B); the mean flagellum lengths for WT, Δatg5 and Δatg5::ATG5 promastigotes were 13.7±3.1 µm, 4.7±2.1 µm, and 9.8±2.5 µm, respectively, and body lengths were 9.8±2.3 µm, 3.6±3.1 µm, and 7.6±5.9 µm, respectively. The mean body lengths and mean flagella lengths were significantly different between Δatg5 and WT (p<0.01 and p<0.05, respectively).
Homology-based genome annotation based on sequence similarity can lead to some interesting predictions on function, but the evolutionary distance between early and late eukaryotes means that predictions for protozoa need to be experimentally validated. A good example is the ATG12-ATG5 pathway in Leishmania. This was originally predicted by others to be absent [18] but subsequently possible putative homologues with very low sequence identity with yeast and human counterparts were identified by us and others [20], [42]. We have now resolved this uncertainty by demonstrating that the L. major ATG5-ATG12 conjugation system can be reconstituted in vitro using recombinant proteins. The conjugate was formed by the enzymatic reactions of ATG7 (E1-like) and ATG10 (E2-like) and required lys185 of ATG5, a free C-terminal glycine residue of ATG12 and ATP (Figure 2). The finding that ATG5 and ATG12 co-localise to puncta induced under starvation (and thus thought to be nascent autophagosomes) are consistent with these observations and support the hypothesis that the ATG5-ATG12 conjugation pathway exists within L. major promastigotes and apparently is, in the main, mechanistically similar to that of higher eukaryotes. Deletion of the ATG5 gene from Leishmania generated mutants that were unable to form autophagosomes (as assessed by the absence of GFP-ATG8 puncta), which is entirely consistent with the ATG5-ATG12 conjugation pathway having a crucial role in autophagy in the parasite. More studies are required, however, to determine the extent to which the process in Leishmania is similar to that in yeast and mammalian cells and whether it shares features with the non-canonical processes of autophagy that are beginning to be elucidated [6]. In eukaryotes such as mammals, yeast and Arabidopsis, both the ATG5-ATG12 and the ATG8–PE conjugates localize at the phagophore to facilitate autophagosome genesis, but ATG5 and ATG12 are not normally observed on the completed autophagosome [5]. Our findings with L. major also show that the ATG5-ATG12 complex does not associate with all ATG8-labeled structures and is not apparently trafficked to the lysosome, consistent with the hypothesis that it assists in driving the expansion and/or curvature of the nascent autophagosome but dissociates from them just before, or immediately after, completion. One important unusual aspect of the process in Leishmania, however, is the apparent origin of the membrane and phospholipid for the phagophore, as we discuss below.
Our ability to generate mutants lacking ATG5 confirmed that the protein is not essential for parasite survival in vitro as promastigotes or as amastigotes in macrophages or mice. Nevertheless, the survival of the Δatg5 parasites in explanted macrophages was very greatly reduced in comparison with WT parasites, as was growth in mice, so it appears that autophagy contributes in a very significant way, either directly or indirectly, to the parasite's virulence and so it should not be ruled out as a target for novel therapies. Autophagy is considered to be important for general cell homeostasis as well as for survival against adverse conditions such as oxidative stresses [43] and it seems very likely that this holds for Leishmania too. Cells in which the normal mechanisms responsible for homeostasis are adversely affected are very likely to be less able to withstand challenges such as those to which Leishmania is exposed when entering a macrophage. Moreover, interference with the normal processes of differentiation between forms, in which we have shown autophagy plays an important part [1], would also adversely affect survival when entering a new host or host cell.
One major mechanism mediating the Δatg5 mutant's reduced ability to withstand stresses of infection is the significant perturbation of mitochondrial function, including a lower membrane potential and so energy production and an enlarged mitochondrial mass, resulting from deletion of the gene encoding ATG5. Global analyses of the metabolome of the mutant revealed marked elevation of the phospholipid levels, in particular greatly elevated levels of PE and PC. Interestingly, levels of PI and cardiolipin (which occurs primarily in the inner mitochondrial membrane) were unaffected, showing specificity in the changes resulting from ATG5 deletion. Phospholipid metabolism, especially that of PE which has many crucial signalling effects, is normally regulated very tightly within the mitochondrion and is inextricably linked to mitochondrial function, although relatively little is known about mechanisms regulating the phospholipid content and integrity of mitochondrial membranes [30]. However, all the evidence suggests that significant alterations to the mitochondrial PE composition would cause dysfunction of the organelle and result in a deficiency in ATP generation.
It seems very likely that the increased PE and PC content and the abnormal mitochondrial properties we observed for Δatg5 are causally linked. The key question, however, is how is ATG5 associated with these changes? As we have confirmed that ATG5 is necessary for autophagy in Leishmania, and as one known involvement of autophagy in mammalian cells and yeast is mitophagy, then one could postulate that the lack of the ATG5 in Δatg5 means that an important mechanism for removal of damaged and unwanted mitochondrial material is absent with resultant damage to the structure. The possibility that the lack of mitophagy is the cause of the observed mitochondrial dysfunction cannot be excluded, although there have been no definitive reports of mitophagy in Leishmania and we could not detect any mitochondrial proteins (using MUP-GFP as an outer membrane protein marker or ROM-GFP as an inner membrane protein marker) being trafficked to the lysosome in autophagosomes under our current experimental conditions. Moreover, the presence of a single large mitochondrion in Leishmania excludes mitophagy occurring as in yeast and mammalian cells in which whole mitochondria are enclosed within autophagosomes [22], [28]. Thus if mitophagy does occur in Leishmania, the mechanism must differ from that thought to operate normally in these other cells. In higher eukaryotes, mitophagy can occur co-ordinately with mitochondrial fission [44] and there must be a mechanism for mitochondrial fission in Leishmania in order to ensure correct partition of the single organelle during cell division; however, this type of mitophagy has not been observed to date. Leishmania must have mechanisms for maintaining mitochondrial performance and one similar to the recently discovered vesicular trafficking pathway between mitochondria and lysosomes in mammalian cells, that is independent of ATG8 and ATG5, is also worthy of consideration as being complementary to mitophagy [45].
There are, however, other possible causes of the elevated PE and PC levels in Δatg5. It has been suggested that Leishmania is different from most eukaryotes in that the only route for PE synthesis is the Kennedy pathway with PS decarboxylase (PSD) being unimportant even though Leishmania do possess a PSD gene [31], [32]. In contrast, mammals use mainly the mitochondrion-located PSD route [46], with the PE produced actively exported to other organelles [30]. The first two steps of the Kennedy pathway in yeast and mammalian cells are exclusively located in the ER with the final step, involving ethanolamine-phosphotransferase (EPT), being located in either the ER or mitochondrion. Interestingly, in T. brucei, a trypanosomatid closely related to Leishmania, the EPT is mitochondrial (Gibellini, F & Smith T.K unpublished). Thus in T. brucei PE is synthesised in the mitochondrion. This could very well be the situation with Leishmania too, with all PE being generated in the mitochondrion. We therefore hypothesised that a second way in which the mitochondrion interplays with autophagy in Leishmania is in the provision of membrane for the developing phagophore and also PE to anchor the ATG8 in the phagophore. In the absence of autophagosome genesis in Δatg5 this phospholipid utilisation would not occur - with the result that phospholipid homeostasis in the mitochondrion would be disrupted and mitochondrial function thus impaired. Our data obtained in this study on the localisation of autophagosomes and phospholipid content of Δatg5 support this hypothesis. The application of dual-labelling of promastigotes with MUP-GFP or ROM-GFP as mitochondrial markers and mC-ATG5 showed that approximately two thirds of the ATG5 puncta were apparently associated with the mitochondrion and the multiple puncta occurring early in starvation with mC-ATG5 labelling alone had a distribution consistent with mitochondrial association too (Figure 3E). It has been recently shown for mammals that the outer membrane of mitochondria can be the source of autophagosome PE [12], [13], but only under unusual circumstances. Our findings on the elevated PE and also PC of Δatg5 are consistent with the hypothesis that in Leishmania the mitochondrion is a normal source of membrane and particularly PE and PC for autophagosome biogenesis and thus in this way Leishmania apparently differs greatly from mammalian cells and yeast.
Thus our results with Leishmania show that the functioning of ATG5, as well as being essential for autophagy itself, is also crucial for mitochondrial homeostasis indirectly as autophagy plays an important role in maintaining phospholipid and especially PE homeostasis. We suggest that the possibility that this is a mechanism contributing to the maintenance of mitochondrial membrane integrity in other eukaryotes warrants further investigation.
Interestingly, PE biosynthesis in Leishmania is elevated in promastigotes undergoing metacyclogenesis [31], [32] - which is when autophagy is most prevalent and required [1]. Moreover, the Kennedy pathway which is central to the provision of PE in Leishmania promastigotes is supplied from sphingolipid metabolism [31]. It is notable that the Leishmania mutants deficient in sphingolipid biosynthesis had a differentiation defect [31], which is consistent with these mutants being unable to synthesise PE and thus autophagy being prevented. Thus this study on sphingolipid synthesis provides further evidence of association between PE synthesis, autophagy and differentiation; it would be interesting to investigate whether these mutants present phenotypic alterations similar to those of Δatg5.
The present study has provided also insights into consequences of the mitochondrial dysfunction that results from lack of ATG5. The inability of the Δatg5 mutants to salvage materials via autophagy presumably adds to the energy deficiency resulting from the mitochondrial dysfunction, and these two limitations together mean that under resource-limiting conditions the Δatg5 cells needed to resort to extreme measures. That the mutants showed morphological abnormalities including much reduced flagellum length in promastigotes suggests that flagellum regression is a mechanism whereby the parasite reduces energy utilisation, or indeed releases additional energy, in time of nutrient stress. Such changes have been reported previously [47], our findings suggest that these changes in Δatg5 could also be a secondary response to energy-generation crises.
The greatly reduced virulence of the Δatg5 mutants could be mediated in a number of ways. These include the lack of autophagy hindering the parasite's ability to transform to amastigotes. However, the changes in the mitochondrion resulting from deletion of ATG5, mediated by the lack of removal of PE and PC and/or the absence of a type of mitophagy that is needed for maintaining mitochondrial homeostasis, seems very likely also to be a key causal factor. The low virulence of the autophagy-deficient line provides evidence that components of the autophagy machinery in Leishmania warrant consideration for drug discovery programmes.
All animal procedures were undertaken in adherence to experimental guidelines and procedures approved by The Home Office of the UK government. All work was covered by Home Office Project Licence PPL60/3929 entitled ‘Mechanism of control of parasite infection’. All animal protocols received ethical approval from the University of Strathclyde Ethics Committee.
L. major (MHOM/IL/80/Friedlin) promastigotes (designated WT for this study) were routinely grown and handled as described previously [48]. In this study, early log, mid log and early stationary phases of promastigote growth normally corresponded to approximately 5×105, 5×106, and 1×107 parasites/ml, respectively. To study the effects of differing conditions, promastigotes at 107 cells/ml were washed and suspended in phosphate-buffered saline (PBS) for starvation or HOMEM either serum-free or supplemented with 10 or 20% (v/v) FCS. The required antibiotics were added to the cultures of the transgenic lines are as follows: hygromycin B (Sigma) at 50 µg/ml; phleomycin (Cayla, France) at 10 µg/ml; puromycin (Calbiochem) at 10 µg/ml; neomycin (G418, Geneticin, Life Technologies) at 25 µg/ml.
The occurrence of metacyclic promastigotes was assessed using peanut agglutinin (PNA) [40] and western blot analysis of markers for metacyclic promastigotes. Infectivity to peritoneal macrophages from CD1 mice was with stationary phase promastigotes at a ratio of 5 promastigotes per macrophage and incubation for up to 5 days at 32°C in 5% CO2/95% air. Non-phagocytosed promastigotes were removed after 24 h and parasite abundance within the macrophages were determined after staining with Giemsa. Infectivity to mice was determined using groups of 5 mice inoculated subcutaneously in the rump with 5×105 stationary phase promastigotes; the width of the resulting lesion in the rump was measured.
Excised rump lesions of mice in cold PBS containing 50 µg/ml gentamycin (Sigma) were homogenised in a glass tissue grinder, large debris was removed (150×g for 1 min at 4°C), amastigotes sedimented (1700×g for 15 min) and suspended in complete HOMEM medium with 50 µg/ml gentamycin and then incubated at 25°C to back-transform the amastigotes to promastigotes or immediately fixed for scanning electron microscopic (SEM) analysis.
Parasite metabolites were extracted and analysed using LC-MS as detailed previously [49]–[51]. For phospholipid profiling, promastigotes at mid log growth phase, cultured in complete HOMEM medium at 26°C in the absence or presence of D3-serine (CDN) for the final 24 h, were extracted according to [52] and analyzed by electrospray mass spectrometry. Samples suspended in chloroform/methanol (1/2 v/v) were analyzed with a Micromass LCT mass spectrometer equipped with nanoelectrospray source. They were loaded into thin-wall nanoflow capillary tips (Waters) and analyzed by ES-MS in both positive and negative ion modes using a capillary voltage of 0.9 kV and cone voltages of 50 V. Where necessary MS/MS daughter ion scanning was performed on a Micromass Quattro Ultima triple quadrupole or a ABSCIEX 4000 Q-Trap mass spectrometer equipped with nanoelectrospray source using argon or nitrogen as a collision gas, respectively, with collision energies between 35–70 V depending upon lipid class. Each spectrum encompasses at least 50 repetitive scans.
The open reading frames (ORFs) of ATG7 (LmjF07.0010), ATG10 (LmjF31.3105), ATG5 (LmjF30.0980), ATG3 (LmjF33.0295) and ATG12 (LmjF22.1300) were amplified by PCR using gene-specific primers (Table S2A). All PCR assays using Taq and Tgo DNA polymerases as part of the High Fidelity PCR system (Roche) were carried out for 30 cycles of denaturation (94°C, 15 s), annealing (65°C, 15 s) and extension (72°C, 2 min) and products cloned into pET28a+ and verified by nucleotide sequencing (Dundee Sequencing Services). Plasmids were transformed into BL21(DE3) for recombinant protein expression. A mutant of ATG5, ATG5K128A, was obtained by site-directed mutagenesis (Strategene) using primers shown in Table S2A, while the truncated proteins, ATG12g and ATG8g were generated by PCR as described above.
The plasmid used to generate the ATG5 null mutant was the pGL345-HYG plasmid [53] modified with fragments of the 5′ and 3′ UTRs flanking the ORF of ATG5. The 5′ (1.0 kb) and 3′ (1.1 kb) flanks amplified from a L. major genomic DNA template by PCR with primers modified with HindIII/SalI and SmaI/BgIII restriction sites, respectively, required for cloning (detailed in Table S2B) were sequentially inserted into the appropriately pre-digested pGL345-HYG to give pGL345ATG5-HYG5′3′. The pGL345ATG5-BLE5′3′ plasmid was generated from plasmid pGL345ATG5-HYG5′3′ by replacing the SpeI/BamHI ORF of the hygromycin resistance gene with a SpeI/BamHI ORF of the bleomycin resistance gene. The ATG5 ‘add back’ construct modified at the C-terminus with a poly-histidine epitope and containing the BglII and BamHI sites was cloned into the pRIB-Pur vector [54] to produce pRIB-Pur-ATG5-His. L. major ATG5 was cloned into the extrachromosomal pNUS-mCherrynH vector to give pN-mC-ATG5 whereas ATG12 and the gene encoding the mitochondrial ubiquitin-like protein (MUP; LmjF26.2070) were cloned into the pNUS-GFPnH vector to give pN-GFP-ATG12 and pNUS-GFPcH vector to give pN-MUP-GFP. The gene encoding the mitochondrial rhomboid (LmjF02.0430) was cloned into the extrachromosomal pNUS-GFPcN to give pN-ROM-GFP. The pN-GFP-ATG8 construct has been described previously [1].
The cassettes used for transfection of promastigotes were linearized by HindIII/BglII digestion and the excised cassette used for electroporation using the Nucleofactor system (Lonza) with the program V-033. Parasite populations recovered after transfection were cloned by serial dilution. Clonal populations of parasites resistant to hygromycin were obtained and two of these heterozygotes (Δatg5+/−) were used for the second round of transfections with the pGL345ATG5-BLE5′3′ construct and parasites were clones by serial dilution. One null mutant clone (Δatg5) was selected for further analysis. Lines re-expressing ATG5 were generated by integration of the pRIB-Pur-ATG5-His plasmid cassette, excised after digestion with PacI and PmeI and used for electroporation of Δatg5 promastigotes, to generate Δatg5::ATG5. Cell lines expressing tagged fluorescence proteins were generated by electroporation of promastigotes with 15 or 30 µg of plasmid and selection using the appropriate antibiotic(s) to give, for example, Δatg5[GFP-ATG8], the nomenclature of the ATG5 null mutant expressing GFP-ATG8 (Table S2B).
Genomic DNA from the Δatg5 clones was extracted and Southern blots performed as described previously [48]. DNA (5 µg) was digested with XhoI, fractionated by agarose gel electrophoresis, nicked, denatured, neutralized and blotted onto Hybond™-N+ membrane (Amersham Pharmacia) by capillary transfer. The probe was prepared from a 1100 bp HindIII/SalI 3′ flank fragment of pGL345ATG5-HYG5′3′.
For live imaging, promastigotes in complete HOMEM medium were mounted on coverslips and the occurrence of puncta were observed using either a Nikon TE2000S or a Delta Vision core (Image Solutions) inverted microscope equipped with FITC and mCherry filter sets. To investigate autophagy induced by starvation, promastigotes were incubated in PBS (designated nutrient-deprived medium, ND) at 2×107 cells/ml for up to 2 h and monitored for puncta similarly. Images were processed using IPlabs 3.7 image processing software (BD Biosciences Bioimaging). The presence and number of puncta within the cells was assessed in at least 100 cells from each of 3 independent experiments. Promastigotes at 1×107 cells/ml were incubated with either 0.1 µM MitoTracker Red CMXRos (MTR, Invitrogen) or 0.2 µM MitoTracker Green TM (MTG, Invitrogen) for 30 min at 26°C or co-stained with both MTR and MTG similarly. Promastigotes were then washed in PBS and either mounted on glass slides for analysis by fluorescence microscopy or assessed for fluorescence (MTR at excitation 579 nm, barrier filter 599 nm; MTG at excitation 490 nm, barrier filter 516 nm) using a microtitre plate reader.
For assessing the levels of ROS, promastigotes at 1×107/ml were harvested by centrifugation, washed once in serum-free HOMEM, and 2×106 cells in 200 µl were incubated with 0.1 mM H2DCFDA (Molecular Probes) for 2 h at 26°C and the fluorescence measured using a microtitre plate reader (excitation 380–420 nm, barrier filter 520 nm). To evaluate metabolic activity and cell viability, Alamar Blue (resazurine salt, Sigma) was added to a final concentration of 0.0125% to 4×106 promastigotes/ml at log phase of growth for one hour and its reduction measured using the fluorescence microtitre plate reader (excitation 550 nm, barrier filter 590 nm).
Expression of L. major ATG proteins, using the plasmids described above, in BL21(DE3) E. coli was carried out overnight at 15°C after induction with 1–2 mM isopropyl-β-D-thiogalactopyranoside (IPTG). Recombinant proteins were purified using an affinity chromatography column (Qiagen) and eluants obtained using 1 M imidazole were dialysed at 4°C overnight as follows: ATG7, ATG3, ATG10 and ATG8 - 50 mm Tris/HCl pH 7.5, 150 mm NaCl with 2 mm dithiothreitol; ATG5 - 20 mm Tris/HCl pH 8.0, 500 mm NaCl. The histidine tags of all ATG proteins except ATG5 were excised using thrombin (Novagen). The cleaved histidine tag and thrombin were subsequently removed by nickel chelate and benzamidine-Sepharose (Sigma) affinity chromatography.
Purified ATG7, ATG10, ATG12g and His-ATG5, each at 0.1 µg/ml, were mixed in reconstitution buffer (50 mm Tris-HCl, pH 8.0, 100 mm NaCl, 2 mm dithiothreitol, 1 mm MgCl2, and 1 mm ATP) and the reaction mixture was incubated for 1 h at 30°C. The conjugation reaction was stopped by boiling in sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) sample buffer. Samples were resolved by SDS-PAGE and subjected to western blot analysis using appropriate antibody.
Western blots were performed as previously described [48] with primary antibodies: the α-His (Promega) and α-GFP (Abcam) antibodies were used at 1∶1000 and 1∶100 dilutions, respectively, and their corresponding secondary antibodies were α-rabbit IgG-horseradish peroxidase (HRP) (Promega) at 1∶20000 and α-rat-HRP (Promega) at 1∶2500. α-HASPB rabbit antibodies (kindly provided by Professor Deborah Smith, University of York) were diluted 1∶5000. α-CS (cysteine synthase) antibodies [55] were used a loading control at 1∶5000.
Parasites were fixed with 2.5% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4 for 40 min and processed for transmission electron microscopy (TEM) as described previously [56]. Sections (80 nm) were examined with the Zeiss 912 TEM. For scanning electron microscopy (SEM), fixed samples were dried prior to coating with a very thin film of gold/palladium before examination. Promastigote body and flagella lengths were measured using the ESI Vision 3.2 Image analysis software (Olympus Soft Imaging Solutions). The cell morphologies noted within the parasite population were classified into groups according to the following criteria: amastigote-like forms that were ovoid and lacking an emergent flagellum; spindle shaped promastigotes with varying flagella and body lengths; and promastigotes that were similar to WT promastigotes. A minimum of 200 cells was examined for each sample.
Experimental data from macrophage infections, mice infectivity and Alamar Blue assays were pooled for comparison using unpaired t-tests. A p value of <0.05 was used as the level of significance.
Gene accession numbers (http://www.genedb.org) of proteins used in this study are:
ATG3, LmjF33.0295; ATG5, LmjF30.0980; ATG7, LmjF07.0010; ATG8, Lmj19.1640; ATG10, LmjF31.3105; ATG12, LmjF22.1300; ubiquitin-like peptidase MUP, LmjF26.2070; serine peptidase rhomboid, LmjF04.0850.
|
10.1371/journal.ppat.1000799 | Kaposi's Sarcoma-Associated Herpesvirus ORF57 Protein Binds and Protects a Nuclear Noncoding RNA from Cellular RNA Decay Pathways | The control of RNA stability is a key determinant in cellular gene expression. The stability of any transcript is modulated through the activity of cis- or trans-acting regulatory factors as well as cellular quality control systems that ensure the integrity of a transcript. As a result, invading viral pathogens must be able to subvert cellular RNA decay pathways capable of destroying viral transcripts. Here we report that the Kaposi's sarcoma-associated herpesvirus (KSHV) ORF57 protein binds to a unique KSHV polyadenylated nuclear RNA, called PAN RNA, and protects it from degradation by cellular factors. ORF57 increases PAN RNA levels and its effects are greatest on unstable alleles of PAN RNA. Kinetic analysis of transcription pulse assays shows that ORF57 protects PAN RNA from a rapid cellular RNA decay process, but ORF57 has little effect on transcription or PAN RNA localization based on chromatin immunoprecipitation and in situ hybridization experiments, respectively. Using a UV cross-linking technique, we further demonstrate that ORF57 binds PAN RNA directly in living cells and we show that binding correlates with function. In addition, we define an ORF57-responsive element (ORE) that is necessary for ORF57 binding to PAN RNA and sufficient to confer ORF57-response to a heterologous intronless β-globin mRNA, but not its spliced counterparts. We conclude that ORF57 binds to viral transcripts in the nucleus and protects them from a cellular RNA decay pathway. We propose that KSHV ORF57 protein functions to enhance the nuclear stability of intronless viral transcripts by protecting them from a cellular RNA quality control pathway.
| In order to replicate efficiently, a virus must ensure that its genes are properly expressed in the context of an infected host cell. Recent work has demonstrated that eukaryotic cells have RNA quality control pathways that degrade improperly processed, aberrant RNAs. Our published findings using an unusual Kaposi's sarcoma-associated herpesvirus (KSHV) nuclear RNA, called PAN RNA, have suggested that intronless polyadenylated transcripts are subject to such a quality control system. Because most KSHV genes lack introns, we hypothesized that KSHV must have evolved mechanisms that bypass this quality control system. In support of this idea, we show that the ORF57 protein, a multifunctional enhancer of KSHV gene expression, binds to and stabilizes PAN RNA. We further define an element called the ORF57-responsive element (ORE) in PAN RNA that is necessary for ORF57-binding and activity on PAN RNA. In addition, we show that the ORE is sufficient to confer ORF57-responsiveness to a heterologous intronless mRNA, but not its spliced counterpart. These observations substantiate the model that ORF57 enhances KSHV gene expression by protecting viral transcripts from host RNA surveillance pathways. More broadly, these data suggest that viruses producing intronless nuclear RNAs require mechanisms to evade host quality control mechanisms.
| Post-transcriptional events in mRNA biogenesis are of central importance to the fidelity and regulation of gene expression. Cellular factors regulate nearly every step of RNA metabolism including transcription elongation, RNA splicing, 3′ end formation, nuclear export, translation, etc. In fact, genome-wide profiling experiments demonstrate that a significant percent of the observed changes in RNA levels are dictated by regulation of the stability of a transcript rather than its transcription (e.g. [1],[2]). RNA half-life can be modulated directly, through the activities of regulatory stabilizing or destabilizing protein factors or small RNAs [3]–[5]. In addition, RNA quality control pathways ensure aberrant transcripts are less stable than their functional counterparts [6]. Given the importance of these pathways for gene expression, it is no surprise that viruses have evolved mechanisms to counteract pathways that otherwise would lead to the destruction of viral transcripts [3],[7].
The Kaposi's sarcoma-associated herpesvirus (KSHV) is a member of the gammaherpesvirus family that causes Kaposi's sarcoma, a common AIDS-associated malignancy, as well as the lymphoproliferative disorders primary effusion lymphoma (PEL) and some cases of multicentric Castleman's disease (MCD) [8]–[10]. The life cycle of KSHV includes a latent phase in which the viral DNA is maintained in infected host cells as a circular episome. During latency, few viral genes are expressed and no viral replication occurs. When the KSHV lytic phase is reactivated, a well-regulated cascade of gene expression is initiated by the viral transactivator ORF50 (Rta) resulting in infectious virus production [11]–[13]. Like all herpesviruses, the KSHV genome is nuclear and its genes are expressed utilizing the host cell transcription, RNA processing, and translation machinery. In many respects, KSHV genes resemble those of their host; that is, they have canonical promoter elements, 3′-end formation signals, and consensus pre-mRNA splice sites.
However, KSHV genes differ from canonical cellular genes in several relevant ways. Some transcripts are bicistronic, KSHV introns are smaller than the average size of a mammalian intron, and genes are more closely arranged in the genome than host genes. Most importantly for the present work, ∼70% of KSHV genes lack introns [14], whereas most human protein-coding genes contain multiple introns [15]. This difference in gene structure has implications for the expression of viral genes. The presence of an intron in a pre-mRNA and/or the changes in ribonucleoprotein particle (RNP) composition that result from splicing promote the efficiency of almost every stage of gene expression, including transcription initiation and elongation, 3′-end formation, mRNA export, RNA localization, and translation [16]–[27]. As a result, transgenes containing an intron are often expressed at significantly higher levels than those same genes lacking an intron [18]. To compensate for the lack of introns or splicing, viruses that express unspliced or intronless transcripts have evolved mechanisms that promote efficient gene expression in the absence of splicing [28]–[33].
KSHV encodes a viral post-transcriptional regulator of gene expression called ORF57 (Mta, KS-SM) that is essential for viral replication [34],[35]. ORF57 is a member of a conserved family of herpesvirus proteins that post-transcriptionally enhance gene expression [29]–[33]. ORF57 has been implicated in a variety of steps of RNA biogenesis from transcription to translation and it increases the efficiency of intronless gene expression [30],[31],[36],[37]. ORF57 has been reported to interact with ORF50 and to enhance transcription in a promoter and cell-type specific manner [37]–[39]. In addition, ORF57 binds cellular export factors and promotes the nuclear export of at least a subset of intronless viral mRNAs [40]–[43]. Unlike the herpes simplex homolog (HSV) ICP27, which contributes to host gene shut-off by inhibiting splicing, ORF57 promotes the splicing of some viral mRNAs [35],[44], and splicing activity is also seen with the Epstein-Barr virus (EBV) ORF57 homolog, SM [45]. ORF57 has further been suggested to play a role in translation of an internal ribosome entry site-containing reporter [46]. Thus, ORF57 is a multifunctional regulator of mRNA biogenesis that may, in part, compensate for the lack of introns in viral gene expression.
ORF57 is critical for the accumulation of the polyadenylated nuclear (PAN) RNA (nut1, T1.1) [34],[35],[37],[42], a non-coding nuclear transcript that accumulates to high levels during the lytic phase of viral infection [47],[48]. The PAN RNA promoter is ORF50-dependent [49],[50], and PAN RNA accumulation further depends on the activity of a 79-nucleotide (nt) RNA element, called the ENE [51]–[53]. Mechanistically, the ENE interacts in cis with the poly(A) tail of PAN RNA resulting in the sequestration of the poly(A) tail from exonucleases. Detailed kinetic analysis of the effects of the ENE on PAN RNA decay in transfected cells showed that PAN RNA is subject to two kinetically distinguishable decay pathways, one with a very short half-life (10–20 min) and another with a longer half-life (3–5 hrs). ENE-lacking or ENE-mutant PAN transcripts are more likely to be degraded in the rapid RNA decay pathway than are their wild-type ENE containing counterparts. Because the ENE is sufficient to increase the nuclear accumulation of heterologous intronless transcripts, we further proposed that this rapid decay pathway is part of a nuclear RNA surveillance system that rapidly degrades inefficiently exported mRNAs.
ORF57-mediated enhancement of the exclusively nuclear PAN RNA suggests that it may be involved in inhibiting the proposed RNA surveillance mechanism. Here, we test this idea and find that ORF57 stabilizes PAN RNA, particularly those transcripts that lack the ENE. We see no ORF57-dependent effect on RNA polymerase II (pol II) density on the PAN RNA gene nor does ORF57 lead to PAN RNA export. Therefore, we conclude that the observed stability enhancement constitutes the major effect of ORF57 on PAN RNA accumulation. In addition, ORF57 binds PAN RNA directly in vivo and a deletion of the 5′ portion of PAN RNA, dubbed the ORF57-responsive element (ORE), reduces ORF57 binding and ORF57 response. We show that tethering of ORF57 to an ORE-deleted PAN RNA restores ORF57-mediated up-regulation. Finally, we show that the ORE is sufficient to confer increased ORF57-response to a heterologous intronless β-globin mRNA, but not its spliced counterpart. We conclude that ORF57 protects viral transcripts from the same cellular RNA decay pathway that the ENE protects from in cis and that its stabilization activity is dependent on ORF57 binding to target RNAs.
If ORF57 protects transcripts from RNA decay pathways in vivo, we reasoned that the effects of ORF57 would be more pronounced on unstable ENE-lacking transcripts than on their ENE-containing counterparts. To test this idea, we compared the RNA levels of PAN RNA containing the ENE (PAN-WT) to PAN RNA lacking the ENE (PAN-Δ79) in the absence of ORF57 or in its presence. We transfected HEK293 cells with constructs that express PAN-WT or PAN-Δ79 and co-transfected ORF57-expression constructs at two concentrations or empty vector. After ∼18-24 hours, we extracted total RNA, and quantified relative RNA levels by northern blot (Figure 1). Consistent with published results [34],[35],[37],[42], ORF57 increases wild-type PAN RNA levels in a dose-dependent fashion (Figure 1A, lanes 1–3). Quantitation of these data show that, at the highest ORF57 concentration tested, PAN RNA is ∼3.4-fold more abundant (Figure 1B). As predicted from our model, ENE-lacking transcripts show an even greater response to ORF57, ∼11-fold (lanes 4–6 and Figure 1B). Because the ENE is involved in RNA stability, these results are consistent with the conclusion that ORF57 increases the half-life of PAN RNA.
To directly examine the effects of ORF57 on PAN RNA half-life, we employed a transcription pulse strategy [54]. In these experiments, we transfected HEK293 Tet-off advanced (293TOA) cells with TRP-Δ79, a plasmid that expresses the ENE-lacking PAN-Δ79 transcript from a tetracycline-responsive promoter [52]. In 293TOA cells, transcription from this promoter is turned off in the presence of doxycycline (dox, a tetracycline analog), and is induced in its absence. In our initial experiments, we examined the effects of ORF57 on PAN RNA decay after a two-hour transcription pulse (Figure 2). As expected, TRP-Δ79 RNA was undetectable prior to transcription pulse, but can be detected after two hours in dox-free media (Figure 2A, top panels). Examination of the decay profiles clearly shows an increase in RNA stability when ORF57 is expressed (Figure 2B, top). Interestingly, the mobility of a portion of remaining transcripts after transcription shut-off is reduced while others show increased mobility. We have determined that these mobility changes are due to differences in poly(A) tail length (data not shown). The relationship between changes in poly(A) length and ORF57 function is currently under investigation and will be described elsewhere (see Discussion). We also examined the effects of ORF57 on TRP-WT, a PAN expression construct containing the ENE [52]. However, in 293TOA cells this plasmid produced an extremely stable transcript (t1/2>24hr), impractical for use in decay assays. Overall, these data demonstrate that ORF57 increases the half-life of unstable PAN RNA transcripts.
Previous studies showed that PAN RNA is subject to two decay pathways with different kinetic properties [51],[52]. That is, one pool of PAN RNA transcripts is degraded very rapidly with half-lives of ∼10-20 minutes, while another pool of transcripts is degraded more slowly (t1/2 ∼3-5hrs). The presence of the ENE appears to protect transcripts from the rapid decay system resulting in a decrease in the fraction of transcripts that are observed in this population. Consistent with these published findings, the decay profiles in Figure 2B are nicely fit by two-component exponential decay curves where the two components represent the two pools of transcripts. Using regression analysis, we can determine the decay parameters in PAN RNA degradation, including the fraction of transcripts undergoing rapid decay and the half-life of each population. Because ENE-lacking transcripts are preferentially up-regulated by ORF57, we predicted that, like the ENE, ORF57 expression would decrease the fraction of PAN transcripts in the rapid RNA decay pathway.
To test this idea, we performed regression analysis of the data for TRP-Δ79 RNA decay in the presence or absence of ORF57 (Figure 2B, Table S1 and Figure S1). Examination of the kinetic parameters shows that in the absence of ORF57, 73% of the transcripts are in the rapidly degrading population (t1/2 ∼7.8 min) (Figure 2C). In contrast, only 51% of the transcripts are degraded rapidly when ORF57 is co-expressed, a statistically significant decrease. Because the more slowly degrading transcripts accumulate over time, the fraction of transcripts observed in the rapidly degrading pool decreases when longer transcription pulse times are employed [51]. If ORF57 decreases the fraction of transcripts that degrade rapidly, it follows that the observed rapidly degrading fraction would decrease more quickly when ORF57 is present. Indeed, the effects of ORF57 on TRP-Δ79 RNA decay are even more apparent after an 18-hour transcription pulse (Figure 2A and Figure 2B). In this case, the apparent half-life (i.e. the time difference at 50% remaining, Figure 2B) is increased ∼8-fold. More importantly, the percent of transcripts degrading rapidly in the presence of ORF57 is reduced to 15%, while 57% is rapidly degraded in its absence (Figure 2C). Taken together, these data strongly argue that ORF57 enhances PAN RNA levels by protecting it from a rapid cellular RNA decay pathway.
Previous reports suggested that ORF57 enhances transcription rates of specific promoters in certain cell types, including the PAN RNA promoter in 293 cells [37]–[39]. Even though we see an increase in PAN RNA half-life in the presence of ORF57, it remains possible that a significant portion of the up-regulation of PAN RNA by ORF57 is at the level of RNA synthesis rather than decay. To test the effects of ORF57 on transcription initiation, we initially examined the response of PAN RNA to ORF57 from three different promoters (Figure 3A). Consistent with the results of others, PAN RNA expression from the cytomegalovirus immediate early (CMVIE) promoter is responsive to ORF57 [36],[37],[39]. We extended this analysis by examining PAN RNA steady-state levels driven by the cellular elongation factor 1α (EF1α) and viral SV40 promoters (Figure 3A). Each of these constructs is 3′ processed using the PAN RNA cleavage and polyadenylation signals. For every promoter tested, ORF57 increased PAN RNA levels in a dose-dependent fashion supporting a post-transcriptional role for ORF57. Interestingly, the magnitude of the change differs among the constructs and this does not necessarily correlate with the strength of each promoter. For example, as judged by overall RNA levels, the PAN and SV40 promoters are the strongest and weakest promoters, respectively (Figure S2). However, PAN-WT RNA driven by each of these promoters is increased by similar margins at the highest ORF57 expression levels (3.5-fold and 2.8-fold) (Figure 1B and Figure 3A). In contrast, the CMVIE and EF1α promoters both show greater increases in steady-state levels in response to ORF57 (9.5 and 6.7-fold, respectively). These data suggest that the effect of ORF57 on PAN RNA is not due to any promoter-specific element, but that magnitude of the ORF57 enhancement may be linked to a qualitative difference in these promoters.
To further test the effects of ORF57 on PAN RNA transcription, we examined the pol II density on the PAN RNA gene by chromatin immunoprecipitation assays (ChIP). In these experiments, ChIP was performed using antibodies to RNA polymerase II and the relative levels of co-immunoprecipitating DNA were compared. We examined DNA from the 5′ and 3′ ends of PAN RNA (Figure 3B) to assess polymerase density across the PAN gene. First, we determined the relative polymerase density on TRP-Δ79 in the presence and absence of ORF57 (Figure 3B) and saw only a slight increase in polymerase density on either the 5′ or 3′ portion of the PAN gene when ORF57 was co-expressed (∼1.3-fold). As expected, in the presence of dox, the polymerase density significantly decreases. Second, we assayed polymerase density on the PAN RNA gene when transcription is driven from the PAN promoter (Figure 3C). Because this promoter depends on the ORF50 viral transactivator, we included samples lacking ORF50 as negative controls. Comparing samples containing or lacking ORF57, we observed little difference in polymerase density at the 5′ end of the PAN gene. Similarly, we saw minimal ORF57-dependent difference in signal from the 3′ end of PAN gene, although there is lower signal in this sample (data not shown). Taken together, these data strongly support the conclusion that the predominant effect of ORF57 on PAN RNA steady-state levels occurs subsequent to transcription initiation.
ORF57, like its homologs in other herpesviruses, has been implicated in the nuclear export of intron-lacking viral mRNAs [29], [32], [40]–[42],[55]. Because the machinery involved in RNA decay differs between the cytoplasm and the nucleus, it stands to reason that a change in subcellular localization would affect RNA decay profiles. Therefore, we performed in situ hybridization to verify that PAN RNA remains nuclear in the presence of ORF57. As shown in Figure 4, PAN-Δ79 RNA localizes to faint spots in the nucleus (top panels), with a few cells demonstrating a more diffuse pattern (data not shown and Figure S3). In the presence of ORF57, PAN-Δ79 RNA is also observed strictly in the nucleus (Figure 4, middle panels). The signal intensity, the number of cells showing signal above background, and the percentage of cells with diffuse nuclear staining were all increased in the presence of ORF57 and account for the higher levels of PAN-Δ79 in the presence of ORF57 (data not shown). Importantly, no cytoplasmic PAN RNA signal was detected. In addition, TRP-Δ79 RNA was observed exclusively in the nucleus (Figure S3), as was wild-type PAN RNA driven from either the PAN or TRP promoters (data not shown). As an additional control, since many in situ hybridization protocols will “wash away” cytoplasmic RNA, we tested whether we could detect cytoplasmic signal by examining the localization of a spliced β-globin reporter mRNA. In this case, cytoplasmic signal was observed (Figure 4, bottom panels). These results demonstrate that the increase in PAN RNA stability in the presence of ORF57 is not the result of PAN RNA export to the cytoplasm. Because ORF57 enhances RNA stability, but has little or no effect on PAN RNA export or transcription, we conclude that the effect of ORF57 on PAN RNA accumulation is the result of increased nuclear RNA stability.
ORF57 could stabilize PAN RNA by one of two non-mutually exclusive mechanisms. First, ORF57 may inhibit the activity of RNA decay enzymes, either by binding and inactivating them directly or by decreasing their expression. Second, ORF57 may interact with transcripts and protect the bound RNAs, directly or indirectly, from decay enzymes. The second model predicts that ORF57 is in a complex with PAN RNA and that this binding is necessary for protection by ORF57. Even though a previous report suggested that ORF57 did not bind to PAN RNA in vitro [56], we investigated whether ORF57 bound to PAN RNA in vivo.
One difficulty in examining RNA-protein interactions using co-immunoprecipitation techniques is that RNA-protein complexes frequently reassort in cellular extract [57],[58]. That is, an RNA-binding protein will associate with specific transcripts in an extract that were not bound in vivo. Therefore, in order to conclude that an RNA-protein interaction occurs in cells, it is imperative to test whether a given RNA-protein complex forms subsequent to lysis. To do this, we employed a “cell-mixing” experiment [57],[58] (Figure 5A). We induced lytic reactivation of KSHV in HH-B2 cells, a latently infected PEL cell line, with sodium butyrate (NaB) to initiate the expression of lytic genes including PAN RNA. After 24 hours, we mixed the lytically reactivated cells with HEK293 cells transfected with Flag-tagged ORF57 (Fl-ORF57) or with untagged ORF57 expression constructs. After combining intact cells, the cells were lysed and subjected to immunoprecipitation (IP) with anti-Flag antibodies. PAN RNA, which is exclusively derived from the HH-B2 cells, was efficiently and specifically immunoprecipitated with the anti-Flag beads (Figure 5B). Because PAN RNA is not produced in the same cells as the Fl-ORF57, this result clearly indicates that ORF57 interacts with RNAs in cell lysate and that RNAs that co-immunoprecipitate with ORF57 do not necessarily reflect RNP composition in vivo. As a result, caution must be taken in the interpretation and experimental design of RNA immunoprecipitation experiments with ORF57.
To control for ORF57-RNA reassortment in cell extract, we employed an ultraviolet light (UV) cross-linking protocol [57]. We transfected HEK293 cells with PAN-WT and Fl-ORF57 or untagged ORF57 expression plasmids and exposed the cells to UV light to covalently cross-link protein to RNA. Cells were lysed under stringent conditions and ORF57 was immunoprecipitated with anti-Flag antibodies. Subsequently, we detected PAN RNA signal by northern blot and quantitated this signal as a percent of input RNA (Figure 5C and 5D). Using this procedure, PAN RNA is immunoprecipitated from extracts containing Fl-ORF57, but not the untagged control (compare lanes 5 and 6, top panel). Most importantly, because living cells were exposed to UV, a UV-dependent interaction reflects an interaction in cells. Because the “no UV” control shows little detectable signal (lane 4), we conclude that ORF57 associates with PAN RNA in cells. Moreover, UV cross-linking is limited to interactions in which the RNA and protein are in close contact, so we can further conclude that ORF57 binds directly to PAN RNA in vivo.
Little is known about the requirements for ORF57 recruitment or RNA-binding in vivo, so we tested whether cis-acting elements in PAN RNA are necessary for the ORF57-PAN RNA interactions. We utilized a series of four previously described PAN RNA expression constructs with 300-nt overlapping deletions in PAN RNA, called PANΔ1-Δ4 [53]. ORF57 cross-linked to PANΔ2, PANΔ3, and PANΔ4 with similar efficiency as PAN-WT (Figures 5C and 5D). In contrast, cross-linking to PANΔ1 RNA was only slightly higher than background signal. We previously reported that PANΔ1 RNA is expressed at lower levels than PAN-WT RNA [53] (see below). To verify that the lower expression of PANΔ1 was not responsible for its lack of ORF57 binding, we performed experiments in which we transfected less of the PAN-WT expression plasmid, to render its steady-state levels similar levels to PANΔ1 (data not shown). In this case, efficient binding to PAN-WT was maintained, so we conclude that reduced ORF57-PANΔ1 cross-linking is not due to lower PANΔ1 expression levels. These results demonstrate that sequences near the 5′ end of PAN RNA are necessary for efficient cross-linking of ORF57 to PAN RNA in cultured cells.
The model that ORF57-RNA interactions are required for activity predicts that diminished binding of ORF57 to PANΔ1 will result in decreased ORF57 activity. Therefore, we examined the accumulation of PANΔ1 in the presence of increasing amounts of ORF57 (Figure 6A). Under the same conditions in which ORF57 increases PAN-WT by 3.4-fold and PAN-Δ79 by 11-fold (Figure 1), we observe no statistically significant change in PANΔ1 levels in the presence of ORF57. Thus, the loss of ORF57 binding to PANΔ1 correlates with loss of PAN RNA up-regulation. Taken with data presented below, our results show that the 5′ end of PAN RNA (nt 1–312) contains an ORF57-responsive element (ORE).
The loss of binding of ORF57 to PANΔ1 could be due to any one of several non-mutually exclusive models. For example, the ORE may contain a high-affinity binding site for ORF57. Alternatively, there may be a cis-acting sequence in the ORE whose activity is necessary for ORF57 function. As a result, loss of this cis-acting sequence disrupts the ORF57 up-regulatory pathway. Consistent with this idea, previously published data suggested that nt 13–312 contain an activity important for PAN RNA accumulation [53]. Therefore, we examined whether ORF57-responsiveness correlates with the RNA accumulation activity provided by this region of PAN RNA. Using northern blots (data not shown), we confirmed that PANΔ1 is expressed at ∼5-fold lower levels than PAN-WT when ORF57 is not co-expressed (Figure 6B). In contrast, when PANΔ1 is placed behind a CMVIE promoter (CMV-Δ1), the expression levels are reduced only slightly, ∼30% (Figure 6B). Thus, this cis-acting PAN RNA accumulation activity found in nt 13–312 is not operative when a CMVIE promoter is used to express PAN RNA. Using the CMVIE-driven PANΔ1 constructs, we next tested the hypothesis that this activity is related to ORF57-responsiveness. Quantitation of northern blot data show that CMVIE-driven PANΔ1 remains significantly less responsive to ORF57 than CMVIE-driven wild-type PAN RNA (Figure 6C). CMVIE-driven WT PAN was up-regulated nearly 10-fold in the presence of ORF57, while the CMVIE-driven PANΔ1 RNA levels increase by a factor of only ∼2-fold. Therefore, we conclude that the lack of ORF57-responsiveness of PANΔ1 is unrelated to its reduced levels from the PAN promoter. Moreover, these data show that the ORE functions in a promoter-independent fashion.
Loss of ORF57 binding to PAN RNA correlates with loss of function, supporting the model that RNA-binding is necessary for ORF57 responsiveness. We next tested whether restoration of ORF57-binding was sufficient to restore ORF57-responsiveness to PANΔ1. To do this, we employed a tethering system using the bacteriophage MS2 coat protein, which binds with high affinity to a well-defined bacteriophage RNA hairpin sequence [59],[60] (Figure 7). We expressed an amino-terminal fusion of the bacteriophage MS2 coat protein with ORF57 (NMS2-NLS-Fl-ORF57) and verified that the fusion did not abrogate ORF57 activity on CMV-driven full-length PAN RNA (lanes 1–8). Expression of this construct increases PAN RNA accumulation similarly to Fl-ORF57, (compare lanes 4 and 5 with 2 and 3), while expression of the MS2-NLS-Fl protein alone did not increase PAN RNA levels (lanes 6 and 7). Therefore, we conclude that NMS2-NLS-Fl-ORF57 maintains ORF57 activity. Next, we co-expressed NMS2-NLS-Fl-ORF57 fusion protein with a CMV-Δ1 derivative that includes six binding sites for the MS2 coat protein (Figure 7, lanes 12,13). In this case, PAN RNA levels increase ∼7-fold. Expression of neither Fl-ORF57 nor MS2-NLS-Fl alone had a similar effect (lanes 10–11 and 14–15, respectively). Importantly, the increase depends on the presence of MS2-binding sites in the RNA: CMV-Δ1 shows minimal response to MS2-NLS-Fl-ORF57 expression (lanes 20, 21). Thus, tethering of ORF57 to PANΔ1 transcripts can complement the lack of ORF57-responsiveness of PANΔ1. We conclude that the stabilization of PAN RNA by ORF57 depends on direct interactions between ORF57 and its target RNA.
The data presented above show that the ORE is necessary for ORF57 responsiveness in the context of PAN RNA. We next tested whether the ORE is sufficient to confer increased ORF57 response to a heterologous transcript. For these experiments, we used a series of previously described β-globin reporter constructs [53]. While wild-type β-globin contains two introns (Figure 8A, β-wt), these reporters delete either the first (βΔ1) or both (βΔ1,2) β-globin introns. Into the 3′ UTR of the βΔ1,2 construct, we cloned ∼300 nt PAN RNA sequences (Figure 8A, right). The sequences, named PF1-PF4 (PAN fragment 1–4), correspond to the sequences deleted in PANΔ1-PANΔ4 (Figure 5C), respectively, except PF1 extends to the 5′-most nucleotide (nt 1-312) and PF4 extends to the nucleotide immediately preceding the polyadenylation hexamer (nt 703–1052). We co-transfected these constructs with the Fl-ORF57 expression plasmid and monitored β-globin mRNA levels by northern blot (Figure 8B, lanes 1–15). Two different schemes were used to normalize the data (Figure 8C). First, to show the response of each mRNA variant to Fl-ORF57, we normalized the β-globin mRNA signal in the presence of Fl-ORF57 to the signal from the same construct in its absence (Figure 8C, top). Second, we normalized all the data to the βΔ1,2 plus 0.4 µg ORF57 (lane 3) to yield information about the expression levels of the reporter mRNAs relative each other (Figure 8C, bottom).
Our data conclusively demonstrate that the ORE increases the response of intronless β-globin mRNA to ORF57. In the absence of any PAN fragment, ORF57 increases the expression levels of intronless β-globin mRNA in a dose-dependent fashion reaching ∼4-fold at the highest ORF57 levels tested (Figure 8B lane 3, Figure 8C). When PF1, which contains the ORE, is placed in the β-globin 3′ UTR, ORF57 has an even greater effect: βΔ1,2-PF1 transcripts are up-regulated ∼23-fold. This effect is specific to PF1 because PF2, PF3, PF4 do not increase ORF57 response. As previously shown [51],[53], the ENE-containing insert (PF4, lanes 13–15) increases the levels of intronless β-globin mRNA. However, as in the case of PAN RNA, ORF57-responsiveness decreases (∼2-fold) when the ENE is included in the transcript (Figure 1). These data show that the ORE is sufficient to confer increased response to ORF57 in a heterologous context. Because the ORE was inserted into the β-globin 3′ UTR, we can further conclude that ORE activity is not strictly dependent upon being at the 5′ end of the transcript.
Published data suggest that intronless mRNAs are subject to degradation by a cellular RNA quality control system [51]–[53],[61],[62] and the data presented above support the model that ORF57 protects transcripts from this RNA decay pathway. If this is the case, we reason that ORF57 should have a limited effect on spliced mRNA abundance, even if the mRNAs contain the ORE, because spliced transcripts are not subject to the cellular RNA quality control pathway. To test this idea, we examined the effects of ORF57 on spliced β-globin mRNA levels containing (βΔ1-PF1) or lacking (βΔ1) the ORE (Figure 8A). Quantitation of β-globin mRNA accumulation by northern blot showed that neither of these mRNAs was significantly altered by ORF57 (Figure 8B, lanes 16–21, and Figure 8C). Examination of the RNA levels shows that the spliced mRNAs accumulate to higher levels than the mRNAs generated from intronless genes, as expected (Figure 8C, bottom). Interestingly, the levels of the ORE-containing intronless transcripts approach those of the spliced mRNAs under the highest levels of ORF57. These results are consistent with the model that ORF57 protects viral mRNAs from cellular RNA decay factors that preferentially degrade transcripts generated from intronless genes.
In this report we uncover three novel findings about the ORF57 protein, an essential KSHV protein involved in viral gene expression. First, these data establish a role for ORF57 in stabilizing nuclear transcripts. Second, our data show that ORF57 binds directly to its target, PAN RNA, in living cells and that binding correlates with function. Third, ORF57 and its homologs can increase the expression of a variety of mRNAs [30]–[33], suggesting a relatively nonspecific effect. However, here we demonstrate the existence of an ORF57-responsive element in PAN RNA suggesting that, at least in some cases, specific cis-acting sequences have evolved to recruit ORF57 to its targets. Consistent with previous reports, our data further show that ORF57 primarily enhances transcript accumulation from intron-lacking genes but not from intron-containing genes, even when the ORE is included in the spliced transcript. Together, these data are consistent with a model in which ORF57 binds to intronless viral transcripts and protects them from a cellular RNA quality control pathway.
The data presented here demonstrate ORF57 stabilizes nuclear RNAs that would otherwise be rapidly degraded. Using steady-state analysis (Figure 1) and a transcription pulse assay (Figure 2), we show that ORF57 increases the stability of an unstable polyadenylated nuclear RNA. Kinetic analysis further supports the conclusion that ORF57 is protecting RNAs, at least in part, from the same rapid RNA decay pathway that the ENE protects transcripts from in cis [51],[52]. Even though ORF57 has been implicated in mRNA export [40]–[42], its expression does not lead to cytoplasmic accumulation of PAN RNA. Thus, we conclude that the effect of ORF57 on RNA stability is independent of its proposed function in RNA export. This conclusion is consistent with previous reports implying that ORF57 stabilizes its target RNAs [39],[42] and with the observations that ORF57 homologs SM and ICP27 can stabilize specific transcripts [63],[64]. However, these data are the first direct demonstration that KSHV ORF57 increases the half-lives of nuclear RNAs and that his function is separable from its proposed roles in mRNA export, transcription, and translation.
We observe only a slight increase in polymerase density on PAN RNA genes driven from either the PAN or tetracycline-responsive promoter (Figure 3). Therefore, we conclude ORF57 increases PAN RNA levels primarily by a post-transcriptional mechanism. We were surprised at the lack of increase in pol II density on the constructs driven by the PAN promoter, which binds the viral transactivator ORF50, an ORF57-interacting protein [38],[39]. Assuming these proteins associate in our experimental system, the interaction appears to have little effect on transcription initiation. It should be noted that we used an antibody (8WG16) that preferentially recognizes the initiating hypophosphorylated form of pol II [65]. Thus, it remains possible that a hyperphosphorylated elongating form of pol II increases on the PAN RNA gene in response to ORF57. Taken with the observation that PAN RNA is up-regulated by ORF57 from four different promoters (Figure 3A), our data strongly support the conclusion that transcription initiation is unaffected by ORF57.
ORF57 enhances the expression of viral mRNAs, the noncoding nuclear PAN RNA, and heterologous reporter mRNAs, so it appeared that its effects were not strongly influenced by cis-acting sequences. On the contrary, our data demonstrate that a specific cis-acting sequence, the ORE, can enhance the effects of ORF57 on both PAN RNA and on a β-globin reporter mRNA. Using a previously described set of deletions [53], we show that ORF57 has reduced binding to PANΔ1 RNA and that this correlates with loss of activity (Figure 5, Figure 6). Moreover, we placed the ORE into an intronless β-globin mRNA and found that it enhances ORF57-responsiveness by ∼5-fold, demonstrating that the ORE affects mRNAs as well as the noncoding PAN RNA (Figure 8). Tethering ORF57 to the ORE-lacking PANΔ1 transcripts in cells is sufficient to complement the ORE deletion, so it seems likely that ORF57 recruitment resides at the core of ORE activity (Figure 7). The simplest interpretation of these results is that the ORE is a high-affinity ORF57-binding site, and that RNA binding by ORF57 is necessary for its stabilization activity. Interestingly, intronless β-globin mRNA levels are enhanced by ORF57 ∼4-fold in the absence of the ORE. Perhaps ORF57 has enough non-specific RNA-binding activity to account for its general effects on reporter RNAs. Alternatively, this effect may be linked to a separate ORF57 activity that functions independently of RNA binding.
Mechanistically, our data support the model that ORF57 binds to its RNA targets and inhibits the activity of nuclear RNA decay enzymes, but we do not yet know the molecular details of ORF57-mediated RNA stabilization. In one model, ORF57 binds RNA making it inaccessible to RNA decay enzymes. Alternatively, ORF57 could indirectly stabilize transcripts by promoting changes in RNP composition or conformation. It remains formally possible that ORF57 increases PAN RNA stability by retaining the transcripts in the nucleus, thereby protecting them from cytoplasmic decay enzymes. However, given the reported role of ORF57 in mRNA export [40]-[42] and its ability to shuttle [66], we think this last hypothesis is unlikely.
Our data are consistent with the model that ORF57 counteracts a nuclear RNA quality control pathway that rapidly degrades transcripts that derived from intronless genes [51]–[53],[61],[62]. It is important to point out that this model does not depend on cells specifically recognizing intronless RNAs. Rather, because of the extensive coupling between the steps of cellular RNA biogenesis including pre-mRNA splicing [16]–[27], we believe it more likely that intronless RNAs are inefficiently processed. If RNA surveillance and RNA maturation are in kinetic competition as formally proposed by Doma and Parker [6], these inefficiently processed intronless transcripts would be predicted to be subject to degradation by RNA quality control pathways. Because the majority of KSHV genes are intronless [14], we propose that ORF57 functions to counteract this RNA decay pathway to promote the robust expression of viral genes. ORF57 has been reported to promote the export of intronless viral mRNAs [40]–[43], so it is easy to imagine that ORF57 allows mRNAs to bypass nuclear decay systems by enhancing the efficiency of their export. However, in the case of PAN RNA, we can uncouple ORF57 mRNA export activities from its function in RNA stabilization. Therefore, our data suggest a more active role for ORF57 in protecting transcripts from degradation in the nucleus. Current work focuses on further testing this model by identifying the cellular decay machinery involved as well as the viral RNAs bound and protected by ORF57.
Because PAN RNA accumulates to such high levels in KSHV lytically reactivated cells [47],[48],[49], it likely performs an important function for the virus. Therefore, studying the biogenesis of this unusual transcript is essential to understanding KSHV biology. In addition, PAN RNA provides a useful a tool to separate ORF57 functions in RNA stability from its role in RNA export. We used an unstable ENE-lacking PAN RNA for our decay studies (Figure 2), but several observations suggest an important role for ORF57 in PAN RNA biogenesis in infected cells. First, steady-state levels of PAN RNA containing the ENE are up-regulated in the presence of ORF57 (Figures 1 and 3). Second, published reports have shown that ORF57-deleted bacmids produce reduced levels of PAN RNA during lytic infection [34],[35]. Finally, we have observed that insertion of multiple copies of the ENE leads to higher levels of PAN RNA (unpublished observations) or βΔ1,2 mRNA [53] than insertion of one ENE, suggesting that a single ENE does not completely block RNA degradation. Thus, the proposed overlapping activities of the ENE and ORF57 may both be essential to fully stabilize PAN RNA during lytic replication. Perhaps more importantly, our studies provide insights into the possible mechanism of ORF57 activity on the accumulation of intronless viral transcripts that lack ENE-like elements. However, further experimentation is necessary to test the role that ORF57 plays on PAN RNA stability in the context of viral infection and on the stability of intronless viral mRNAs.
A particularly interesting component of ORF57-mediated RNA stabilization is the role of the poly(A) tail in regulation of transcript stability. After transcription shut-off in the presence of ORF57, we found that some transcripts become hyperadenylated, while others are partially deadenylated (Figure 2, data not shown). Because both the hypo- and hyperadenylated PAN transcripts are present 8 hrs subsequent to transcription shut-off, it seems that ORF57 stabilizes both forms. Indeed, when we over-expose our northern blots, we observe transcripts resembling these hyper- and hypoadenylated forms in the samples from cells lacking ORF57, but at significantly reduced levels (data not shown). Several different non-exclusive roles for the poly(A) tail in nuclear RNA stability can be imagined that are consistent with our observations. In the first model, PAN RNA is recognized by the cell as an aberrant transcript, presumably due to its lack of export. In manner analogous to yeast and bacterial systems, hyperadenylation of the transcripts is linked to quality control [67],[68]. Interestingly, the host-shutoff mechanism employed by KSHV appears to involve destabilization of hyperadenylated cellular mRNAs [69]. Moreover, recent work implicates mRNA export factors as regulators of poly(A) length in cells and in polyadenylation assays performed in vitro [70]–[73]. In the second model, ORF57 promotes polyadenylation, which then leads to greater transcript stability. In a third model, ORF57 stabilizes nuclear PAN RNA, and the hyperadenylation results from promiscuous polyadenylation of the stabilized nuclear transcripts. Distinguishing among these models will shed light into host-virus interactions between KSHV ORF57, cellular poly(A) machinery, and cellular RNA decay pathways.
HEK293, 293TOA, (Clontech) and 293A-TOA cells were grown in Dulbecco's Modified Eagle's Medium (Sigma) supplemented with 10% fetal bovine serum (FBS), 1X penicillin-streptomycin (Sigma), and 2 mM L-glutamate. TOA media utilized tetracycline-free FBS (Clontech) and was supplemented with 100 µg/mL G418 (Fisher Scientific). HH-B2 cells [74] were cultured in RPMI-1640 media (Sigma) supplemented with 15% FBS, 1X penicillin-streptomycin (Sigma), and 2 mM L-glutamate. Transfections were performed using TransIT-293 reagent as per the manufacturer's protocol (Mirus). Most ORF57 titration experiments were performed in 12-well tissue culture plates with a total of 0.7–0.8 µg plasmid DNA. A typical transfection with PAN-promoter constructs contained 0.15 µg of the PAN construct, 0.15 µg of an ORF50 expression construct, which is necessary for transcription from the PAN promoter, 0.1 µg of a control (mgU2-19/30) and 0.4 µg of Fl-ORF57 plus pcDNA3. The control plasmid generates two products; one is a spliced non-coding transcript and the other is an intron-derived scaRNA important for methylation of U2 snRNA [75]. These transcripts show little or no response to ORF57, under our typical transfection conditions. However, they may slightly increase (<2-fold) in the presence of ORF57 under the conditions used in Figure 7 (see below). The result of this slight increase would yield a relative underestimation of the effects of ORF57 in Figure 7, so it does not alter our conclusions. When other promoters were used, 0.3 µg of PAN RNA or β-globin expression plasmids were transfected. Total RNA was harvested 18–24 hrs post-transfection using TRI Reagent (Molecular Research Center). The experiments shown in Figure 7 were slightly different. In this case, we transfected 0.1 µg of the PAN RNA expression plasmids, 0.1 µg of the control plasmid and 0.4 and 0.8 µg of the Fl-ORF57, MS2-NLS-Fl-ORF57, or MS2-NLS-Fl expression constructs. Total RNA was harvested 44–50 hrs post-transfection. The 293A-TOA cell line was generated by transfecting 293A cells (Invitrogen) with a tTA-expressing plasmid (Clontech). Stable transformants were selected in 300 µg/ml G418 and cloned by limiting dilution.
We have previously described the transcription pulse assay and the quantitation of PAN RNA decay kinetics [51],[52]. We previously used HeLa Tet-off cells, but the expression of ORF57 in HeLa tet-off cells led to abrogation of the tetracycline-responsiveness. The reason for this observation remains unknown. In the current studies, we utilized 293TOA cells.
PAN-WT, PANΔ1, PANΔ2, PANΔ3, PANΔ4 and β-globin reporters were described in [53], while PAN-Δ79, TRP-WT, and TRP-Δ79 were described in [52]. PcFl-ORF57II (used to express Fl-ORF57) was generated by PCR amplification ORF57 from KSHV DNA using primers NC495 (5′ ATTAGCGGATTCATGGTACAAGCAATGATAGAC 3′) and NC496 (5′ AAAAGGCTCGAGTTAAGAAAGTGGATAAAAGAATAAACCC 3′). Underlined sequence show relevant restriction sites. The product was digested with BamHI and XhoI and inserted into pcDNA-Flag (gift of Jens Lykke-Andersen, University of Colorado) cut with the same. The sequence of all constructs using PCR-based methods was verified. Consistent with previous reports [36],[66], this construct had a silent mutation (CAT to CAC) encoding for the His at amino acid 261.
The CMVIE-WT (pcPAN) was made by PCR amplification of PAN transcribed region and downstream sequence using primers NC39 (5′ ATTTCCAAGCTTACTGGGACTGCCCAGTCACCTTGGCTGCCGCTTCACC 3′) and NC42 (5′ TAAAGCGGGCCCCCATCCCAATCGACGCAA 3′). The product was cut with ApaI, blunted, cut with HindIII and inserted into pcDNA3 digested with BbsI, blunted, and then digested with HindIII. The CMV-Δ1 derivative of this construct (pcPANΔ1) was generated by PCR amplification with primers NC494 (5′ CTCCGAAAGCTTACTGGGACTGCCATTCAATC 3′) and NC10 (5′ GGGGGCCCGTCACATTTAGGGCAAAGTGG 3′) using PANΔ1 as a template. The product was digested with XbaI and HindIII and inserted into pcPAN. The SV40 and EF1a-driven constructs were also derivatives of pcPAN. In this case, pcPAN was digested with NruI and HindIII to remove the CMVIE promoter and SV40 or EF1a promoters were inserted using the same restriction sites. Promoter sequences for SV40 and EF1a were amplified using primers NC418 (5′ GATCTCGCGACTCCCCAGGCAGGCAGAAGT 3′) and NC419 (5′ GATCAAGCTTTGGATATACAAGCTCCCGGG 3′), and NC416 (5′ GATCTCGCGAGGCTCCGGTGCCCGTCAGTG 3′) and NC417 (5′ GATCAAGCTTGAACGTTCACGGCGACTACT 3′), respectively.
The NMS2-NLS-Fl expression vector (pcNMS2-NLS-Fl) was constructed by amplifying the MS2 coat protein coding sequence using primers NC448 (5′ AGACCCAAGCTTGCCACCATGGCTTCT 3′) and NC449 (5′ GGATCCAAGCTTAGATCCACCCTTGTCATCGTCGTCCTTGTAGTCTACCTTTCTCTTCTTTTTTGGTCCACCTCCACCTCCGTAGAT 3′) and pcNMS2 [59] as a template. The resulting fragment was digested with HindIII and inserted into pcNMS2 digested with HindIII. NMS2-NLS-Fl-ORF57 was generated by insertion of the BamHI-XhoI fragment of pcFl-ORF57II into pcNMS2-NLS-Fl. The CMV-Δ1-XMS2 construct (pcPANΔ1-XMS2) was generated by amplifying six MS2-binding sites using primers NC584 (5′ AAATGCTCTAGAAACTACCAAACTGGGTCTAG 3′) and NC134 (5′ GACCCTAGATCTACTATAGAATAGGGCCCTCT 3′), digestion of the resulting product with XbaI, and insertion into pcPANΔ1 digested with XbaI.
For ChIP assays, one 10 cm plate of HEK293 cells was used (∼107 cells) per sample. Twenty-four hours post-transfection (plus/minus dox as indicated), methanol-free formaldehyde was added to the media at final concentration of 0.75%. Plates were incubated for 10 min and formaldehyde was quenched with 125 mM glycine for 5 min. After washing three times in ice-cold 1X phosphate buffered saline (Sigma), cells were harvested with a rubber policeman and collected by centrifugation at 3500×g for 3 min. Pellets were resuspended in 500 ml RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS) with 1X protease inhibitors (cocktail V, Calbiochem), and 1 mM phenylmethanesulphonylfluoride (PMSF), sonicated 6 times for 5 seconds using a Branson Sonifier 450 with a 4.8 mm diameter micro tip producing an average DNA size of ∼250–1000 bp. The extracts were then centrifuged at 800×g for 5 minutes at 4° and the supernatant was pre-cleared for one hour with 20 µl of Protein-A agarose (Pierce). After centrifugation to remove the beads, absorbance 260 was determined. While this does not accurately reflect DNA concentration due to the complexity of the extract, the value can be used to equilibrate extract concentrations for immunoprecipitation (5% was placed at −20° as “input”. Approximately 8 µg of 8WG16 antibody (Abcam) was added to extract (except no antibody control) and the mixture was nutated overnight at 4°. In addition, 20 µl of Protein-A agarose were blocked overnight at 4° with 0.5 mg/ml sheared salmon sperm DNA and 0.1 mg/ml bovine serum albumin (BSA) in RIPA. The next day, the beads were added to the antibody-extract mixture and the nutated for 1.5 hr at 4°. The beads were then washed a total of six times by nutating at room temperature for 3 minutes in 1 mL of the following solutions: 1) RIPA, 2) low salt wash (0.1 SDS, 1% TritonX100, 2 mM EDTA, 20 mM TRIS (pH 8), 150 mM NaCl), 3) high salt wash (0.1 SDS, 1% TritonX100, 2 mM EDTA, 20 mM TRIS (pH 8), 500 mM NaCl), 4) LiCl wash (0.25 M LiCl, 1% NP40, 1% sodium deoxycholate, 1 mM EDTA, 10 mM Tris pH 8.0), 5) TE (10 mM Tris pH 8.0, 1 mM EDTA), 6) TE. After washing, the beads were nutated in 150 mL of elution buffer (1% SDS, 100 mM NaHCO3, pH 9.0) for 15 min. The elution step was repeated, the eluted fractions were combined, 60 µl 1M Tris-HCl (pH 6.8) was added to the eluted complexes, proteinase K was added to 0.2 mg/ml and the samples were incubated at 37° for 60 min. The crosslinks were subsequently reversed at 65° for 5–18 hr. The samples were extracted with phenol-chloroform isoamyl alcohol (25∶24∶1), ethanol precipitated in the presence of 0.3M sodium acetate and 20 µg GlycoBlue (Ambion). The pellets washed with 70% ethanol and resuspended in 20 µl of water.
Input and pellet DNA was diluted 1∶100 and 1∶5, respectively and 2 ml of this was used as template for a 20 µl real-time PCR reaction containing iTAQ fast SYBR green supermix (Bio-Rad) with a final concentration of 100 nM primers. Real-time PCR parameters were 40 cycles of 95° for 3 sec and 60° for 30 sec using a 7500 Fast Real-time PCR system (Applied Biosystems). Primers used were: PAN 5′, NC527 (5′ CGCCGATTGTGGGTTGA 3′) and NC528 (5′CGAAAGCCAGGATGGGTATATT 3′); PAN 3′, NC550 (5′ TGTTTTAATGTGTATGTTGTGTTGGAAGT 3′) and NC551 (5′ TTCACCTACAAGAAAACATCGTTAGTC 3′); GAPDH, NC533 (5′ ATGGAAATCCCATCACCATCTT 3′) and NC534 (5′ CTAGTTGCCTCCCCAAAGCA 3′). Efficiency of the PAN 5′, PAN 3′ and GAPDH amplification was determined to be 80%, 63%, and 72%, respectively [76].
Quantitation of the ChIP signals was performed as follows. First, the relative quantities of Inputs and Pellets for each experiment were determined based on their amplification efficiency and the pellet/input ratio was determined. Each experiment included a “no antibody” control and this pellet/input value was subtracted from the other three samples. Subsequently, the background corrected pellet/input PAN 5′ and PAN 3′ ratios were normalized to the corresponding GAPDH ratios for the same experiment and the “no ORF57” values were set to one. The p-values reported throughout the manuscript are unpaired equal variance Student's t-test values.
UV cross-linking and cell-mixing experiments were performed as previously described [57], except poly(U) was omitted from the protocol. Northern blotting was performed as in [53]. For in situ hybridization analyses, PAN RNA was detected using a mix of three DNA oligonucleotides, NC29 (ATCGGCGGCACCAATGAAAACCAGAAGCGGCAAGAAGGCA), NC30 (CCAATGTTCTTACACGACTTTGAAACTTCTGACAAATGCC), and NC31 (GCACGTTAAATTGTCAAAAGTATAACATGTTTTTCCAATA). In situ hybridization was performed as previously described [53],[77], except in the experiments shown in Figure S3, direct labeling using FAM-conjugated oligonucleotides was employed as an alternative to digoxygenin-tailing protocols. Confocal and fluorescence microscopy was used in Figures 4 and S3, respectively.
|
10.1371/journal.pgen.1004283 | A Synthetic Community Approach Reveals Plant Genotypes Affecting the Phyllosphere Microbiota | The identity of plant host genetic factors controlling the composition of the plant microbiota and the extent to which plant genes affect associated microbial populations is currently unknown. Here, we use a candidate gene approach to investigate host effects on the phyllosphere community composition and abundance. To reduce the environmental factors that might mask genetic factors, the model plant Arabidopsis thaliana was used in a gnotobiotic system and inoculated with a reduced complexity synthetic bacterial community composed of seven strains representing the most abundant phyla in the phyllosphere. From a panel of 55 plant mutants with alterations in the surface structure, cell wall, defense signaling, secondary metabolism, and pathogen recognition, a small number of single host mutations displayed an altered microbiota composition and/or abundance. Host alleles that resulted in the strongest perturbation of the microbiota relative to the wild-type were lacs2 and pec1. These mutants affect cuticle formation and led to changes in community composition and an increased bacterial abundance relative to the wild-type plants, suggesting that different bacteria can benefit from a modified cuticle to different extents. Moreover, we identified ein2, which is involved in ethylene signaling, as a host factor modulating the community's composition. Finally, we found that different Arabidopsis accessions exhibited different communities, indicating that plant host genetic factors shape the associated microbiota, thus harboring significant potential for the identification of novel plant factors affecting the microbiota of the communities.
| The leaves of plants are inhabited by a diverse community of microorganisms. These leaf inhabitants influence their hosts with respect to growth and resistance to abiotic and biotic stresses. Recent studies revealed that the bacterial communities associated with leaves undergo selection, resulting in conserved microbial communities. However, the factors that are involved in structuring of bacterial communities are not well understood. In order to uncover host genetic factors that determine the community composition and to exclude confounding environmental effects, we inoculated Arabidopsis thaliana with a synthetic bacterial community under controlled conditions We screened a panel of Arabidopsis mutants defective in various traits for alterations in community structure and abundance and were able to show that cuticle synthesis and ethylene perception affect the bacterial community. In addition, we identified plant ecotypes with drastic differences in the community composition. Our system can thus be used to identify additional host genes and to broaden insights into plant microbe interactions, potentially providing a basis for applied plant protection through the identification of traits that enhance growth of plant probiotic bacteria.
| The aerial parts of the plants, which are dominated by leaves, represent one of the largest terrestrial habitats for microorganisms [1]–[3]. This habitat, called the phyllosphere, is occupied by a diverse community of bacteria and fungi, which is important for plant health and growth [1]–[3]. Microorganisms in the phyllosphere can promote plant growth through the production of hormones. They can also be involved in plant protection, which is due to direct interactions of microorganisms through the production of antibiotic compounds and competition for resources [3]. Additionally, microorganisms may protect plants against pathogens by inducing systemic resistance [4], [5]. Commensals belonging to the genus Sphingomonas and their closely related species might represent part of the core phyllosphere community that protect plants against pathogens [6]. In addition, certain Pseudomonas strains have been shown to be plant protective agents [7].
Given the functional importance of the phyllosphere community on plant traits, it is relevant to understand the processes that are responsible for determining the composition of this community. This pertains to the fundamental question in community ecology of what principles underlie the assembly of strains into communities. A large body of theoretical and empirical work addresses this question (for a recent review see [8]), and has implied a vast number of different processes that play a role in community assembly [9] A recent synthesis groups this diversity of processes in just four classes – selection, drift, speciation and dispersal [10]. While this synthesis has not been specifically developed for microbial communities, it is well suited as a conceptual framework to describe and analyze the assembly of microbial communities [11]; furthermore, the suitability of the phyllosphere to test ecological concepts has been pointed out [12].
Here, we focus on selective factors that shape the assembly of the phyllosphere community, that is, on factors that have a consistent and reproducible effect on the composition of the microbial community on plant leaves. Previous studies have established that the bacterial phyllosphere communities are dominated by few phyla: Proteobacteria, Actinobacteria and Bacteroidetes [1], [13]. It is assumed that different factors contribute to the shaping of bacterial communities in the phyllosphere, including environmental cues, microbial interactions, the plant genotypes and phenotypes [1], and environmental factors such as temperature, water availability [14], [15], and geographic location (for example, [16]). The effects of plant factors on community composition have been demonstrated for leaf age [17], plant species [18], and cultivars [19]–[21]. Moreover, total population size is also affected by the plant species [22]. Several quantitative trait loci (QTL) have been identified as associated with bacterial diversity in corn [23] or with disease suppression in tomato [24]. However, no direct effects of specific genes on the composition of the phyllosphere community could be established in these studies. Using the model plant Arabidopsis thaliana, one study identified jasmonic acid synthesis as a factor driving epiphytic diversity in the phyllosphere [25], whereas another study did not reveal any effect for trichomes on bacterial diversity [26].
There are a number of plant factors that could potentially have selective effects on phyllosphere microbial communities. A first potentially important factor is the hosts' innate immune system. Plants recognize bacteria at two levels of their immune system: the first level is pattern-triggered immunity (PTI), whereby plant receptors recognize microbial-associated molecular patterns (MAMPS), for example, flagellin [27]; and the second level is effector-triggered immunity (ETI), where intracellular plant receptors recognize microbial effectors, which are virulence factors transferred by pathogens into the host cytoplasm to dampen PTI [28]. It is not known how plants discriminate between pathogenic and commensal or beneficial microorganisms and whether plant receptors recognize these non-pathogenic phyllosphere bacteria and trigger plant immune signaling networks downstream of PTI or ETI activation, with potential effects on community structure. The habitat is scarce in nutrients [1] so other potential traits that may influence the presence of plant-associated microorganisms include, for example, mutants in sugar transporters [29] or amino acid transporter [30]. Similarly, it is not known whether mutants defective in secondary metabolites used for defense, such as camalexin, glucosinolates [31], and flavonoids [32], affect their associated microbial populations. More specifically, mutants in pectin synthesis are hypothesized to affect the abundance of methylotrophic bacteria because methanol, as a by-product of pectin synthesis, is an important factor for bacterial growth under competitive conditions [1], [33].
In general, identifying host genetic factors using field experiments is challenging because of the confounding influence of the external environment as well as the diversity of natural microbial communities. To reduce environmental complexity, gnotobiotic model systems with well-defined communities represent an alternative approach. The advantages of such controlled systems are that they allow for reproducible experimentation and the use of molecular fingerprinting methods to characterize the defined community. In mice, bacterial synthetic communities have been successfully used to study how diet impacts the microbiota [34], [35]. To our knowledge, synthetic bacterial communities have not yet been used to identify plant host genotypes that shape the associated microbiota.
Using a synthetic community approach, we aimed to identify plant genetic factors that influence community composition and/or the bacterial abundance of the leaf-associated community of A. thaliana. A set of 55 plant mutants was screened for such phenotypic effects, resulting in the identification of three mutants with significant community alterations. In addition, of the nine natural accessions tested, four were found to modify community composition and abundance, indicating that natural variation can be used for future experiments with the synthetic community to identify novel host genes affecting phyllosphere microbiota.
Knowledge of the overall composition of the microbiota of the A. thaliana phyllosphere [1], [13], [36] provides invaluable information for formulating a core microbiota based on cultivated model strains. A laboratory strain collection was used to establish a bacterial synthetic community, which allowed for the reproducible colonization of the phyllosphere in a gnotobiotic system. First, 20 strains were tested as individual inoculates. To be included in the synthetic community, strains were chosen that met two criteria: i) they were able to colonize the phyllosphere (higher than 107 CFU/g leaf fresh weight upon single inoculation, see Table 1), and ii) they did not induce disease symptoms nor cause a reduction in growth. In addition, the strains needed to represent the most abundant phylogenetic groups detected in the phyllosphere. Because Alphaproteobacteria is the most abundant sub-phylum in the phyllosphere of A. thaliana [1], four species were selected to represent this phylogenetic group: Sphingomonas phyllosphaerae and Sphingomonas sp. Fr1, which both have a plant-protective effect on A. thaliana [6], and Methylobacterium radiotolerans and M. extorquens PA1, which are efficient colonizers of the phyllosphere [37] but do not show a plant-protective effect [6]. In addition, two representatives of the Actinobacteria and one Betaproteobacteria were chosen for these abundant phyla of the phyllosphere (Table 1). Although five different strains of Gammaproteobacteria were tested as single isolates (Table S1), none could be included in the community because those strains either reduced plant growth or induced a strong disease phenotype under the experimental conditions. Mixing these strains with the rest of the community did not mask the disease phenotype.
To monitor changes in community composition in a high-throughput manner, automated ribosomal intergenic spacer analysis (ARISA) was used. Briefly, the 16S–23S rRNA intergenic spacer region was amplified by PCR using fluorescence-tagged universal primers. The PCR products were separated using a capillary sequencer. Each species in the community could be distinguished from the others based on its unique ARISA profile. Because some species were characterized by multiple peaks due to multiple 16S rRNA gene copies and variable length of the intergenic spacer regions, one representative peak was chosen for each species (Table 1). In addition, the peak area was normalized by the 16S rRNA gene copy number so that the abundance of each species was roughly proportional to the peak area in a semi-quantitative approach [38], [39] (for a validation experiment in which the DNA of one species was diluted against a mixture of DNA background see Figure S1).
A time course experiment was performed where the seven-member synthetic community (Table 1) was assessed immediately after spray inoculation of wild-type Col-0 plants and once a week for four weeks thereafter. Community composition was compared using the Bray-Curtis dissimilarity index (the more different two communities are, the closer to 1 their index is). Figure S2A shows that community comparisons of the inoculum to leaves sampled immediately after spraying were indistinguishable from community comparisons of leaf samples with each other (P>0.05). After one week, community comparisons of plant samples to inoculation solution was significantly greater relative to community comparisons made within plant samples (P = 0.0048). Paralleled determination of the population sizes by leaf washings revealed that the number of bacteria per plant was 1.18*103 immediately after spraying and increased steadily through time (Figure S2B). Based on the time-course experiment, we decided for the remainder of the study to harvest the leaves two weeks post-inoculation to allow growth of and competition between bacteria.
Using ARISA to analyze the community associated with wild-type Col-0 samples from ten independent biological experiments, we found that the synthetic community colonizes the phyllosphere in a reproducible manner (Figure S3). The average relative fluorescence intensity after colonization ranged from 3% for S. phyllosphaerae to 40% for Rhodococcus (Table 1).
A real-time qPCR method was developed to estimate the bacterial abundance in the phyllosphere. First, we confirmed that the PCR primers amplify the 16S rRNA gene in a linear fashion in the absence (Figure S4A) and in the presence of plant DNA (Figure S4B). The relative abundance of the 16S rRNA gene was calculated by normalizing with a plant gene and is proportional to the amount of bacterial DNA (Figure S4C).
To identify plant host genetic factors that influence community composition, a priori candidate genes were selected from six different classes: cuticle and trichome, cell wall and pectin synthesis, secondary metabolism, sugar and amino acid transporters, defense signaling and pattern-triggered immunity (see Table S2 for a complete list of mutants). In addition, a small panel of natural accessions was screened. For each plant genotype, community composition and the 16S rRNA gene copy number were assessed (Figure 1).
A total of 55 A. thaliana mutants and accessions were tested in 10 independent experiments, each including Col-0 as a control and Landsberg erecta (Ler) as well as Wassilewskija (WS) when needed, dependent on the genetic background of each mutant. Results are shown in Figure S5 for individual screens and Figure S6 for all plant genotypes as a clustering analysis based on the Bray-Curtis dissimilarity index. We observed that the communities from some of the selected mutants clustered together and separate from Col-0. To compare independent experiments, community comparisons of each genotype to wild-type samples were calculated with the Bray-Curtis Dissimilarity index (Figure S7). Col-0 was used as the wild-type, except for mutants with a different background. In order to exclude genotypes where samples showed high variability from each other, community comparisons were also made within plant samples. Figure 2A shows the ratio of comparisons between each genotype and the corresponding wild-type over the comparison within each genotype. For each of these experiments, the 16S rRNA gene copy number was determined by qPCR (Figure 2B).
From the initial screen, the ten plant genotypes showing the highest dissimilarity to the wild-type (Figure 2A) and the six ecotypes with higher or lower bacterial abundances (Figure 2B) were selected for validation experiments using ARISA and qPCR. In addition, colony forming units were determined to verify altered community abundances and compositions using a cultivation-dependent method. From these sets of genotypes, 3 mutants (lacs2, pec1, and ein2) and 4 accessions (Mr-0, Ler, RRS-7, and Ct-1) with an altered community composition and/or overall abundance could be verified (see below). In contrast, results with the ‘sweet’ mutants could not be confirmed in validation experiments, and no genotype effect for community composition nor for bacterial abundance was observed (Figure S8).
A different community composition of lacs2 and pec1 samples was confirmed with independent replicate experiments using two independent alleles each (Figure 3A). Multivariate analysis of variance confirmed a significant effect of these genotypes across replicate experiments (Table 2 and Table S3). LACS2 encodes for long-chain acyl-coenzyme A synthetase 2, an enzyme involved in cutin biosynthesis. The mutant plants are characterized by the absence of the cuticular membrane and a reduction of cuticular polyesters, which normally compose the wild-type cuticle [40]. The pec1 mutant has an intermediate phenotype between Col-0 and lacs2 and carries a mutation in an ATP-binding cassette transporter involved in the export of cuticle precursors [41]. Statistical tests indicated that there was a genotype effect for the relative fluorescence intensity (RFI) of Rhodococcus, Sphingomonas sp. Fr1, S. phyllosphaerae and Variovorax. Compared to the wild-type, both mutants harbored more Variovorax and less Rhodococcus, whereas pec1-3 had less Sphingomonas sp. Fr1. Using the qPCR method to estimate relative 16S rRNA gene copy numbers, we found that lacs2 harbored a higher bacterial abundance compared to the wild-type (Figure 2B). Multivariate analysis of variance indicated that there is a genotype effect for 16S rRNA gene copy numbers of both lacs2 and pec1 (Table 2 and Table S4) which was further confirmed by the fact that both lacs2 alleles carried a higher bacterial abundance compared to wild-type plants (Figure 4).
The results of the ARISA and qPCR analysis were partially confirmed using bacterial enumeration (Figure S9). For example, Variovorax cells were more abundant on the lacs2 and pec1 mutants, in-line the ARISA results. Contrary to the ARISA results, Sphingomonas were more abundant on both the pec1 and lacs2 mutants, suggesting that the relative abundance (ARISA results) does not necessarily reflect the absolute abundance (bacterial enumeration). This discrepancy might be due to differences in the protocols. ARISA profiles were determined using DNA extracted from the whole plants (both epiphytic and endophytic communities), whereas the protocol for leaf enumeration possibly extracts more epiphytic than endophytic bacteria.
Independent replicate experiments confirmed a shift in the synthetic bacterial community colonizing the ein2 mutant plants compared to the wild-type plants (Figure 3B). The adonis test validated a significant effect of genotype across replicate experiments (Table 2 and Table S3). There was a plant genotype effect for the RFI of Variovorax, which was higher on the ein2 mutant. The RFI of other bacterial species were not affected by plant genotype. The ein2 mutant is ethylene insensitive [42] and carries a mutation in EIN2 [43], which plays a central role in the ethylene response, an important hormone for response to the environment and plant defense [44]. On the contrary, when we tested other mutants in the plant defense signaling pathways, we found that the jasmonate mutant aos and the salicylic acid mutant sid2 harbored a similar community composition compared to the wild-type plants (Figure 3B). In the initial screen, bacterial abundance, as measured by relative 16S rRNA gene copy number, was found to be higher on the ein2 mutant (Figure 2B). ANOVA confirmed a significant effect for this genotype (Table 2 and Table S4); however, this effect was weak and not significant for a single experiment (Figure S10). The ARISA results were confirmed using bacterial enumeration (Figure S9). Variovorax cells were more abundant on the ein2 mutant, whereas the abundance of other bacterial species was not affected by this mutation.
The four Arabidopsis accessions identified in the first round of screening with a different community composition, Mr-0, Ler, Ct-1, and RRS-7, were confirmed in independent replicate experiments (Figure 5A). Multivariate analysis of variance confirmed a significant effect of each genotype on community composition (Table 2 and Table S3). Statistical tests revealed that there was a genotype effect for the RFI of Arthrobacter (lower in Mr-0), M. extorquens PA1 (higher in RRS-7), M. radiotolerans (higher in Ler and RRs-7), and Sphingomonas sp. Fr1 (lower in Ct-1, Ler and RRS-7. In addition, bacterial abundance was also different on the natural accessions as indicated by quantitative qPCR of the 16S rRNA gene (Table 2 and Table S4). The 16S rRNA gene copy numbers were higher on Mr-0 and lower on Ler, Ct-1, and RRS-7 (Figure 5B).
In this study, we established a synthetic community approach to examine the effect of host genotype on the bacterial community composition and total abundance on plant leaves. We demonstrate that a model microbiota developed in a reproducible manner in the phyllosphere, allowing for the monitoring of perturbations dependent on the host's genotype. This system has several advantages, including its relatively short time scale and small space requirements, facilitating the independent biological repetition of experiments. Although in the future, more complex communities might be tested to link plant genotype to bacterial abundance, the low complexity community applied in this study has already allowed for the identification of three A. thaliana mutants with effects on community composition and/or total abundance. The plant mutants with the strongest effect on the associated bacteria were lacs2 and pec1, which are characterized by a more permeable cuticle compared to wild-type plants [40], [41], [45]. The cuticle is present on the outside surface of epidermal cells and is composed of the aliphatic polyester cutin and waxes [46]. In addition to its function as a diffusion barrier that diminishes water loss and as a protection against abiotic stresses, such as UV radiation [47], the cuticle also serves as a key interface for plant-microbe interactions. First, the cuticle represents the initial interaction surface with microorganisms colonizing the phyllosphere, therefore, features of the cuticle affect adhesion and, thus, microbial immigration [48]. Second, the cuticle controls transpiration, thus reducing water availability, which is a limiting factor for the growth of phyllosphere bacteria [15]. Third, the cuticle is involved in the transport of polar solutes and lipophilic organic compounds, thus reducing nutrient availability [47]. Furthermore, the question of whether components of the cuticle themselves can be used as substrate by bacteria is open. Evidence for the importance of the cuticle for the phyllosphere community comes from both sides of the study of plant-bacteria interactions. In terms of bacteria, epiphytic bacteria have been shown to alter the leaf surface permeability of isolated intact cuticles [49]. In terms of the plants, epicuticular wax synthesis has been shown to impact the colonization of single bacteria inoculates using maize mutant plants [50], and to impact the phyllosphere community composition, as shown very recently in Arabidopsis [51]. Moreover, epidermal thickness was correlated with the bacterial population size colonizing eight Mediterranean plant species [52].
In this study, we present evidence that cuticle permeability affects not only bacterial abundance but also community composition. We observed that lacs2 and pec1 mutants harbored more Variovorax. The average RFI of Variovorax in Col-0 was 16%; however, for the cuticle mutants, the average RFI ranged from 27% (lacs2-4) up to 54% (pec1-3). One explanation is that an enhanced permeability of the cuticle leads to an increase in nutrient availability, which favors the growth of the Betaproteobacterium Variovorax, the name of which notably refers to its ability to consume many different substrates. In contrast, the RFI of Rhodococcus decreased for both cuticle mutants, ranging from 23% (pec1-3) to 30% (lacs2-4) compared to 40% in Col-0. One possible explanation is that Rhodococcus feeds on cuticle components, which are less abundant in the mutants. In contrast, the abundance of both Methylobacterium strains was not different on the cuticle mutants, which might indicate that methanol availability is not affected by either mutation.
We found that the lacs2 mutant carried a higher bacterial abundance compared to both the wild-type and pec1 mutant (Figure 4). The pec1 mutant was shown to have an intermediate phenotype to lacs2 in terms of cuticular permeability, as measured using toluidine blue staining, sensitivity to herbicides and water loss [41]. Moreover, analysis of the cuticle ultrastructure revealed that pec1 retains a thick layer of electron-dense material, representing insoluble lipid-derived polymers, that is missing in lacs2 [40]. Analyses of the leaf polyester monomers demonstrated that the monomer composition of the lacs2 mutant was reduced by 20–25% compared to wild-type amounts [40], whereas only minor changes were observed for pec1 [41]. Interestingly, both lacs2 and pec1 mutants are more resistant to the fungal pathogen Botrytis cinerea, with the increased resistance proposed to be due to the induction of antifungal compounds by elicitors diffusing through the cuticle [40]. In contrast, lacs2 is more susceptible to avirulent Pseudomonas syringae, with the increased susceptibility hypothesized to be due to enhanced tissue collapse upon infiltration of the pathogen [53]. However, tissue collapse likely does not play a role in our study because the synthetic community was not syringe-infiltrated but rather sprayed onto the leaves. Therefore, we hypothesize that the higher bacterial abundance phenotype measured on lacs2 was due to the increased leaching of nutrients from this mutant compared to the wild-type and pec1 mutant.
The ethylene-insensitive mutant ein2 harbored a different community composition. In particular, Variovorax was more abundant in the phyllosphere of the mutant compared to the wild-type plants. Ethylene is a plant hormone with multiple roles in development, such as seed germination, fruit ripening, and root hair formation [54], [55]. In addition, ethylene modulates plant resistance to pathogens, which is dependent on the type of attacker. Generally, ethylene is found to reduce the appearance of diseases caused by necrotrophic and hemibiotrophic pathogens and to increase disease symptoms caused by other types of pathogens [56]. Interestingly, several pathogenic bacteria and fungi interfere with the plant defense-signaling pathway by producing ethylene [57], [58], which, in this case, can be considered a virulence factor. In contrast, some plant growth-promoting bacteria colonizing roots can degrade the compound 1-aminocyclopropane-1-carboxylic acid (ACC), the precursor of ethylene, using the enzyme ACC deaminase, thereby increasing root length [59]. In this study, we found the phyllosphere of the ein2 mutant to show quantitative differences in the community composition. In the A. thaliana rhizosphere, the ein2 mutant was found to harbor a lower bacterial abundance and was not associated with any changes in bacterial community composition [60]. In contrast, in the tobacco rhizosphere, Long et al. found a lower bacterial diversity and a different community in ethylene-insensitive transgenic plants compared to wild-type plants [61]. The distinct functions of ethylene in roots and leaves might affect the associated communities differently, for example, ethylene is involved in the formation of root hairs, which are hypothesized to serve as an entry point for the bacterial colonization of roots. Wild-type tobacco plants have also been demonstrated to have more root hairs compared to ethylene-insensitive plants and have been found to be associated with a different bacterial community [61].
Natural accessions of A. thaliana are a source of genetic diversity that can be harnessed to identify novel genes underlying phenotypic variations, which can then be used for quantitative trail locus (QTL) analyses of recombinant inbred lines (RIL) [62]. Furthermore, the recent development of cheaper SNP arrays and sequencing technology has enabled genome wide associations (GWA) in A. thaliana [63]. Natural accessions thus provide a valuable resource to begin identifying the intricate relationship of plants and associated microorganisms. Recently, using 16S rRNA gene amplicon sequencing of root samples and analyzing eight Arabidopsis accessions, Lundberg et al. [64] identified a small subset of 12 operational taxonomic units (OTU) out of 778 that showed host genotype dependent quantitative differences. In another study only one OTU of the root endophyte community showed significantly different quantitative enrichment when analyzing two Arabidopsis accessions [65]. Field and sample types (rhizosphere versus bulk soil) were found to be more important than the plant genotype for bacterial root community composition [65], [66].
Notably, here using a synthetic community approach applied to the phyllosphere of Arabidopsis we found that 4 out of the 9 accessions tested harbor a different community composition compared to Col-0, indicating that natural accessions offer significant potential for the discovery of new genes affecting community composition and/or abundance. In addition, several of the tested accessions harbor different Methylobacterium and Arthrobacter abundances that were not affected by the lacs2, pec1 and ein2 mutants. Interestingly, we found accessions with both lower (Ct-1, Ler, and RRS-7) and higher (Mr-0) bacterial abundances compared to the Col-0 accession. Future experiments with the model synthetic community and methods developed in this study will represent a valuable approach to map and identify novel genes affecting community composition and bacterial abundance.
The isolates and type strains used for the synthetic community are listed in Table 1. Variovorax sp., Arthrobacter sp. and Rhodococcus sp. were isolated from wild plants growing at different sites located near Madrid, Spain and described in [16]. The 16S rRNA genes of these three strains were sequenced for verification. The 16S rRNA gene copy number was determined by Southern blot.
A. thaliana plants were cultivated on half-strength MS nutrient medium including vitamins and 0.55% plant agar (both from Duchefa, Haarlem, Netherlands) and supplemented with 1% sucrose. Seventy milliliters of medium was poured into microboxes outfitted with a XXL filter (Combiness, Nazareth, Belgium). To avoid leaves touching the medium, a sterile Lumox Film 25 (Sarstedt, Nümbrecht, Germany) with 6 holes (diameter, 4 mm) was placed on the agar surface. A. thaliana seeds were surface sterilized using a standard protocol [6] and stratified for 3 days (at 4°C) before being placed at the holes. Plants were grown under short-day conditions (a 9-h photoperiod) in a standard growth chamber, as previously described [6].
Both Methylobacterium strains were grown on mineral salt medium [67] supplemented with 0.5% succinate as a carbon source. All other strains were grown on nutrient broth (NB) without additional NaCl (Sigma-Aldrich, St. Louis, MO, USA). The two Sphingomonas strains were grown in liquid cultures, whereas the other strains were grown on solid media. Strains were grown at 28°C for three days (both Methylobacterium strains), two days (Rhodococcus sp., Variovorax sp., Arthrobacter sp. and S. phyllosphaerae), or one day (Sphingomonas sp. Fr1). Before inoculation, cells from the liquid cultures were washed once and resuspended in 10 mM MgCl2 solution. For cultures grown on solid medium, a loop of material was resuspended in 10 mM MgCl2 solution. The optical density at 600 nm (OD600) of each solution was adjusted to 0.2. The synthetic community was obtained by mixing the seven strains at 1∶1∶1∶1∶1∶1∶1 OD600. This solution was then diluted to an OD600 of 0.02. Plants were inoculated by spraying 200 µl of bacterial suspension with an airbrush paint gun [68].
For the time course experiment, samples were harvested immediately after spraying (four DNA pools, each with ten plants), and 1, 2, 3, and 4 weeks after inoculation (for each time point, four DNA pools, each with five plants). For the screening of plant genotypes, plants were harvested two weeks after inoculation (five plants from five different microboxes were pooled for one DNA extraction). In total, between 3 and 6 DNA pools per genotype were sampled depending on the experiment. Plants were taken out of the microboxes, and the roots and cotyledons were removed using flame-sterilized scalpels and forceps. In addition, for the validation experiments, the population size colonizing individual plants was determined using a dilution series (see below for the protocol).
DNA was extracted from the plant tissues using the NucleoSpin Plant II kit (Macherey-Nagel, Düren, Germany). Plant samples were lyophilized, and one metal bead was added to each sample in a 2 ml- centrifuge tube before chilling in liquid nitrogen. Samples were homogenized for 2 min at 25 Hz using a Retsch TissueLyser (Retsch, Haan, Germany). The SDS-based lysis buffer PL2 was added to the homogenized samples, and the standard protocol according to the manifacturer's instructions was used thereafter.
Primer 1492F (reverse complement of 1492R, [69]), and 23Sr [70] were used to amplify the 16S–23S rRNA intergenic space region. These primers were tested with DNA from plants grown axenically; no product was detected, indicating that they do not amplify mitochondrial or chloroplast DNA. Primer 1492F was labeled at the 5′ end with fluorescein. The reaction volume was 25 µl and contained 1-fold Phusion HF reaction buffer, 200 µM dNTP, 250 nM of each primer, 3% DMSO, 0.4 units of Phusion polymerase and approximately 20 ng DNA. The PCR program consisted of an initial denaturation step of 4 min at 94°C, followed by 35 cycles of denaturation at 94°C for 30 sec, annealing at 60°C for 30 sec, elongation at 72°C for 1 min followed by a final elongation step at 72°C for 7 min. PCR products were verified on agarose gels before preparing for ARISA. PCR products (2 µl, diluted 10- and 20-fold) were mixed with 8 µl of HiDi formamide (Applied Biosystems) and 0.2 µl MapMarker 1000-ROX (BioVentures, Murfreesboro, USA). After denaturation at 95°C for five minutes, the samples were analyzed using a 3130 ABI capillary sequencer. Genemapper version 3.7 (Applied Biosystems) was used for the data analysis. Sizing tables were exported for analysis with R.
The reaction volumes were 20 µl and contained 1-fold FastStart Universal SYBR green Master (Roche Applied Science), UltraPure DNase/RNase-free water (Life Technologies), 600 nM primer mix (16S rRNA primers) or 300 nM (plant gene primers), and approximately 5 ng DNA. The 16S rRNA gene was amplified using primers 799F [71] and 904R [72], see Table S5 for the primer sequences. Although these primers were designed to exclude organelle DNA, a product is amplified with DNA from plants grown axenically (deltaCt = 6–9 between inoculated Col-0 plants and axenically grown plants). This PCR product was cloned and sequenced and found to include sequences from organelles (chloroplast and mitochondria) confirming that the plants were axenic. Because we compared the samples with each other, we assumed that the small amount of plant DNA amplified by those primers did not affect our estimations. The primers ExpF1 and ExpF2 (Table S5) amplify the plant gene AT4G33380, a reference gene used for transcript normalization [73], which was used to normalize the 16S rRNA gene copy number. Controls, including no template, were included for each run. PCR assays were run in duplicate on a Rotorgene 3000 (Corbett Life Science, Qiagen). The PCR program consisted of a touchdown program with an initial denaturation step of 10 min at 95°C, followed by 35 cycles of 15 sec of denaturation at 95°C, 25 sec for the annealing step with the temperature decreasing from 65°C to 55°C (2 degrees per cycle), and 45 sec of elongation at 72°C followed by a melting curve analysis.: The raw data were exported directly from Corbett Research Software version 1.7 and imported into LinRegPCR version 12.8 [74] to determine cycle number to threshold (Ct) and efficiency (E). The 16S rRNA gene copy number was normalized to the plant gene AT4G33380 and calculated as follows: 16S rRNA/plant gene = Eplant gene Ct plant gene/E16SCt 16S, where Ct is the mean of the 2 duplicate reactions and E is the mean for all reactions with a particular primer pair for each run. To test for linearity of the qPCR method, a two-fold dilution series of Variovorax DNA was prepared (starting concentration 1 ng/µl DNA). qPCR was run following the standard protocol with 5 µl of bacterial DNA in the absence (Figure S4A) or presence of plant DNA (5 ng per reaction) (Figure S4B).
Cell numbers were determined on randomly selected plants from several microboxes using a previously described protocol [6]. Briefly, leaves were washed in 100 mM phosphate buffer (pH 7) containing 0.2% Silwett by shaking for 15 minutes on a Retsch TissueLyser and sonicating for 5 minutes in a water bath. This protocol has been demonstrated to release both the epiphytic and endophytic Pseudomonas syringae associated with leaves [75]. Ten-fold dilution series were plated on different media. Naturally rifampicin-resistant Variovorax cell numbers were determined on King's B plates containing rifampicin (50 µg/ml). Sphingomonas cell numbers were determined on NB plates containing streptomycin (20 µg/ml). Methylobacterium extorquens PA1 could be distinguished from other members of the community because of its pink colony color on minimal media supplemented with 0.5% methanol. Total bacterial cell counts were determined by counting cell numbers on minimal media supplemented with 0.5% succinate.
The R statistical environment was used for all the statistical analyses and plotting (R Development Core Team; http://www.R-project.org). Relative fluorescent intensity (RFI) was calculated by dividing individual peak area by the total peak area for each sample using the R binning script written by Ramette [38]. Parameters for the script were: a range from 500 bp to 1000 bp, a minimum RFI cutoff of 0.2%, a window size of 5 bp and shift of 1 bp. One peak was chosen to represent each species of the community (Table 1). Furthermore, RFI was normalized by the 16S rRNA gene copy number to take into account variations in copy numbers among strains. Abundance tables were analyzed using the package vegan [76]. The function vegdist with default parameters (binary = FALSE) was used to calculate the Bray-Curtis index. This function calculates the Bray-Curtis index based on proportions of different types in a sample, in contrast to the binary version of the vegdist function, which only takes into account the presence and absence of different types. Hclust was used for hierarchical clustering with the average method. The Wilcox rank sum test was used to contrast Bray-Curtis indices of comparisons between plants samples and the inoculum and within plant samples (Figure S7). Multivariate analysis of variance was conducted with the vegan functions adonis [77] to assess the effect of genotype and experiment on community composition (Table 2 and Table S3). To test for the effect of genotype on the RFI of each bacterial population, a generalized linear model was used with a quasibinomial distribution to correct for overdispersion. The Dunnet's test was used to test significance in the comparison to the appropriate wild-type. Multivariate analysis of variance (ANOVA) was used to test the effect of genotype and experiment on 16S rRNA gene copy numbers (Table 2 and Table S4). After normalizing to the wild-type, the numbers were log-transformed. A qqplot indicated that the standardized residuals were normally distributed; furthermore, this was confirmed by the Shapiro-Wilk test. The Student's t-Test was used to test whether each mutant differed significantly from the wild-type. Similarly, ANOVA was used to test the effect of genotype on CFU/g FW, for which the data were log10 transformed (Figure S9). The P values were adjusted for multiple testing using the Bonferroni correction.
The full-length 16S rRNA gene sequences of the Variovorax sp., Arthrobacter sp. and Rhodococcus sp. strains used in this study have been deposited in the European Nucleotide Archive under accession numbers HG737356, HG737357, HG737358, respectively.
|
10.1371/journal.pntd.0007325 | Pharmacokinetics, safety, and efficacy of a single co-administered dose of diethylcarbamazine, albendazole and ivermectin in adults with and without Wuchereria bancrofti infection in Côte d’Ivoire | A single co-administered dose of ivermectin (IVM) plus diethylcarbamazine (DEC) plus albendazole (ALB), or triple-drug therapy, was recently found to be more effective for clearing microfilariae (Mf) than standard DEC plus ALB currently used for mass drug administration programs for lymphatic filariasis (LF) outside of sub-Saharan Africa. Triple-drug therapy has not been previously tested in LF-uninfected individuals from Africa. This study evaluated the pharmacokinetics (PK), safety, and efficacy of triple-drug therapy in people with and without Wuchereria bancrofti infection in West Africa.
In this open-label cohort study, treatment-naïve microfilaremic (>50 mf/mL, n = 32) and uninfected (circulating filarial antigen negative, n = 24) adults residing in Agboville district, Côte d’Ivoire, were treated with a single dose of IVM plus DEC plus ALB, and evaluated for adverse events (AEs) until 7 days post treatment. Drug levels were assessed by liquid chromatography and mass spectrometry. Persons responsible for assessing AEs were blinded to participants’ infection status.
There was no difference in AUC0-inf or Cmax between LF-infected and uninfected participants (P>0.05 for all comparisons). All subjects experienced mild AEs; 28% and 25% of infected and uninfected participants experienced grade 2 AEs, respectively. There were no severe or serious adverse events. Only fever (16 of 32 versus 4 of 24, P<0.001) and scrotal pain/swelling in males (6 of 20 versus 0 of 12, P = 0.025) were more frequent in infected than uninfected participants. All LF positive participants were amicrofilaremic at 7 days post-treatment and 27 of 31 (87%) remained amicrofilaremic 12 months after treatment.
Moderate to heavy W. bancrofti infection did not affect PK parameters for IVM, DEC or ALB following a single co-administered dose of these drugs compared to uninfected individuals. The drugs were well tolerated. This study confirmed the efficacy of the triple-drug therapy for clearing W. bancrofti Mf and has added important information to support the use of this regimen in LF elimination programs in areas of Africa without co-endemic onchocerciasis or loiasis.
ClinicalTrials.gov NCT02845713.
| Lymphatic filariasis is a mosquito-borne infection that causes disability in the form of lymphedema, hydroceles, and elephantiasis. It has been targeted for global elimination based on mass drug administration in the total population at risk including many people uninfected with LF. Recently, a single co-administered dose of IVM + DEC + ALB has been shown to be much more effective than the standard treatment with DEC + ALB for sustained clearance of Mf for 3 years based on studies in Papua New Guinea. This study confirms the efficacy and safety of triple-drug therapy for clearing of Wuchereria bancrofti Mf in an African population. The presence of LF did not affect drug levels and the medicines were well tolerated, with 28% and 25% rate of moderate AEs in infected and uninfected individuals respectively, and no severe or serious AEs, supporting the use of triple-drug therapy for mass drug administration. This study shows for the first time that triple-drug therapy also has a potent macrofilaricidal effect, as determined by the reduction in circulating filarial antigen and inactivation of worm nests one year following treatment.
| Lymphatic filariasis (LF) is a mosquito-borne parasitic disease caused by nematode parasites. Host responses to the adult worms in lymphatic vessels cause stigmatizing morbidity (lymphedema, hydrocele, and elephantiasis) that can lead to chronic disability. Parasites that cause LF (Wuchereria bancrofti, and less commonly Brugia malayi and Brugia timori) are currently estimated to infect 80 million people in 52 tropical countries, with about 850 million at risk [1]. The World Health Organization (WHO) has targeted LF for global elimination by 2020 [2, 3]. The elimination effort is based on a mass drug administration (MDA) approach that uses one of three anti-filarial drug regimens; i) ivermectin (IVM) plus albendazole (ALB) in regions of Africa where Onchocerca volvulus is co-endemic, ii) ALB alone in areas where LF is co-endemic with Loa loa, and iii) diethylcarbamazine (DEC) plus ALB in LF endemic areas outside of Africa and in regions of Africa where L. loa and O. volvulus are not present [4]. Often, five or more annual rounds of MDA are required to reduce the community microfilarial reservoir to a level that cannot support sustained transmission of new infections by mosquitoes. Recently, a one time co-administered dose of IVM plus DEC plus ALB, or triple-drug therapy, was shown to achieve sustained microfilariae (Mf) clearance for three years in 96% of individuals with moderate to heavy LF infections in Papua New Guinea (PNG), compared to a lower cumulative clearance of Mf with standard therapy of DEC plus ALB administered once a year over the same period of time [5, 6]. Although the frequency of mild adverse events (AEs) was higher in the triple-drug regimen compared to the standard treatment of DEC plus ALB (27% versus 5%) [5], there were no serious AEs. Finally, it is unknown how LF infection itself might affect the pharmacokinetics (PK) and pharmacodynamics of these drugs. For example, inflammatory responses to chronic infections and the killing of Mf could affect the P450-mediated metabolism and PK of ALB in its first-pass conversion to ALB sulfoxide, the active metabolite [7]. This study evaluated the PK, safety, and efficacy of triple-drug therapy in men and women with and without Wuchereria bancrofti infection.
This was an open-label cohort study of treatment naïve Wuchereria bancrofti-infected (n = 32) and uninfected (n = 24) adults residing in the Agboville district of Côte d’Ivoire, which is endemic for LF and non-endemic for onchocerciasis. The primary outcomes were drug levels and safety. The secondary outcomes were reduction in circulating Mf and parasite antigen levels, and inactivation of adult worm nests. Individuals who assessed adverse events and measured drug levels were blinded to LF infection status. The study protocol and related documents were approved by institutional review boards in Cleveland, USA (University Hospitals Cleveland Medical Center IRB #03-16-09) and in Côte d’Ivoire (Comité National d’Ethique et de la Recherche, CNER, N/Ref:022/MSLS/CNER-kp). This trial is registered at Clinicaltrials.gov (NCT02845713).
DEC (Banocide GlaxoSmithKline) was purchased for the study. ALB (GlaxoSmithKline) and IVM (Merck & Co. Inc.) were obtained from Ministry of Health stocks in Côte d’Ivoire used for the LF MDA program. A fixed dose of 400 mg ALB was used for all participants. IVM and DEC doses were 200 μg/kg and 6 mg/kg, respectively.
Individuals were prescreened in their home villages for W. bancrofti infection with a rapid diagnostic test that detects circulating filarial antigenemia (the Alere Filariasis Test Strip, FTS, Alere, Inc, Waltham, MA, USA) [8]. Persons with positive FTS results had blood collected between 21:30 and 23:00 for Mf testing by membrane filtration with 1 mL of anticoagulated venous blood (5μM, Nuclepore Corp., Pleasanton, CA, USA). Two microscopists independently read Giemsa-stained filters to assess Mf load.
The study took place at the Centre de Recherche de Filariose Lymphatique d’Agboville, located at General hospital of Agboville, Côte d’Ivoire. Eligible participants included adults 18–70 years, with no acute illness, and no treatment with ALB or IVM within the past two years. Infected participants required >50 Mf/mL. Participants were considered uninfected if FTS strip was negative in whole blood and confirmed with plasma. Exclusion criteria included a positive pregnancy test; a history of chronic kidney or liver disease; a serum alanine transaminase, aspartate transaminase, or creatinine level >1.5 times the upper limits of normal; or blood hemoglobin <7 gm/dL. Biochemical tests were performed with a Piccolo biochemistry machine (Abbott Labs, Lake Bluff, IL, USA) and hemoglobin with a Hemocue Hb 201+ machine (HemoCue America, Brea, CA, USA). Individuals were also excluded if they had taken medications that could interfere with test drug metabolism within one week of study onset or if they had evidence of urinary tract infection on a spun urine sample (>10 neutrophils per high powered field by microscopy, 400x) or 3+ nitrate on dipstick (Diastix, Bayer Inc.). Because DEC can cause serious AEs in people infected with onchocerciasis [9], all individuals were tested by microscopic examination of skin snips taken from both iliac crests and by the presence of antibodies to a recombinant O. volvulus antigen (Ov16 Rapid Test, Standard Diagnostics Inc. Youngin, South Korea) [10]. Persons with microfiladermia or a positive Ov16 antibody test (n = 3) were excluded. All participants signed a written consent prior to screening and enrollment into the study.
Enrollment began on April 17, 2015 and continued through June 4, 2015. Groups of five participants of the same sex were brought to the research center the night prior to treatment for screening with baseline laboratory tests. For men,ultrasound examinations were performed. Investigators evaluating AEs and performing ultrasound examinations and laboratory tests were blinded to participants’ infection status. Participants remained at the research center until 72 hours post-treatment, then returned to their village for passive follow up until repeat examinations on day 7.
The first treatment dose of triple-drug therapy was given on April 18, 2015. Starting at 7 a.m., and within about 30 minutes after eating a typical Ivorian breakfast of wheat bread and eggs, all participants were treated with a single co-administered dose of triple-drug therapy. Venous blood samples were collected at 1, 2, 3, 4, 6, 8, 12, 24, 36, 48, 72 hours and 7 days after treatment, with aliquots of plasma stored at -20°C for later testing of drug levels. A peripheral intravenous catheter was placed for the first 12 hours due to frequent blood draws. Additional venous blood was collected between 21:30 and 23:00 at 39 hours, 7 days and 1 year for Mf testing with two 1 mL samples by membrane filtration. Biochemistry and urine tests were repeated at 24 and 48 hours and 7 days post-treatment.
After treatment, participants were monitored for AEs every 6 hours for the first 48 hours, then every 12 hours until 72 hours, and again at day 7 post-treatment. Passive surveillance for potential AEs was conducted by trained community health workers located in the participants’ home villages on days 4–6. New or worsening symptoms, changes in vital signs, or new abnormal findings on physical examination were considered to be treatment emergent adverse events (TEAEs) and were scored using a modified version of the National Cancer Institute Common Terminology Criteria for Adverse Events, v4.0. Blood pressure and heart rate were taken with an Omron-7 automatic blood pressure cuff with the participant in a seated position. Auricular temperatures were obtained using a digital thermometer.
Scrotal ultrasound examinations were performed on men in the supine position using a SonoScape S8 portable ultrasound system equipped with a 5–7.5Hz liner array transducer (International Diagnostic Devices, Inc, Las Vegas, NE, USA). Color and pulsed wave Doppler modes were used to differentiate lymphatic vessels from blood vessels. Adult worm nests were identified based on the characteristic bidirectional movement of the worms (the “filarial dance sign”) [11]. Abnormal ultrasound findings were digitally recorded. The presence or absence of worm nests was recorded, along with the number and size of worm nests, lymphatic dilatation, and hydroceles. Ultrasounds were repeated at 24, 48, 72 hours and 7 days and 1 year after treatment.
Microfilarial counts were expressed as Mf/mL and log transformed after adding 1, and geometric mean values (GM) were used as measures of central tendency to normalize the results. Baseline sample characteristics between individuals infected or uninfected with LF and the impact of treatment of Mf and worm nests were compared using the chi-squared test or the Mann-Whitney test using GraphPad Prism, version 6. For the analysis of the impact of treatment on the inactivation of worm nests, only men who had detectable worm nests at baseline were included. For the sample size calculation we used methods as previously described [14]. We wanted to observe less than 50% difference in clearance of drugs between LF-infected and uninfected individuals. Considering a variance of ALB-OX (the most variable of the drugs [6]) of approximately 50% from the mean, with an alpha = 0.05 and power = 0.8, the number of participants required would be 20–24 per study population. The PK parameters were determined by noncompartmental analysis using Phoenix WinNonlin version 6.3 (Pharsight Corporation, CA, USA). For analytes that did not have at least eight calculated values, the mean and standard deviation were not calculated. Where R2adj was <0.85, this parameter was reported as NE (not estimated). Total drug exposure up to the last measured concentration (AUC0-last) was calculated using the linear trapezoidal method for ascending concentrations and the logarithmic trapezoidal method for descending concentrations. The AUC0-last was defined as the area under the concentration-time curve from the time of dose until the last concentration above lower limit of quantitation. Area under the concentration-time curve from 0 to infinity (AUC0-inf) was calculated using the formula AUC0-t + C(last) / Kel, where C(last) is the last quantifiable concentration. Half-life (t1/2) was calculated using the formula ln(2)/ Kel. Maximum concentration (Cmax) and time to reach maximum concentration (Tmax) were taken directly from the observed data. Apparent volume of distribution (Vz/F) was calculated as dose/(Kel *AUC0-inf) and apparent clearance (CL/F) calculated as dose/AUC0-inf, according to standard procedures.
PK estimate comparisons between uninfected and infected participants were performed using the Kruskal-Wallis test using the JMP software version 13 (Cary, NC, USA).
Screening of 1,534 adults yielded 70 eligible individuals, with 36 LF-infected (FTS+, Mf+) and 34 uninfected (FTS-, Mf-). These individuals then underwent full screening (Fig 1). Three participants were excluded due to evidence of onchocerciasis infection and one individual was excluded because of elevated creatinine. Excluded participants were treated with IVM + ALB and returned to their village. Ten participants initially treated in the LF-uninfected (FTS-) group were subsequently found to be LF-infected based on repeated examination of plasma with FTS assay and excluded from subsequent analysis per protocol. Baseline demographics were similar in the two groups (Table 1). Thirty-two (57%) of all participants were men. One year following treatment, 31 of 32 (97%) infected and 22 of 24 (92%) uninfected participants were re-examined for the presence of LF infection (FTS, Mf, and ultrasound in males). The three individuals not examined had all moved from the area. Two uninfected men seen at follow-up did not have a repeat ultrasound.
The mean plasma concentration-time profiles of ALB-OX (the active metabolite of ALB), DEC, and IVM are shown in Fig 2, and of ALB and ALB-ON are shown in S1 Fig. There was no difference in plasma concentration for all the drugs between infected and uninfected individuals for all time points examined. Maximum concentrations of ALB-OX were substantially higher than corresponding concentrations for ALB, but showed less variability compared to ALB and ALB-ON, a secondary metabolite. The main PK parameters (median and range) of ALB, ALB-OX, ALB-ON, DEC, and IVM for all 56 subjects are shown in S1 Table. The main PK parameters (median and range) stratified by infection status or by sex are shown in S2 and S3 Tables. Overall, the DEC Tmax occurred at a median time of 4.0 hours post-treatment, with a reported Cmax of 1,522 ng/mL. The median t1/2 for DEC was 9.5 hours. The CL/F and Vz/F of DEC were 8.1 L/hour and 111 L, respectively. The median Cmax for DEC was not different based on infection status (Fig 3A); however, Cmax was higher in female compared to male participants (P< 0.05, Fig 4A). The AUC0-t for DEC with LF and without LF (Fig 3B), or based on sex (Fig 4B) was not different (P>0.05). The median Tmax for IVM was 6.0 hours, with a median t1/2 of 48.1 hours (S1 Table). The CL/F and Vz/F of IVM were 6.8 L/hour and 466.7 L, respectively. IVM Cmax and AUC0-t was not significantly different (P>0.05) when comparing participants with and without LF (Figs 3C and 3D), as well as treatment group comparisons based on sex (Fig 4C and 4D). The median Tmax for ALB-OX was 5.0 hours, with a median t1/2 of 8.9 hours (S1 Table). The CL/F and Vz/F of IVM wer2 L/hour and 848 L, respectively. For ALB-OX, the Cmax and AUC0-t parameters were not significantly different (P>0.05) in the presence or absence of LF (Fig 3E and 3F) and for sex basis (Fig 4E and 4F).
Since mild subjective complaints were common at baseline, new subjective findings (e.g. symptoms) and objective findings (e.g. fever, presence of hematuria, hemodynamic changes), or worsening after treatment of objective and/or subjective observations compared to baseline, were considered to be study-related AEs. Every participant had at least one AE (Table 2). Headache, abdominal pain, and muscle/joint pain were the most common symptoms, followed by diarrhea and fatigue. There was no difference in the frequency of subjective AEs between LF-infected and uninfected individuals. Women were more likely to have multiple subjective AEs (21 (88%) versus 22 (69%) for men, P = 0.02), but there was no difference in the severity of AEs between sexes. For all participants, AEs were most common between 18 and 48 hours post-treatment. With respect to objective AEs, 16 of 32 (50%) LF-infected subjects had fevers versus 4 of 24 (17%) uninfected participants post-treatment (P = 0.01). Individuals with fevers were not treated with antipyretics to allow for evaluation of the kinetics of post-treatment fever. Fevers occurred most commonly between 18 and 42 hours. Frequency and severity peaked at 36 hours in LF-infected subjects. All fevers resolved by 72 hours after treatment. Scrotal swelling and pain occurred in 6 of the 20 (30%) LF- infected men, but this was not observed in uninfected men (P = 0.04). The onset of scrotal swelling and pain occurred from 48 to 96 hours post-treatment. Hematuria (based on urine dipstick, but not by microscopy) was detected in 10 of 32 (31%) LF- infected individuals, compared to 3 of 24 (13%) uninfected participants (P = 0.1, Table 2). Hematuria was predominantly seen at 24 and 48 hours and was resolved by day 7.
Pre-treatment infection parameters are shown in Table 1. Infected men and women had very similar Mf counts and circulating filarial antigen test scores (geomean = 141 Mf/mL and average FTS of 2.3, and geomean = 141.82 Mf/mL and mean FTS of 2.4, respectively). At 39 hours and 7 days post-treatment, all LF-infected participants were Mf negative, and 27 of 31 (87%) were Mf negative 1 year after treatment (Fig 5A). FTS scores decreased significantly after treatment (Fig 5B, P<0.001), and 6 of 31 (19%) participants were FTS negative. All 32 men enrolled in the study had scrotal ultrasound at baseline. Adult worm nests were detected in 14 of 20 (70%) infected men and in 0 of 12 uninfected men. The mean number of worm nests at baseline in positive men was 2.6 (range 1–6). The mean maximum diameter of worm nests was 3.8 mm (range 1.4–7.6 mm). Ultrasounds were repeated on all men at 24, 48, 72 hours and 7 days following treatment. No significant changes were seen in the number or size of worm nests, the degree of lymphatic dilatation or the development of hydroceles at those time points. Ultrasound examinations were performed for 30 men at 1 year after treatment (20 of 20 infected men and 10 of 12 uninfected men). No new worm nests were seen in men of either group. Of the 14 men who had worm nests at baseline, 11 had no detectable worm nests at one year (79% reduction, p<0.001, Fig 5C). Of the three men who had worm nests visible at 1 year, two had a reduced number relative to baseline, and one had no change. One of the men who still had detectable worm nests, though decreased, also had detectable Mf. The other two men were amicrofilaremic at 1 year.
Four individuals failed to have sustained clearance of Mf at 1 year following triple-drug treatment (Fig 5). To determine whether incomplete Mf clearance might result from reduced drug levels compared to levels in individuals with sustained clearance, we calculated the variance of all three drugs from the mean AUC0-inf for each participant (Fig 6). Although there was considerable variability in drug levels among individuals, persons who were microfilaremic at 1 year (circled) had similar overall drug levels to those observed in participants with sustained Mf clearance.
Results from this study show that the presence of moderate to heavy LF infection does not affect IVM, DEC or ALB drug levels or their PK parameters following a single co-administered dose of the triple-drug regimen. Triple-drug therapy was well tolerated in both LF-infected and uninfected individuals and was effective for clearing Mf of W. bancrofti in Ivorian participants for up to 1 year after treatment. These studies provide important additional information in support of the use of triple-drug therapy for MDA in LF endemic areas where onchocerciasis and loaisis are not present.
LF infection did not affect the PK and pharmacodynamics (PD) of IVM, DEC or ALB, based on the observation that the kinetics of drug levels and derived PK parameters for all three drugs were the same irrespective of whether an individual was LF-infected or not. There was considerable variability in plasma ALB and IVM drug levels among individuals. Both drugs, and especially ALB, have been shown to have highly variable gastrointestinal absorption [15], although this has not been shown to affect drug efficacy. By contrast, plasma levels of DEC showed relatively little variation among individuals. This is likely a consequence of good drug bioavailability of DEC. Drug levels did not differ between sexes with the exception of Cmax for DEC, which was higher in women. Compared to ALB and IVM, which are lipophilic and thus have a large Vz/F [16], DEC has a low Vz/F that is more closely associated with the ideal body weight of an individual [17]. This suggests that high DEC concentrations might occur in persons who are overweight if they are dosed based on actual body weight with no maximum dose. However, future studies are needed to evaluate the relation with weight, DEC dosing, and systemic concentrations. Of note, previous studies have shown that the addition of IVM to DEC plus ALB or IVM to ALB does not significantly alter the PK parameters of individual drugs compared to those observed in various combinations, including the triple-drug combination [6, 18].
All participants complained of one or more AEs, which is comparable to that observed in PNG study participants who were treated with the triple-drug regimen (83%) in a similarly designed PK study [6]. The high rate of reported AEs in infected and uninfected participants is probably related to the high frequency of symptom assessment and to effects of confining normally active adults to a hospital ward for three days. One of the most common AEs was back pain, which was relieved by getting up and walking around. Another common AE was dyspepsia, which could possibly be attributed to changes from the normal village diet to that provided while in the study clinic. The two AEs that differed between infected and uninfected adults were fever and scrotal pain and swelling. Fever is a well-known side effect of LF treatment that is associated with the host inflammatory reaction to dying Mf [19, 20]. The inflammatory response is exacerbated in both prevalence and severity with higher blood Mf counts [6, 21, 22]. Scrotal pain is likely a reaction to dying adult worms. The ingestion of the drugs themselves have well-known side effects independent of their impact on helminth infections; for IVM this can include dizziness, joint pain, and skin irritation, and for DEC and ALB, common symptoms are nausea, abdominal discomfort, and headache.
Treatment with triple-drug therapy rapidly cleared Mf in all participants by 39 hours, and all but four individuals remained amicrofilaremic (87% clearance) at 1 year following treatment. This is comparable to, although somewhat less than, the sustained blood Mf clearance levels (96%) of LF-infected subjects in PNG 1 year post-treatment with triple-drug therapy, even though participants from PNG had 4.5 to 10 times higher Mf levels [6, 22]. Failure to sustain Mf clearance was not attributable to reduced drug levels (Fig 6). It is possible that some participants were re-infected during the follow-up period, although this is less likely in one year because the prepatent period for W. bancrofti is about 4 months [23] and we failed to observe any new worm nests in participants. Results from a larger clinical trial of triple-drug therapy in Côte d’Ivoire may shed further light on variability in responses to this regimen. Our results suggest that triple-drug therapy killed many adult worms, as evidenced by reduced circulating filarial antigen levels as assessed by lower FTS scores and the inactivation of worm nests observed by ultrasound after treatment. Inactivation of worm nests following treatment has been interpreted as evidence of a macrofilaricidal effect [24], and one study confirmed this by histological examination of surgically removed worm nests after DEC treatment [25]. Measurement of circulating filarial antigen levels provides a second means of assessing macrofilaricidal activity of antifilarial medications, and antigen levels are believed to correlate with adult filarial burdens [26].
Limitations of this study include the exclusion of children and people with chronic disease (who would normally be included in MDA programs) and the relatively small sample size. Community-wide safety trials including almost 26,000 people in five countries have now been completed. These showed that triple-drug therapy was well tolerated and suggested that it should be as safe as the two-drug MDA regimen DEC + ALB that has been widely used outside of sub-Saharan Africa. Based on these and other studies, WHO changed its policy to recommend triple-drug therapy for MDA in areas that are non-endemic for onchocerciasis or loiasis and unlikely to eliminate LF by the year 2020 [27].
|
10.1371/journal.pntd.0001562 | Early Clinical Features of Dengue Virus Infection in Nicaraguan Children: A Longitudinal Analysis | Tens of millions of dengue cases and approximately 500,000 life-threatening complications occur annually. New tools are needed to distinguish dengue from other febrile illnesses. In addition, the natural history of pediatric dengue early in illness in a community-based setting has not been well-defined.
Data from the multi-year, ongoing Pediatric Dengue Cohort Study of approximately 3,800 children aged 2–14 years in Managua, Nicaragua, were used to examine the frequency of clinical signs and symptoms by day of illness and to generate models for the association of signs and symptoms during the early phase of illness and over the entire course of illness with testing dengue-positive. Odds ratios (ORs) and 95% confidence intervals were calculated using generalized estimating equations (GEE) for repeated measures, adjusting for age and gender.
One-fourth of children who tested dengue-positive did not meet the WHO case definition for suspected dengue. The frequency of signs and symptoms varied by day of illness, dengue status, and disease severity. Multivariable GEE models showed increased odds of testing dengue-positive associated with fever, headache, retro-orbital pain, myalgia, arthralgia, rash, petechiae, positive tourniquet test, vomiting, leukopenia, platelets ≤150,000 cells/mL, poor capillary refill, cold extremities and hypotension. Estimated ORs tended to be higher for signs and symptoms over the course of illness compared to the early phase of illness.
Day-by-day analysis of clinical signs and symptoms together with longitudinal statistical analysis showed significant associations with testing dengue-positive and important differences during the early phase of illness compared to the entire course of illness. These findings stress the importance of considering day of illness when developing prediction algorithms for real-time clinical management.
| Dengue virus causes an estimated 50 million dengue cases and approximately 500,000 life-threatening complications annually. New tools are needed to distinguish dengue from other febrile illnesses. In addition, the natural history of pediatric dengue early in illness in a community-based setting has not been well-defined. Here, we describe the clinical spectrum of pediatric dengue over the course of illness in a community setting by using five years of data from an ongoing prospective cohort study of children in Managua, Nicaragua. Day-by-day analysis of clinical signs and symptoms together with longitudinal statistical analysis showed significant associations with testing dengue-positive and important differences during the early phase of illness compared to the entire course of illness. These findings are important for clinical practice since outside of the hospital setting, clinicians may see dengue patients toward the beginning of their illness and utilize that information to decide whether their patient has dengue or another febrile illness. The results of these models should be extended for the development of prediction algorithms to aid clinicians in diagnosing suspected dengue.
| Dengue virus (DENV) causes the most prevalent mosquito-borne viral disease affecting humans, with 2.5–3 billion people at risk for infection and approximately 50 million cases of dengue each year [1], [2]. The four DENV serotypes are transmitted to humans by Aedes aegypti and Ae. albopictus mosquitoes, primarily in urban and peri-urban areas in tropical and subtropical countries worldwide. Most cases present as classic dengue fever (DF), a debilitating but self-limited illness that manifests with high fever, retro-orbital pain, severe myalgia/arthralgia, and rash. However, in some cases, mainly children, illness progresses to life-threatening dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS), characterized by vascular leakage leading to hypovolemic shock and a case fatality rate up to 5% [1], [3], [4]. Currently, no licensed vaccine or antiviral therapy exists for dengue. Early identification of patients at risk of developing severe dengue is critical to provide timely supportive care, which can reduce the risk of mortality to <1% [1], [2]. However, distinguishing dengue from other febrile illnesses (OFIs) early in illness is challenging, since symptoms are non-specific and common to other febrile illnesses such as malaria, leptospirosis, rickettsiosis, and typhoid fever [5]–[7] in dengue-endemic countries. In addition, many distinguishing clinical features of DHF/DSS generally emerge only after 4–5 days, at defervescence, when the patient is already critically ill.
Although the World Health Organization (WHO) has recently established new clinical guidelines to classify dengue severity [1], serological, virological, and molecular biological tests are required to definitively diagnose DENV infection. In many endemic countries, laboratory diagnosis of dengue is often problematic due to lack of reagents, expense, or delay in obtaining results. Patients with suspected dengue are often hospitalized for close monitoring to ensure proper treatment if they begin to develop severe dengue; however, up to 38–52% are later diagnosed with OFIs [8], [9] and thus were hospitalized unnecessarily at great financial cost to their family and society [10]. New tools are therefore needed to distinguish dengue from OFIs to prevent deaths from severe dengue and to mitigate the economic burden of excess hospitalization.
Recent approaches using multivariable logistic or linear regression models have shown that petechiae, thrombocytopenia (platelet count ≤100,000 cells/mm3), positive tourniquet test, rash, and other signs and symptoms can distinguish dengue from OFIs [11]–[17]; however, results were not consistent across studies. Only two studies considered clinical and laboratory features according to day of illness [18]–[20], but as these were hospital-based studies, the results likely reflect patients with more severe symptoms and not the clinical spectrum of all symptomatic cases in dengue-endemic populations. Furthermore, none of these studies analyzed data using longitudinal statistical methods, which account for correlations between repeated measures on individuals over time. The use of longitudinal statistical methods to analyze cohort data is essential to utilize all of the data available for analysis and appropriately estimate the within-person and between-person variance in measures over time.
In this study, we used five years of data from an ongoing prospective cohort study of approximately 3,800 children aged 2–14 years in Managua, Nicaragua, to examine the frequency of clinical signs and symptoms by day of illness and to generate models for the association of signs and symptoms during the early phase of illness and over the entire course of illness with testing dengue-positive. In order to account for the longitudinal structure of the data, odds ratios (ORs) and 95% confidence intervals were calculated using generalized estimating equations (GEE), adjusting for age and gender.
In August and September 2004, a community-based pediatric cohort was established in District II of Managua, a low-to-middle income area with a population of approximately 62,500 [21]. Study activity was based in the Health Center Sócrates Flores Vivas (HCSFV), a public facility that is the primary source of health care for District II residents. Briefly, participants aged 2–9 years were recruited through house-to-house visits, and additional two year-olds were enrolled each year to maintain the age structure of the cohort [21]. Children were eligible to remain in the study until age 12 or until they moved from the study area. The parent/legal guardian of each participant signed an informed consent form, and children ≥6 years old provided verbal assent. In 2007, participants ≤11 years old were given the opportunity to continue for an additional 3 years, and a second informed consent was performed.
The study was approved by the Institutional Review Boards of the University of California, Berkeley, the Nicaraguan Ministry of Health, and the International Vaccine Institute in Seoul, Korea. Parents or legal guardians of all subjects in both studies provided written informed consent, and subjects 6 years of age and older provided assent.
Upon enrollment, parents/legal guardians of all participants were encouraged to bring their child(ren) to the HCSFV at first sign of illness or fever. Study physicians and nurses, trained in identification of possible dengue cases, provided medical care for study participants. Febrile illnesses that met the WHO criteria for suspected dengue (Table 1) and those without other apparent origin (undifferentiated febrile illnesses) were treated as possible dengue cases and followed daily while fever or symptoms persisted through visits with study medical personnel (Figure 1). Complete blood counts (CBCs) were completed every 48 hours or more frequently as necessary, as indicated by the physician. Cases were monitored closely for severe manifestations and were transferred by study personnel to the Infectious Disease Ward of the Manuel de Jesús Rivera Children's Hospital, the national pediatric reference hospital, when they presented with any sign of alarm (Table 1). In addition, an annual healthy blood sample was collected to identify all DENV infections during the previous year and for baseline CBC values. Study physicians in both the hospital and HCSFV completed systematic data collection forms that contained approximately 80 variables (Table 1). In the hospital, additional clinical data, including fluid balance and treatment, were collected daily during hospitalization or through ambulatory follow-up visits by a team of study physicians and nurses. Data were also recorded on medical tests ordered and treatments prescribed.
A case was considered laboratory-confirmed dengue when acute DENV infection was demonstrated by: detection of DENV RNA by RT-PCR; isolation of DENV; seroconversion of DENV-specific IgM antibodies observed by MAC-ELISA in paired acute- and convalescent-phase samples; and/or a ≥4-fold increase in anti-DENV antibody titer measured using Inhibition ELISA [22]–[25] in paired acute and convalescent samples. DENV serotypes were identified by RT-PCR and/or virus isolation.
Laboratory-confirmed dengue cases were further classified by severity. DHF and DSS were defined according to the traditional WHO criteria (Table 1) [26]. Additional categories of severity were included for those cases presenting with shock without thrombocytopenia and/or hemoconcentration (dengue with signs associated with shock (DSAS)) [23] or dengue fever with compensated shock (DFCS) [27] (Table 1). Laboratory-confirmed cases were defined as primary DENV infections if acute-phase antibody titer, as measured by Inhibition ELISA, was <1∶10 or if convalescent phase antibody titer was <1∶2560, and as secondary infections if the acute titer was ≥1∶10 or convalescent titer was ≥1∶2560 [22]–[25].
Data from the first five years of the study (August 30, 2004–June 30, 2009) were used for analysis. The first three days after onset of fever were considered the early febrile phase of illness. Day of illness at presentation was determined by the date of fever onset, which was defined as the first day of illness as reported by the parent/guardian. Variable definitions are described in Table 1. Positive tourniquet test was examined using cut-offs of ≥10 petechiae/in2 and ≥20 petechiae/in2. Platelet count was dichotomized using a cut-off of ≤150,000 cells/mm3 to enable comparisons during days 1–3. Only data from days 1–8 of illness were included for analysis.
The frequency of dengue testing results (laboratory-confirmed dengue-positive versus dengue-negative) and disease severity (DF versus severe dengue) was examined by year, demographics, serotype and immune response. The frequency of the WHO case definition for suspected dengue was examined by dengue testing results and age, and a chi-square test for trend was performed. The frequency of clinical signs and symptoms by day of illness and dengue severity was also examined using chi-square tests.
To examine the association between clinical signs and symptoms and the odds of testing dengue-positive versus dengue-negative, odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using GEE models assuming an exchangeable correlation structure with robust standard errors to account for the correlations between repeated measures on the same patients over time. First, ORs were calculated using bivariable models that included only dengue testing results and each of the signs or symptoms. All signs and symptoms were then examined in multivariable models that adjusted for age and gender. Data from the first three days of illness and from all days of illness only were analyzed separately. Finally, for comparison, we used traditional logistic regression models to analyze the association between signs and symptoms and testing dengue-positive with data collapsed by illness episode to disregard repeated measures on the same patients, using the same model generation process as for the GEE models. All analyses were conducted using STATA 10 (StataCorp LP, College Station, TX).
From August 2004 to June 2009, 22,778 episodes of febrile illness were evaluated, of which 1,974 episodes were suspected dengue or undifferentiated fever (Figure 1). Of the 1,974 possible dengue cases, 1,793 (91%) tested negative and 181 (9%) were laboratory-confirmed as dengue-positive, of which 161 (89%) were classified as DF, 9 (5%) as DHF, 4 (2%) as DSS, 3 (2%) as DSAS and 4 (2%) as DFCS (Table 1). Nearly all (95%) of the severe dengue cases but only 116 (72%) of the DF cases met the WHO case definition for dengue. The proportion of laboratory-confirmed DENV infections that met the WHO case definition significantly increased by age (chi-square test for trend 5.977, p = 0.01), while younger children experienced significantly more undifferentiated febrile illness due to DENV infection (Figure 2). The median age for cases meeting the dengue case definition was 8 years (range 2–13) and that of undifferentiated febrile illness due to DENV infection was 6 years (range 2–10).
The number of confirmed dengue-positive cases varied by year, as expected (Table 2) [28]. Both genders were equally represented, with a slightly higher percentage of females experiencing severe dengue, though this difference was not statistically significant. The majority of DF cases were DENV-2 (58%), followed by DENV-1 (21%) and DENV-3 (9%), while 60% of severe dengue cases were DENV-2, followed by DENV-3 (25%) and DENV-1 (10%). In addition, there were nearly equal proportions of primary and secondary immune responses among DF cases, whereas the majority (70%) of severe dengue cases were secondary DENV infection (Table 2). The median day of illness at presentation was day 2 for all patients, and almost all presented on days 1–3 of illness (90%). The total follow-up time of all children in the cohort was 17,931 person-years with a median follow-up of 3.9 years per child.
As shown in Figure 3, several signs and symptoms appeared to differentiate OFIs from DF cases, and DF cases from severe dengue cases, according to day of illness. In particular, higher proportions of DF and severe dengue cases experienced petechiae, platelets ≤150,000 cells/mm3, leukopenia, and positive tourniquet test compared to patients with OFIs. Higher proportions of severe cases experienced petechiae, platelets ≤150,000 cells/mm3, myalgia/arthralgia and abdominal pain compared to DF cases and patients with OFIs. Abdominal pain differentiated severe dengue cases from DF and OFI only beginning on day 3 of illness (for severe dengue compared to DF: chi-square 0.144, p = 0.70 for days 1–2 versus chi-square 16.910, p<0.0001 for day ≥3).
Bivariable and multivariable analyses were performed using GEE models to examine signs and symptoms early in illness and over the course of illness (Table 3). On days 1–3 of illness, dengue-positive cases had up to 2-fold increased odds of fever, headache, retro-orbital pain, myalgia, arthralgia, and vomiting compared to patients with OFIs. They also had from 3-fold to 9-fold increased odds of rash, petechiae, positive tourniquet test with cut-offs of ≥10 and ≥20 petechiae/in2, leukopenia, platelets ≤150,000 cells/mm3, poor capillary refill, cold extremities and hypotension compared to patients with OFIs. In contrast, they had decreased odds of abdominal pain, likely because this feature appears later in the entire course of dengue illness. On all days of illness, dengue-positive cases had increased odds of the same signs and symptoms as on days 1–3 of illness; however, the magnitude of the point estimates tended to be higher. This difference was most pronounced for rash and platelets ≤150,000 cells/mm3, which had ORs approximately double in magnitude. In addition, dengue-positive cases had increased odds of three additional signs and symptoms: poor appetite, absence of cough, and increased hematocrit. When GEE analyses on data with the longitudinal structure preserved were compared to traditional logistic regression analyses on data collapsed on febrile episode, the point estimates for the ORs were similar, although the 95% confidence intervals for the logistic regression models tended to be slightly narrower (data not shown).
In this study, we describe the clinical spectrum of pediatric dengue starting early in illness in a community setting. Longitudinal statistical analysis of day-by-day clinical signs and symptoms revealed significant associations with testing dengue-positive and important differences during the early phase of illness compared to the entire course of illness. These results stress the importance of considering day of illness when developing prediction algorithms for real-time clinical management.
The early identification of dengue cases and particularly those at risk for severe dengue is critical for preventing severe illness and death. We found that 25% of laboratory-confirmed dengue cases did not meet the WHO case definition, suggesting that the WHO criteria are not sufficient to identify dengue at younger ages. Younger children may experience different signs and symptoms from adults or may be unable to communicate their symptoms to their parents, health care providers, or both. Previous studies demonstrated that children may experience significantly more cough, vomiting, abdominal pain, rash, epistaxis, oliguria, thrombocytopenia, hepatomegaly, and shock compared to adults, although the direction of these differences was not consistent across studies [13], [15], [29]–[34]. A recent study of dengue in adults showed significant differences in clinical features and outcomes across ten-year age groups, indicating that signs and symptoms associated with DENV infection may continue to evolve past childhood [12]. If these differences are confirmed, the WHO case definition may need to be adjusted to be age-specific to function effectively for younger children and older age groups.
Retro-orbital pain and low platelets were among the clinical features independently associated with DENV infection in this study. These results are supported by a study of dengue patients in Puerto Rico in which data were recorded at the time of initial consult rather than at hospitalization [15], and by a study of Thai children [11]. Moreover, our results showing increased frequency of abdominal pain in patients beginning at day 3 of illness are consistent with a prospective study of adults admitted to an emergency department in Martinique [35]. A positive tourniquet test using cut-offs of ≥10 and ≥20 petechiae/in2 was also independently associated with DENV infection. Both cut-offs were used because studies have indicated that a cut-off of ≥10 may improve discrimination of DENV infection [20], [36]; however, the 1997 WHO classification scheme specified a cut-off of ≥20 [26]. Our results support using a cut-off of ≥10 petechiae/in2, and this cut-off has been specified in the 2011 WHO clinical guidelines [37].
A major strength of this study is the use of statistical models designed for analysis of longitudinal data. Few other prospective community-based cohort studies have analyzed early clinical features in pediatric dengue compared to OFI [20], [38]–[40], and none that we are aware of were analyzed using longitudinal statistical methods that account for correlations between repeated measures on patients. Here, we preserved the longitudinal structure of the dataset by using statistical models that support repeated measurements on subjects over time and account for correlations between signs and symptoms experienced within the same individual on different days of illness and in multiple episodes. Longitudinal data have long been collected in dengue research but have rarely been analyzed using appropriate statistical methods. This may introduce bias into findings, as studies may overestimate the magnitude of association or reduce the statistical power of the study as data are lost when they are collapsed for non-longitudinal analysis.
An additional strength of this study is that it is community-based [21], enabling day-by-day capture of information on the early course of illness and on the full clinical spectrum of symptomatic dengue. In contrast, nearly all previous studies enrolled patients upon presentation to a hospital [18], where patients present later; thus, these studies were unable to capture information on the early days of illness or on mild disease. By examining the clinical spectrum of dengue by day of illness, we were able to detect differences in the prevalence of signs and symptoms that could not be revealed by simply analyzing whether they ever occurred over the course of illness. In addition, through multivariable longitudinal models, we were able to identify distinguishing features of dengue during the early phase of illness compared to the entire course of illness. These findings are important for clinical practice since outside of the hospital setting, clinicians may see dengue patients toward the beginning of their illness and utilize that information to decide whether their patient has dengue or another febrile illness. The results of these models should be extended for the development of prediction algorithms to aid clinicians in diagnosing suspected dengue.
This study was not without its limitations. Some participants migrated out of the study area or withdrew from the study; however, our retention rate was approximately 95% per year [21], suggesting that any bias from loss to follow-up would be minimal. It is also possible that we did not capture all symptomatic dengue cases. However, in yearly participant surveys, only an average of 2–3% of participants reported having attended a health-care provider outside of the study or having an illness and not attending any medical provider [21], and approximately 20-fold more laboratory-confirmed dengue cases were captured in the cohort study than by the National Surveillance System [41]. Unfortunately, due to the low number of severe dengue cases, this study did not have sufficient statistical power to compare severe dengue cases to DF cases using GEE models, and these low numbers may have influenced the lack of significant association of signs of severe dengue with testing dengue-positive. For this study, we used the 1997 WHO classification scheme for disease severity. In 2009, the WHO updated its guidelines for classification of dengue disease severity [1], [37]; it would be interesting to re-analyze the data in a future study using the new classification scheme. Studies of the usefulness and applicability of the revised guidelines have been recently performed [42], [43].
In summary, this study is one of the few cohort studies to provide early data on the full clinical spectrum of pediatric dengue. Though we found significantly increased odds for association of several clinical signs and symptoms with testing dengue-positive, these increases were more modest for the early phase of illness compared to the course of illness, suggesting that caution should be taken when using the results from the entire course of illness to develop prediction algorithms. Non-parametric methods such as decision tree analysis overcome some of the limitations of traditional logistic regression models and have recently been applied to develop algorithms for prediction of dengue diagnosis and disease severity [9], [44], [45]. These and other data-adaptive approaches such as Super Learner [46] that are less subject to bias should be further explored to develop prediction algorithms for early identification of dengue cases and improved clinical management.
|
10.1371/journal.ppat.1005053 | Activation of TLR2 and TLR6 by Dengue NS1 Protein and Its Implications in the Immunopathogenesis of Dengue Virus Infection | Dengue virus (DV) infection is the most prevalent mosquito-borne viral disease and its manifestation has been shown to be contributed in part by the host immune responses. In this study, pathogen recognition receptors, Toll-like receptor (TLR) 2 and TLR6 were found to be up-regulated in DV-infected human PBMC using immunofluorescence staining, flow cytometry and Western blot analyses. Using ELISA, IL-6 and TNF-α, cytokines downstream of TLR2 and TLR6 signaling pathways were also found to be up-regulated in DV-infected PBMC. IL-6 and TNF-α production by PBMC were reduced when TLR2 and TLR6 were blocked using TLR2 and TLR6 neutralizing antibodies during DV infection. These results suggested that signaling pathways of TLR2 and TLR6 were activated during DV infection and its activation contributed to IL-6 and TNF-α production. DV NS1 protein was found to significantly increase the production of IL-6 and TNF-α when added to PBMC. The amount of IL-6 and TNF-α stimulated by DV NS1 protein was reduced when TLR2 and TLR6 were blocked, suggesting that DV NS1 protein is the viral protein responsible for the activation of TLR2 and TLR6 during DV infection. Secreted alkaline phosphatase (SEAP) reporter assay was used to further confirm activation of TLR2 and TLR6 by DV NS1 protein. In addition, DV-infected and DV NS1 protein-treated TLR6-/- mice have higher survivability compared to DV-infected and DV NS1 protein-treated wild-type mice. Hence, activation of TLR6 via DV NS1 protein could potentially play an important role in the immunopathogenesis of DV infection.
| Despite the prevalence of dengue virus infection and the heavy economic burden it puts on the endemic countries, the immunopathogenesis of dengue virus infection remains unclear. Plasma leakage in dengue hemorrhagic fever (DHF) develops not when the viremia is at its peak in infected patients but when viremia has been significantly reduced or cleared. This suggests that host immune response is responsible for the development DHF. The interactions of the viral factors with host factors which trigger the host immune responses are likely to play a significant role in the development of dengue diseases, thus are of great interests. In this study, we found that dengue NS1 protein activates TLR2 and TLR6, leading to increase proinflammatory cytokine production. In addition, the interaction of viral factor with TLR6 was found to play an important role in the manifestation of dengue virus infection. Our study provides new insights into the involvement of TLR6 in dengue virus infection and the potential of using TLR6 anatagonist in therapeutic treatment for DV infection.
| Dengue virus (DV) is a member of the Flavivirus genus of the Flaviviridae family. Dengue virus is a positive-sense, single-stranded RNA virus and it has four distinct serotypes (DV1 to 4). Infection by one serotype only confer resistance to the other serotypes for a few months and subsequent secondary infection of a different serotype has a higher risk of developing into the more severe forms of dengue infection; the dengue hemorrhagic fever or dengue shock syndrome [1–5]. Dengue virus genome encodes for a single polyprotein that consists of 3 structural proteins (capsid, premembrane and envelope) that form the physical structure of the virus particle and 7 non-structural proteins (NS1, NS2a, NS2b, NS3, NS4a, NS4b, NS5) which are necessary for the replication of the virus.
Dengue is a mosquito-borne viral disease transmitted through a human-to-mosquito-to-human transmission cycle typically by the Aedes mosquitoes: Aedes aegypti and Aedes albopictus. DV infection remains the most prevalent mosquito-borne viral disease and the geographical regions at risk are continually growing due to globalisation and climate change [6]. It is estimated that 100 million cases of dengue infection occur worldwide each year with 2.5 billion people at risk [7–9]. Till now, no effective treatment and vaccine are available for DV infection.
The pathogenesis of dengue is not well-understood. The mechanism underlying the wide range of dengue manifestations remain largely unknown. However, the observation that plasma leakage in DHF develops not when the viremia is at its peak in infected patients but when viremia has been significantly reduced or cleared, suggesting that host immune response is responsible for the development DHF [10–13]. In addition, studies have demonstrated that the host immunological mechanism could play a key role in the manifestation of dengue infection [3,14–16]. Up-regulation of proinflammatory cytokines and immune cells during dengue virus infection could lead to increased vascular permeability and leakage [17–20]. The hypotheses of antibody-dependent enhancement of infection and original antigenic sin have been proposed to explain the underlying mechanism that contributes to the manifestation of the more severe forms of the dengue infection during secondary infections [21–24].
Toll-like receptors (TLRs) are pathogen recognition receptors (PRRs). PRRs are a group of receptors that play a key role in immune surveillance. Pathogen recognition receptors are important as they alert the immune system of the presence of foreign microbes by recognizing pathogen-associated molecular patterns (PAMPs) and activating the immune system upon binding to PAMPs. In human, 10 functional TLRs are documented and each recognizing a group of PAMPs. When TLR is activated, adapter molecules like myeloid differentiaton primary response gene 88 (MyD88), toll-interleukin 1 receptor domain containing adaptor protein (Tirap), TIR-domain-containing adaptor-inducing interferon-β (Trif) and toll-like receptor 4 adaptor protein (Tram) are recruited. These adapter molecules in turn activate other downstream transcriptional gene regulators like activating protein-1, NFκB and interferon regulatory factors which induce expression of chemokines, proinflammatory cytokines [tumor necrosis factor alpha (TNF-α), IL-6, IL-1β and IL-12] or costimulatory molecules [25]. The up-regulation of costimulatory molecules is essential for the induction of pathogen-specific adaptive immune responses [26]. Thus, TLRs can activate both the innate and adaptive immune responses.
TLR can recognize viral pathogen via a number of different viral ligands. Generally, TLR3 detects double-stranded viral RNA, TLR2 and TLR4 sense the presence of virus via their proteins, TLR7/8 binds single-stranded viral RNA and TLR9 recognizes viral CpG DNA [27]. Among the TLRs, TLR3 and TLR7 have been found to trigger IL-8 production when stimulated by dengue viral RNA [28,29]. TLR6 was found to be up-regulated in DV2-infected K562 cells using Human Th1-Th2-Th3 RT2 Profiler PCR arrays in our previous study [30]. Although TLR6 was previously known only to be activated by diacylated lipoprotein of bacteria [31], recent studies have found that TLR6 can be activated by viruses including hepatitis C virus and respiratory syncytial virus [32,33]. Viral infection induces various TLR-mediated innate responses, which subsequently play a pivotal protective or pathogenic role in conjunction with virus-specific adaptive immune responses [34,35]. In the current study, dengue virus infection was found to activate and up-regulate TLR2 and TLR6 of human PBMC and DV NS1 protein was shown to be the viral protein responsible. Knockout of TLR6 increased the survivability of mice infected by dengue virus. Prolonged activation of TLR6 by DV NS1 protein during DV infection could be responsible for the lower survivability observed in wild-type mice compared to the TLR6-/- mice. Hence, TLR6 may play an important role in the immunopathogenesis of dengue virus infection.
In our previous study [30], several genes involved in the TLR6 pathway have been found to be significantly up-regulated during dengue virus infection in K562 cells on day 3 post-infection which include TLR6, IL-6, TNF-α and CD80. TLR6 pathway activation is well-documented to play an important part in activating both the innate and adaptive immunity. Human PBMC were found to express the whole range of human TLRs (TLR1-10) [36]. The expression of TLRs has been found to increase following inflammations and exposure to pathogens or specific ligands [37–40].
The susceptibility of the PBMC to DV2 infection was first determined using plaque assay (Fig 1A). The virus titer peaked on day 2 post-infection (3.82 Log10PFU/ml) and decreases from day 3 to day 5 post-infection. The increase in virus titer on day 2 post-infection of PBMC provided evidence of replication of DV2 in PBMC.
Next, flow cytometric analyses were performed to determine the expression of TLR6 of mock-infected and DV2-infected PBMC. Gating was used to exclude the cell debris (Fig 1B). PBMC were also stained with anti-CD14 FITC conjugated antibody to serve as a marker for human monocytes [41,42]. Monocytes are the main cells in PBMC that express TLR2 and TLR6 [31]. Upon dengue virus infection, higher percentages of TLR6+CD14+ cell population was observed compared to the mock-infected PBMC on day 2 and 3 post-infection (Figs 1C, 1E, and S1). TLR6 requires heterodimerization with TLR2 to recognize ligand and trigger cytokine production [43–45]. Hence, TLR2 expression was also investigated. TLR2+CD14+ cell population was up-regulated upon DV infection on day 3 post-infection (Figs 1D, 1F, and S2). CD14+ monocytes expressing TLR2 were reported by Azeredo and colleagues (2010) to be increased in peripheral blood of dengue patients. In addition, DV2-infected PBMC expressed higher level of TLR6 and TLR2 on day 3 post-infection but not day 1 and day 2 post-infection (S3 Fig).
PBMC were also stained for both TLR2 and TLR6 simultaneously for flow cytometric analysis. In addition, the CD3 and CD20 coexpression on PBMC were analyzed as high percentage of CD3+CD20+ cell population could suggest neoplastic transformation. The percentage of PBMC expressing both CD3 and CD20 were 3.84% and 3.94% which are within the range detected by other research groups using healthy donors [46–48]. The percentages of mock-infected PBMC which were TLR2+ and TLR6+ were lower than the percentages of DV2-infected PBMC which were TLR2+ and TLR6+ (Fig 1G, 1H and S1 Table). The median fluorescence intensity of the TLR2 and TLR6 were also higher for the DV2-infected than the mock-infected (S1 Table).
After affirming the up-regulation of TLR2 and TLR6 of PBMC when infected by DV, activation of these receptors during DV infection were investigated by measuring the amount of IL-6 secreted into the extracellular milieu by the DV2-infected PBMC. The amount of IL-6 in the culture media of DV2-infected PBMC increased significantly from day 2 to day 5 post-infection as compared to that of the mock-infected PBMC (Fig 2A). Similarly, the DV2-infected PBMC significantly up-regulated the amount of TNF-α secreted into the extracellular mileu from day 2 to day 4 post-infection, compared to that of the mock-infected PBMC (Fig 2B). UV-inactivated DV did not induce up-regulation of IL-6 (Fig 2A) and TNF-α (Fig 2B) when added to PBMC culture. This could suggest that viral replication is required for the up-regulation of IL-6 and TNF-α. To determine if the up-regulation of IL-6 and TNF-α detected were contributed by TLR2 and TLR6 activation, TLR2 and TLR6 blocking antibodies were used. Blocking of TLR2 and TLR6 reduced the amount of IL-6 produced by PBMC when stimulated by LPS (5 μg/ml) or MALP-2 (50 ng/ml), compared to that of the isotype control (Fig 2C). MALP-2 is a 2-kDa synthetic derivative of the macrophage-activating lipopeptide and it is a specific agonist for TLR2 andTLR6. Blocking of TLR2 and TLR6 also reduced the amount of IL-6 produced by PBMC during dengue virus infection, this suggested that TLR2 and TLR6 are activated during dengue virus infection and this activation led to increase in IL-6 secretion. Similar observation was made for TNF-α production by PBMC (Fig 2D). This suggested that TLR2 and TLR6 are the receptors activated during DV infection to result in the increase in IL-6 and TNF-α expression.
To determine if any specific viral protein is responsible for the activation of TLR2 and TLR6, IL-6 and TNF-α expression by PBMC after treatments with individual viral proteins were assayed. The detection of up-regulation of IL-6 and TNF-α by PBMC would indicate possible activation of receptors by the dengue viral proteins. ELISA was performed to quantify the amount of IL-6 secreted into the supernatant by PBMC after treatment with the individual dengue viral proteins on day 2 post-treatment. Among the dengue viral proteins, DV NS1 protein is the only viral protein which stimulated high amount of IL-6 expression (5864 pg/ml) (Fig 3A). IL-6 expression was slightly down-regulated by dengue envelope protein (296 pg/ml) and NS3 protein (297 pg/ml) compared to His-tag-treated PBMC (346 pg/ml). The IL-6 level of His-tag-treated PBMC was comparable to that of the untreated, suggesting that His-tag did not trigger IL-6 production and the IL-6 detected in the culture supernatant of the His-tag-treated PBMC was due to basal expression. The IL-6 level of UV-inactivated DV2-treated PBMC was also comparable to that of the untreated. This suggested that non-replicative virus cannot induce IL-6 expression. The positive control, LPS was found to stimulate IL-6 production by PBMC. ELISA was also performed to determine which of the viral protein can induce up-regulation of TNF-α by PBMC. DV NS1 protein is the only viral protein which stimulated high amount of TNF-α expression (293 pg/ml) compared to that of the His-tag-treated PBMC (30 pg/ml) (Fig 3B). Similar to IL-6, the TNF-α level of His-tag-treated PBMC was comparable to that of the untreated and the TNF-α level of UV-inactivated DV2-treated PBMC was also comparable to the untreated. The positive control, LPS was also found to stimulate TNF-α production by PBMC. The result suggested that DV NS1 protein is the viral protein that stimulates IL-6 and TNF-α production by PBMC during DV infection.
Next, lower concentration of DV NS1 protein (1 μg/ml) was used to stimulate PBMC. This concentration of DV NS1 protein is within the concentration range detected in dengue patients (several nanograms per millilitre to several micrograms per millilitre) [49]. ELISA was performed to quantify the amount of IL-6 in the supernatants of DV NS1-treated PBMC on day 1 to day 3 post-treatment. DV NS1 protein-treated PBMC produced significantly higher amount of IL-6 compared to that of the His-tag-treated PBMC from day 1 post-treatment (Fig 3C). The up-regulation was faster than that of the DV2-infected PBMC which only produced significantly higher IL-6 from day 2 post-treatment (Fig 2A). The delay observed in the DV2-infected PBMC could be due to time required for DV NS1 protein synthesis during DV replication and secretion. The secreted DV NS1 protein can then be detected by the cell surface receptors, TLR2 and TLR6. The IL-6 produced by DV NS1 protein-treated PBMC peaked on day 2 post-treatment (2917 pg/ml).
Similarly, the amount of TNF-α produced by DV NS1 protein-treated PBMC was quantified using ELISA. DV NS1 protein-treated PBMC produced significantly higher amount of TNF-α compared to that of the His-tag-treated PBMC from day 1 post-treatment (Fig 3D). The amount of TNF-α produced by DV NS1 protein-treated PBMC peaked on day 1 post-treatment (226 pg/ml). The amount of TNF-α decreased from day 2 to day 3. To determine if the IL-6 production by PBMC upon DV NS1 protein stimulation is through TLR2 and TLR6, TLR2 and TLR6 neutralizing antibodies were used. The specificity of TLR2 and TLR6 blocking antibodies were tested using TLR4 specific ligand, ultrapure LPS (S4 Fig). TLR2 and TLR6 blocking antibodies did not affect TLR4. TLR2 and TLR6 of PBMC were blocked by the neutralizing antibodies prior to addition of DV NS1 protein into the PBMC culture (1 μg/ml). The supernatant of the treated PBMC were collected on day 1 post-treatment and IL-6 was quantified using ELISA. Day 1 was chosen as the time point as IL-6 was found to be significantly up-regulated from day 1 post-treatment in Fig 3C. With both TLR2 and TLR6 blocked, IL-6 secreted by DV NS1 protein-treated PBMC was significantly reduced compared to that of the isotype control (Fig 3E). With only TLR6 blocked, IL-6 secreted by DV NS1 protein-treated PBMC was comparable to that of both TLR2 and TLR6 blocked (Fig 3E). Therefore, DV NS1 protein stimulation of IL-6 requires both TLR2 and TLR6. The stimulation is inhibited when one of the receptors is blocked. With only TLR2 blocked, IL-6 secreted by DV NS1 protein-treated PBMC was significantly reduced and surprisingly lower than that of the His-tag-treated PBMC (Fig 3E). High amount of TLR2 neutralizing antibody may have some effect on the basal IL-6 expression of PBMC. TLR2 neutralizing antibody was found to be more effective than TLR6 neutralizing antibody. The expected level of IL-6/TNF-α for DV NS1 protein-treated PBMC with 500 ng/ml of TLR2 and 500 ng/ml of TLR6 neutralizing antibodies should be between the level of IL-6/TNF-α for DV NS1 protein-treated PBMC with 1000 ng/ml of TLR2 only and those with 1000 ng/ml of TLR6 only. The presence of the more efficient TLR2 blocking antibodies in the treatment group with both blocking antibodies should be able to more efficiently block the TLR2/6 pathway compared with the treatment with only TLR6 blocking antibody. However, the observation was not the case. The expected result would only happen if the TLR2 and TLR6 antibodies can sterically hinder the binding of each other to prevent one TLR2/6 complex from binding both TLR2 and TLR6 antibodies at the same time. The observed result suggested that the two antibodies did not sterically hinder each other. Thus, one TLR2 and one TLR6 antibodies can bind and inhibit the same TLR2/6 complex, an inhibition which can be achieved initially with just either one TLR antibody.
In summary, DV NS1 protein-treated PBMC which were also treated with TLR2 or TLR6 neutralizing antibodies or both secreted less IL-6 compared to the isotype control. Together, these data implied that TLR2 and TLR6 are the receptors activated by DV NS1 protein. Similar to what was observed for IL-6, the TNF-α amount secreted by His-tag-treated PBMC in general, was not affected by the neutralizing antibodies (Fig 3F). With both TLR2 and TLR6 blocked, TNF-α secreted by DV NS1 protein-treated PBMC was significantly reduced compared to that of the isotype control. With only TLR6 blocked, TNF-α secreted by DV NS1 protein-treated PBMC was comparable to that of both TLR2 and TLR6 blocked. With only TLR2 blocked, TNF-α secreted by DV NS1 protein-treated PBMC was significantly reduced. With 1000 ng/ml of TLR2 neutralizing antibody, the basal TNF-α expression of His-tag-treated PBMC was affected, as shown by the lower TNF-α level of the TLR2 blocked His-tag-treated PBMC compared to that of the isotype control His-tag-treated PBMC. This may suggest that TLR2 activation partially contributed to the basal expression of IL-6 and TNF-α detected in our PBMC culture. In summary, PBMC which were treated with TLR2 and/or TLR6 neutralizing antibodies secreted less TNF-α compared to the isotype control. Together, the data implied that TLR2 and TLR6 are the receptors activated by DV NS1 protein.
In addition, SEAP reporter assay was used to further confirm if DV NS1 protein is activating TLR2 and TLR6 using the HEK 293 cells. HEK 293 cell line which naturally does not possess many of the TLRs was also used for the reporter assay. HEK 293 cells have good transfection efficiency to allow expression of desired TLR and the SEAP reporter plasmid for investigating specific TLR ligand. HEK 293 cells were found to express low level of endogenous TLR6 but not TLR2 [50,51]. The reports of low level of expression of TLR6 and no expression of TLR2 in HEK 293 cells were further confirmed in our western blot results (S5 Fig). LPS and MALP-2 were used as positive control. For HEK 293 cells transfected with only SEAP reporter plasmid, the SEAP secretion by HEK 293 cells treated with DV NS1 protein and the positive controls were not significantly different from the negative controls (untreated and His-tag-treated HEK 293 cells) (Fig 3G). This suggested that DV NS1 protein, LPS and MALP-2 cannot stimulate NFκB activation in the absence of TLR2. DV2-infected HEK 293 cells produced significantly higher SEAP than mock-infected HEK 293 cells. This suggested that DV2-infection can trigger NFκB activation through pathway independent of TLR2. For HEK 293 cells transfected with SEAP reporter, TLR2 and TLR6 expression plasmids, the SEAP secretion by HEK 293 cells treated with DV NS1 protein and the positive controls were significantly different from the negative controls from day 1 post-treatment (Fig 3H). The result suggested that activation of NFκB by DV NS1 protein is dependent on both TLR2 and TLR6.
DV2-infection was found to up-regulate TLR6 in PBMC (Fig 1E and 1H). To determine if this up-regulation is contributed by DV NS1 protein effect on the cells, Western blot analyses were used. PBMC were treated with His-tag (1 μg/ml), DV NS1 protein (1 μg/ml), LPS (5 μg/ml), mock-infected or DV2-infected (M.O.I of 10). TLR6 bands of DV2-infected, DV NS1 protein-treated and LPS-treated PBMC were of higher intensity than those of mock-infected and His-tag-treated PBMC (Fig 4A). The relative density of the TLR6 bands normalized against the actin bands was plotted on a graph (Fig 4B). The result suggested that DV NS1 protein can stimulate up-regulation of TLR6 in PBMC. Similarly, TLR2 bands of DV2-infected, DV NS1 protein-treated and LPS-treated PBMC were of higher intensity than those of mock-infected and His-tag-treated PBMC (Fig 4C). The relative density of the TLR2 bands normalized against the actin bands was plotted on a graph (Fig 4D). The results suggested that DV NS1 protein can stimulate up-regulation of TLR2 and TLR6 in PBMC.
TLR2 and TLR6 expression on PBMC were further investigated using immunofluorescence analyses. Mock-infected and DV2-infected PBMC were harvested on day 3 post-infection and stained for TLR2/TLR6 and CD14. The staining of untreated PBMC (Fig 4E and 4J) were comparable to that of the mock-infected (Fig 4F & 4K) and His-tag-treated PBMC (Fig 4H and 4M). Similar to the results obtained in flow cytometric and western blot analyses, TLR2/TLR6 was up-regulated in DV2-infected PBMC (Fig 4G and 4L), indicated by the denser red spots compared to the mock-infected PBMC (Fig 4F and 4K). TLR2/TLR6 was also up-regulated in DV NS1 protein-treated PBMC (Fig 4I and 4N) compared to the His-tag-treated PBMC (Fig 4H and 4M). Colocalization of both TLR2 and TLR6 with CD14 were observed for untreated, mock-infected, DV2-infected, His-tag-treated and DV NS1 protein-treated PBMC (yellow stains). Hence, the colocalization of the receptors could be independent of infection or DV NS1 protein stimulation. It was reported that TLR2 and TLR6 heterodimers pre-exist and are not induced by ligand [45].
The activation of TLR2 and TLR6 could be a double-edged sword that is both beneficial and detrimental to the host. To assess the potential role of the activation of TLR2 and TLR6 plays during dengue virus infection, the use of cell model is not sufficient, an animal model is required. Wild-type and TLR6 knockout (TLR6-/-) C57BL/6 mice were used in this part of the study. In order to determine if TLR6 activation during dengue virus infection contributes to the pathogenesis of the disease, wild-type and TLR6-/- mice were injected with 2.7 x 108 PFU of DV2 on day 1–2 day-old (Fig 5A). The survival rate of the DV2-infected wild-type mice was 61.4% at the end point of the study. The survival rate of the TLR6-/- DV2-infected mice was 83.0% at the end point of the study. Knockout of TLR6 increased the survival rate of the mice at the end point of the study by 21.6%, suggesting that activation of TLR6 may contribute to the pathogenesis of the disease, leading to higher fatality observed in the DV2-infected wild-type mouse population. Using Log-rank test, DV2-infected wild-type mice survival curve was found to be statistically different from DV2-infected TLR6-/- mice. Hence, knockout of TLR6 significantly enhanced the survival rate of the DV2-infected mice.
Next, we investigated what could have resulted in the difference in survival rate of wild-type and TLR6-/- mice. Pups which were 1–2 day-old were injected with 2.7 x 108 PFU of DV2 and quantified for virus titer in the sera and livers. DV2 were detected in all the DV2-infected pups from day 1 to day 2 post-infection. The average virus titer detected in the sera of the DV2-infected wild-type mice on day 1 was 1.51 x 105 PFU/ml while that on day 2 was 9.17 x 102 PFU/ml and that on day 3 was 1.81 x 102 PFU/ml (Fig 5B and Table 1). This suggested that the pups were susceptible to dengue virus infection. 1–2 day-old TLR6-/- mice were also infected in the same way as the wild-type. Virus in the sera of TLR6-/- mice was also quantified. The average virus titer detected in the sera of the DV2-infected TLR6-/- mice was 2.73 x 106 PFU/ml on day 1 while that on day 2 was 2.40 x 103 PFU/ml and that on day 3 was 2.54 x 101 PFU/ml (Fig 5B). Comparing the virus titers obtained in the sera of DV2-infected wild-type and TLR6-/- mice, virus titers were not statistically significantly different. This suggested similar susceptibility of wild-type and TLR6-/- mice to DV2. Viremia persisted in both wild-type and TLR6-/- mice till day 3. By day 4 post-infection, virus can no longer be detected in the sera of mice except for one TLR6-/- mice whose serum detected presence of DV2 on day 5 post-infection. Virus titers in the livers of wild-type and TLR6-/- mice were also quantified using plaque assay (Fig 5C). Unlike what was observed for the sera, DV2 was not detected in every liver of the DV2-infected mice on day 1 and 2 post-infection. On the contrary, DV2 were detected in the livers of DV2-infected mice more frequently on day 4 and day 5 post-infection compared to that of the sera. This may suggest that though not all the DV2 can establish infection in the liver organ, for those that established, it can persist longer in the liver than in the sera. The sera and livers of both the wild-type and TLR6-/- mock-infected mice were detected to be absent of DV2.
DV2-infected mice developed some abnormal signs like enlarged belly, hind limb paralysis, moribundity and death (Table 2). Hindlimb paralysis was also observed in DV2-infected mice. Symptoms of paralysis of extremities which has been observed in some dengue patients were also observed in the murine model [52]. However, such occurrences were rare. From the observations of the mice on a daily basis, the occurrence of paralysis was observed on day 10–14 post-infection.
Some of the DV2-infected mice were found to succumb to the infection. No viable virus was detected in the tissues of the dead mice which could be due to decomposition. Blood of the dead mice could not be harvested due to the clotting of the blood. High virus titers were detected in the brain, liver and limbs of the moribund mice, the virus titers were higher than the average virus titers detected in the brain, liver and limbs of the asymptomatic DV2-infected mice (Table 2). However, no virus was detected in the sera of the moribund mice which was similar to what was observed for the asymptomatic DV2-infected mice on day 5 post-infection. The amount of IL-6 in the sera of one of the moribund mice was assayed and high amount of IL-6 was detected (2690.5 pg/ml) (Table 2).
Paraplegia was observed in some of the DV2-infected mice but not for any of the mock-infected mice. As paraplegia was observed in the DV2-infected mice, hindlimbs of the mice were also harvested and quantify for DV2 titer using plaque assay (Fig 5D). Similar to what was observed for the DV2 titers of the liver, DV2 was also not detected in the limbs of every DV2-infected mouse on day 1 and day 2 post-infection and virus was detected on day 4, day 5 and day 21 post-infection. No virus was detected in the limbs of mock-infected wild-type and TLR6-/- mice.
Paralysis of the limb could be due to presence of virus in the central nervous system [53]. In view of that, the brains of DV2-infected wild-type and TLR6-/- mice were harvested. DV2 was able to gain entry into the brain and persisted there in both the wild-type and TLR6-/- mice (Fig 5E). DV2 was able to replicate in the brain of TLR6-/- mice from day 1 post-infection while DV2 was only detected in the brain of DV2-infected wild-type mice from day 3 post-infection. DV2 was only detected from day 3 post-infection in the brains of wild-type mice. DV2 can still be detected at the endpoint of the experiment in both wild-type and TLR6-/- mice. No virus was detected in the brain of the mock-infected mice.
When the virus titers of mice which displayed lower limb paralysis were titered using plaque assay, it was found that most of the viruses were localized in the brain rather than the limb, liver or the serum (Table 2). This suggested that DV2 persisted in the brains of these mice and affected the central nervous system, leading to the paralysis. High virus titers were found in the homogenized brains of the mice which displayed symptoms of paralysis, much higher than the asymptomatic mice. One of the mice was found to have only one limb paralyzed, the virus titer in each of the limb was titered separately to determine if there would be a difference in virus titers in the two limbs. The virus titer in the paralyzed limb (1.4 x 103 PFU/g) was comparable to that of the normal limb (4.0 x 103 PFU/g), suggesting that paralysis was not due to virus replication in the limb (Table 2). Moreover, some of the mice which exhibited limb paralysis were not detected with DV in the limbs, further substantiating that paralysis was not due to DV in the limbs (Table 2).
It has been shown using the PBMC cell model that DV2 infection up-regulates IL-6 expression. IL-6 expression in sera of DV2-infected mice was investigated using ELISA. It was noticed that not all the mice up-regulated IL-6 expression upon dengue virus infection (Fig 6A). Some of the mice, both DV2-infected wild-type and TLR6-/-, remained unresponsive to the infection. The amount of IL-6 in the sera of those mice was comparable to that of the mock-infected mice. Among those that responded, DV2-infected wild-type mice secreted higher amount of IL-6 compared to that of the DV2-infected TLR6-/- mice, indicating that TLR6 activation contributed to the IL-6 expression during dengue virus infection. This observation is similar to what was seen in the human PBMC cell model. In addition, it was noticed that IL-6 up-regulation in the sera of DV2-infected TLR6-/- mice subsided by day 5 post-infection while that of responsive wild-type mice remained up-regulated. In general, there was an increasing trend observed in the IL-6 expression of those responsive DV2-infected wild-type mice from day 1 to day 5 post-infection.
The amount of TNF-α in the sera of DV2-infected mice was also determined using ELISA. TNF-α expression in sera of DV2-infected mice was similar to IL-6 expression (Fig 6B). Among those that responded, DV2-infected wild-type mice secreted higher amount of TNF-α compared to that of the DV2-infected TLR6-/- mice, indicating that TLR6 activation contributed to the TNF-α expression during dengue virus infection.
DV NS1 protein was found to be the viral protein responsible for activating TLR2 and TLR6 using the PBMC model. As DV NS1 protein could be the viral protein responsible for the IL-6 and TNF-α up-regulation in the mice as well, the presence of DV NS1 protein in the sera of mice after intraperitoneal injection of DV2 was determined using Bio-Rad Platelia Dengue NS1 Antigen detection kit.
DV NS1 protein persisted in the sera of mice after injection of DV2 for both wild-type and TLR6-/- mice (Fig 6C and 6D). In general for both wild-type and TLR6-/- mice, the DV NS1 protein level started to decrease from day 4 post-infection and on day 5, DV NS1 protein level in one of the wild-type mice fell close to the relative OD of the mock-infected mice. The amount of DV NS1 protein in the sera of the DV2-infected wild-type and TLR6-/- mice was comparable. This could be due to the comparable virus titers in the DV2-infected wild-type and TLR6-/- mice, suggesting comparable replication level and thus similar DV NS1 protein production.
Next, we determined if TLR6 of the mice was activated by DV NS1 protein during DV infection. The effect of DV NS1 protein on IL-6 expression of murine peritoneal macrophages was investigated. Similar to what was observed for human PBMC cell model, without the presence of TLR6, DV2-infected murine peritoneal macrophages secreted significantly lesser amount of IL-6 (Fig 7A). The amount of IL-6 produced by DV NS1 protein-treated wild-type murine peritoneal macrophages was significantly much more than DV NS1 protein-treated TLR6-/- murine peritoneal macrophages. The level of IL-6 produced by DV NS1 protein-treated TLR6-/- murine peritoneal macrophages was comparable to that produced by the mock-infected TLR6-/- murine peritoneal macrophages. In the absence of TLR6, DV NS1 protein cannot stimulate the production of IL-6 by murine peritoneal macrophages. The amount of IL-6 produced by DV2-infected TLR6-/- murine peritoneal macrophages was higher than that of the mock-infected, suggesting that the stimulation by DV NS1 protein only contributed partially to the amount of IL-6 detected in the DV2 infection. In addition, the effect of MALP-2 on the secretion of IL-6 by the murine peritoneal macrophages was eliminated in the absence of TLR6. Similar to IL-6, TNF-α production by the murine peritoneal macrophages upon stimulation with DV NS1 protein was significantly reduced in the absence of TLR6 (Fig 7B). Ultrapure LPS, a TLR4-specific agonist can induce IL-6 and TNF-α production by murine peritoneal macrophages of both wild-type and TLR6 knockout mice. These results suggested that TLR6 of mice can be activated by DV NS1 protein.
As DV NS1 protein can induce IL-6 and TNF-α production by the murine peritoneal macrophages, the effect of introducing DV NS1 protein into the mice was investigated. Wild-type and TLR6-/- mice were injected with 20 μg of DV NS1 protein via intraperitoneal injection. The control mice were injected with 20 μg of His-tag protein. The survivability of the DV NS1 protein-treated and His-tag-treated mice were monitored for 7 days post-treatment (Fig 7C). At the endpoint, 94.4% of the His-tag-treated wild-type mice survived the treatment while only 27.8% of the DV NS1 protein-treated wild-type mice survived. At the endpoint, 100% of the His-tag-treated TLR6-/- mice survived the treatment while 88.9% of the DV NS1 protein-treated TLR6-/- mice survived. The survival rate of the DV NS1 protein-treated wild-type and TLR6-/- mice were significantly different. The knockout of TLR6 increased the survivability of mice after treatment with DV NS1 protein. IL-6 (Fig 7D) and TNF-α (Fig 7E) of the treated wild-type mice were assayed. Similar to DV-infected mice, IL-6 and TNF-α in the DV NS1 protein-treated wild-type mice were found to be significantly higher than that of the His-tag-treated wild-type mice.
Among the PBMC, monocytes have been implicated in both the protection and immunopathogenesis of dengue [54]. Depletion of monocytes and macrophages in mice using clodronate-loaded liposomes resulted in 10-fold higher systemic DV titers, highlighting the important roles of monocytes and macrophages in DV control during an infection [55]. Ironically, monocytes were found to be the cells among PBMC that supported DV infection and the cells responsible for antibody-dependent enhancement of DV infection [56,57]. Although other cell types in the PBMC like T cells and B cells were found to be less susceptible to DV infection, they are likely to play important roles during DV infection [56]. There are evidences that cell-cell cross-talks between various immune cells in PBMC affect cytokine production during an infection [58,59]. Hence, PBMC culture would be a more representative and physiological model of infection than using monocytes alone.
In the recent years, it has become evident that PRRs play a major role in infectious and even in non-infectious diseases [60,61]. One family of PRRs, the TLR family has emerged as a key component of the innate immune system and it can activate signals which are crucial for the initiation of adaptive immune responses [61]. Studies in recent years have shown the presence of mRNA and protein expression of TLRs in various immune and non-immune cells [62–64]. In our study, TLR2 and TLR6 were found to be up-regulated in PBMC upon DV infection. This up-regulation suggested the possible involvement of TLR2 and TLR6 in dengue virus infection.
TLR6 was found to be up-regulated by PBMC on day 3 post-infection (Fig 1C). As TLR2 is partner of TLR6, its expression by PBMC during dengue virus infection was also investigated. Like what was observed for TLR6, TLR2 was found to be up-regulated by PBMC on day 3 post-infection (Fig 1D). The mechanism of TLR2 regulation has not been fully elucidated [65–68]. It was reported that chromatin remodelling involving DNase I and restriction enzyme occurs at TLR2 promoter region following infection [67]. This remodelling of chromatin increases accessibility of transcription factors resulting in greater transcription of TLR2 [67]. In addition, two pathways were found to be important for TLR2 regulation. IKKβ-IκBα-dependent NFκB pathway activation and MKK3/6-p38α/β pathway inhibition are essential for TLR2 expression [65]. One study supported the involvement of NFκB in TLR2 expression. Pyrrolidine dithiocarbamate (PDTC), a pharmacologic inhibitor of NFκB was shown to prevent the up-regulation of TLR2 by TLR2 and TLR4 agonist [66]. On the other hand, the up-regulation of TLR6 is not well-studied and remained unclear.
Upon dengue virus infection, PBMC secretes both IL-6 and TNF-α (Fig 2A and 2B). This is consistent with what was observed in the sera of dengue patients. Dengue patients’ sera have been found to contain high level of IL-6 [20,69,70]. Similarly, TNF-α level was also increased in the sera of dengue patients [20,69,70]. Similar to monocytes, IL-6 and TNF-α are implicated in both the protection and immunopathogenesis of dengue virus infection [71,72]. Upon sensing the presence of foreign microbes through recognition of PAMPs, biological mediators like IL-6 and TNF-α are released. Although these mediators initiate and regulate the inflammatory response and adaptive immune response to eliminate foreign microbes, they have also been found to be involved in lethal manifestations like septic shock syndrome, vascular leakage and cachexia, resulting from disease, infection or injury [73–75]. This provided evidence that the manifestation of illness could also be caused by the host own immune system, not necessarily by exogenous pathogens.
It was noticed that IL-6 production by the DV2-infected PBMC (Fig 2A) was lower than that of the antibody-treated DV2-infected PBMC (Fig 2C). This difference could be due to the presence of antibodies which are originated from rabbit and mouse. The presence of foreign proteins can trigger immune response. Another possibility contributing to the difference in IL-6 production observed could be donor variability. The PBMC used for the two set of experiments were from different donors.
The dengue viral protein responsible for the activation of TLR6 and TLR2 were first screened using ELISA. Among the 10 dengue viral proteins, only DV NS1 protein up-regulated both IL-6 and TNF-α expression of PBMC (Fig 3A and 3B), making it the most likely candidate. DV NS1 protein was documented to be secreted by infected cells and the presence of DV NS1 protein was detected in the sera of patients [76–78]. Moreover, the amount of DV NS1 protein in the sera of patients was found to correlate with the severity of the dengue disease [76]. This correlation suggested that DV NS1 protein plays an important role in the pathogenesis of dengue disease. In addition, DV replication was found to be critical for the up-regulation of IL-6 and TNF-α during DV infection as UV-inactivated DV did not induce the up-regulation. DV NS1 protein being a non-structural protein requires DV replication to be synthesized. Hence, DV NS1 protein being the dengue viral protein fits the results we obtained using the PBMC in vitro cell model.
In this study, we have developed a murine model for dengue virus infection using 1–2 day old C57BL/6 mouse. Although consistent viremia was detected in the mice infected at 1–2 day old for both wild-type and TLR6-/- mice, the virus titers were observed to decrease as the day of infection progressed (Fig 5B). This suggested that the DV2-infected mice were able to mount an effective immune response to fight the infection. The fast clearance of DV may suggest that the innate immunity is sufficient for the clearing. Published studies have demonstrated that innate immunity was sufficient to clear DV infection using human cell-engrafted scid mice [79,80]. Similar trend was observed for the virus titers in the livers of the DV2-infected mice for both wild-type and TLR6-/- mice. However, the virus titers of the livers were not as consistent compared to that of the sera, only a few of the livers of DV2-infected mice were detected to contain infectious DV2 (Fig 5C). A point to be taken into consideration is that the DV detected in the liver could be contributed partially by the virus found in the blood. However, the blood contamination has been minimized as the blood of the pups was harvested before the harvest of the liver. Hence, the liver should contain minimal amount of blood when harvested. The observation that virus was detected in all the serum samples but not in all the liver samples of infected pups on day 1 and day 2 post-infection showed that the above mentioned contamination was kept to a minimum.
One clinical sign observed in dengue patients and the murine model is paralysis of extremities [52]. The occurrence of paralysis was observed to be between day 10 to day 14 of infection (Table 2). Paralegia was not unique to the murine model used in our studies. Paralegia was also observed in other murine models of DV infection and the time of development was similar [81,82]. AG129 mice were reported to develop paralysis within 7 to 14 days post-infection [82]. The development of paralysis was faster than the wild-type counterpart of the mice and thus the authors attributed the difference to the AG129’s deficiency in IFN receptors.
Among the asymptomatic mice, virus was detected in the brain as early as day 1 post-infection for the DV2-infected TLR6-/- mice and the virus persisted in the brains of the mice till the endpoint of the experiment. This suggested that virus clearance is least efficient in the brains of the mice. Similar trend was observed for AG129 murine model [82]. AG129 mice harboured DV in the extraneural tissues and neural tissues on day 3 post-infection, with higher viral titers in the extraneutral sites than the neural sites. By day 7 post-infection, virus was only detected in the neural tissues. This result supported our data (Fig 5B–5E). It was noticed that the viral loads in the brain of the asymptomatic mice (< 104 PFU/g) were much lower than that of the mice displaying hindlimb paralysis (2.8 x 104 PFU/g, 5.2 x 105 PFU/g, 8.2 x 105 PFU/g, 7.1 x 105 PFU/g) (Table 2). Hence, extensive replication of DV in the brain of the mice could have resulted in the paralysis observed in the mice. Similarly, AG129 mice with paralysis were reported to carry high viral loads in the brain [82]. The viral loads reported were comparable to ours, between 104 to 106 PFU/g (Table 2).
Mice which were not sacrificed but were monitored for disease progression, recovered two days after the onset of paralysis. The brain, liver and limb of one of the mice were harvested and quantified for virus. No virus was detected in the liver and limbs of the recovered mouse while virus was still detected in the brain. The viral load (2.12 x 104 PFU/g) was still higher than that of the asymptomatic mice but lower than that of the symptomatic mice (Table 2 and Fig 5E). This suggested that viral load in the brains of the mice can be controlled by the mouse immune system even though the clearing of the virus in the brain was not as efficient as compared to the sera, livers and limbs. The microglial cells are the main cell type of the innate immune system in the brain [83]. The microgial cells also express TLRs and produce pro-inflammatory mediators in response to TLR ligands [84,85]. Human microgial cells express high levels of TLR2 and TLR3, moderate levels of TLR4, TLR5, TLR6, TLR7 and TLR8 but low level of TLR9 [86]. Mouse microgial cells express similar TLRs except for TLR5 [87]. As there is no lymphatic system in the brain for immune cells to migrate through and microglial cells are poor antigen-presenting cells, the immune responses in the brain are limited [83]. This may explain why the virus can persist in the brain for a longer time in comparison to other organs and sera.
As human IL-6 and TNF-α were detected in our human cell-based model upon DV infection, murine IL-6 and TNF-α were assayed for in the sera of the DV2-infected mice. Unlike what was observed for the cell-based model, IL-6 and TNF-α were only detected to be up-regulated in the sera of some of the DV2-infected mice (Fig 6A and 6B). For both wild-type and TLR6-/- mice, the level of IL-6 and TNF-α detected in some of the DV2-infected mice were comparable to that of the mock-infected mice. This suggested only some of the DV2-infected mice responded to the DV-infection by up-regulation of IL-6 and TNF-α. This high variability of hyporesponsiveness of young mice to stimulation was also documented by other research groups [88–90]. Among the responsive mice, the IL-6 and TNF-α of the DV2-infected wild-type mice were higher than the DV2-infected TLR6-/- mice and the up-regulation lasted for a longer time. Knockout of TLR6 reduced IL-6 and TNF-α production, suggesting TLR6 activation contributed to IL-6 and TNF-α production in mice during DV infection.
As DV NS1 protein was found to be the viral protein that is activating TLR6, the duration in which DV NS1 protein remained in circulation in the mice injected with DV was investigated. The presence of DV NS1 protein was detected in all the sera of DV2-infected wild-type and TLR6-/- mice using Bio-rad platelia kit DV NS1 antigen detection kit from day 1 to day 5 (Fig 6C and 6D). The level of DV NS1 protein detected from the DV2-infected wild-type mice was not significantly different from that of the DV2-infected TLR6-/- mice. This was probably a consequence of similar virus titers in the sera of the DV2-infected wild-type and TLR6-/- mice. DV NS1 protein remained in circulation in the mice longer than DV (Figs 5B, 6C and 6D). DV NS1 protein remained detectable in the sera of DV2-infected mice on day 5 post-infection while DV were no longer detected in most of the sera by day 4 post-infection. Slower rate of DV NS1 protein clearance compared with DV from the plasma of dengue patients were also reported [76]. DV NS1 protein can be detected for a longer period of time in the sera of dengue patients compared to DV [91].
The presence of DV NS1 protein level in the mice contributes to IL-6 and TNF-α level in the mice. DV NS1 protein level remained relatively high from day 1 to day 5 post-infection. This could be the reason why wild-type mice still observed high IL-6 and TNF-α expression even when virus titer dropped to 0 PFU/ml in the sera for most of the mice while IL-6 and TNF-α of TLR6-/- mice dropped after the elimination of DV from the sera (Fig 6A and 6B). This may suggest that IL-6 and TNF-α in the sera at the later part of infection was mostly contributed by DV NS1 protein activating TLR6.
The Kaplan-meier survival plot was used to estimate the survival of the wild-type and TLR6-/- mice population after DV infection for over 21 days. The survival plots of the DV2-infected wild-type and TLR6-/- mice intercept, indicating that the probability of survival for one population of the mice were higher for a period of time during DV infection but became lower compared to the other population as the infection progresses (Fig 5A). The DV2-infected TLR6-/- mice have a lower survival probability at earlier time points and the DV2-infected wild-type mice have a lower survival probability at later time points. This could be due to the replication of DV in the brain of the TLR6-/- mice. Viral loads were detected in the brain of the TLR6-/- mice on day 1 post-infection but not for wild-type mice. This suggested that the brains of pups were more vulnerable to DV infection in the absence of TLR6. Sensing of DV through other PRRs may be more limited for the microgial cells during the early development of the mice and thus TLR6 appeared to play a more critical role. This vulnerability decreased with age as TLR6-/- mice suffering from paraplegia were found to be able to recover from it. The overall survival probability of wild-type mice during DV2 infection was lower than TLR6-/- mice. In the absence of TLR6 activation, the overall survival probability of the mice to DV infection increased. This suggested the involvement of TLR6 in the immunopathogenesis of DV infection. Activation of TLR2 and TLR6 by DV NS1 protein up-regulates IL-6 and TNF-α. High expression of IL-6 and TNF-α have been shown to be associated with fatality of mice [92,93]. Prolonged up-regulation of IL-6 and TNF-α due to stimulation of TLR6 by DV NS1 protein may be the cause of death for the wild-type mice. Prolonged up-regulation of IL-6 and TNF-α may increase the risk of the mice developing complications from the proinflammatory cytokines.
Murine peritoneal macrophages from 4-week old wild-type and TLR6-/- C57BL/6 mice were used to further verify the involvement of TLR6 in IL-6 and TNF-α expression during DV infection. Murine peritoneal macrophages were widely used to elucidate TLR ligands and TLR6 ligands were among those tested [32,33,94,95]. MALP-2 was also used to further confirm the functionality of TLR6 of the wild-type and TLR6-/- murine peritoneal macrophages. Wild-type murine peritoneal macrophages up-regulated both IL-6 and TNF-α upon stimulation by MALP-2 while TLR6-/- murine peritoneal macrophages were non-responsive. During DV infection, TLR6-/- murine peritoneal macrophages produced significantly less IL-6 and TNF-α compared to that of the wild-type murine peritoneal macrophages. This corroborates the result obtained from the sera of the mice. Knockout of TLR6 did not completely abrogate IL-6 and TNF-α up-regulation during DV infection, suggesting TLR6 activation only partially contributed to the IL-6 and TNF-α detected during DV infection. The redundancy of pathogen sensing pathways was documented [96]. One pathogen can be recognized by multiple PRRs and the signalling pathways activated downstream of TLRs have redundancy [97]. The synergistic effect of activating more than one PRR has also been reported. Synergy between TLR2 and TLR4 can potentiate the up-regulation of cytokine production [98]. This observation may also provide some explanation on why wild-type and TLR6-/- mice did not have significant difference in virus detected in the sera of the mice. Knockout of TLR6 did not prevent the activation of macrophages. The macrophages can still sense the presence of pathogen through other PRRs and gets activated to produce IL-6 and TNF-α. Similar to our human PBMC model, DV NS1 protein stimulated the production of IL-6 and TNF-α by wild-type murine peritoneal macrophages. TLR6-/- murine peritoneal macrophages were unresponsive to DV NS1 protein stimulation, suggesting DV NS1 protein activates TLR6 of murine peritoneal macrophages to produce IL-6 and TNF-α. Using TLR6-/- murine cellular model, TLR2 and 6 antibody blocking assay and SEAP reporter assay, DV NS1 protein has been shown to be the viral protein responsible for TLR2/6 stimulation during DV infection and both receptors are required. However, whether the stimulation is direct or indirect has not been elucidated. Studies have demonstrated that host-derived molecules may also stimulate TLR signalling [99]. Hence, there is a possibility that DV NS1 protein may stimulate the release of endogenous ligands to trigger TLR2 and TLR6 activation rather than binding to TLR2/6 complex itself
It was found that mice injected with DV NS1 protein alone without DV can induce up-regulation of IL-6 and TNF-α (Fig 7D and 7E). Hence, the result suggested that DV NS1 protein contributed to the up-regulation of IL-6 and TNF-α production observed in DV2-infected wild-type mice (Fig 6A and 6B). In addition, results from the murine peritoneal macrophages suggested that DV NS1 protein stimulates IL-6 and TNF-α production primarily through TLR6. Together, these results suggest that the higher survivability of the TLR6-/- mice during DV infection could be due to their non-responsiveness to DV NS1 protein. The survival rate of the DV NS1 protein-treated wild-type mice (27.8%) was lower than that of the DV-infected wild-type mice (61.4%). This could be due to the amount of DV NS1 protein injected was more than what was produced in the mice by the DV infection.
It was shown in our studies that DV NS1 protein is able to activate TLR2 and TLR6 to induce up-regulation of IL-6 and TNF-α. This production of cytokines could be the cause of the development of dengue hemorrhagic fever as cytokines were found to play important roles in several viral hemorrhagic fevers [100,101]. Furthermore, cytokines were found to have prognostic value in DV infection in other studies [20,102,103]. All these findings suggest that a possible treatment for dengue would be to control the proinflammatory cytokine production during DV infection. It was reported that when an immunomodulator, tetracycline hydrochloride was administered into Tick-Borne Encephalitis virus patients, the concentration of IL-6 and TNF-α were reduced and the patients have a faster clinical recovery [101]. This study suggested that modulation of the amount of IL-6 and TNF-α can have a positive effect on patients with viral hemorrhagic fevers. Since TLR6 activation during DV infection can contribute to the production of proinflammatory cytokines, immunomodulation approaches that target TLR6 can reduce the proinflammatory cytokines. The reduction of proinflammatory cytokines can potentially prevent the progression of the disease to the more severe forms.
Recent studies have shown that TLRs may be responsible for the manifestation of autoimmune diseases, allergy, cancer, infectious diseases and sepsis [104,105]. In our study, the activation of TLR6 decreases the survival of mice during DV infection, suggesting a role for TLR6 in the immunopathogenesis of DV infection. The roles of TLRs in human diseases are not fully understood but there are in vitro and animal model data to support TLR roles in disease initiation and progression [97,106]. There is a growing interest in exploring TLRs as the therapeutic targets for these diseases [97,104–107]. It was proposed that inhibition of TLR function might limit disease pathogenesis in conditions such as sepsis, rheumatoid arthritis and systemic lupus erythematosus, in which the immune system is inappropriately overactive [97,104,106]. Antimalarial drugs such as hydroxychloroquine which act as a TLR7, TLR8 and TLR9 antagonist are used for the treatments of rheumatoid arthritis and systemic lupus erythematosus [106,108]. TLR2 has been implicated in the pathogenesis of systemic lupus erythematosus, diabetes, Alzheimer’s disease [109,110]. A TLR2-specific monoclonal antibody, OPN-305 which inhibits TLR2-mediated proinflammatory cytokine production is being tested for the potential treatment of inflammatory diseases [106]. Drugs or antibodies that target TLR2 are likely to have an effect on TLR2 and TLR6 signaling as shown by PBMC model, in which inhibition of IL-6 and TNF-α was achieved by the blocking of either TLR2 or TLR6. The host may not be vulnerable to pathogens in the duration of TLR6-targeted therapy, due to the redundancy of PRR pathways. TLR6-targeted therapies have a potential for intervention in dengue virus infection and amelioration of disease symptoms. Besides using small-molecule agonists or antagonists for targeting TLRs, the use of microRNAs in the regulation of TLRs may be available in the near future [111,112].
In our study, we have found that TLR2 and TLR6 were involved in the detection of the presence of DV during DV infection. However, mice without TLR6 were still able to secrete IL-6 and TNF-α during DV infection. Therefore, other PRRs are also likely to be involved. It would provide a better understanding of the DV infection if the identities of those PRRs were elucidated. Some of the proposed PRRs are TLR3, TLR7 and TLR8 [28].
In conclusion, DV NS1 protein is found to be responsible for triggering TLR2 and TLR6 during DV infection in our study. This stimulation partially contributes to IL-6 and TNF-α expression during DV infection. Activation of TLR6 may play a role in the immunopathogenesis of DV infection in the mice as survivability of the mice increased in the absence of TLR6. Lastly, our results provide an insight into the possibility of using TLR6 antagonist in therapeutic treatment for DV infection.
Human peripheral blood mononuclear cells (PBMC) were isolated with informed consent from healthy blood donors as whole blood donation, from the Division of Haematology, Department of Laboratory Medicine, National University Hospital, Singapore and was approved by National University of Singapore Institutional Review Board (NUS-IRB: 10-072E). Animal research was approved by NUS IACUC (protocol no: 090/10, R15-0033, BR023/10, BR14-1255). The mice were anesthesized using isoflurane. Euthanasia was performed using carbon dioxide asphyxiation, followed by cervical dislocation.
The Baby Hamster Kidney (BHK) cells (ATCC), Human Embryonic Kidney (HEK) 293 cells and Aedes albopictus C6/36 cells were grown in RPMI-1640 (Sigma Aldrich) supplemented with 10% fetal calf serum [(FCS), PAA], DMEM (Sigma Aldrich) supplemented with 10% FCS and L-15 media supplemented with 10% FCS respectively. Human peripheral blood mononuclear cells (PBMC) were isolated with informed consent from healthy blood donors as whole blood donation, from the Division of Haematology, Department of Laboratory Medicine, National University Hospital, Singapore and was approved by National University of Singapore Institutional Review Board (NUS-IRB: 10-072E). PBMC were isolated from the donors’ buffy coats by centrifugation on a density gradient (400x g/30 mins in Ficoll-Paque Plus, GE Health Science) according to manufacturer’s procedures. Isolated PBMC were grown in RPMI-1640 supplemented with 10% FCS and 1% penicillin-streptomycin of concentration: 10, 000 units penicillin and 10 mg streptomycin/ml. LPS-treated PBMC were added lipopolysaccharide (Sigma Aldrich, L-2630) into PBMC culture medium. Ultrapure LPS-treated PBMC were added ultrapure lipopolysaccharide (InvivoGen, LPS-EB Ultrapure). His-tag-treated PBMC were added 1 μg/ml of His-tag (Abcam ab14943) into PBMC culture medium. DV NS1 protein-treated PBMC were added 1 μg/ml of DV NS1 protein (Abcam ab64456) into PBMC culture medium.
Dengue virus serotype 2 (DV2), strain (Den2STp7c6), a low passage isolate from a dengue-infected patient in Singapore and DV2 strain 16681, a kind gift from Professor Gubler from DUKE NUS were used in this study. The virus was propagated in C6/36 cells. PBMC were transferred separately into 50 ml falcon tubes and centrifuged at 300x g for 5 mins to remove the culture medium. PBMC were infected with DV2 at a multiplicity of infection (MOI) of 10 and incubated at 37°C for 1.5 hour with intermittent shaking. The cells were washed with PBS once to remove residual virus before RPMI-1640 medium with 10% FCS was added to the cells. PBMC were seeded into each well of 24-well plates (NUNC). HEK 293 cells were seeded in each well of 24-well plates 1 day before infection. Prior to infection, the culture medium was aspirated from the wells and the HEK 293 cells were infected with DV2 at an MOI of 10 with incubation at 37°C for 1.5 hour with intermittent shaking. The cells were subsequently washed with PBS before culture medium was added to the cells. Supernatant from uninfected C6/36 culture was denoted as the mock-infected controls. RPMI-1640 supplemented with 10% FCS and 1% penicillin-streptomycin was the culture medium used for PBMC while DMEM supplemented with 2% FCS was used for HEK 293 cells. UV-inactivated virus was obtained by irradiation of the virus under the ultraviolet lamp for 1.5 hours. The UV-inactivated virus in the medium was then purified in 100 kDa nominal molecular weight limit centricons (Millipore, UFC910096) and centrifuged at 4000x g for 25 mins in a swing-out centrifuge. PBS was then added into the centricons to wash the virus and centrifuged again at 4000x g for 25 mins. The virus was then collected and reconstituted with L-15 medium. The inactivation of the virus by UV-irradiation was confirmed by performing virus plaque assay.
Plaque assay was carried out to quantify the number of infectious virus particles using BHK cells. Briefly, BHK cells were cultured to approximately 80% confluency in 24-well plates. The virus stock was 10-fold serially diluted from 10−1 to 10−6 dilution in RPMI 1640. BHK monolayers were infected with 100 μl of each virus dilution. After incubation in 5% CO2 atmosphere at 37°C for 1 hour with rocking at 15 mins intervals, the medium was aspirated and 1 ml of 1% (w/v) carboxymethyl cellulose in RPMI supplemented with 2% FCS was added to each well. After 6 days of incubation at 37°C in 5% CO2 incubator, the cells were fixed and stained for 1 hour with 200 μl of 1% crystal violet in staining solution. After thorough rinsing with water, the plates were dried and the virus plaques were scored visually.
Cell pellets were lysed using CelLytic M cell lysis reagent (Sigma Aldrich) with EDTA-free protease inhibitor cocktail (Roche) for 10 mins on ice. The total cellular protein in samples was quantified by Bradford Assay (Bio-Rad). 15 μg of protein was loaded in each lane and separated by SDS-PAGE before being transferred onto a nitrocellulose membrane via the semi-dry transfer system (Bio-Rad). Western blot was performed to detect human TLR6, using rabbit IgG anti-TLR6 (sc-30001, Santa Cruz Biotechnology) (1:200 dilution) and TLR2, using rabbit IgG anti-TLR2 (ab86754, Abcam) (1:200 dilution). Blots were incubated with HRP-conjugated goat anti-rabbit IgG (H+L) secondary antibody (Pierce) (1:2500 dilution). Analyses were performed using enhanced chemiluminescence detection system with Pierce ECL Western Blotting Substrate. The density of the bands was quantified using GelQuant.NET software provided by biochemlabsolutions.com.
PBMC infected with DV2 at an MOI of 10 or mock-infected were transferred from 24-well plates into 15ml falcon tubes. The cells in the falcon tubes were centrifuged at 300x g for 5 mins. The supernatant were then discarded and 5ml of PBS were added for the washing of the cells. The cells were spun down once more to remove the PBS. PBS containing 5% BSA was used to resuspend the cells before the cells were incubated on ice for 20 mins. Primary antibody [anti-TLR6 (Santa Cruz SC-30001), anti-TLR2 (Santa Cruz SC-21759) or anti-CD14 (Millipore CBL453F)] was then added at a dilution of 1: 200 to the cell suspension and incubated on ice for 30 mins. The cells were then spun down and the supernatant removed. 5 ml of PBS was used to wash the cells before the cells were spun down again to remove the PBS. Following that DyLight 633/FITC-conjugated goat anti-rabbit or anti-mouse IgG (H+L) secondary antibody (Pierce), was added at 1: 200 and incubated on ice for 30 mins. The cells were then washed twice with 5ml of PBS before fixing using 4% paraformaldehyde at room temperature for 10 mins. After which the cells were washed and resuspended in 1 ml of PBS. For the staining of two different antigens in the same sample, the above procedure of staining was repeated once more using primary antibodies derived from a different species. The cells were analyzed using Beckman Coulter CyAn ADP Analyzer. Samples were gated to exclude cell debris.
PBMC were stained with anti-TLR6 antibody (Santa Cruz, SC-30001) or anti-TLR2 (Santa Cruz SC-21759) and anti-CD14 (Millipore CBL453F) at a dilution of 1:200 for 30 mins, followed by FITC-conjugated goat anti-rabbit IgG (H+L) secondary antibody for 30 mins and fixed using 4% paraformaldehyde. The cells were then incubated with 4’-6-Diamidino-2-phenylindole (DAPI) at a concentration of 300nM for 5 mins at room temperature. The cells were spun down at 300x g for 5 mins and washed twice in 5 ml of PBS. The cells were resuspended in 10 μl of PBS. 10 μl of the cell suspension was placed onto a glass slide, mounted on coverslip and viewed under the microscope (IX81 Olympus, Japan) at 1000x magnification.
Quantification of cytokines (IL-6 and TNF-α) was carried out using sandwich enzyme-linked immunosorbent assay (ELISA) which was performed in 96-well plate. ELISA for human and murine IL-6 and TNF-α were performed using commercial kits (BD biosciences, Pharmingen) and according to manufacturer’s protocol. Briefly, 100 μl of anti-IL-6 or anti-TNF-α antibody diluted 1:250 with coating buffer were added into each well to coat the antibody onto the plate through an overnight incubation at 4°C. The plates were then washed three times using wash buffer (PBS with 0.05% Tween-20) before blocking using 200 μl of assay diluent per well. After adding the standards and the samples, the plates were washed three times using wash buffer and incubated with 100 μl of anti-IL-6 or anti-TNF-α biotinylated antibody and streptavidin-conjugated horseradish peroxidase diluted 1:250 with assay diluent for an hour. The plate was then washed seven times, followed by adding tetramethyl benzidine substrate solution to each well. Absorbance was measured using ELISA reader (Tecan) at wavelength of 450 nm with reference wavelength of 570 nm. The concentrations of the cytokine in experimental samples were determined from a standard curve with known concentrations of the cytokine. Samples were performed in triplicates.
PBMC were incubated with TLR2 or TLR6 blocking antibodies (IgG1) (InvivoGen, San Diego, USA) at a concentration of 1000 ng/ml for 30 mins on ice. Unbound antibodies were washed off with PBS before infection or mock-infection was performed. For PBMC to be blocked by both TLR2 and TLR6 blocking antibodies, 500 ng/ml of each antibody were used instead. Normal mouse IgG1 (Santa Cruz Biotechnologies, Santa Cruz, USA) from unstimulated mice was used as isotype control at a concentration of 1000 ng/ml.
Activation of NFκB was determined using a reporter plasmid (pNF-κB/SEAP, IMGENEX) which expresses secreted alkaline phosphatase (SEAP) protein under the control of the NFκB promoter. These plasmids were transfected into HEK 293 cells using Invitrogen Lipofectamine LTX according to manufacturer’s protocol. SEAP catalyzes the hydrolysis of p-Nitrophenyl phosphate producing a yellow product that can be read using ELISA reader at 405 nm. Stable cell clones of the transfected cells were obtained by selection using G418 (PAA) at 500μg/ml. In brief, 2 x 105 transfected HEK 293 cells were seeded into 24-well plate 1 day prior to treatment. The cells were then treated under various conditions [infected with DV2 at an MOI of 10, mock-infected, DV NS1 recombinant protein (1 μg/ml), LPS (5 μg/ml) or MALP-2 (50 ng/ml) (Imgenex, IMG-2206) added into the culture medium]. Supernatant were harvested 1 to 3 day post-treatment. The amount of SEAP in the supernatant was assayed according to manufacturer’s protocol and read using an ELISA reader (Tecan) at wavelength of 405 nm.
TLR2 and TLR6 in HEK 293 were expressed using a plasmid co-expressing the human TLR2 and TLR6 genes (pDUO-hTLR6/TLR2, InvivoGen) which was transfected into the NFκB-SEAP HEK 293 cell clones obtained as mentioned in the previous paragraph. Stable cell clones of the transfected cells were obtained by selection using both blasticidin (Invitrogen) at 10 μg/ml and G418 (PAA) at 500 μg/ml. The expression was confirmed with immunoblotting after the transfected cells were stained for TLR2 or TLR6 using the antibodies mentioned above.
Dengue viral recombinant proteins (Capsid, PrM, envelope, NS1, NS2A, NS2B, NS3, NS4A, NS4B, NS5) with 6x his-tag and protein tag (6x his-tag) at concentration of 1 mg/ml were expressed using BaculoDirect Baculovirus Expression System (Invitrogen) according to manufacturer’s protocols.
Sera of mock-infected and DV2-infected mice were harvested for DV NS1 antigen detection assay using the Platelia Dengue NS1 Antigen detection kit (Bio-Rad, #72830) for day 1 to day 5 and day 21 post-infection. The relative amount of DV NS1 protein was measured in optical density and was read using Tecan plate reader at wavelength of 450 nm with reference wavelength of 620 nm.
C57BL/6 mice used in this study were obtained from InVivos and the former NUS CARE (NUS, Singapore). TLR6 knock-out C57BL/6 breeder mice were obtained from Oriental BioService, Kyoto, Japan and bred in NUS, Singapore under NUS IACUC approved breeding protocols, BR023/10 and BR14-1255. The use of mice for this study was approved by NUS IACUC under protocols, 090/10 and R15-0033.
One to two days old C57BL/6 mice were infected with 5.4 x 108 PFU/ml of 16681 DV2 via intraperitoneal injection (IP) at a volume of 0.05 ml/g. For mock-infection, C6/36 culture supernatant of the same volume as the virus was injected instead. For DV NS1 protein treatment, one to two days old C57BL/6 mice were injected with 20 μl of DV NS1 recombinant protein of concentration 1 mg/ml. For His-tag treatment, 20 μl of His-tag of concentration 1 mg/ml was injected instead.
Mice were euthanized before blood was collected by cardiac puncture. The blood was left to clot at room temperature and centrifuged at 3300x g for 5 mins to obtain the serum. It was observed that some mice had a bulge at the site of injection on day 1 post-infection. Peritoneal fluid was extracted from the bulge using 27 G needle and syringe for virus quantification as well. Brains of mice were harvested by removing the skin on top of the head and making an incision at the centre of the scalp using scissors. Livers of mice were harvested by making an incision at the abdomen. Hindlimbs of mice were harvested by cutting the hind limbs of the mice at the pelvis joint. Brains, livers or hind limbs of mice were placed in hard tissue homogenizing tube containing ceramic beads (Precellys, Bertin, Germany). The weight of the tissues in each of the tubes was recorded and 0.5 ml of PBS was added to each tubes. The tissues in the tubes were homogenized using a tissue homogenizer (Precellys, Bertin, Germany). The conditions used were 6500 rpm for 10 secs with 3 repetitions and 5 secs rest in between. The tubes were then centrifuged at 3500x g for 10 mins. The supernatant was collected in a new eppendorf tube and centrifuged at 10,000x g for 10 mins. DV2 in the serum or peritoneal fluid or supernatant of homogenized tissues were determined using plaque assays. 1% penicillin-streptomycin, 1% amphotericin B (MP Biomedicals, Southern California, USA) and 0.5% gentamycin (PAA, GE Healthcare, Piscataway, USA) were added to the overlay medium. As the volume of the serum or peritoneal fluid harvested from each of the pups may be less than 100 μl, there may not be neat sample and the calculation of PFU/g was adjusted according to the volume of sample used. IL-6 and TNF-α in the supernatant of homogenized tissues were determined using ELISA.
4% thioglycollate medium was prepared and autoclaved. 4-week old mice were injected with 1 ml of 4% thioglycollate medium via intraperitoneal injection. Four days after injection, the mice were euthanized and the skin around the abdomen of the mice was removed to expose the intraperitoneal cavity. Ice cold PBS was injected into the intraperitoneal cavity without bursting the peritoneal membrane. Precaution was taken to avoid puncturing any organ or intestine. The abdomen of the mice was gently massaged before withdrawing the PBS containing macrophages from the intraperitoneal cavity. The murine peritoneal macrophages in PBS were collected and centrifuge at 450x g for 5 mins at 4°C. One ml of Red blood cell lysing buffer Hybri-Max (Sigma-Aldrich) was added to the cell pellet and resuspended for 3 mins. Fourteen ml of PBS was added and the tube was centrifuged at 450x g for 5 mins at 4°C. The cells were washed again with PBS before culturing in RPMI-1640 supplemented with 10% heat-inactivated FCS, 1% penicillin/streptomycin, 1% amphotericin B and 0.5% gentamycin.
DV2-Infected or mock-infected wild type or TLR6-/- C57BL/6 mice were monitored daily and observed for any abnormal signs which could be symptoms of infection for up to day 21 post-infection. DV NS1 protein-treated or His-tag-treated wild type or TLR6-/- C57BL/6 mice were monitored daily and observed for any abnormal signs which could be symptoms of infection for up to day 7 post-infection.
The statistical comparisons were carried out using two tailed Student’s t-test for repeated measurements when applicable. The significance level was set at *: p < 0.05, **: p < 0.005, ***: p < 0.0001. Data shown are obtained from three independent experiments unless stated otherwise. Kruskal-Wallis test was used for non-parametric data set. Log-rank test was used to compare the survival curves of mice.
|
10.1371/journal.pntd.0000991 | Water Use Practices Limit the Effectiveness of a Temephos-Based Aedes aegypti Larval Control Program in Northern Argentina | A five-year citywide control program based on regular application of temephos significantly reduced Aedes aegypti larval indices but failed to maintain them below target levels in Clorinda, northern Argentina. Incomplete surveillance coverage and reduced residuality of temephos were held as the main putative causes limiting effectiveness of control actions.
The duration of temephos residual effects in household-owned water-holding tanks (the most productive container type and main target for control) was estimated prospectively in two trials. Temephos was applied using spoons or inside perforated small zip-lock bags. Water samples from the study tanks (including positive and negative controls) were collected weekly and subjected to larval mortality bioassays. Water turnover was estimated quantitatively by adding sodium chloride to the study tanks and measuring its dilution 48 hs later.
The median duration of residual effects of temephos applied using spoons (2.4 weeks) was significantly lower than with zip-lock bags (3.4 weeks), and widely heterogeneous between tanks. Generalized estimating equations models showed that bioassay larval mortality was strongly affected by water type and type of temephos application depending on water type. Water type and water turnover were highly significantly associated. Tanks filled with piped water had high turnover rates and short-lasting residual effects, whereas tanks filled with rain water showed the opposite pattern. On average, larval infestations reappeared nine weeks post-treatment and seven weeks after estimated loss of residuality.
Temephos residuality in the field was much shorter and more variable than expected. The main factor limiting temephos residuality was fast water turnover, caused by householders' practice of refilling tanks overnight to counteract the intermittence of the local water supply. Limited field residuality of temephos accounts in part for the inability of the larval control program to further reduce infestation levels with a treatment cycle period of 3 or 4 months.
| Dengue is currently the most important viral disease of humans transmitted by arthropods worldwide. Aedes aegypti, a human-biting mosquito dwelling in artificial domestic containers, is the main vector of dengue. Ae. aegypti larval control programs are frequently based on the application of the insecticide temephos. A five-year larval control program in northeastern Argentina significantly reduced infestations but could not maintain them below target levels, especially during summer. Identifying the underlying processes responsible for such shortcomings is important for improving dengue prevention strategies. Large water-holding containers were the most productive container type and the main targets for control. We found that the duration of temephos residual effects in household-owned large tanks was much shorter than expected and allowed early reinfestation post-treatment. The main factor limiting temephos residuality was fast water turnover, caused by householders' practice of refilling tanks overnight to counteract the intermittence of the local water supply.
| Dengue is currently the most important arboviral disease in the world; it affects an estimated 50 million people and causes 30,000 deaths per year [1], [2]. In Argentina, the regional context of increasing dengue incidence led in 2009 to the most severe and extended DF epidemic ever recorded [3]. Aedes aegypti (Diptera: Culicidae) (L.), the main vector of dengue, urban yellow fever and chikungunya virus, is a highly domestic and anthropophilic mosquito found inside or around human dwellings in urban settings [4]. In the absence of a vaccine, efforts to reduce dengue transmission frequently rely on vector control actions targeting immature stages through chemical or biological treatment of artificial water-holding containers. An international panel recently concluded that strategies for vector control and disease prevention need to be greatly improved [5].
A five-year, city-wide control program for the prevention of dengue transmission applied temephos in granular formulation to water-holding containers using spoons every 3 or 4 months in Clorinda, northeastern Argentina, from 2003 to 2008 [6]. The program successfully limited dengue transmission and significantly reduced larval indices but failed to maintain them below target levels (the city-wide Breteau index was rarely <5%). Large tanks were found to be the most productive type of water-holding container [7], as in several studies around the globe [8]–[11]. Significant larval resistance to temephos was not recorded locally [12]. Incomplete surveillance coverage and reduced residuality of temephos under local conditions were held as the main putative factors limiting the effectiveness of control actions [6].
The duration of the residual effects of a given treatment (i.e. the amount of time the treatment is effective for vector control after its application) is a very important metric needed to estimate the frequency of treatment applications required to achieve control objectives. Under field conditions, treatments are affected by site-specific processes that modify the duration of residual effects relative to what is measured in controlled experiments under more artificial conditions. Therefore, the ultimate evaluation of treatment effectiveness is under field conditions [13].
Temephos, an organophosphate insecticide not toxic for humans at recommended doses, has been extensively used as a larvicide against Ae. aegypti during the past 40 years [14]–[18]. It is generally applied in granular formulation and delivered into containers using spoons, and more recently, inside permeable bags for slow release and reintroduction after householders clean treated containers [19]. Reference publications traditionally considered that temephos residual effects lasted between 8 and 12 weeks [20] or about 5 weeks [21]. A recently published guide for dengue vector control indicated: "Two or three application rounds carried out annually in a timely manner with proper monitoring of efficacy may suffice, especially in areas where the main transmission season is short." [22]. Recent studies using different temephos formulations, application procedures and experimental conditions have shown widely variable durations of residual effects ranging from 1 to 6 months [19], [23], [24]. The actual residuality of temephos under field conditions has rarely been documented. Infestation was detected within 7 days post-treatment with temephos in Brazil [25] and Nicaragua [26], but in the former study the number of experimental units was very limited (<18) whereas in the latter containers from 1,903 study houses were treated without the supervision of the investigators and observed only once post-treatment. In Peru, the larvicidal effect of temephos started to decline 7 weeks post-treatment but the field study only included eight experimental units [23]. None of these studies sought to identify the processes that caused such limited, widely variable effectiveness of temephos.
Our field-based study conducted in Clorinda had four objectives: (i) Estimate the duration of the residual effects of temephos in large water-storage tanks by means of larval mortality bioassays; (ii) Compare the effectiveness of temephos applied with spoons or inside permeable zip-lock bags; (iii) Identify factors and processes associated with the eventual decay of temephos residuality, and (iv) Describe the temporal pattern of Ae. aegypti immature infestation in containers treated with temephos.
A larval control program was run by Fundación Mundo Sano (FMS) and other organizations in Clorinda (lat 25°17′S, long 57°43′W), northern Argentina, from 2003 to 2008 [6]. The city had nearly 50,000 inhabitants in 2008. This study was carried out in Primero de Mayo, a large neighborhood with 2,500 houses (20% of the city) and relatively high infestation levels [7]. This neighborhood has an intermittent piped (tap) water service; ground-level, water-storage tanks (300-1,000 L) made of fibrocement or plastic are found in almost 50% of the lots (mean, 1.3 ground-level tanks per lot). The water harbored by these containers is used for many different purposes (e.g., washing, drinking, bathing, cooking, watering plants).
Two longitudinal studies (pilot and main trial) assessed the larvicidal effects of temephos in large water-holding tanks. Mortality bioassays of Ae. aegypti larvae exposed to water samples collected from treated and control containers at several occasions post-treatment were performed. Containers were selected randomly from a database that included approximately 1,300 large tanks identified by lot, size and material in 2007 [7].
During each trial, lots with selected containers were visited and the proposed activity was explained to the head of each household who was asked to give oral consent for temephos treatment, following customary practices of the ongoing larval control program since 2003 [6]. If permission was granted, consent was recorded in a form and each tank was treated with temephos at the recommended dose of 1 ppm (1% granular formulation, Fersol) by experienced FMS field personnel who regularly conduct vector control operations in the area. Following treatment, water samples from each study tank were collected weekly into 500 ml glass jars. Prior to collection, the water of each tank was stirred. Each jar was placed in expanded polystyrene thermic boxes and transported to the local FMS laboratory. This procedure is very similar to the one described by Palomino and others [23]. Two control tanks, one positive (treated with temephos) and one negative (untreated), were prepared in 300 liter-fibrocement tanks filled with piped water and then kept fully lidded and protected from rain and direct sunlight at the backyard of the laboratory.
Immediately after the arrival of water samples to the laboratory, mortality bioassays were performed by exposing 20 third- or fourth-instar larvae of Ae. aegypti to each water sample and recording mortality 24 hs later. The glass jars were left unlidded during the bioassays. Similar methodologies have been used previously [17], [23]–[27]. The larvae used were the second generation of larvae collected in a randomly selected block of Primero de Mayo in 2007 and were reared at the local FMS laboratory. A temephos-treated container was considered to have lost larvicidal effects when bioassay larval mortality was <70% [17], [24], [27].
A pilot trial was conducted in order to test the procedures and estimate how many water samples per container were necessary. A total of 18 selected tanks was treated with temephos applied using spoons on February 4, 2008 (mid-summer). Duplicate water samples were successfully collected weekly during 5 weeks post-treatment; positive controls were tested twice on weeks 7 and 8 post-treatment. Half of the containers were not sampled in the first week post-treatment due to unforeseen operational constraints (i.e., not enough larvae were produced). Meteorological data were collected by a local weather station run by Cooperativa de Provisión de Obras y Servicios Públicos Clorinda Limitada. During this trial, the mean temperature was 26.1°C and mean daily maximum and minimum temperatures were 31.8°C and 23.8°C, respectively. Cumulative rainfall was 151.6 mm.
Sixty water-storage tanks not included in the pilot trial (mean volume, 400 L; standard deviation, 198 L) were treated with temephos applied either using spoons or inside permeable zip-lock bags on November 5, 2008 (mid-spring), and followed up for 14 weeks until February 14, 2009 (mid-summer). Containers were randomly assigned to each treatment in a balanced design. When zip-lock bags were used, holes were punched with a paper clip prior to treatment, and householders were instructed to reintroduce the bag after eventually cleaning or emptying the container. Further discussion on water use practices was not engaged in order to minimize behavioral changes or create any bias toward the study containers.
Water samples for bioassays were collected immediately before treatment, at 2 days post-treatment, and at weekly intervals until reaching 14 weeks post-treatment; samples were not collected at week 8 post-treatment (extending over Christmas). Based on bioassay results from the pilot trial and other tests performed, only one sample of water per container was taken at each time point (data not shown). During the first visit to each study lot, each container was scored for sun exposure (considered low if any structure such as a ceiling or tree overshadowed the container, or high otherwise); container material (fibrocement or plastic), and water type (only rain water, only piped water, or rain and piped water). Pump water was rarely used in the study neighborhood, and not used at all in the randomly selected tanks included in the main trial. Positive and negative control tanks were stirred weekly because preliminary results suggested that the larvicide became attached to the fibrocement of the tanks [28].
The presence or absence of larvae and the number of pupae were registered every time a water sample was collected during the follow-up. All pupae and larvae were collected with large-mouth pipettes; frequently the operators used small sieves in order to strain the container and collect samples of immatures. Samples were placed in labeled test tubes and transported to the laboratory for processing. Larvae were identified to species level using an entomological magnifying glass and an illustrated key [29]. Pupae were kept in small water-filled plastic vials until emergence to allow accurate species identification as adults.
During the first seven weeks of the main trial, the larval control program team visited the rest of the houses in the study neighborhood and either removed, emptied or treated all water-holding containers found with temephos; the study tanks included in the main trial were excluded from these regular operations. The larval control program had not treated the neighborhood during the previous six months. During the main trial, the mean temperature was 25.9°C and mean daily maximum and minimum temperatures were 31.9°C and 20.4°C, respectively. Mean monthly cumulative rainfall was 123.5 mm/month.
To describe (re)infestation patterns, for each of the tanks in the main trial the temporal differences between the following events were calculated: treatment; first occurrence of infestation post-treatment; treatment of the block (by other larval control program teams) where the tank was located, and loss of residual effects.
The intensity of water turnover in tanks of the main trial was estimated by a procedure based on adding 100 ppm of sodium chloride to the containers and measuring the dilution of chloride during the subsequent 48 hs after concluding the follow-up (at week 16 post-treatment). Sodium chloride was selected because both sodium and chloride ions are natural water components and are not expected to suffer significant chemical transformations within 48 hs. The selected concentration was 100 ppm because preliminary estimates of typical chloride concentration in the study tanks were <50 ppm, and the final concentration would be below half of the taste threshold of sodium chloride (considered to be 300 ppm [30]). Tanks were excluded from water turnover assays if a householder was hypertense and drank water from the tank or if the tank was filled with less than 10% of its capacity. The water turnover assays were performed in 32 of the 60 tanks. Prior to the application of sodium chloride, the procedures and purpose of the study were explained to the head of each household who signed an informed consent. The procedures were approved by the Ethics Committee “Doctor Virgilio G. Foglia” in Buenos Aires, Argentina.
Sodium chloride was added to the tanks from a highly concentrated stock solution. Prior to the addition of the salt, the volume of water held by each container was estimated based on tank diameter and height of the water column. Three samples of water were collected from each tank: the first one at five days before addition of sodium chloride; the second one immediately before addition (both samples were used to estimate the mean basal concentration of chloride), and the third one at 48 hs post-addition of the salt. Water samples were placed in 15 ml-plastic tubes sealed hermetically, stored frozen and transported to Buenos Aires.
Chloride concentration was determined by ionic chromatography in a DIONEX DX-100 equipment with conductivity detector, suppressor, sample injection valve and 25 µL sample loop. Two plastic anion columns were coupled in series to serve both as pre-column and analytical chromatographic column (DIONEX AG-4 and AS-4, respectively). A mixture of HCO3-/CO32- (1.7 mM/1.8 mM) was chosen as eluent with a flow rate of 2 mL/min. The typical experimental error was lower than 5% for all results.
The volumetric output flow rate () was calculated by considering mass balance equations in each tank [31]. Simplifying flow rates as continuous processes:
By solving these equations, the volumetric output flow rate may be expressed as:where mcl is the mass of chloride; is the volumetric input flow rate; C is the concentration of chloride in the water added during the 48 hs post-addition of chloride; C(t) is the concentration of chloride over time; V1 is the volume of water measured immediately before the addition of sodium chloride; V2 is the final volume of water 48 hs post-addition of sodium chloride; C1 is the concentration of chloride after adding the salt (calculated as the sum of the concentration observed immediately before addition and 100 ppm −the added concentration), and C2 is the final concentration of chloride 48 hs post-addition of sodium chloride. C (unknown) was assumed to equal the mean basal concentration of chloride in each tank over the two pre-application samples. Water turnover intensity was estimated as the volumetric output flow rate multiplied by 48 hs (e.g., the estimated volume of water removed from the tank during the 48 hs) divided by the volume of each tank.
To confirm that these procedures provided valid estimates of water turnover, controlled experiments were performed using small plastic and fibrocement containers filled with running water from Buenos Aires. Five containers were filled with running water and 100 ppm of chloride were added to each of them. A known fraction of the water was replaced with piped water in each container 24 hs later. The resulting concentration of chloride was measured 48 hs post-addition of chloride. The estimated results differed by less than 5% in all cases (Table S1).
The intensity of water turnover was also assessed during the inspection of all houses in the study neighborhood in November-December 2008. At each house visit, the larval control team asked householders on how often they added water to each water-storage tank inspected.
Ninety-five per cent confidence intervals (CI) for mean larval mortality in the collected water samples were calculated with nonparametric bootstraps using the percentile method implemented with package Boot in R 2.70 [32].
To estimate the association between measured covariates and the larvicidal effect of temephos treatments over time, larval mortality in the bioassays was modeled using generalized estimating equations (GEE) [33]; the individual study tanks were considered as the experimental units. GEE models can serve as an extension of generalized linear models to analyze correlated data [34]. The models were computed with procedure xtgee in STATA 9.0 [35]. The best correlation structure was estimated by computing quasi-likelihoods under the independence model criterion (QIC) [36], a statistic analogous to Akaike's Information Criterion but suitable for GEE models. All QICs were computed following the algorithm of Hardin and Hilbe [34] in STATA. The explanatory variables included in the model were temephos application type, sun exposure, container material, water type, water turnover and the two-way interaction between temephos application type and water type or water turnover. The interactions were selected a priori because both types of temephos application were expected to be affected differently by water use practices. Because water type and water turnover were significantly associated, multicollinearity problems were detected when all the covariates were included in the model. Since water turnover had nearly 50% fewer observations than water type, only the latter was included in the model initially. Subsequently, we replaced water type with water turnover to check whether the outcome was robust to the exact specification of water management practices. An association with reduced larval mortality indicates shorter residual effects of temephos. The duration of residual effects for each tank in the main trial was calculated as the number of days until the loss of larvicidal effects (i.e., bioassay larval mortality <70%) was detected.
In the water turnover assays, the estimated final concentration of chloride 48 hs post-addition of sodium chloride (C2) was lower than the mean basal concentration of chloride in 25% of the cases. In such cases C was assumed to equal the minimum value of the two pre-application samples, because the basal concentration of chloride in piped water was highly variable over time and variations in the concentration of chloride during the potabilization process are typically substantial.
In the pilot trial, the duration of the residual effects of temephos was much shorter than expected and very heterogeneous between individual field tanks. The water collected from one of the tanks did not kill any larvae at the first week post-treatment. Mean larval mortality successively declined to 78% (CI, 46–100%) at one week post-treatment to 83% (CI, 66–100%) at week two, 68% (CI, 50–85%) at week three, and 50–52% (CI, 31–72%) at four or five weeks post-treatment. Only 1% of the larvae from the negative control tank died, whereas the positive control tank showed complete loss of larviciding power at four and five weeks post-treatment. A week later, its water was stirred and subsequent bioassays evidenced full larviciding power. Larval mortality between duplicate water samples was highly correlated (r = 0.97, n = 81, P<0.001).
In the main trial, water samples were successfully collected on 94% of the occasions (n = 849); 4% of the times the container was found dry, and in 2% the owner either refused access or was absent at the time of visit. The positive control tank had 99% larval mortality and the negative control only 1% mortality during the entire follow-up. Larval mortality in water samples collected prior to treatment was nil or close to zero in every container.
The residual effects of temephos in spoon-treated containers were shorter than in the pilot trial and also very heterogeneous between individual tanks (Fig. 1). Two of the 30 treated containers showed no larvicidal effects at two days post-treatment. Mean larval mortality was only 25% at four weeks post-treatment. The duration of residual effects of temephos applied with spoons had a median of 2.4 weeks (first quartile = 2.0 weeks, third quartile = 4.1 weeks; range 0.3 to 9 weeks).
Temephos applied inside zip-lock bags had longer-lasting residual effects than when applied with spoons (Fig. 1). The first evidence of a container losing all larviciding power occurred at two weeks post-treatment. Mean larval mortality was 50% at four weeks post-treatment. Mean larval mortality was consistently higher relative to the spoon-based application until week 8 post-treatment, when almost all containers showed null larvicidal power. The duration of residual effects of temephos applied with zip-lock bags had a median of 3.4 weeks (first quartile = 2.3 weeks, third quartile = 7.7 weeks; range 1 to 10 weeks).
Estimated water turnover and water type were highly significantly associated (linear regression, n = 32, F = 8.73, P = 0.001). Containers scored as holding piped water had a very fast, widely variable water turnover; 73% of these containers had an estimated turnover in 48 hs equal to or greater than the container's volume (Fig. 2). In contrast, 85% of tanks containing rain water had an estimated turnover close to 0; an outlier value with high turnover was probably a misclassified container. Containers with rain and piped water showed intermediate turnover rates. No rain occurred during the 48 hs after application of sodium chloride. In the house-to-house survey conducted in late 2008, householders responded that water was added every day in 61% of tanks filled with piped water (n = 582), and every 3 days or less in 86% of them. Among tanks filled with rain water (n = 170), in 94% of the cases householders responded that they waited until it rained for refilling the tanks.
The association between bioassay larval mortality and selected covariates was assessed with multivariate GEE models. The best correlation structure found was a second-order autoregressive correlation. Samples taken from containers with piped water had significantly lower larval mortality than those with other water types (OR = 0.60, CI = 0.48–0.75) (Table 1). Type of temephos application (spoon or zip-lock bag) significantly modified larval mortality depending on water type; reduced larval mortality occurred in spoon-treated containers filled with piped water (OR = 0.43, CI = 0.31–0.60) or with rain and piped water (OR = 0.39, CI = 0.26–0.59) relative to spoon-treated containers filled with rain water. Larval mortality was marginally significantly associated with sun exposure (OR = 1.17, CI = 1.01–1.36). Container material did not exert significant effects on larval mortality when other factors were taken into account. When water turnover was included in the model instead of water type (which reduced sample size by nearly 50%), the results obtained were qualitatively similar except for sun exposure which had insignificant effects (Table S2).
The duration of residual effects was used as a summary value of treatment effectiveness according to water type and temephos application type (Table 2). In containers filled with rain water, median residuality was substantially higher than in containers with other types of water. Bag-based applications had a higher median residuality than spoon-based applications in containers filled with rain and piped water.
In the study containers, infestation before treatment was high (container index = 33% and >10 pupae per container) and dropped to zero immediately after treatment (Fig. 3). The first signs of infestation post-treatment were detected at week 9 (i.e., only two weeks after the larval control program finished treating the entire neighborhood), when bioassay larval mortality dropped below 20%. Infestation continued to rise until the end of the follow-up. There was a significant association between infestation of individual containers before and after treatment with temephos (χ2 test, d.f. = 1, P<0.001). Among containers ever found to be infested after treatment, 88% had been infested before treatment. Among containers that were infested just before treatment, 73% became reinfested later. Only 9% of the containers not infested before treatment became infested after treatment. Containers were found to be newly infested at an average of 7 weeks (range 2–12 weeks) after estimated loss of larvicidal effects. However, in containers filled with rain water new infestations appeared on average 3 weeks after loss of larvicidal effects, whereas in containers filled with rain and piped water or only piped water the average time elapsed was much longer (7 and 11 weeks, respectively).
Both trials independently showed that the median duration of temephos residual effects in large water-holding tanks under field conditions (2–3 weeks in the main trial) was much shorter than the expected 5 to 8–12 weeks [20], [21], and widely variable between tanks regardless of type of temephos application. This is consistent with early (re)infestations in Clorinda being recorded 5–7 weeks after treatment with temephos during 2003–2005 [6] and 3 weeks post-treatment in 2008 (Garelli et al., unpublished data). These are important results for larval control programs, especially because temephos is widely used and the very few published reports of residual effects under field conditions show conflicting outcomes for undefined reasons. Secondly, the duration of residuality was modified substantially by water management practices (represented by water type and water turnover) —a novel finding in the dengue mosquito control literature. This short, widely variable residuality of temephos coupled with incomplete surveillance coverage [6] are probably the main causes of the inability of the larval control program to bring infestation below target levels with a treatment cycle period of 3–4 months.
Water type and water turnover were strongly and significantly associated. Multivariate analyses showed that both variables were associated with reduced larval mortality and shorter residuality of temephos. Containers filled with piped water had high water turnover rates and significantly shorter residual effects, especially when temephos was applied with spoons. Conversely, containers filled with rain water had lower turnover and much longer residual effects despite the occurrence of rainfall during each of the first six weeks of the main trial. Tanks filled with rain and piped water had intermediate water turnover rates, and residuality significantly differed between bag- and spoon-based applications. The interaction between water type and temephos application type was expected because zip-lock bags with undissolved insecticide were probably reintroduced after containers were emptied (bags were observed inside the tanks throughout the follow-up) [19] and thus provided extended residuality. Similar associations are expected to occur regardless of the larvicide used because high water turnover implies increased clearance of larvicides and shorter residual effects.
The intense water turnover rate in tanks filled with piped water may be explained by current water management practices determined by the intermittent local water supply, especially during the hot summer months when water consumption increases. In the study area, piped water was generally available during the night and was very limited or unavailable during daytime. Many householders reported refilling their tanks every night or so and, in some cases, emptying the tanks before refilling them. Conversely, containers filled with rain water had a much more irregular filling regime dependent on rainfall and almost no water turnover. These observations explain the strong association between water type and water turnover. Based on direct and indirect estimates of water use (Text S1) and householders' reports, we infer that local water management practices determined fast water turnover rates and caused the short-lasting residual effects of temephos in both field trials.
The effects of water management practices on Ae. aegypti abundance under recurrent larval control actions are complex and probably nonlinear. Intensely managed containers with fast water turnover (e.g., as those filled with piped water) are expected to have short-lasting chemical residual effects, but its suitability for adult Ae. aegypti production also depends on the process by which the tank is refilled. If the container is emptied often then it may not become suitable for adult mosquito production because immature stages are eliminated before adult emergence, but if water is added without removing or overflowing immature stages from the container it may become a suitable breeding site. Containers with low or very sporadic water turnover, such as those filled with rain water, should (and did) retain temephos residual effects for much longer periods. However, in the absence of effective surveillance and treatment or after residual effects wane, containers filled with rain water would become the most productive type if other conditions for suitability hold. Rainwater-filled containers became reinfested faster (mean, 3 weeks) than containers with at least piped water (mean, 7–11 weeks), and in a previous survey they had greater probability of being infested and produced more pupae [7].
The most likely sources of reinfestation post-treatment were breeding sites left untreated in closed premises or where householders refused interventions [6]. Even though the existence of putative cryptic sites cannot be completely excluded [37], intensified searches for them yielded negative results [6]. Studies with molecular markers are needed to provide concluding evidence on whether the detected (re)infestations post-interventions were persistent residual foci from eggs surviving treatment or new infestations from genetically different mosquitoes.
Containers infested before treatment were the most likely (73%) to become infested post-treatment, whereas a small fraction of those not infested before treatment (9%) became newly infested. This pattern suggests that the determinants of container suitability for mosquito breeding remained mostly invariant after interventions. In a previous study, containers located in yards rather than indoors, at low sun ex posure, unlidded, filled with rain water, and holding polluted water were found to be positively associated with infestation by larvae or pupae [7]. Most of these factors are related to householders' practices and may likely remain stable over time and space.
Our study has several limitations. Traditional bioassays measure larval survival after 24 hr whereas inhibition of adult production would be the epidemiologically significant metric for assessing transmission risk. If the larvae that survived treatment failed to develop into adults, the actual duration of temephos residual effects with respect to adult production would be underestimated. A larger number of study tanks would have allowed increased precision of larval mortality and associated parameter estimates. Although conclusions drawn from water turnover estimates would have been better supported with a larger sample size and more replicates for each container, the outcome was consistent with householders' reports on how often they managed their tanks and other indirect estimates. Water turnover estimates were probably imprecise because basal concentrations of chloride were widely variable between and within tanks. However, this relative imprecision would not compromise our main conclusions given the large differences in temephos residuality according to water type or water turnover rate. Post-treatment infestation in the main trial was probably underestimated because weekly operations removed all pupae and a sample of larvae, and perhaps very frequent house visits may have promoted householders' awareness and elimination of immature stages from the containers. These processes may also explain the lower number of pupae per container after treatment relative to pre-treatment levels. Had oviposition in the study tanks been monitored, it would have provided valuable information on the links between mosquito vital parameters and temephos treatment.
The positive control tank during the pilot trial had a transient loss of larviciding effects that were recovered after stirring its water. The tendency of temephos to attach to container walls was regarded as a positive feature derived from slow-release properties [38]. In practice, completely unmanaged tanks (i.e. without water movement) were very rare and therefore attachment of temephos to walls may only occur marginally. In the main trial, the water in the positive control tank was periodically stirred and achieved 99% larval mortality through the 14-week period, thus proving the efficacy of the larvicide and the absence of temephos resistance in Clorinda [12].
The wide gap between expected and actual durations of temephos residual effects under field conditions relative to average treatment cycle periods (3–4 months) accounts in part for not meeting larval control targets in Clorinda and probably elsewhere in northern Argentina.
Our results underline the importance of considering water use practices for the case of dengue, and local specificities in general, when designing, testing and implementing control interventions. Most results in epidemiology are context-dependent [39]. Unconditional recommendations may be misleading and undermine the effectiveness of larval control programs. More importantly, our present results and the outcome of the five-year larval control program [6] cast serious doubt on whether two or three application rounds of larvicides carried out annually in a timely manner [22] would be sufficient to achieve control target levels in many settings such as ours.
Water use practices depend on cultural patterns and water availability, and all three constitute a complex set of factors affecting dengue transmission dynamics [40], [41]. Environmental modifications such as the installation of a reliable piped water service are one of the principal actions for dengue vector control [22], [42]. Further research is needed to better understand the links between water management practices, dengue vector control, mosquito abundance and viral transmission. Temephos application inside small zip-lock bags extended the duration of residual effects relative to spoon-based applications; it was well received by householders, and was easily and inexpensively implemented. However, it is insufficient to achieve larval control goals with a treatment cycle period of 3 or 4 months. Considering the observed rate of reinfestation, a treatment cycle of 2 months would greatly improve larval control status at the expense of almost doubling labor costs and increasing community fatigue. The feasibility and sustainability of such high-frequency cycle periods in cities the size of Clorinda remain questionable. Novel forms of applying the larvicide specifically designed to cope with fast water turnover or new slow-release agents and formulations are needed to improve current larval control tactics. Biological control agents such as fish or cyclopoid copepods [43] may also be appropriate for this type of context. A different approach derived from present findings would seek to incorporate water use practices as control measures. Depending on how the intense management of tanks is performed, a strategy based on community participation aiming at healthier household water management practices may reduce infestations substantially because most water-storage tanks had piped water (77%, 582/752) and therefore were subjected to intense water management. Integrated control interventions capturing the multifaceted nature of Ae. aegypti population dynamics have the potential to achieve a much improved vector control status and prevention of dengue transmission.
|
10.1371/journal.pcbi.1002538 | Integrative Approach to Pain Genetics Identifies Pain Sensitivity Loci across Diseases | Identifying human genes relevant for the processing of pain requires difficult-to-conduct and expensive large-scale clinical trials. Here, we examine a novel integrative paradigm for data-driven discovery of pain gene candidates, taking advantage of the vast amount of existing disease-related clinical literature and gene expression microarray data stored in large international repositories. First, thousands of diseases were ranked according to a disease-specific pain index (DSPI), derived from Medical Subject Heading (MESH) annotations in MEDLINE. Second, gene expression profiles of 121 of these human diseases were obtained from public sources. Third, genes with expression variation significantly correlated with DSPI across diseases were selected as candidate pain genes. Finally, selected candidate pain genes were genotyped in an independent human cohort and prospectively evaluated for significant association between variants and measures of pain sensitivity. The strongest signal was with rs4512126 (5q32, ABLIM3, P = 1.3×10−10) for the sensitivity to cold pressor pain in males, but not in females. Significant associations were also observed with rs12548828, rs7826700 and rs1075791 on 8q22.2 within NCALD (P = 1.7×10−4, 1.8×10−4, and 2.2×10−4 respectively). Our results demonstrate the utility of a novel paradigm that integrates publicly available disease-specific gene expression data with clinical data curated from MEDLINE to facilitate the discovery of pain-relevant genes. This data-derived list of pain gene candidates enables additional focused and efficient biological studies validating additional candidates.
| The mechanisms underlying pain are incompletely understood, and are hard to study due to the subjective and complex nature of pain. From a genetics perspective, the discovery of genes relevant for the processing of pain in humans has been slow and genome-wide association studies have not been successful in yielding significantly associated variants. Targeted approaches examining specific candidate genes may be more promising. We present a novel integrative approach that combines publicly available molecular data and automatically extracted knowledge regarding pain contained in the literature to assist the discovery of novel pain genes. We prospectively validated this approach by demonstrating a significant association between several newly identified pain gene candidates and sensitivity to cold pressor pain.
| A significant number of diseases are associated with pain, thereby affecting the quality of life of many individuals. The Institute of Medicine's recent report titled, “Relieving Pain in America” presented pain as a public health challenge, and emphasized the need for an integrative approach to understand mechanisms underlying pain [1]. Such understanding is critical for developing more effective and individualized strategies targeting the prevention and treatment of pain.
Studies in rodents and humans have established the importance of genetic factors in the processing of pain [2], [3], [4]. However, identifying genes important to complex phenotypes such as pain using genome-wide association studies has been challenging [5]. Candidate gene studies have identified many gene variants associated with susceptibility to pain [6], [7]. Despite these advances, genetic discoveries in the domain of pain have been slow in forthcoming compared to other fields [8].
Pain is among the most difficult phenotypes to study due to its complex and subjective nature. The perception of pain is influenced by a multitude of variables including gender, age, mood, ethnicity and genetic factors [9], and a recent meta-analysis highlighted the overall small effect size attributable to any gene variant associated with the processing of pain [10]. The polygenetic nature of pain and the small effect size of gene variants pose significant challenges for pain gene discovery.
Candidate gene studies have proven successful in the identification of pain genes. A particularly promising approach used gene expression microarray analysis to select candidate genes [11], [12]. More recently a meta-analysis of publicly available microarray data from rodents exposed to neuropathic or inflammatory pain was able to efficiently prioritize pain-related genes [13]. A similar approach using human gene expression data could be highly beneficial in generating data-driven hypotheses for pain genetics. However, there is currently a paucity of public gene expression data related to specific human pain conditions.
In this study, we describe an integrative approach exploiting publically available gene expression data for a large set of disease conditions to develop a disease-specific pain index (DSPI). This approach is based on the hypothesis that differences at the gene expression level correlating with pain indices would allow identifying novel pain gene candidates [14]. We validated this approach through a targeted genetic association study in an independent human cohort, where variants of selected pain gene candidates were evaluated for associations with experimental pain sensitivity measures in humans.
We built the disease-specific pain index (DSPI) using our literature-based approach, in which 2962 diseases were ranked according to their disease-pain ratio. Table 1 displays the 20 diseases with the highest pain indices. Diseases included Prinzmetal's angina, neuralgia, causalgia, chronic plantar fasciitis and polyarthralgia; all conditions associated with severe pain. Among the diseases ranked at the bottom of the pain index list were fetal alcohol syndrome, cretinism, hermaphroditism and fetal erythroblastosis; all conditions not primarily associated with pain (Table S1). Inspection of the DSPI indicates that diseases with a high pain index are typically associated with significant clinical pain, while pain is not a hallmark of diseases with a low pain index. As such, the pain index captures relevant aspects of disease-related pain. However, the DSPI relies on the fraction of disease-related publications in PubMed that are associated with the Medical Subject Heading (MESH) term “pain” and therefore, is subject to some bias. For example, a disease associated with significant pain but hardly studied in the context of pain would rank inappropriately low on the DSPI list. A practical example is the pain index of cholera with a rank of 2936, which is at the bottom of the list. Cholera is clearly associated with painful symptoms. Inspection of the DSPI generally revealed relatively low ranks for infectious diseases, likely indicating that the research community predominantly focuses on the most relevant aspects of the condition under study. This suggests that the DSPI also captures to some extend the relevance of pain across multiple diseases.
As previously described, the raw microarray data for 311 diseases were extracted from public gene expression databases [15], [16], [17]. A list of 3812 differentially expressed (DE) genes was then compiled (see Materials and Methods). Pain indices were available for 121 of the 311 diseases with suitable microarray data. The 121 disease-related gene expression changes were ordered according to the DSPI. For each of the 3812 differentially expressed genes, the gene expression fold change across every disease was correlated with the DSPI. This allowed identifying genes whose expression changes were significantly correlated with pain.
The sensitivity and accuracy of this strategy for capturing genes implicated in the processing of pain was first evaluated with the aid of the Pain Gene Database (PGD) [18]. The PGD catalogs genes whose transgenic or knockout mouse counterparts have exhibited changes in pain-related phenotypes. The PGD is actively maintained and, to our knowledge, is the only pain-related gene database. Figure 1 shows the receiver operating characteristic (ROC) curve with confidence intervals. The area under the curve (AUC) was 60.5% indicating a prioritization of known pain genes from the PGD by our method.
We evaluated the significance of the association of the 3812 genes with the DSPI using a threshold-based estimated false discovery rate. Forty-seven genes were significantly associated with the DSPI (pFDR<0.01; Table 2). Among the 47 genes, two genes, DLG4 (PSD-95) and CHRNA4, were referenced in the PGD [19], [20]. DLG4 and CHRNA4 were both found to have expression changes in 13 of 121 diseases that were positively correlated with pain indices (Figure 2A–B).
In light of this significant but modest prioritization of pain related genes, we applied our pipeline to another medically relevant concept: “Inflammation”. As described above for pain we extracted from MEDLINE a Disease-Specific Inflammation Index (DSII) and retrieved gene significantly associated with this index. Using genes belonging to the Gene Ontology category “Inflammatory Response” (GO:0006954) as gold standard we computed the area under the ROC curve. Figure S1 shows a clear prioritization of known “Inflammatory response” genes through our pipeline with an AUC of 73.2%.
Selected genes from the candidate list were prospectively tested for variants that may be associated with differential pain sensitivity in an independent human cohort. These genes were chosen based on their high correlation with the DSPI and plausible biology as assessed by the available literature and human expression profile across tissue using The Scripps Research Institute BioGPS database [21]. The selected genes were: (i) ABLIM3 (actin binding LIM protein family, member 3), PDE2A (phosphodiesterase 2A, cGMP-stimulated), CREB1 (cAMP responsive element binding protein 1), NAALAD2 (N-acetylated alpha-linked acidic dipeptidase 2), and NCALD (neurocalcin delta) (Figure 2C–G).
ABLIM3 was selected as our top candidate as it showed the highest correlation with the DSPI. The cGMP-sensitive phosphodiesterase PDE2A localizes at the neuronal membrane in synapses and has been described as being regulated by TNFα, a known proinflammatory cytokine shown to sensitize primary nociceptors [22]. Additionally, a recent study on Grueneberg ganglion neurons, that are proposed thermosensors, revealed a key role of cGMP enzyme in cold temperature sensing [23]. Interestingly, PDE2A provides a mechanism for nitric oxide-mediated cGMP synthesis to control intracellular concentrations of cAMP [24]. cAMP is a key second messenger that activate numerous downstream protein, notably cyclic-AMP-response element (CRE)-binding protein (CREB) that activate classical immediate-early genes such as c-Fos, which are associated with nociceptive afferent activation [25], [26]. NAALAD2 is highly similar in sequence to NAALAD1 and both hydrolyze N-acetyl-L-aspartate-L-glutamate (NAAG) to N-acetyl-aspartate and glutamate, a neuropeptide that activates and antagonizes neuronal N-methyl-D-aspartate (NMDA) receptors [27]. Based on nociceptive tests in rats, NAALAD1 was found to plays a role in maintaining mechanical allodynia after carrageenan injection [28]. Of note, NAALAD1 was also positively correlated with the DSPI but not to the same level of significance as NAALAD2. Finally, NCALD (neurocalcin delta) is a calcium-binding protein abundantly and almost exclusively expressed in the central nervous system that has not previously been associated with pain [29].
The genotyping study was conducted in samples obtained from twins enrolled in an ongoing independent IRB-approved pharmacogenomic study testing subjects' sensitivity to experimental heat and cold pressor pain among other outcomes (see Materials and Methods for details). The association study was performed using a generalized least square (GLS) test. GLS allowed us to model different variances between monozygotic twin pairs (MZ), dizygotic twin pairs (DZ), and the sexes, as each of these factors has previously been shown to influence pain measures [4], [30], [31], [32], [33].
Within the five selected genes, 251 tag SNPs were tested. Polymorphisms in ABLIM3 (rs4512126) and NCALD (rs12548828, rs7826700, and rs1075791) showed significant association with the cold pressor pain threshold after Bonferroni correction (Figure 3A–B). Linkage disequilibrium (LD) analysis of the genotyped SNPs revealed a relatively weak LD structure around these polymorphisms. The LD structure in both genes was similar between the study cohort and the HapMap CEU population for the same region (Figure S2 and S3). Interestingly, the influence of the rs4512126 loci on the cold pressor pain threshold was tested, which revealed a male specific effect for individuals with the T/T allele (Figure 4A). Males with homozygous T/T alleles exhibited a significantly higher mean pain cold threshold than all other groups (p = 0.005, 4×10−4, 0.02, 0.005, 0.01, for A/A Males, A/A Females, A/T Male, A/T Females and T/T Females, respectively). The largest effect sizes (Cohen's d) were observed between T/T Males and A/A Males and Females (0.38 and 0.39, respectively). Effect sizes between T/T Males and the other groups were below the small effect size threshold (< = 0.2) with 0.16, 0.11 and 0.17 for A/T Males, A/T Females and TT Females respectively.
The primary objective of this study was to demonstrate the utility and validity of a novel, data-driven approach for generating a list of pain gene candidates. Such a list could facilitate the discovery of pain genes. We first validated our approach by demonstrating a statistically significant sensitivity and specificity prioritization of known pain-related genes contained in the Pain Gene Database (PGD). In addition, further genotyping of a human cohort revealed a significant association between variants of the newly discovered pain gene candidates ABLIM3 and NCALD with measures of pain sensitivity in an independent human cohort.
A major emphasis of this study was to document the utility of the principal approach and highlight its future potential. The ever growing amount of publically available molecular and clinical data should allow for expanding and refining this approach to generate more comprehensive and specific lists. For example, as more data becomes available, it may be possible to link gene expression of diseases to specific types of pain, such as neuropathic pain. Similarly, the outlined approach can be expanded to include proteomic data sets, which should provide additional insight into signaling pathways relevant to the processing of pain. Finally, the pain associated with a specific disease can be construed differently. For example, disease-specific pain ratings could be retrieved from databases of large health care organizations [34].
There are a few limitations in our approach and study. First, among the 47 candidate pain genes significantly correlating with the DSPI, only two are referenced in the PGD. While the PGD is a valuable resource of curated information and likely represents the best available reference, it is not yet a globally accepted master repository containing all pain genes, especially those resulting from human studies. The database is constrained by the fact that it only catalogs genes revealed by studies examining nociception in mechanistic – but not disease-related – models in knock-out mice. It should also be noted that gene expression data for diseases and matched controls were only available for 121 diseases. As a result only 130 of the 300 genes listed in the PGD could be explored in the current study.
The presented paradigm did not capture genes such as KCNS1, GCH1, COMT or OPRM1, each of which has been implicated in the processing of pain [9]. This may partially be due to the fact that the current algorithm favored the discovery of genes exhibiting gradual gene expression change across different diseases. Additionally, our approach relied on gene expression changes in diseased tissue, which may not always capture important changes in secondary tissues relevant for the processing of pain, such as neuronal tissues or blood vessels. Additionally, some of these genes, such as KCNS1, are thought to be important in specific types of pain like neuropathic pain, but might not participate genetically in determining pain of other etiologies represented in the 121 diseases. There is considerable potential for more refined approaches in the near future to resolve some of these limitations, as there are a constantly growing number of publicly available repositories containing molecular and phenotypic data sets.
ABLIM3 is a newly characterized protein-coding gene belonging to the actin binding LIM protein family, which is composed of 3 members (ABLIM1-3) and shows a high degree of conservation throughout evolution in vertebrates. ABLIM3 is expressed in various tissues, most prominently in muscle and neuronal tissue [35], [36]. While relatively little is known about the biological function of the ABLIM protein family, conservation of key structural features suggests comparable biological function as linkers between the actin cytoskeleton, cell signaling pathways and transcription events [36]. For example, the ortholog of ABLIM1 in C. elegans (UNC-115) has been implicated in axonal guidance during outgrowth through interaction with Receptor for Activated C Kinase (RACK-1) [37].
Presently, a potential functional role for ABLIM3 in the perception or processing of pain is not apparent. ABLIM3 could potentially affect nociceptive signaling by regulating synaptic strength through actin rearrangement and modulation of synaptic spine density [38]. Neuroplasticity has been shown to play a role in pathological pain and to happen both at the molecular and cellular levels [39]. However, the association of ABLIM3 with pain is a novel finding that is based on a data-driven approach but is not anchored in our current understanding of pain biology. While this approach may offer the advantage of making unexpected and important discoveries, it requires establishing the biological relevance of such discoveries in subsequent experimental steps. The SNP rs4512126 (5q32) is located in the second and largest intron of ABLIM3. This variant was found in weak linkage disequilibrium (>0.6) with five other SNPs and in perfect linkage disequilibrium with rs4546368 located in the same intronic region of ABLIM3. All were non-coding SNPs. Similar to ABLIM3, NCALD has never been reported to be associated with pain. However, several polymorphisms in the 3′ UTR have been associated with mRNA instability and diabetic nephropathy [40]. Individuals carrying the A/A allele possessed a higher cold pain threshold. Nevertheless, we acknowledge that our discovered association between NCALD and pain cold was modest, demonstrated in only a single cohort, and barely above the Bonferroni corrected threshold. The parallel approach using quantile normalized phenotypical pain measures did not sustain the association for NCALD (Figure S4). Further genotyping in alternative cohorts and deep sequencing of these regions would be needed to reveal a potentially causal SNP.
Interestingly, only males with homozygous ABLIM3 T/T showed a significant association with cold pressor pain sensitivity in our study. The sex-specific association of a gene variant with the cold pressor pain threshold is not surprising. Genetic polymorphisms associated with pain in humans and animals have identified a striking number of sexual dimorphisms with either male- or female-specific genetic effects, or a significant difference between the sexes [7], [9], [30], [34], [41].
We present a novel paradigm linking publically available molecular data to clinically relevant phenotypic data for generating a list of candidate genes relevant to the processing of pain. Algorithms for accessing and integrating such data to examine disease-relevant mechanisms are of growing interest as publically available data sets grow at an ever-increasing rate. The outlined approach can complement existing research efforts by assisting the formulation of data-driven hypotheses, and may serve as a template to discover genetic components of other clinically important phenotypes.
The twin study was approved by the Institutional Review Boards of Stanford University and SRI International. All subjects gave written informed consent prior to participation.
MeSH is a comprehensive vocabulary thesaurus organized in a hierarchical structure allowing the indexing of publications with various levels of specificity. MeSH terms are used by trained human curators to annotate publications referenced in MEDLINE. We first built a thesaurus of 3743 disease-related MeSH terms using the Unified Medical Language System (UMLS); restricting ourselves to terms belonging to the following semantic type: Pathologic Function (T046), Disease and Syndrome (T047) and Neoplastic Process (T191) [42].
Searching MEDLINE for each of these MeSH terms gave us the number of publications published on each disease. For each disease returning a result, we conducted a second search in MEDLINE to count the number of papers published that were also annotated with the MeSH heading term “pain”[mh]. Searching MEDLINE for “pain”[mh] includes publications annotated with any terms hierarchically below “pain” in MeSH, such as Aches, Burning pain, and others. MEDLINE searches were automated using the EUtils programming tools available from the NCBI (http://eutils.ncbi.nlm.nih.gov). The ratio of these two counts formed the disease-pain ratio, as shown in equation (1).(1)
The comprehensive disease-specific pain index (DSPI) was established by ranking all 2962 diseases by their respective disease-pain ratio (Table S1). Implicit to our algorithm is the assumption that each disease provides unique qualitative information that may be diluted if weighting results by publication frequency. Thus, the only criteria for inclusion of the disease in the DSPI was to have at least one co-citation with a “pain” related MeSH term (Figure 5A–B).
A similar approach was followed to establish the Disease-Specific Inflammation Index (DSII) by searching for publication annotated with the MeSH term “Inflammation”[mh].
We annotated all datasets from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) and the European Bioinformatics Institute (EBI) ArrayExpress (AE) public databases with UMLS identifiers for diseases as previously published [43], [44], [45], [46], [47]. We further evaluated these data sets to determine whether or not the submitted biological experiments measured a normal control state (disease free tissues) complimentary to the annotated disease state. This was done to ensure that differentially expressed (DE) genes could be extracted for each disease. Drug treated samples were excluded from the study. Disease, tissue and substance annotations were manually reviewed in a post-processing step to ensure accuracy. Extraction of data from GEO and AE according to outlined steps revealed 311 diseases explored across 456 publicly available data sets and comprising 14,457 individual microarrays from 169 different tissues. In this study, microarrays were pre-processed and DE genes lists were generated using Rank Product (Figure 5C) [48]. We kept only genes with a q-value (gene-specific false-discovery rate) level ≤0.05 [49].
One-hundred twenty-one of the 311 diseases retrieved from GEO and AE were also present in the DSPI list, and thus could be associated with disease-specific pain indices (see Table S1 for the DSPI and Table S2 for the overlapping list of 121 diseases). A small number of animal diseases were present in our DSPI since MEDLINE also covers veterinary and animal diseases. These were automatically excluded from the rest of the analysis, as retrieved gene expression data were limited to humans. Fold change values of the DE genes for each of 121 diseases were organized into a matrix with diseases as columns and genes as rows (Figure 5D). Because the lists of DE genes varied from one disease to another, we defined an arbitrary threshold of minimum gene representation. Information on a gene had to be present in at least 10% of the listed diseases. The final matrix contained 3812 genes as rows and 121 diseases arranged in columns. Finally, disease-columns in the matrix were ordered according to their DSPI rank, i.e., from the lowest to the highest pain index.
The Spearman rank correlation was then computed for each gene using the fold-change values against the DSPI ranking. We computed the positive false discovery rate (pFDR) values for each gene by permuting the diseases rank and re-computing the Spearman correlation for each gene. The operation was repeated 1000 times to obtain a null distribution of the correlation coefficients. The pFDR values were calculated as the ratio of the expected proportion of false positive V over the total number of hypothesis rejected R (2) [50].(2)
We evaluated the sensitivity and accuracy of our method in prioritizing pain genes using Receiver Operating Characteristic (ROC) curves. We compared the DSPI-based pain-gene list against a list of known pain genes from the PGD (www.jbldesign.com/jmogil/; accessed October 2010) [18]. We acknowledge that the PGD only contains data from studies in knockout mice. However, the PGD is to our knowledge the only available repository of pain genes. All 308 mouse genes in the PGD were retrieved manually and translated to 300 human homologs using the NCBI Homologene database (www.ncbi.nlm.nih.gov/homologene). All 300 mouse genes were translated to unique human genes except 8 genes lacking homologs. These eight genes were not further considered. We established the confidence intervals for the ROC curve using a leave-one-out resampling method by repeatedly recalculating the pain-gene rankings with nine-tenths of the 121 diseases. These alternative pain-gene lists were then used to compute the confidence interval of the standard error of our pain-gene ranking. We used the ROCR package from R to compute the ROC curves [51]. Of note, the gene expression measurements on the 121 diseases only included 130 of the 300 known pain genes from the PGD.
Gene extracted using the DSII were compared to a gold standard made from the genes belonging to the “Inflammatory response” Gene Ontology category (GO:0006954). The gene list was retrieved from the Molecular Signature Database [52].
Our study used samples and data from a pre-existing large pharmacogenomic study in twins examining the heritability of various opioid effects [32]. More specifically, data on subjects' sensitivity to heat and cold pressor pain before drug exposure were retrieved. The twin study was approved by the Institutional Review Boards of Stanford University and SRI International. All subjects gave written informed consent prior to participation.
In total, 228 healthy pain-free twins (114 twin pairs) of diverse ethnicities were genotyped and phenotyped. We considered only individuals with self-declared European ancestry, which represented 179 individuals with an age range of 18–68 years. Zygosity status of identical (MZ) and fraternal (DZ) twins was assessed by genotyping concordance using a panel of 47 SNPs [53]. Relevant covariates known to potentially confound measures of pain were also assessed and included demographic factors, age, sex, education, depressed mood, anxiety, sleep, and blood pressure. A detailed description of methods has been published elsewhere [32]. Overall, the cohort used for this analysis was not balanced for sex (107 females and 72 males) and consisted of 54 dizygotic and 125 monozygotic twins.
Experimental pain measurements were performed in the Human Pain Laboratory of the Department of Anesthesia at Stanford University School of Medicine. Heat pain was induced with a thermal sensory analyzer (TSA-II, Medoc Advance Medical System, Durham, North Carolina). A 3×3 cm thermode was placed in contact with skin at the volar forearm. Starting at 35°C, the thermode temperature was increased at a rate of 1°C/s. Study participants pushed a button of a hand-held device at the onset of pain. This procedure was repeated 4 times with an inter-stimulus interval of 30 seconds. The average temperature (C°) eliciting pain was recorded as the pain threshold.
Cold pressor pain is thought to mimic important qualities of clinical pain, since verbal descriptors for both types of pain are strikingly similar [54]. Cold pressor pain is more sustained than heat pain and is associated with a much stronger affective response [55]. The cold-pressor pain model can be viewed as a tool examining an integrated pain response with a strong affective component, while the heat pain model is better suited to explore sensory-discriminative aspects of pain. Sensitivity to cold pressor pain was tested by having subjects immerse their hand to mid-forearm in ice-water (1–2°C) continuously recirculated within a 12-liter container. The palm of the hand was in full contact with the bottom of the container. Subjects were asked to indicate the onset of dull, aching pain typically perceived in the wrist and to withdraw the hand once pain became intolerable. The time (seconds) to the onset of pain was recorded as the cold pain threshold and the time to withdrawing the hand was recorded as the cold pain tolerance.
The twin cohort was genotyped for 251 SNPs across 5 genes selected from the list of candidate pain genes (Table 2). We selected LD tag SNPs (r2 = 1) based on the HapMap CEU population using the Tagger software from the Broad Institute (http://www.broadinstitute.org/mpg/tagger/) [56]. Twin's DNA was extracted from peripheral blood lymphocytes and genotyped using a custom-designed Oligo Pool for Methylation Assay (Golden Gate Genotyping Assay, Illumina, Inc, San Diego, CA) and BeadXpress (iGenix Inc., Bainbridge Island, WA). We filtered out all SNPs with a call rate <90%.
Out of 251 SNPs assayed only 216 yielded successful results following Illumina quality controls. Some successfully genotyped loci had missing genotypes for a small number of twins and were imputed using the homozygous wildtype allele from the population. We filtered out all SNPs with a minor allele frequency <5% or those whose genotype frequency departed from Hardy-Weinberg equilibrium at p<0.01. In summary, 207 SNPs were tested against heat pain threshold, cold pain threshold and cold pain tolerance.
While twin individuals were not required to test association of our candidate genes with pain, twins allowed us to control for environmental variability in pain measurements. Rather than utilize methods to correct for the relatedness of observations coming from the two members of the same twin pair, we used a model in which each pair was treated as a single observation by using within-pair genotype and phenotype averages. Genotypes were transformed to numbers according to the allele frequency, 0, 1, 2 for homozygous wildtype, heterozygous and homozygous rare, respectively. Then, the genotypes of twins were averaged within each pair. The genotype of single twin was not altered. Categorical covariate data such as sexes were discretized to −1 and 1 for male and female respectively and then averaged. DZ twin pairs of different sex were coded 0.
To test association of measures of pain sensitivity with each SNP, we used a generalized least square (GLS) regression model [57] examining the null hypothesis that pain and genotype are not associated. We regressed on the pain score measured, y, against the genotype for the SNP considered while controlling for depression of the individual (Beck Depression Index) and sex (3).(3)
GLS allowed us to model different variances of measured traits in MZ and DZ twins and in males and females. We related genotype to the heat pain threshold (degree C°) and the log of both the cold pressor pain threshold and cold pressor pain tolerance. P-values were corrected for multiple hypotheses testing using the Bonferroni correction. One twin with the T/T allele for rs4512126 reached the maximum allowed time in the cold threshold test (3 min). This result was considered an outlier and was removed from the analysis. In parallel, we also evaluated the effect of the pain phenotype normally distributed through quantile normalization. This analysis revealed similar results. The rs4512126 (ABLIM3) polymorphism remained significant for cold pain threshold (Figure S4A). Additionally, rs4512126 and rs7715362 (ABLIM3) showed a significant association with cold pain tolerance (Figure S4B) and rs7720260 (ABLIM3) showed a significant association with heat pain threshold (Figure S4C). However, previously found significant polymorphisms in NCALD did not pass the Bonferroni threshold in this analysis.
We computed the Cohen's d effect sizes for the difference observed between men and women homozygous for the minor allele of the rs4512126 variant as follows d = [xm−xf]/pooled standard deviation, where xm and xf are the average cold pressor pain threshold in males and females, respectively [58]. Positive values indicate a higher male average pain threshold, and negative values indicate a higher female average pain threshold. The result is unit free and Cohen proposed that benchmark values for what should be considered a ‘small’, ‘medium’ and ‘large’ effect (d> = 0.2, 0.5, 0.8, respectively) [59]. We analyzed differences in mean cold threshold for rs4512126 by performing two-sample t-tests with unequal sample size and unequal variances (Welch two sample t-test).
|
10.1371/journal.pgen.1004726 | Fat-Dachsous Signaling Coordinates Cartilage Differentiation and Polarity during Craniofacial Development | Organogenesis requires coordinated regulation of cellular differentiation and morphogenesis. Cartilage cells in the vertebrate skeleton form polarized stacks, which drive the elongation and shaping of skeletal primordia. Here we show that an atypical cadherin, Fat3, and its partner Dachsous-2 (Dchs2), control polarized cell-cell intercalation of cartilage precursors during craniofacial development. In zebrafish embryos deficient in Fat3 or Dchs2, chondrocytes fail to stack and misregulate expression of sox9a. Similar morphogenetic defects occur in rerea/atr2a−/− mutants, and Fat3 binds REREa, consistent with a model in which Fat3, Dchs2 and REREa interact to control polarized cell-cell intercalation and simultaneously control differentiation through Sox9. Chimaeric analyses support such a model, and reveal long-range influences of all three factors, consistent with the activation of a secondary signal that regulates polarized cell-cell intercalation. This coordinates the spatial and temporal morphogenesis of chondrocytes to shape skeletal primordia and defects in these processes underlie human skeletal malformations. Similar links between cell polarity and differentiation mechanisms are also likely to control organ formation in other contexts.
| Little is known about the mechanisms of cell-cell communication necessary to assemble skeletal elements of appropriate size and shape. In this study, we investigate the roles of genetic factors belonging to a developmental pathway that affects skeletal progenitor behavior: the atypical cadherins Fat3 and Dachsous2 (Dchs2), and REREa/Atr2a. We show that cartilage precursors fail to rearrange into linear stacks and at the same time misregulate expression of sox9a, a key regulator of cartilage differentiation, in zebrafish embryos deficient in Fat3 or its partner Dchs2. Similar cartilage defects are observed in rerea−/− mutants, and Fat3 interacts physically and genetically with REREa. Our results suggest that Fat3, Dchs2 and REREa interact to control polarized cell-cell intercalation and simultaneously control skeletal differentiation through Sox9. By transplanting cartilage precursors between wild-type and Fat3, Dchs2 or REREa deficient embryos we demonstrate that all three factors exert long-range influences on neighboring cells, most likely mediated by another polarizing signal. We propose a model in which this coordinates the polarity and differentiation of chondrocytes to shape skeletal primordia, and that defects in these processes underlie human skeletal malformations.
| What are the mechanisms of cell-cell communication that mediate organ morphogenesis? Bones, for example, have many different sizes and shapes yet the individual and collective cell behaviors necessary to assemble these shapes remain largely unknown. During development, cartilage serves as the blueprint for much of the adult skeleton. Cartilage models of long bones, including the digits, are aligned into columns of discoid cells that resemble stacks of coins [1]. This basic arrangement is also found in other endochondral bones, including in the craniofacial skeleton where studies in zebrafish suggest that these stacks form by oriented cell intercalations [2]. Notably, cartilage differentiation and morphogenesis are initiated by sox9, but the morphogenetic pathway(s) activated by this transcription factor remain unknown [3], [4].
Cell-cell intercalations such as those that occur in cartilage are often regulated by planar cell polarity (PCP) pathways [5]. First described in Drosophila epithelia, PCP refers to coordinated polarity within a cell sheet [6]. More recently, vertebrates have been shown to utilize PCP factors not only in polarizing epithelia but also in orienting cell divisions and movements, such as the cell intercalations driving tissue convergence and extension during gastrulation, neurulation and kidney formation [5], [7]–[11]. Hallmarks of PCP in epithelia are the locations of polarized hairs or cilia protruding from cells. Similarly, primary cilia and their associated basal bodies/microtubule organizing centers (MTOCs) orient towards the leading edges of intercalating cells [12], and distally in chondrocytes in the digits of mouse limbs [13]. These coordinated polarity dynamics observed in migrating cells and chondrocytes suggest that PCP may also be part of the mechanism that controls skeletal morphogenesis.
Two main pathways regulate PCP independently in Drosophila: the Frizzled (Fz) pathway and the Fat/Dachsous (Dchs) pathway [14]. Both pathways are conserved in vertebrates and variously required for polarity of diverse tissues, including cochlear hair cells and hair follicles [5], [15]. In mice, components of the Fz pathway regulate oriented divisions and intercalations of chondrocytes in the growth plates of long bones [16], and Fat3/4 cooperate during fusion of vertebral arches [17]. However, little is known about requirements for the Fat/Dchs pathway in skeletal morphogenesis.
Four relatives of components of the Drosophila Ft/Ds pathway function in vertebrate PCP: 1) Ft, 2) Ds, 3) Four-jointed (Fj) and 4) Atrophin (Atro) [18]–[21]. In Drosophila, heterophilic binding of the protocadherins Ft and Ds mediates cell-cell adhesion and communication, while the Golgi kinase Fj phosphorylates their cadherin domains to modulate their binding affinity [22]–[24]. Atro also modulates signaling and binds the intracellular domain of Ft [20], but has a dual role as a transcriptional co-repressor that interacts with histone deacetylase (HDAC) [25], [26]. Ft or Ds mutant clones induced within imaginal discs trigger reversals of cell polarity outside the clone in one direction. In addition, Ds and Fj form opposing gradients across fly epithelia and interact with uniformly expressed Ft molecules [23], [27], [28]. This suggests a model in which the linear gradient of Ft/Ds heterodimers polarizes cell fields [29]. However, more recent quantitative analysis of these polarity reversals in the fly eye instead suggests that Ft and Ds interact to modulate a secondary signal that regulates long-range polarity [30]. Whether or not vertebrate Fat/Dchs signaling propagates polarity at a distance, utilizes molecular gradients, or interacts with other polarizing signals remains completely unknown in any tissue or organ.
Here we use the accessibility and miniature organization of the zebrafish jaw skeleton to investigate the genetic mechanisms of cartilage morphogenesis. We show that morphogenesis of polarized chondrocyte stacks results from oriented cell intercalations that depend upon Fat3, Dchs2 and REREa/Atr2a and their regulation of sox9a expression. Chimaeric analyses show that all three are required non-cell autonomously and over several cell-diameters for cartilage stacking and polarity, consistent with activation of a secondary signal that regulates polarized cell-cell intercalation. Fat3 and REREa interact physically and genetically, and our results suggest that Fat3 indirectly induces sox9a by preventing REREa from repressing it, while Dchs2 induces sox9a expression. Sox9a in turn activates fat3 and dchs2 expression. We propose a model in which Fat/Dchs signaling coordinates morphogenesis and differentiation of cartilage by the non-cell autonomous regulation of polarized cell-cell intercalation and sox9a expression.
To understand the cellular basis of cartilage morphogenesis in the zebrafish pharyngeal skeleton we focused on pharyngeal arch 1 (PA1, mandibular), which in larvae consists of two elements, the ventral, lower – Meckel's cartilage (Mc) - and dorsal, upper – palatoquadrate (pq) - jaw cartilages. We conducted time-lapse analysis of pre-cartilage morphogenesis during the jaw-elongation period in a sox10:lyn-tdTomato transgenic driving membrane-localized red fluorescence in pharyngeal neural crest (NC) cells (Fig. 1A, B; Video S1) [31], [32]. Cell-cell rearrangements drive cartilage morphogenesis between 48-56 hpf. During this period, morphogenesis of the sheet-like pq (Fig. 1A′ B′) and rod-like Mc (Fig. 1A″, B″) was driven by a combination of radial and medio-lateral cell intercalations (Fig. 1C), while little cellular rearrangement occurred at the presumptive joint (arrowheads in Fig. 1A,B). Cell division did not contribute to growth of cartilage during this period but was observed in surrounding tissues (Video S1). EdU labeling confirmed the near absence of proliferation in intercalating prechondrocytes, as previously reported [2](Fig. S1A). Coupling of chondrocyte intercalation and differentiation was revealed in sox10:eGFP transgenics, where increased GFP fluorescence provides a readout of cartilage differentiation (Fig. 1D–F). A stable arrangement of chondrocytes in PA1 was achieved by 66 hpf. Quantification of chondrocyte morphology in pq revealed that in stacks the cell length to width ratio [LWR] is typically 3.6 +/− 1, with 78% of chondrocytes oriented perpendicular to the long axis of pq (n = 91 cells, 5 embryos) (Fig. 2A, B).
To characterize cell polarity during and after PA1 cartilage morphogenesis, sox10:eGFP transgenics were stained using an anti-gamma tubulin antibody to reveal the positions of microtubule organizing centers (MTOCs) [12], [33]–[35]. This revealed a dynamic pattern of polarity during the cell-cell intercalation period – 48-54 hpf, which stabilized by 66 hpf (Fig. 1D–F). Co-staining of acetylated alpha-tubulin showed that most MTOCs were associated with primary cilia (Fig. S1B). MTOCs in prechondrocytes within pq and Mc were initially oriented towards the center of each condensation (Fig. 1D″-E″). As Mc cell rearrangements stabilized by 66 hpf, three zones of uniform polarity became apparent along its dorsal-ventral (D–V) axis: 1) ventrally oriented near the jaw joint, 2) ventrally oriented near the midline joint with the contralateral Mc, and 3) dorsally oriented throughout the highly stacked region of Mc in between (Fig. 1F, G). In contrast, MTOCs were uniformly ventrally localized in pq at 66 hpf (Fig. 1F, G). Quantification of MTOC orientation in pq revealed 81% of chondrocytes were ventrally polarized (n = 113 cells, 5 embryos, Fig. 2A, B). Additional cryptic polarity reversals were observed throughout the cartilaginous skeleton of arches 2–7 at this stage (Fig. 1G).
To determine if such patterns of cartilage polarity are conserved across vertebrates, we stained MTOCs in Mc in mice. Similar to our data in zebrafish, chondrocytes in vibratome sections of Mc in mouse embryos at stages E12.5 and E13.5 were polarized (81% of MTOCs were ventrally oriented, n = 243, 2 embryos)(Fig. 1H, I and Fig. 2C). Furthermore, a distinct reversal in MTOC orientation was detected near the ventral (distal) end of Mc (Fig. 1H, J). Thus chondrocyte stacking, domains of coordinated polarity and cryptic boundaries between them are conserved in Mc in both fish and mammals.
Cartilage stacking defects were previously reported in PA1 of zebrafish atr2a/rerea−/− mutants [36]. Drosophila Atro interacts with Ft in a common PCP pathway [20]. Therefore, we examined expression of multiple zebrafish Ft-, Ds- and Fj- orthologues – rerea is ubiquitously expressed in the head [36]. We found that fat3, dachsous2 (dchs2) and four-jointed1 (fjx1) are expressed in skeletogenic populations of all pharyngeal arches and the pectoral fin and co-expressed with col2a1, a marker for chondrocyte differentiation, between 48–72 hpf (Fig. 3A–F and Fig. S2A, B). Fat3 expression is also detected at lower levels in cells surrounding skeletogenic areas (Fig. S2C, C′). Both fat3 and dchs2 are expressed in rerea−/− mutants (Fig. S2E, G). In order to visualize Fat3 protein localization, we generated an anti-Fat3 polyclonal antibody using the intracellular domain as epitope. Our antibody revealed that Fat3 localizes to the membranes of chondrogenic and non-chondrogenic pharyngeal NC cells at 54 hpf (Fig. 3G–H′), with stronger signal in non-chondrogenic areas (arrow).
To test for potential roles for Fat/Dchs signaling in cartilage PCP, we investigated prechondrocyte intercalation and polarity in REREa-, Fat3- or Dchs2-deficient embryos in a sox10:eGFP background between 48–66 hpf. Stacking defects were observed in rerea−/− mutants as early as 60 hpf, suggesting a failure in cell-cell intercalation (Fig. 4A–F). Patterns of intercalating cell polarity, as assayed by MTOC position, in pq were generally similar in rerea−/− mutants to siblings at 48–60 hpf - MTOCs were oriented towards the center of the condensation. However, by 66 hpf, most chondrocytes in rerea−/− mutants failed to orient their MTOCs ventrally (Fig. 4F and Fig. 2D).
To test requirements for Fat3 in cartilage morphogenesis we designed two antisense morpholino oligonucleotides (MO) – fat3-MO1 and fat3-MO2, targeting the translation start site and a splice acceptor site, respectively. These MOs caused severe reductions in Fat3 protein levels, as assayed by whole mount immunostaining of PA1 in injected embryos (Fig. 4I, J), and fat3-MO1 suppressed eGFP expression when co-injected with fat3-5′UTR-eGFP mRNA (Fig. 4K, L). Both MOs caused similar cartilage differentiation defects at 66 hpf, with greater severity in dorsal elements, as revealed by loss of fluorescence in sox10:eGFP transgenics under epifluorescence microscopy (Fig. S3B; Table 1). A mixture of fat3-MO1 (23.1 nM) and fat3-MO2 (69.2 nM) gave consistent phenotypes when coinjected with p53-MO (7.7 nM) to eliminate non-specific apoptosis [37], [38] and was used in all subsequent experiments. To quantify defects in prechondrocyte shape and orientation we focused on pq at 66 hpf under confocal microscopy (Fig. 4G): Fat3-deficient cells were significantly less elongated, lost their perpendicular orientation, and coordinated polarity (Fig. 2D).
Similarly, we designed two MOs against distinct regions of the Dchs2 translation start region, dchs2-MO1 and dchs2-MO2. Knock-down efficiency was demonstrated by coinjection of either MO with dchs2-5′UTR-eGFP mRNA, which suppressed eGFP expression (Fig. 4M–O). Both MOs (92.3 nM) caused differentiation defects of the pharyngeal skeleton at 66 hpf (Fig. S3C; Table 1). Similar to Fat3, Dchs2-deficient cells were less elongated and without coordinated orientation or polarity (Fig. 4H; Fig. 2D).
These results demonstrate that REREa, Fat3 and Dchs2 are required for cartilage stacking and polarity, consistent with a role in PCP. Fat3 and Dchs2 are also required for cartilage differentiation.
A key aspect of PCP pathways is their role in coordinating the polarity/behavior of cell populations through cell-cell communication. We thus proceeded to test the non-cell autonomous requirements of REREa, Fat3 and Dchs2 in cartilage morphogenesis and differentiation by chimaeric analysis. WT NC cells from sox10:lyn-tdTomato transgenic donors were transplanted into sox10:eGFP hosts deficient in REREa, Fat3 or Dchs2 using the detailed fate map for the zebrafish gastrula (Fig. 5A). Chimaeras were screened for tdTomato+ NC cells at 24 hpf and raised to 66 hpf, when cartilage polarity and differentiation are spatially and temporally stabilized (see Fig. 1). The length-width ratio (LWR) and orientation of the long axes of cells were measured as assays of intercalation, while MTOC position was used to determine cell polarity.
WT transplants rescued intercalation in rerea−/− embryos (Fig. 2E; Fig. 5B–D). In chimaeras, both cartilage stacking and polarity of rerea−/− cells were rescued when compared with contralateral cartilages in the same animals, serving as an internal control. Notably, WT NC cells rescued stacking and polarity of REREa-deficient cartilage even when small numbers of transplanted WT cells contributed to the pq – 3 out of 5 chimaeras, exemplifying the long-range action of REREa in regulating cartilage morphogenesis.
WT NC transplants similarly rescued cartilage intercalation and polarity, but also differentiation, non-cell autonomously in both Fat3- and Dchs2-deficient embryos (Fig. 2F, G; Fig. 5E-J). WT transplants rescued stacking, polarity and differentiation - even when very small numbers of WT cells contributed to the pq - and long-range rescue was observed in 5 out of 21 Fat3-deficient embryos, and 6 out of 21 Dchs2-deficient embryos.
sox10:eGFP fluorescence is strongly reduced in cartilage precursors of Fat3- or Dchs2-deficient embryos, indicating a differentiation defect. To examine this in more detail we assayed sox9a and col2a1 expression by ISH. Expression of both genes was reduced or absent during the normal time-course of cranial cartilage differentiation (54–72 hpf) in both Fat3- and Dchs2-deficient embryos, with the dorsal arches affected at a greater frequency than ventral arches (Fig. 6A–C). In contrast, sox9a expression appeared spatially expanded in rerea−/− mutants, including ectopic expression in the presumptive jaw-joint interzone (Fig. 6D), as previously reported [36]. These results suggest that Fat3 and Dchs2 promote and REREa represses sox9a expression.
Previous studies have reported stacking and differentiation defects in sox9a−/− mutant embryos [4], which we confirmed by injecting a Sox9a-MO mix into sox10:eGFP embryos [4](Fig. S4). Sox9a-deficient embryos showed stacking and polarity defects comparable to Fat3- or Dchs2- deficient embryos (Fig. 2D). Analysis of fat3 and dchs2 mRNA by ISH revealed that both were reduced in the presumptive pharyngeal skeleton of Sox9a-, as well as Fat3- and Dchs2-deficient embryos (Fig. 6E–H, Fig. S5). These results show that Sox9a is required for the expression of fat3 and dchs2, which in turn positively regulate sox9a expression.
To further investigate the functional relationship between Fat3, Dchs2 and REREa, we assayed sox9a and col2a1 expression in embryos deficient in both REREa and Fat3, or REREa and Dchs2. rerea+/+ and +/− embryos injected with Fat3-MO showed loss or reduction of sox9a/col2a1 expression similar to WT embryos (n = 121)(Fig. 7C, G). In contrast, robust sox9a/col2a1 expression was detected in the presumptive pharyngeal skeleton of all rerea−/− mutants deficient in Fat3 (n = 35) (Fig. 7D, G). This loss-of-function interaction suggests that REREa represses sox9a expression in Fat3-deficient tissues. rerea−/− embryos (n = 27) injected with Dchs2-MO showed reduced sox9a/col2a1 expression, similar to Dchs2-MO injected siblings (n = 47)(Fig. 7E–G).
Drosophila Atro binds Fat [20], as does Atr1 to Fat1 in mouse vascular smooth muscle cell primary cultures [39]. To test if zebrafish REREa/Atr2a binds Fat3 we performed in vitro binding assays with 2 non-overlapping fragments of the full-length Fat3 intracellular domain, together with a subset of the Atro domain of REREa containing the highly conserved Atr-box required for strong Fat-binding [39] (Fig. 8). We found that the N-terminal fragment of the Fat3 intracellular domain bound the Atro domain of REREa (Fig. 8). Our results are consistent with direct interactions between Fat3 and REREa upstream of sox9a expression. Taken together, these results suggest that Fat3 and REREa interact directly in a common pathway regulating sox9a expression. Further, the repressor activity of REREa on sox9a transcription appears to be inhibited by Fat3.
In this study, we show that zebrafish pharyngeal cartilages – which form the blueprint for much of the adult craniofacial skeleton - are composed of polarized arrays of stacked chondrocytes. This neat organization is achieved through cell-cell intercalation and requires Fat3, Dchs2 and REREa (Fig. 9, Table 2). Our chimaeric analyses show that all three factors are required to promote polarized intercalation of prechondrocytes non-cell autonomously and over several cell diameters. While Fat3-Dchs2 bridges may rescue cells in contact with WT transplants in Fat3- or Dchs2-deficient embryos, the long-range rescue we observe likely reflects the activation of an unknown secondary signal that regulates long-range polarity. Furthermore, in our model Fat signaling plays more of a permissive role in cartilage morphogenesis and polarity, since we do not detect intercalation and/or polarity perturbations across boundaries between WT transplants and Fat3-, Dchs2- or REREa-deficient cells. Our results also provide a novel transcriptional link between Fat3/Dchs2/REREa and Sox9 in cartilage morphogenesis: REREa represses sox9a transcription (either directly or indirectly), which is antagonized by its interaction with Fat3, while Dchs2 activates sox9a expression. Sox9a is in turn required for fat3/dchs2 expression.
Whether or not vertebrate Fat/Dchs signaling controls a PCP mechanism analogous to fly epithelia continues to be debated, and whether or not it can do so in a mesenchymal tissue like the skeleton, remains unclear. However, fat3, dchs2 and fjx1 are all expressed in craniofacial skeletal precursors and both Fat3 and Dchs2 deficient cartilages lose coordinated polarity and fail to intercalate (similar to rerea−/− mutants), consistent with a PCP-based model. Further, Fat3, Dchs2 and REREa regulate polarized cell-cell intercalation at long-range, likely through the modulation of a secondary signal, as recently demonstrated for Drosophila Ft/Ds signaling [30]. Lastly, defects in oriented precartilage cell intercalations resemble defects in convergence and extension movements during gastrulation [8], [40], [41], which are considered a hallmark of impaired vertebrate PCP [5].
Sox9 is a well-known regulator of cartilage differentiation, but its roles in morphogenesis remain unclear. Human Sox9 heterozygous mutations cause Campomelic Dysplasia (CD), a condition characterized by several skeletal dysmorphologies including cleft palate and hypoplasia/bending of many endochondral bones [42]. While differentiation defects in CD can be explained by disruption of Col2a1 expression, a direct transcriptional target of Sox9 [43], dysmorphologies of precartilage condensations of Sox9 mutant humans, mice and zebrafish remain unexplained [4], [42]. Our results demonstrate that Sox9a is part of a regulatory loop consisting of Fat3, Dchs2, and the transcriptional co-repressor REREa, which we propose coordinates differentiation and morphogenesis. In our model, Sox9a regulates fat3 and dchs2 levels, which in turn control polarized cell-cell intercalations and also feedback through REREa to allow differentiation. This model is consistent with all of our chimaeric results and points to a novel role for Sox9a in cartilage polarity and stacking.
Our analysis of the intracellular localization of MTOC's/ciliary basal bodies reveals unexpected domains of cartilage polarity throughout the pharyngeal skeleton, both in fish and mice. It also provides a read-out of cell polarity – MTOCs organize the microtubule cytoskeleton, Golgi complex, and primary cilium [5] – that is governed by PCP signaling in gastrulating cells [9], [12] as well as cartilage cells of the digits in mouse limbs [13]. We find that MTOCs orient towards the core of skeletal condensations during cell intercalation, similar to MTOCs in cultured cells [44], and cells converging during gastrulation [12]. In later stages, MTOCs in cartilage relocate to either the dorsal or the ventral side of each stacked chondrocyte in both zebrafish and mice. Notably, distinct zones of ventrally- or dorsally-oriented MTOCs are juxtaposed within individual cartilage elements, and these zones do not correlate with any obvious anatomical boundaries such as joints. Do these switches in MTOC orientation reflect local sources of signaling molecules? Are they instructive for setting up domains of proliferation, local tissue interactions, or muscle-skeleton attachments?
Fat3 is a vertebrate orthologue of Drosophila fat2/fat-like [45], which regulates PCP of actin fibers within the ovarian follicle epithelium, without demonstrated links to Ds or Atro [46]. In contrast, Drosophila ft – the orthologue of vertebrate fat4, regulates PCP in the eye, abdominal cuticle and wing discs, together with Ds, Fj, and Atro [18], [20], [27], [47]. We provide new genetic evidence for a requirement for Fat3, Dchs2 and REREa in a common pathway regulating morphogenesis and differentiation of vertebrate pharyngeal cartilage. A physical interaction between RERE/Atr2 and Fat1 was previously reported in murine smooth muscle cells [39], suggesting that the direct interaction between Fat and Atro orthologues may be generally conserved. Our study also suggests that some form of signal transduction occurs downstream of both Fat3 and Dchs2, with Fat3 modulating REREa activity and Dchs2 activating sox9a transcription. While a prevalent model proposes that Fat-Ds-mediated cell-cell communication is unidirectional, with Ds acting as ligand and Fat as receptor [29], our finding parallels that of recent studies suggesting signal transduction downstream of Ds in Drosophila [48]–[50].
The roles we have found for the Fat/Dchs pathway in cartilage cell intercalation and morphogenesis resemble those of the Fz-PCP pathway that have been described in various vertebrate contexts [5], [7]–[11], including the zebrafish pharyngeal skeleton [51], [52]. However, studies in Drosophila suggest that Fat-PCP and Fz-PCP pathways regulate tissue polarity independently [14]. What seems to set the Fat/Dchs pathway apart in cartilage, based on our findings, is its role in coordinating morphogenesis and differentiation through Sox9. Similarly, in the fly eye, Ft and Atro coordinate R3/R4 photoreceptor fate determination with polarity, suggesting a conserved role for this pathway in coupling polarity and differentiation [18], [20], [53]–[55]. These processes must be coordinated during the development of all tissues and organs, yet are rarely studied together and the underlying mechanisms remain unclear. Future studies are needed to clarify whether or not Fat signaling plays similar roles in other tissues (e.g. cranial neural tube, renal tubules, etc.) where it has been implicated [11], [17].
All animals were handled in strict accordance with good animal practice as defined by the relevant national and/or local animal welfare bodies, and all animal work was approved by the University of California, Irvine Institutional Animal Care and Use Committee.
Adult zebrafish of the *AB strain, carrying alleles of rerea/atr2atb210 (bab) [36], the transgenic reporter line sox10:eGFP [56], [57] or sox10:lyn-tdTomato [32] were maintained and staged as described [31], [58]. CD-1 and Rosa26flox-mTRed-Stop-flox-mGFP mice were maintained as described [59]. Embryonic day 12.5 and 13.5 (E12.5 and E13.5) timed matings were setup by crossing CD1 females (Charles River) with Rosa26flox-mTRed-Stop-flox-mGFPmales [60].
Whole-mount in situ hybridization was carried out as previously described (Thisse et al., 1993). fat3, dchs2 and fjx1 amplicons were amplified from 54 hpf cDNA using the following primers: Fat3f: CTTCATCGCCTTCAGGAAGA, Fat3r: GGCGGGTAGTCAC TGTCAAT, Dchs2f: CCGAGGAAGAGACAGCAGAGG, Dchs2r: CGTATTCCTGGCTGGGCA AC, Fjx1f: GAGCAGCGGGTGTTCTGGAG, Fjx1r: CATCAATCCTGCTCTGCAATGTG. Each amplicon was subcloned into pCR4-TOPO (Invitrogen) following manufacturer's instructions, transformed into BL21 competent cells and sequenced. Published probes include col2a1 [61] and sox9a [62].
Immunohistochemistry was performed with rabbit anti-gamma-Tubulin (1/100 Genetex GTX113286), mouse anti alpha-acetylated Tubulin (1/100 Sigma T6557) and Alexa Fluor 568 Phalloidin (1/50 Life technologies A12380). Fixed mouse heads were embedded in 5% agarose and 100 µm sections were cut on a vibratome.
Visualization of mouse embryonic cartilages was achieved by performing alcian blue staining as previously described [63]
Morpholino antisense oligos (MO) were designed to block translation or splicing (Gene Tools, Inc.) including: fat3-MO1, 5′-CCTTCACCTGTGCAAACAGAGAACA-3′; fat3-MO2, 5′-TGCCCTCTTGCTCAGTTCGGCTCAT-3′; dchs2-MO1, 5′-CATGTTCATGC-GAAAACATTAGCAG; dchs2-MO2, 5′-AGAAAGTCCGTGTGTAAAACTCCAT-3′; Sox9a i1d, 5′-AATGAATTACTCACCTCCAAAGTTT-3′ [62]; Sox9a i2d, 5′-CGAGTCAAGTTT-AGTGTCCCACCTG-3′ [62]. MOs were prepared at 1 mM in dH2O and stored at room temperature. To construct fat3-5′UTR-eGFP and dchs2-5′-UTR-eGFP reporter genes, cDNA amplicons containing target sites of MOs were subcloned in frame into the pCS2-eGFP vector. MO- and mRNA injection volumes were approximately 500 pL. A 4 hr developmental delay was usually observed with MO-injected- and rerea−/− embryos, which was corrected for throughout the study.
Embryos labeled by in situ hybridization were photographed on a Zeiss Axioplan 2 microscope, equipped with a MicroPublisher 5.0 RTV camera using Volocity software (Improvision). Fluorescent immunostained embryos were photographed on a Zeiss LSM780 confocal microscope using a 63x/1.15 W C-APO objective. For time-lapse imaging, embryos were imaged on a Nikon Eclipse Ti spinning disk microscope equipped with a 40x/1.15 WI Apo LWD objective. Approximately 100 um z-stacks were captured at 0.5 um intervals every 5 minutes for 8 hours. ImageJ/Fiji was used for image processing. Cell contours were hand-drawn in ImageJ and measured for length-width ratio (LWR) and orientation. Each cell was divided into 4 quadrants to determine MTOC position. Cell orientation and MTOC position were plotted as rosette diagrams and Watson's U2 tests for significance were conducted using Vector Rose (PAZ software). In Fat3-MO or Dchs2-MO embryos, LWR and orientation were recorded within a 3-to-4 cell-thick presumptive palatoquadrate (pq) region bordered posteriorly by the mandibular aorta, dorsally by the adductor mandibularis muscle (amm) and anteriorly by the presumptive jaw joint – at amm mid-length.
WT, sox10:lyn-tdTomato donor embryos were injected with 3% rhodamine-dextran at the 1-2-cell stage and cells were transplanted into sox10:eGFP hosts at the shield stage (6 hpf). Host embryos with red fluorescent cells in the pharyngeal arches were sorted at 24 hpf and reared up to 66 hpf for immunochemistry. rerea−/− mutant hosts were identified by lack of pectoral fins and eye coloboma at 66 hpf.
For the GST pull-down assay, 2 fat3 fragments were PCR-amplified and cloned in frame with the N-terminal Myc tag provided in the pCS2-MT vector. Myc-Fat3 N (aa 4184-4366) and C (aa 4357–4500) were then synthesized by in vitro translation using TNT Quick (Promega). GST-REREa (aa 1007–1227) was produced by PCR-amplification and cloned in frame into pGEX-4T-1 (Amersham) GST. GST-REREa protein was produced in BL21 cells, extracted in PBS with protease inhibitor cocktail (Sigma) and purified using glutathione-coupled beads (GE healthcare).
For binding assays, equal amounts of Myc-Fat3 were added to 100 µL of pre-equilibrated beads containing GST-fusions in HMK buffer [64] and rotated at room temperature for 3 hours. Beads were recovered, washed in HMK and analyzed by SDS-PAGE gels followed by western blotting with ECL chemoluminescence (Amersham). The following primary antibodies were used: mouse anti-Myc (a gift from J. Sosnik, 1∶5000), rabbit anti GST (GeneScript 1∶10,000).
|
10.1371/journal.ppat.1005846 | Evidence and Role for Bacterial Mucin Degradation in Cystic Fibrosis Airway Disease | Chronic lung infections in cystic fibrosis (CF) patients are composed of complex microbial communities that incite persistent inflammation and airway damage. Despite the high density of bacteria that colonize the lower airways, nutrient sources that sustain bacterial growth in vivo, and how those nutrients are derived, are not well characterized. In this study, we examined the possibility that mucins serve as an important carbon reservoir for the CF lung microbiota. While Pseudomonas aeruginosa was unable to efficiently utilize mucins in isolation, we found that anaerobic, mucin-fermenting bacteria could stimulate the robust growth of CF pathogens when provided intact mucins as a sole carbon source. 16S rRNA sequencing and enrichment culturing of sputum also identified that mucin-degrading anaerobes are ubiquitous in the airways of CF patients. The collective fermentative metabolism of these mucin-degrading communities in vitro generated amino acids and short chain fatty acids (propionate and acetate) during growth on mucin, and the same metabolites were also found in abundance within expectorated sputum. The significance of these findings was supported by in vivo P. aeruginosa gene expression, which revealed a heightened expression of genes required for the catabolism of propionate. Given that propionate is exclusively derived from bacterial fermentation, these data provide evidence for an important role of mucin fermenting bacteria in the carbon flux of the lower airways. More specifically, microorganisms typically defined as commensals may contribute to airway disease by degrading mucins, in turn providing nutrients for pathogens otherwise unable to efficiently obtain carbon in the lung.
| Persistent CF lung infections are composed of hundreds of microbial taxa whose interactions contribute to respiratory failure. Despite their importance, the complex interplay between the lung microbiota and host environment is poorly understood. For example, the nutrients that sustain bacterial growth in vivo, and how those nutrients are derived, are not well characterized. We reveal that a subset of CF microbiota is capable of fermenting mucins for carbon and energy which, in-turn, can support the carbon demands of other respiratory pathogens in co-culture. Moreover, we show that metabolites consistent with mucin fermentation are abundant within airway secretions, highlighting a potential key role for fermentative metabolisms in CF lung disease. A thorough understanding of pathogen ecology in the CF airway and the nutritional dynamics that sustain their growth will have important implications for the design of new therapeutic strategies and the management of disease progression.
| Mutations in the gene encoding the cystic fibrosis transmembrane conductance regulator (CFTR) protein cause an imbalance of ion transport that leads to mucus hyperviscosity and impaired mucociliary clearance [1]. Within the airways, prolonged residence time of mucus provides a stagnant nidus for chronic bacterial infections–the predominant cause of mortality in CF patients [2]. Traditionally, culture-based studies have focused on a small number of taxa associated with CF lung disease (e.g. Pseudomonas aeruginosa, Staphylococcus aureus), however, culture-independent surveys of the CF lung microbiome have revealed a far more complex bacterial community than previously appreciated [3–5]. While the temporal dynamics of these communities and their association with disease states have been studied in detail, the in vivo host environment, and microbial metabolism therein, is relatively understudied.
For example, the means by which CF pathogens obtain sufficient energy for growth is not known. Bacterial numbers within the CF lung can reach 108−109 cells gm-1 of sputum [6], which are comparable to densities found within the distal colon [7]. However, unlike the gut where dietary sources provide a constant influx of nutrients, carbon within the airways must be predominately host-derived. The respiratory tract contains a number of host compounds that can be used by microbes as nutrient sources, including immunoglobulins, cytokines, defensins and lactoferrin [8,9], yet these are unlikely to be present at concentrations to support the dense microbiota of the CF airways [10–13]. Additionally, studies of CF sputum from adult patients have shown an abundance of small molecules that can support the growth of pathogens in vitro, including sugars, fatty acids, phospholipids, and amino acids [14–17]. However, the mechanism by which these compounds reach high abundance in airway mucus remains poorly defined.
The accumulation of mucus secretions in the CF airways represents an abundant nutrient source. The major macromolecular constituents of mucus, mucins, are a large reservoir of both carbon and nitrogen, and have been measured at concentrations of up to 10 g L-1 in sputum [18]. Mucins are high molecular weight (2–20 x 105 Da) glycoproteins composed of an amino acid backbone with O-linked oligosaccharide side chains that form 50–90% of the molecular mass [19]. Carboxyl and sulfate groups decorate their terminal sugars conferring a net negative charge, while terminal cysteine-rich domains form disulfide bonds with neighboring polymers, forming a highly cross-linked gel-like structure that is resistant to rapid bacterial degradation [20].
Despite their recalcitrance, mucins are a main nutrient source for niche-specific microbiota of the gut and oral cavity. For example, oral streptococci produce a variety of glycolytic and proteolytic enzymes that liberate bioavailable carbohydrates from salivary glycoproteins [21]. Few single species are known that can completely degrade mucins when grown in monoculture [22], though specific consortia of oral bacteria have been shown to co-operatively degrade both the polysaccharide and peptide structures of salivary mucins [21,23]. In turn, these primary degraders are thought to modify the nutritional landscape of the oral cavity and stimulate the growth of secondary colonizers [24]. Similar interactions between commensal gut microbiota and the mucosal layer of the large intestine are well known [25]. By contrast, very little is known about the degradation of airway mucins by opportunistic pathogens and their role as a nutrient source in vivo [26].
Here, we investigated the role of mucins in the carbon flux of the CF airways and characterized their potential to stimulate pathogen growth. Our results demonstrate that P. aeruginosa uses mucins inefficiently in monoculture. This lack of growth raised the question: can other members of the CF microbiota degrade mucins and alter the nutritional reservoir available for pathogens? We subsequently found that co-culture of pathogens with an anaerobic bacterial consortium composed of taxa commonly found in the lower airways [3,5,27,28] can facilitate robust pathogen growth using mucins as a sole carbon source. We also confirmed that fermentation-derived metabolites are abundant within expectorated CF patient sputum, consistent with previous studies [14,29]. Finally, we found that genes required for the catabolism of mucin fermentation byproducts are highly expressed by P. aeruginosa in vivo. Taken together, these data support a central ecological role for commensal anaerobes in the nutritional dynamics of the lower airways and the progression of CF lung disease.
We first wanted to determine if mucins alone could sustain the growth of P. aeruginosa. To do so, we assayed strain PA14 for growth in a defined minimal medium containing intact mucins from porcine gastric mucin (PGM) as the sole carbon source (Fig 1A). PGM was first dialyzed and filtered to remove impurities and small metabolites that could potentially support growth (see Methods). Interestingly, 15g L-1 of purified mucins (sugar equivalent of ~65mM glucose assuming 80% sugar by weight) only resulted in a moderate OD600 gain of 0.1 after 24 h. By contrast, PA14 grew to 0.65 OD on glucose (13mM) alone, which underscores the inability of PA14 to efficiently break down and utilize PGM. Recognizing that PGM and human respiratory mucins (MUC5AC and MUC5B) are structurally diverse [30], we also assayed Pseudomonas growth in a minimal medium containing purified MUC5B as the sole carbon source (S1 Fig). PA14 reached a two-fold lower density when utilizing MUC5B relative to PGM, suggesting that P. aeruginosa PA14 cannot efficiently utilize complex mucin glycoproteins, including the most abundant mucin of the lower airway.
Given that PA14 was isolated from a burn wound, we then reasoned that P. aeruginosa clinical isolates derived from a mucin-rich sputum environment would have enhanced ability to degrade and utilize mucins as a growth substrate. To test this, clinical isolates derived from CF patients at various stages of disease were grown in minimal mucin medium. Growth yield was assayed long past the onset of stationary phase (48h) to account for slow growth phenotypes. Under these conditions, each isolate achieved a moderate density on PGM, comparable to PA14 (final OD600 between 0.06 and 0.19)(Fig 1B). Growth on mucin was also compared to growth with glucose alone, in addition to growth with a complete amino acid supplement to account for auxotrophy, a common trait among CF lung isolates [31]. Indeed, isolates JMF2, JMF3, and JMF5 grew poorly on glucose, suggesting auxotrophies that were corrected by the supplement. The ability of these isolates to grow on glucose supplemented with amino acids to a greater extent than the more nutrient-dense mucin medium suggests that the lack of robust growth on PGM was not due to slow-growth phenotypes. Rather, the fact that each isolate grew when sufficient carbon was made available to them demonstrates that P. aeruginosa has a general inability to efficiently utilize complex mucin glycoproteins in monoculture.
The inefficiency of P. aeruginosa to use mucins as a growth substrate motivated us to consider recent culture-dependent and culture-independent studies of CF microbiota for insights into the carbon flux of the airways. Notably, obligately anaerobic taxa have gained recent attention for their abundance in expectorated sputum, bronchoalveolar lavage fluid and explanted lungs [5,27,28,32–34]. A role for these organisms in CF disease has not yet been established; however, some have been characterized for their ability to degrade and ferment salivary mucins in the oral cavity [21,23,35]. Based on these observations, we hypothesized that oral anaerobes, once aspirated into the lower airways, could alter the nutrient pool by degrading respiratory mucins. More specifically, we predicted that the degradation and fermentation of mucin glycoproteins would liberate sugars, amino acids and short chain fatty acids (SCFAs), all of which are abundant components of CF sputum that are readily utilizable by P. aeruginosa [36].
To test whether fermentative bacteria are able to generate metabolites from mucin that could simultaneously stimulate CF pathogen growth, a saliva-derived bacterial community was first enriched on PGM (S2 Fig). This enrichment culture was then used to inoculate an anaerobic minimal mucin medium supplemented with 1.0% agar to mimic a high-viscosity sputum gel [37] (Fig 2A). Once the lower agar phase containing the mucin-enriched bacterial community had solidified, PA14 was suspended in buffered 0.7% agar medium without mucin (i.e. no carbon source) and was placed in the upper portion of the tube. This experimental setup a) establishes an oxygen gradient allowing anaerobes to grow, and b) restricts the movement of microbes but allows metabolites to freely diffuse. Under these conditions, P. aeruginosa would be expected to achieve a higher cell density if provided with diffusible growth substrates from the lower phase.
Co-culture growth was monitored over a 72h period. After 24h, turbidity was noticeable in the lower phase and a diffusible blue-green pigment (pyocyanin) characteristic of P. aeruginosa growth was observed throughout the co-culture tube (Fig 2B). By contrast, no observable pigment was produced in the absence of oral anaerobes. Colony counts of the upper phase revealed that P. aeruginosa reached its maximum density after 48h of co-culture (Fig 2C), and achieved an order of magnitude (10X) increase (p = 0.007) in cell density relative to the tubes in which oral anaerobes were omitted. These data demonstrate that while PA14 can achieve a moderate density using PGM as a sole growth substrate (consistent with Fig 1A), its growth and production of a known virulence factor is significantly enhanced via mucin breakdown and cross-feeding by oral-associated anaerobes.
We then tested the growth of several P. aeruginosa clinical isolates in addition to the CF-associated pathogens Achromobacter xylosoxidans, Burkholderia cenocepecia, Stenotrophomonas maltophilia and Staphylococcus aureus using the agar co-culture model (Fig 2D and 2E). For each bacterium, a notable increase in growth over the inoculum was observed in the presence of mucin fermenters after 48h. With the exception of JMF3, the final cell density was significantly higher for each co-culture relative to monoculture tubes in which anaerobes were omitted. Collectively, these data support the hypothesis that oral-derived microbiota can serve as primary mucin degrading organisms, in turn liberating metabolites that stimulate the growth of P. aeruginosa and other CF lung pathogens.
To determine if there exists a fraction of the CF lung microbiota that has the ability to degrade and ferment mucin glycoproteins, we then performed mucin enrichment experiments on expectorated sputum from 14 stable, non-exacerbating CF patients. To do so, a small fraction of sputum was used to inoculate an anaerobic culture medium supplemented with PGM as the sole carbon and nitrogen source. Following anaerobic growth, genomic DNA was isolated from the initial sputum sample and the corresponding enrichment culture followed by bacterial 16S rRNA gene sequencing to identify enriched taxa. If taxa from sputum become enriched in a medium with mucin as the sole carbon source, it would demonstrate the presence of mucin-degradation capacity in the lower airway environment. As expected, sputum microbiota (prior to enrichment) was highly variable between patients (Fig 3A), and Pseudomonas made up the highest percentage of sequence reads with a per-patient average of 31.3% across the cohort (Fig 3B). Notably, taxa previously characterized for their mucin-degrading activity (Prevotella, Veillonella, Streptococcus and Fusobacterium) also made up 35.1% of the population (11.4%, 9.7%, 9.9% and 4.1% of normalized sequence reads, respectively).
Post-enrichment, sputum-derived communities were predominated by fermentative organisms commonly associated with both the oral cavity and lower airways (Fig 3A and 3B). On average, enrichment communities were composed of 66% of taxa known to have salivary mucin degradation ability (27.4% Prevotella, 19.2% Veillonella, 10.7% Streptococcus, 8.4% Fusobacterium) with all other genera present at 4% or below (Fig 3B). Lachnoanaerobaculum and Prevotella were significantly enriched while Neisseria, Staphylococcus, and Pseudomonas were selected against (p<0.05). The relative abundance of mucin-fermenting taxa in the initial sputum samples and their ability to grow on mucin in vitro suggests a suitable niche space exists in the CF lung for mucin-fermenting anaerobes.
We then tested representative isolates of the four most abundant mucin-enriched species for their ability to cross-feed P. aeruginosa PA14 (Fig 3C). Using the co-culture model, mucin degradation by P. melaninogenica and V. parvula alone did not stimulate an increase in P. aeruginosa growth relative to anaerobe-free controls. By contrast, F. nucleatum and S. parasanguis both supported a significant increase in PA14 growth; however, final PA14 cell density achieved in the presence of any individual mucin-fermenter was significantly less (p<0.0001) than when all four mucin-degraders (P.m., V.p., F.n. and S.p.) were added at an equal density. Notably, the PA14 cell density achieved with this defined, four-species anaerobic consortium was comparable to growth achieved using the undefined enrichment culture (S2 Fig) used in Fig 2. These data demonstrate that while P. aeruginosa growth can be stimulated by select sputum-enriched anaerobes individually, the collective activity of a mucin-degrading consortium associated with the lower airways results in a significantly higher growth yield.
To study the cross-feeding ability of mucin fermenting bacteria in further detail, we then characterized the metabolites that were generated during sputum enrichment. Fermentative anaerobes are known to produce mixed acid metabolites so we quantified a number of organic acids via high performance liquid chromatography (acetate, butyrate, citrate, formate, isobutyrate, isovalerate, ketobutyrate, ketoisovalerate, lactate, 2-methylbutyrate, propionate, succinate and valerate), and quantified amino acid production using gas chromatography. Enrichment cultures generated detectable levels of only three short-chain fatty acids (SCFAs) measured: high amounts of acetate (30.2 +/- 9.7 mM) and propionate (15.4 +/- 3.8 mM) were found in each sample, while only one enrichment sample containing an abundance of Streptococcus sp. (#7025, see Fig 3A) produced a detectable concentration of lactate (7.8 mM) (Fig 4A). No other SCFAs assayed were detectable in the mucin-enrichment supernatants.
Not surprisingly, free amino acids were also present in each sample (1.37 +/- 1.32 mM total) (Fig 4B), most likely due to their liberation from the mucin polypeptide backbone. The abundance of amino acids correlates well with previous studies of sputum composition where they were found to be present in appreciable quantities [15]. Altogether, these data demonstrate that acetate, propionate, and amino acids are the major byproducts of mucin fermentation by sputum-derived anaerobes in vitro, suggesting that they may be carbon sources that are bioavailable to pathogens in vivo.
To assess the contributions of propionate and acetate to P. aeruginosa growth in our in vitro cross-feeding model, we generated mutants lacking the genes encoding AcsA (acetyl-coA synthetase) and PrpB (methylisocitrate lyase) that are required for the catabolism of acetate and propionate, respectively. We also generated a double mutant (ΔprpBΔacsA) that was defective in both pathways. Each mutant grew normally on glucose, yet demonstrated a predictable growth limitation when its cognate substrate was provided as the sole carbon source (S3 Fig). Mutants were then tested for their ability to grow in the agar co-culture model with the oral-derived anaerobic consortium (S2 Fig) provided as the mucin-degrading inoculum. ΔacsA, ΔprpB, and ΔacsAprpB demonstrated significant growth defects relative to PA14 (p<0.05) when grown in co-culture with mucin fermenters (Fig 5). These data demonstrate that P. aeruginosa growth is partially dependent on acetate and propionate catabolism in a model bacterial community characteristic of the CF lower airways.
Finally, to provide evidence of fermentative activity in vivo, we used two complementary techniques to study bacterial metabolism within expectorated sputum: mass spectrometry and quantitative reverse transcription PCR (qRTPCR). First, using gas chromatography mass spectrometry (GC-MS), acetate and propionate were quantified in paired saliva/sputum samples collected from 7 stable CF patients. Given the presence of fermenting taxa and SCFA previously identified in saliva [38], an oral rinse was first performed prior to sample collection to reduce the residual metabolite background from the mouth (see Methods). SCFAs were found at millimolar concentrations (5.9 +/- 1.8 mM and 48.2 +/- 47.2 μM, for acetate and propionate, respectively) in all sputum samples. Despite our washing effort, the saliva sample also contained detectable levels of both acetate and propionate. Acetate concentrations were significantly higher in sputum relative to saliva (5.9 versus 2.4 mM, p = 0.004)(Fig 6A). Propionate, on the other hand, showed no significant difference (p = 0.89) between sample pairs (Fig 6B). Thus, we cannot rule out propionate contamination from the oral cavity. However, given that the propionate concentration in saliva was not diluted compared to sputum, it is reasonable to approximate that sputum contains a comparable concentration of propionate to the saliva samples. Our results are supported by recent reports of acetate and propionate in CF bronchoalveolar lavage fluid [14,29]. The ratio of propionate to acetate (1:100) in our samples was unexpected given the much higher ratio (1:2) generated during in vitro mucin fermentation. This disparity may suggest that propionate is either produced at low levels in vivo, or that propionate, over acetate, is preferentially consumed by CF microbiota in a cross-feeding relationship.
As a complement to our GC-MS measurements, we then used qRTPCR to assess whether Pseudomonas senses and responds to acetate and/or propionate within the airways. As a proxy of the use of these mucin-derived metabolites by P. aeruginosa, we targeted the expression of both acsA and prpD in sputum relative to their expression levels under controlled laboratory conditions. In vitro, acsA and prpD were differentially expressed by both PA14 and JMF5 in the presence of acetate (4.0-fold higher, p = 0.01) and propionate (10.8-fold, p<0.001), respectively, relative to growth on glucose alone (Fig 7). When compared to in vitro cultures, analysis of sputum revealed that prpD (5.4-fold, p = 0.005) but not acsA (no change, p = 0.73) was highly expressed. Expression of prpD was highly variable, however this was not unexpected given the variable nature of patient samples. These qRTPCR data are consistent with the active catabolism of propionate by Pseudomonas in vivo, and may account for the disparity seen in propionate:acetate ratios between sputum and mucin enrichment cultures. If SCFAs were simultaneously produced and consumed by the airway bacterial community, we would expect the concentration of these metabolites to remain low, and genes required for their catabolism to be highly expressed. Though we cannot rule out the possibility of low rates of propionate production in vivo, the expression of prpD by Pseudomonas within sputum is a reliable biomarker of the presence of mixed-acid fermentation byproducts. It is noteworthy that acetate catabolism via AcsA is important to PA14 yields during cross-feeding in vitro, yet the qRTPCR data suggest that acsA is minimally expressed in vivo. This discrepancy may be due to strain-specific differences in PA14 compared to clinical isolates, or may be influenced by unknown environmental cues in vivo.
Collectively, data presented here demonstrate the presence of fermentation metabolites and evidence consistent with their catabolism in the airways of CF patients. These results, coupled with previous reports of SCFA in bronchoalveolar lavage fluid [14,29] and degraded mucins in CF sputum [18,39], provide compelling evidence that anaerobic organisms can contribute to CF lung disease by degrading mucins and consequently providing utilizable substrates for opportunistic airway pathogens.
In this study, we investigated the nutritional role of mucins in the growth of P. aeruginosa in the CF airways. While Pseudomonas was found to inefficiently utilize mucins as a carbon source on their own, we determined that mucin fermentation by oral anaerobes can stimulate the growth (10X) of P. aeruginosa and other opportunistic CF pathogens. Moreover, we revealed that in vitro mucin fermentation generated high concentrations of SCFAs and amino acids, which were also abundant and bioavailable for P. aeruginosa within patient sputum. Together, these results suggest that the high levels of utilizable metabolites present in sputum reported previously [14,15,29] may be derived from bacterial mucin degradation. In this context, fermentative anaerobes aspirated from the oral cavity that become established in the lower airways may play a central role in the progression of CF lung disease.
Expectorated sputum and many of its specific constituents–lipids [15,17], amino acids [16], and modified sugars [40], for example–are known to support bacterial growth in vitro and have been studied in detail. Yet, how the majority of these compounds are made available within the CF airways has not been defined. Mucins represent an abundant source of both amino acids and sugars, and play a key role in shaping the microbial community structure of the gastrointestinal tract. However, the process of mucin degradation, and its potential contribution to airway disease has not been addressed in detail. Previous studies have shown that mucins can support the carbon demands of P. aeruginosa in vitro [41,42]; however, these studies included autoclaved preparations of commercial porcine gastric mucin (PGM) that contain low molecular mass compounds that are readily utilized. In fact, when PGM preparations were filtered and dialyzed in our study (leaving only large, intact glycoproteins) appreciable growth of P. aeruginosa isolates was not observed.
The near ubiquitous presence of mucin-fermenting anaerobes in sputum [5, 27, 28, this study], coupled with numerous studies demonstrating that the respiratory mucins MUC5B and MUC5AC are degraded in CF patients compared to healthy controls [18,39] supports the idea that bacterial mucin degradation is commonplace within the CF airways. Streptococcus sp., for example, which were consistently abundant in mucin-enriched sputum cultures, have been extensively characterized for their ability to degrade salivary mucins [21]. By doing so, oral streptococci modify the nutritional landscape of the oral cavity and stimulate the growth of secondary colonizers [24]. Here we demonstrate that by degrading mucins, commensal anaerobes can also stimulate the growth of opportunistic pathogens found within the respiratory tract, supporting an ecological role for mucin-fermenting anaerobes in the development of CF airway infections.
In vitro, degradation of glycan sugars and the polypeptide backbone of mucin by sputum-derived anaerobes generated an abundance of SCFAs and amino acids. Consistent with recent studies [14, 29], SCFAs were also found in CF patient sputum. Because SCFAs serve as a reliable biomarker of fermentative metabolism [43], the universal presence of SCFAs across our patient cohort provides strong supporting evidence for the existence of fermentation within the CF airways. Specifically, our data demonstrate that the fermentation of mucins generates the same metabolites that have been shown to support growth of Pseudomonas in sputum [15]. Moreover, the expression of prpD in sputum suggests that P. aeruginosa is both sensing and utilizing propionate in vivo; however, the significance of propionate catabolism in disease is not known. While propionate can support growth of P. aeruginosa, it is also a potent microbial inhibitor [44]. As such, we do not yet know if the in vivo degradation of propionate helps to satisfy the carbon requirements of P. aeruginosa or whether it is being selectively degraded for detoxification purposes.
Altogether, our data point to a compelling model for the role of oral anaerobes in the development of CF lung disease (Fig 8). In this model, opportunistic pathogens that cannot degrade mucins (e.g. P. aeruginosa, S. aureus) do not become established in the lower airways until mucin-fermenting bacteria have colonized. In healthy subjects, anaerobes that are routinely aspirated into the lower airways encounter functional host defenses and are effectively cleared. In CF, however, impaired mucociliary clearance and defective immunological responses [2] increase the likelihood of oral anaerobe colonization (Fig 8A and 8B). Their increased residence time allows anaerobes to degrade respiratory mucins and condition the lung environment into a niche that is suitable for pathogens to thrive (Fig 8C). Several lines of clinical evidence exist in support of our proposed model: (i) pediatric patients often have asymptomatic primary colonization by oral anaerobes [45,46] prior to the establishment of chronic P. aeruginosa infections, (ii) routine administration of broad-spectrum antibiotics in the absence of respiratory infection symptoms is an effective therapy to delay the onset of colonization by P. aeruginosa and reduce the frequency of acute exacerbations [47,48], and (iii) the in vitro antibiotic susceptibility of lung pathogens does not always correlate with clinical outcomes [49]. In the latter instance, we propose that antibiotics do not solely target the pathogen, but rather disrupt the complex metabolic interactions that supply them with substrates for growth and stimulate their pathogenicity.
While this work suggests a role for mucin fermenters in the progression to pathogen colonization, it also raises a number of questions. Most importantly, it does not address the role of oral anaerobes or bacterial nutrient acquisition in late stages of CF disease (Fig 8D). In chronic airway infections, bacterial diversity often declines and lung microbiota becomes predominated by P. aeruginosa [50,51]. Therefore, it is probable that P. aeruginosa does not rely upon mucin fermentation in late stages, but rather contributions from the host inflammatory response. It is known that chronic airway infections can lead to a ‘leaky’ lung epithelium that allows bulk plasma, with an abundance of bioavailable metabolites, to reach the epithelial surface [52]. Additionally, persistent P. aeruginosa infections incite a neutrophil-dominant inflammatory response that is associated with increased concentrations of nutrients and proteases in the airway milieu. Human neutrophil elastase, for example, is capable of cleaving mucins glycoproteins into free amino acids, and has been implicated in the progression of lung disease [39]. Given these other potential nutritional sources, it is probable that P. aeruginosa uses multiple carbon sources in vivo as infections progress, and we suspect that pathogens may become independent from mucin fermenters in late stages of CF lung disease.
This study underscores the importance of identifying the underlying microbial ecological dynamics that give rise to CF lung disease progression. In particular, our data warrant further studies of targeted therapies towards fermentative organisms and their metabolisms that may contribute to the establishment and progression of chronic airway disease. In a broader context, the presence of oral anaerobes in other, mucus-rich disease environments where both oral anaerobes and P. aeruginosa are major players–chronic obstructive pulmonary disease, sinusitis, and ventilator-associated pneumonias–suggests that mucin fermentation and metabolic cross-feeding may be a more widespread phenomenon. We are currently investigating this interspecies dynamic in a range of infectious contexts.
Bacterial strains and primers are shown in S1 Table. P. aeruginosa PA14 and Staphylococcus aureus MN8 were obtained from D.K. Newman (California Institute of Technology). S. parasanguis ATCC 15912 was obtained from M.C. Herzberg (University of Minnesota). F. nucleatum subsp. nucleatum ATCC 25586, P. melaninogenica ATCC 25845, and V. parvula ATCC 10790 were obtained from Microbiologics (St. Cloud, MN). Clinical strains of P. aeruginosa, B. cenocepecia, S. maltophilia, and A. xylosoxidans [53] were isolated from patients enrolled in this study. Aerobes were routinely cultured in Luria Bertani medium or a minimal mucin medium containing 60mM KH2PO4 (pH 7.4), 90mM NaCl, 1mM MgSO4, 15 g L-1 porcine gastric mucin (PGM; Sigma), and a trace minerals mix described elsewhere [54]. During preparation, PGM was first autoclaved, dialyzed using a 13 kDa molecular weight cutoff membrane, clarified by centrifugation, followed by passage through a 0.45 μm Millipore syringe filter to sterilize and isolate soluble intact glycoproteins. MUC5B was used in place of PGM where specified, though was used sparingly due to its purification difficulty [55]. Glucose (13mM), NH4Cl (60mM), acetate (20mM) and propionate (20mM) were supplemented where specified. MEM essential and non-essential amino acid mixes (Sigma) were added at a final concentration of 0.5X the manufacturer suggested concentration. S. parasanguis, F. nucleatum, P. melaninogenica, and V. parvula were cultured in Brain-Heart Infusion media supplemented with hemin (0.25 g L-1), vitamin K (0.025 g L-1) and laked sheep’s blood (5% vol/vol)(BHI-HKB). Growth was monitored in 96-well plates (Corning) in 250 μL volumes using a BioTek Synergy H1 plate reader. For enrichment growth, approximately 100 μL of sputum was then used to inoculate minimal mucin medium under anoxia (95% N2, 5% CO2) and the remainder was frozen at -80°C. Following 48h of incubation at 37°C, 100 μL of culture was used to inoculate a second anaerobic culture tube and allowed to grow for 48h at 37°C. Genomic DNA was isolated from both the initial sputum sample and enrichment cultures, and bacterial community composition was determined using 16S rRNA gene sequencing.
Forty-eight adult participants with CF were recruited at the University of Minnesota Adult CF Center. Inclusion criteria were a positive diagnosis of CF and ability to expectorate sputum. For enrichment cultures, sputum was expectorated following an oral rinse into 50mL conical tubes, placed on ice, and processed within three hours. Sputum used for qRTPCR analysis was placed in RNALater (Sigma) immediately following expectoration, and stored at -80°C. For metabolite analysis (n = 7), subjects first performed an oral rinse, followed by paired saliva and sputum collection and immediate storage at -80°C. To obtain clinical isolates, sputum aliquots were cultured on Pseudomonas Isolation Agar (Oxoid) for 72 hours at 37°C. Colonies were screened using PCR, sequenced to confirm identity, and stored in 15% glycerol at -80°C.
Sputum and enrichment cultures were thawed to room temperature, and 500 μL of each sample was used for genomic DNA extraction using the PowerSoil DNA isolation kit (MoBio, Carlsbad, CA). Purified DNA was submitted to the University of Minnesota Genomics Center (UMGC) for 16S library preparation using a two-step PCR protocol described previously [56]. A defined mock community was also submitted for sequencing, as were water and reagent controls that did not pass quality control step due to 16S rRNA gene content below detection thresholds. Raw sequence reads were obtained from UMGC and analyzed using a QIIME [57] pipeline developed by the UMGC. The average number of reads per sample after filtering and taxonomic assignment was 2.2 x 105, with the minimum and maximum reads per sample of 5.6 x 103 and 5.1 x 105, respectively. Read pairs were stitched together and 16S amplicon primers were removed using PandaSeq (version 2.7)[58]. Fastq files were merged and sequence IDs converted to QIIME format using a custom perl script. Chimeric sequences were detected using the QIIME (version 1.8.0) script identify_chimeric_seq.py function, using the usearch61 method. Open reference OTU picking was performed using the pick_open_reference_otus.py script, using the usearch61 method and the Greengenes 13_8 16S rRNA reference database [59] clustered at 97% similarity. MetaStats was used to detect differentially abundant features of the mucin-enriched bacterial community [60].
Mucin fermentation cultures (diluted 1/100) were used to inoculate 2 mL of mucin (15g L-1) minimal medium in molten 1.0% agar at 50°C in a glass culture tube under anoxic conditions (Fig 2A). Mucin fermentation cultures contained one of the following: (1) individual anaerobic species (V. parvula, F. nucleatum, P. melaninogenica, or S. parasanguis) grown from single colony picks, grown overnight in anaerobic BHI-HKB broth and washed twice with PBS; (2) a defined four-species anaerobic consortium (V.parvula, F. nucleatum, P. melaninogenica, and S. parasanguis together) prepared as described above; or (3) a saliva-derived mucin-fermenting bacterial community (S2 Fig) revived from glycerol freezer stocks and passaged twice in mucin minimal medium. Upon solidification of the mucin-fermenting fraction, an additional 1 mL of molten minimal medium agar (without mucin) was inoculated with a 1/1000 dilution of an overnight culture of P. aeruginosa (or other strains where specified). Inoculum sizes are shown in S2 Table. Samples were then poured over the mucin fermenting community fraction and allowed to solidify. Tubes without mucin-fermenters were used as negative controls. After solidification, co-cultures were placed at 37°C for 48 h or other specified time points. Agar plugs were then removed from the upper phase and homogenized by pipetting in 10 mL of phosphate buffered saline. Colony forming units per tube were determined by serial dilution and plating on LB agar.
Unmarked deletions were generated for the genes prpB (PA14_53940) and acsA (PA14_52800) in the PA14 wild-type background. Flanking regions (~1kb in length) containing the first and last codons of for acsA and prpB were generated using primers listed in S1 Table. The flanking regions and the deletion vector pSMV8 [61] (linearized by digestion with XhoI and SpeI) were assembled by Gibson assembly [62]. The resulting plasmid was transformed into E. coli WM3064 [61] and mobilized into PA14 by conjugation. Single recombinants were selected on LB agar containing 50 μg mL-1 gentamicin. Double recombinants were selected for on LB agar containing 6% sucrose. Potential prpB and acsA mutants, or prpBacsA double mutants were identified by PCR, and markerless deletions were confirmed by sequencing.
Targeted quantification of short-chain fatty acids was performed via high performance liquid chromatography (HPLC). The system consisted of a Shimadzu SCL-10A system controller, LC-10AT liquid chromatograph, SIL-10AF autoinjector, SPD-10A UV-Vis detector, and CTO-10A column oven. Separation of compounds was performed with an Aminex HPX-87H guard column and an HPX-87H cation-exchange column (Bio-Rad [Hercules, CA]). The mobile phase consisted of 0.05 N H2SO4, set at a flow rate of 0.5 mL min-1. The column was maintained at 50°C and the injection volume was 50 μL. Amino acid and metabolite (acetate and propionate) quantification from enrichment supernatants were performed by Millis Scientific, Inc. (Baltimore, MD) using liquid chromatography-mass spectrometry and gas chromatography-mass spectrometry (GC-MS). Samples for amino acid quantification were spiked with 1 μL of uniformly labeled amino acids (Cambridge Isotope Labs) and derivatized using AccQ-Tag reagent (Waters Corp.) for 10 min at 50°C. A Waters Micromass Quatro LC-MS interfaced with a Waters Atlantis dC18 (3 μm 2.1x100 mm) column was used. Reverse-phase LC was used for separation (mobile phases A:10mM ammonium formate in 0.5% formic acid, B:methanol) with a constant flow rate (0.2 mL min-1) and a column temperature of 40°C. Electrospray ionization was used to generate ions in the positive mode and multiple reaction monitoring was used to quantify amino acids. Samples (~100 μL) for acetate and propionate quantification were first diluted (150 μL water), spiked with internal standards (10 μL of 1000ppm acetate [13C2] and 1000 ppm propionate [13C1]) and acidified using 2 μL of 12N HCl. After equilibration at 60°C for 2h, carboxen/ polydimethylsiloxane solid phase microextraction (SPME) fiber was used to adsorb the headspace at 60°C for 30min. Acids were then desorbed into the gas chromatograph inlet for 2 min. A 30 m x 0.32 mm ID DB-624 column attached to a Thermo Electron Trace gas chromatograph with helium carrier gas (2.0 mL min-1) was used for separation of analytes. A Waters Micromass Quatro GC mass spectrometer was used for detection and quantification of target ions. Significance between sputum and saliva samples were determined by paired Student’s t-test.
To quantify P. aeruginosa gene expression in vivo, sputum (n = 17) was expectorated into RNAlater and immediately frozen to preserve the gene expression profile. Frozen sputum was thawed in TriZol (Life Technologies), homogenized using ceramic beads and purified according to the manufacturer’s protocol. RNA was concentrated using the Clean & Concentrator kit (Zymo) and de-salted using Turbo DNA-free (Life Technologies). Bacterial RNA was enriched (only in sputum samples) using the MicrobEnrich kit (Life Technologies), and purity was confirmed using Qubit (LifeTechnologies) spectrophotometry. qRTPCR was performed as previously described [63]. Briefly, DNA was reverse transcribed from 1 μg of total RNA using the iScript cDNA synthesis kit (BioRad). cDNA was then used a template for quantitative PCR on an iQ5 thermocycler (BioRad) using iTaq Universal SYBR Green Super Mix (BioRad). Triplicate measurements were made on each sputum sample. For control cultures, P. aeruginosa was grown in 4-morpholinepropanesulfonic acid (MOPS) minimal medium to an OD600 of~0.6 supplemented with glucose (12mM), acetate (20mM), propionate (20mM), or acetate + propionate (20mM each) where specified. After growth, cells were harvested by centrifugation, frozen at -80°C, and RNA was extracted as described above. Primer pairs are listed in S1 Table. For all primer sets, the following cycling parameters were used: 94°C for 3 min followed by 40 cycles of 94°C for 60s, 55°C for 45s, and 72°C for 60s. Primer efficiencies were tested for clpX (91.6%), acsA (91.6%) and prpD (96.8%). Relative RNA values were calculated from the Ct values reported and the experimental primer efficiencies, and were normalized to the expression of clpX. clpX was compared to oprI values to ensure constitutive expression levels. Significance between treatments was determined by two-tailed unpaired Student’s t-test.
16S rRNA gene sequences generated as part of this study were deposited as fastq files in NCBI GenBank under BioProject accession number SRP067035.
These studies were approved by the Institutional Review Board at the University of Minnesota (UMN IRB nos. 1401M47262 and 1404M49426). All subjects (adults) provided informed written consent prior to sample collection.
|
10.1371/journal.pbio.2000237 | Plasticity in Single Axon Glutamatergic Connection to GABAergic Interneurons Regulates Complex Events in the Human Neocortex | In the human neocortex, single excitatory pyramidal cells can elicit very large glutamatergic EPSPs (VLEs) in inhibitory GABAergic interneurons capable of triggering their firing with short (3–5 ms) delay. Similar strong excitatory connections between two individual neurons have not been found in nonhuman cortices, suggesting that these synapses are specific to human interneurons. The VLEs are crucial for generating neocortical complex events, observed as single pyramidal cell spike-evoked discharge of cell assemblies in the frontal and temporal cortices. However, long-term plasticity of the VLE connections and how the plasticity modulates neocortical complex events has not been studied. Using triple and dual whole-cell recordings from synaptically connected human neocortical layers 2–3 neurons, we show that VLEs in fast-spiking GABAergic interneurons exhibit robust activity-induced long-term depression (LTD). The LTD by single pyramidal cell 40 Hz spike bursts is specific to connections with VLEs, requires group I metabotropic glutamate receptors, and has a presynaptic mechanism. The LTD of VLE connections alters suprathreshold activation of interneurons in the complex events suppressing the discharge of fast-spiking GABAergic cells. The VLEs triggering the complex events may contribute to cognitive processes in the human neocortex, and their long-term plasticity can alter the discharging cortical cell assemblies by learning.
| Many microscale features in the human neocortex—a part of the brain involved in higher functions such as sensory perception, generation of motor commands, spatial reasoning, and language—are closely similar to those reported in experimental animals commonly used in neuroscience, like mice. However, the human neocortical neurons also exhibit specializations only reported in our species. One such feature is the capacity of excitatory principal cells to elicit firing in local inhibitory interneurons with a single action potential via very strong excitatory synapses. It has been suggested that this feature has specifically evolved to enhance coordinated firing of neuronal ensembles in higher brain functions. However, it is unknown how these circuits are modified by learning. Therefore, we investigated how these very strong excitatory synapses are changed, and if their impact on the firing of local inhibitory neurons is altered by repetitive action potentials mimicking learning-related activity. By recording in human neocortical slices, we report that the strong excitatory synapses on interneurons exhibit robust activity-dependent long-term plasticity. The plasticity also regulates the discharge of local interneurons driven by these synapses. Although these specialized synapses have only been reported in the human neocortex, their plasticity mechanism is evolutionarily conserved. We suggest that the strong synapses with robust plasticity have evolved to enhance complex brain functions and learning.
| Evolution has shaped the human neocortex producing microcircuit features that are specific to our species [1]. Neuronal density, ultrastructural features, and functional properties of neurons [2–4] reflect specific adaptations in the human neocortex to perform complex and fast signal processing [5–13]. A remarkable feature in the human neocortex is that single pyramidal cell (PC) action potentials (APs) are able to generate di- and polysynaptic GABAergic interneuron discharge known as complex events [10,11]. The events emerge from the activity of a small subset of excitatory connections forming very large glutamatergic excitatory postsynaptic potentials (VLEs), specifically to GABAergic interneurons in supragranular layers of the frontal, the temporal, and the prefrontal cortices [10,11]. Similar strong excitatory connections between individual neocortical neurons have not been found in nonhuman brains. Therefore, it has been proposed that the VLEs and the complex events participate in cortical information encoding in high order cognitive processes [10]. However, this would predict that these events are dynamically modulated by learning [14,15]. Yet, it is unknown whether the VLEs show use-dependent long-term plasticity, and if their specific modulation indeed affects the complex events. We hypothesize that the immense strength of VLEs is generated and regulated by common activity-driven synaptic long-term plasticity processes, and that they may occur in various different inhibitory interneuron types [16,17]. Alternatively, these connections could be hard-wired selectively in a specific, yet unknown, subset of postsynaptic GABAergic interneurons without prominent lasting plasticity in the adult neocortex [18,19].
We asked whether VLEs show activity-induced long-term plasticity, and if their selective modulation had impact on the local complex events. By performing triple and dual whole-cell recordings of synaptically connected identified neurons, we found that VLEs exhibit metabotropic glutamate receptor (mGluR)-dependent long-term depression (LTD) that converts them to common weak excitatory postsynaptic potential (EPSP) connections. In addition, this alters the neocortical complex events suppressing the cell assemblies activated by the PC. Thus, the VLEs occur in various interneuron types, and their occurrence is regulated by the synapse’s activity history. To the best of our knowledge, this is the first study reporting synaptic plasticity in human neocortical interneurons.
By performing triple and dual whole-cell recordings from identified L2–3 human neocortical neurons, we found that some fast-spiking GABAergic interneurons (FSINs) receive glutamatergic input from individual afferent PCs showing VLEs (Fig 1A and 1B). Simultaneous recording from three neurons demonstrated that a fast-spiking GABAergic cell can receive VLEs (average amplitude 9.60 ± 0.20 mV, showing no failures) from one L2–3 PC and small amplitude glutamatergic EPSPs (average 3.29 ± 0.12 mV, showing no failures) similar to EPSPs between PCs from another PC [10]. Recordings from 21 synaptically connected PC–FSIN pairs revealed vast differences in the single AP-evoked EPSP amplitudes between the pairs (Fig 1B, S1 Table, S1 Data). In FSINs, the EPSP averages showed a range from 0.62 mV to 16.49 mV, with nonparametric distribution (failures excluded, evoked with 10 s interval at Em −68.5 ± 1.4 mV, n = 21, Shapiro-Wilk test). Despite the amplitude difference, the excitatory postsynaptic currents (EPSCs) in FSINs similarly exhibited fast time-to-peak kinetics (0.59 ± 0.04 ms, n = 18) (Fig 1B, S1 Data), indicating that the amplitude variability is unlikely to derive from different electrotonic filtering of the glutamatergic synaptic inputs. Likewise, distribution of the average EPSP amplitudes in PCs to non–fast-spiking interneuron (non-FSIN) pairs showed nonparametric distribution with a range from 0.7 mV to 6.9 mV (failures excluded, at Em −70.3 ± 1.5 mV, n = 9, Shapiro-Wilk test) (S1 Table). Thus, VLEs are not occurring solely in FSINs, but are exhibited in various types of GABAergic neurons including fast- and non–fast-spiking cells. On the contrary, PC–PC connections showed parametric distribution of average EPSPs with small amplitude (2.01 ± 0.02 mV at Em −69.4 ± 1.8 mV, failures excluded, Shapiro-Wilk test, n = 16) (Fig 1B, S1 Table, S1 Data). The interneuron EPSPs were defined as VLEs when their average (failures excluded) was larger than mean + 2 x standard deviation (SD) of the EPSPs in PC–PC connections (4.21 mV, failures excluded) in baseline conditions (mean ± SD = 2.01 ± 1.10 mV, n = 480 in 16 cells). The postsynaptic interneurons in the triple and paired recordings were immunohistochemically confirmed positive for vesicular GABA transporter (vgat+) (n = 31). Cells that in addition were immunopositive for parvalbumin (pv+) showed rapid axon currents (spike inward current width [SW] of 0.43 ± 0.02 ms, n = 11) characteristic of the FSINs [10,11]. The pv+ cells, together with vgat+ interneurons showing similar fast spike kinetics (SW 0.49 ± 0.02 ms, n = 11), but with nonconclusive or untested pv reaction, were considered FSINs (n = 22). Ten FSINs were further identified as putative basket cells by their axon morphology [10]. The non–fast-spiking vgat+ interneurons and the PCs had significantly longer spike kinetics with SW of 0.96 ± 0.04 ms (n = 9) and 1.18 ± 0.06 ms (n = 16), respectively (p < 0.01 between all groups, ANOVA with Tukey’s posthoc test) [13]. The non-FSINs with intact soma were immunohistochemically tested for somatostatin (sst) for further identification of the cells [20]. Detailed results on the EPSPs excluding and including failures, the EPSCs, and the immunohistochemical reaction analyses with cell type identification are shown in S1 Table. In all potential connections tested between neurons (n = 1,056), we found (including connections lost during baseline) a monosynaptic response in 11.0% of cases. Success rate for identified PC–FSIN pairs was 4.0%, PC–non-FSINs connections: 1.9%, PC–PC pairs: 1.1%, and FSIN–PC connections: 3.8%, showing similar or slightly lower connectivity rates than reported in the rodent neocortex L2–3 [21–23].
We asked whether glutamatergic connections to interneurons showed long-term plasticity akin to that reported in the rodent cortex [24,25]. To test this, we performed experiments applying high frequency bursts of APs (5 APs at 40 Hz, x 40 with 0.5 s interval) in the presynaptic PC after a baseline of EPSPs (at least 5 min, but less than 10 min, analyzed including failures) [26,27]. Postsynaptic FSINs (SW 0.40 ± 0.02 ms) were held in resting membrane potential in current clamp (–67.6 ± 2.2 mV, n = 9) (Fig 2). First, we tested VLEs (average in baseline 5.85 ± 0.59 mV, did not show failures, n = 5) in control conditions and found that the afferent PC bursts firing generated an LTD of the EPSPs (amplitude to 0.52 ± 0.02 of baseline at 20–25 min after 40 Hz bursts, n = 5 cells, p < 0.01, Wilcoxon test) (Fig 2A, S1 Fig, S2 Data, S7 Data). The LTD was associated with a reduced paired-pulse EPSP ratio (1st/2nd EPSP amplitude with 50 ms interval) to 0.74 ± 0.06 of baseline (p < 0.05, Mann-Whitney test, baseline mean 1.41 ± 0.22) and a decrease in the EPSP amplitude CV−2 (1/squared coefficient of variation) value (to 0.44 ± 0.12% from baseline, p < 0.05, Mann-Whitney test) (S3 Data), indicating presynaptic site of depression [28]. The results on the paired-pulse ratio (PPR) and altered EPSP amplitude variation by LTD are summarized in histograms in Fig 2B. Likewise, corresponding experiments in occasional PC–non-FSIN pairs with VLEs (n = 2, averages in the baseline including failures 5.81 mV and 6.89 mV, failure rates 0% and 13%, respectively) showed that single fiber burst firing can also generate LTD in some non-FSINs (S2 Fig).
We next tested whether LTD in VLEs requires group I mGluRs as various forms of long-term plasticity in glutamatergic synapses to FSINs depend on these receptors in the rodent cortex [24,29–32]. We studied four PC–FSIN pairs with VLEs (average in baseline 8.42 ± 2.83 mV at Em −69.5 ± 2.1 mV, did not show failures, n = 4) as above, but in the presence of LY367385 (100 μM) and 2-Methyl-6-(phenylethynyl)pyridine hydrochloride (MPEP, 25 μM). Thus, LTD was blocked in the VLE connections (Fig 2A) with no significant change in the amplitude (1.03 ± 0.04 of baseline at 20–25 min, n = 4 cells, Wilcoxon test), PPR (1.21 ± 0.11 of baseline, baseline mean 1.07 ± 0.17) or 1/CV2 (1.63 ± 0.39 of baseline) (Mann-Whitney test) (Fig 2B, S2 and S3 Data). To conclude, PC–FSIN connections with large EPSPs show activity-driven LTD, which requires group I mGluRs.
In contrast to VLEs, the PC–FSIN pairs with small EPSPs (average in baseline with failures 1.89 ± 0.43 mV at Em −69.2 ± 3.5 mV, failure rate 11.2 ± 9.5%, SW 0.48 ± 0.04 ms, n = 5) failed to show lasting plasticity following the 40 Hz bursts (amplitude 1.07 ± 0.06 of baseline at 20–25 min) (Fig 3A, S1 Fig, S4 Data). Likewise, no lasting plasticity was seen in any (paired t-test) of the three vgat+ non-FSINs (SW 0.91 ± 0.07, n = 3) with small amplitude EPSP (2.08 ± 0.58 mV, n = 3) in similar experiments. Given that LTD in the VLEs was accompanied by stronger postsynaptic depolarization during the presynaptic spike bursts, we studied whether small amplitude EPSP connections would show plasticity if the postsynaptic FSIN was depolarized during the PC bursts. We reproduced experiments above with a separate set of PC–FSIN (SW 0.47 ± 0.03 ms, n = 5) pairs with small amplitude EPSP (average with failures 1.44 ± 0.22 mV, failure rate 11.0 ± 3.3%, n = 5), and paired presynaptic PC spike bursts with postsynaptic cell depolarization (20–30 mV, 250 ms steps from Em, see Methods for details) beyond the firing threshold (Fig 3A). This protocol also failed to generate long-lasting change in EPSPs in the PC–FSIN pairs (0.94 ± 0.04 of baseline at 20–25 min, n = 5, Wilcoxon test) (Fig 3A, S4 Data). Interestingly, a small but significant LTD was observed with this configuration in two (0.71 ± 0.10 and 0.83 ± 0.05 at 20–25 min compared to baseline, p < 0.05 for both cells, paired t-test) of three individual PC–non-FSIN (vgat+) pairs tested. Finally, we studied the synaptic connections between L2–3 PCs applying presynaptic 40 Hz bursts while the postsynaptic cell was at resting membrane potential. The PC–PC pairs were connected with small amplitude EPSPs (average with failures 1.40 ± 0.30 mV at Em −65.9 ± 5.4 mV, failure rate 4.2 ± 3.2%, n = 4), and the 40 Hz bursts failed to generate lasting plasticity in the EPSP (1.01 ± 0.05 of baseline at 20–25 min, n = 4 cells, Wilcoxon test) (Fig 3B, S4 Data), possibly because LTP and LTD in human PCs require strong postsynaptic depolarization for either activation of glutamate NMDA receptors or L-type voltage-gated calcium channels [7]. Input resistance in the plasticity recordings showed small increase to 1.09 ± 0.01 (baseline-normalized) at 20–25 min from baseline (n = 26, p < 0.01, t-test) [33]. The results show that following just a single PC burst firing, LTD specifically occurs in large EPSPs between PCs and interneurons, and not in other investigated synaptic connections.
Because studies in rodents have reported LTD in glutamatergic synapses to cortical interneurons either by chemical or strong synaptic activation of group I mGluRs, we studied whether FSINs with weak excitatory inputs showed the LTD when multiple glutamatergic fibers were simultaneously activated (Fig 3C) [27,29,34]. Evoking compound EPSCs from many small glutamatergic inputs with extracellular electrical stimulation (see Methods), we applied 40 Hz bursts to the glutamatergic pathway as above after baseline (at least 5 min, but less than 10 min). Focusing the EPSC analysis in the FSINs on the monosynaptic component of the current (see Fig 3C) [35,36] we found that the 40 Hz burst stimulation resulted in LTD (EPSC 0.72 ± 0.02 from baseline at 20 min, n = 7 cells, p < 0.001, t-test). Blockers for glutamate N-methyl-D-aspartatereceptor (NMDARs) (DL-2-Amino-5-phosphonopentanoic acid [DL-APV], 100 μM) and GABAARs (PiTX, 100 μM) were present in the experiments. The EPSC amplitude LTD was accompanied by decreased CV−2 (baseline-normalized to 0.72 ± 0.16%, n = 7, t-test). This LTD was blocked in experiments with group I mGluR antagonists LY367385 (100 μM) and MPEP (25 μM) (n = 7, p < 0.05, t-test) (Fig 3C, S5 Data). The FSINs in these extracellular stimulation experiments showed narrow SW 0.62 ± 0.04 ms, n = 14. Accordingly, we reproduced these experiments with rat glutamatergic fibers to FSINs in L2–3 (SW 0.64 ± 0.06 ms, n = 10) and confirmed LTD (EPSC 0.64 ± 0.04 from baseline at >15 min, n = 5, p < 0.01, Wilcoxon test) and its blockade with the group I mGluR antagonists (EPSC from baseline 0.95 ± 0.5, n = 5, Wilcoxon test) (Fig 3C, S5 Data). The EPSC amplitude in LTD showed reduced CV−2 (baseline-normalized to 0.41 ± 0.09%, n = 5, Mann-Whitney test), but not when LTD was blocked with the mGluR antagonists (1.04 ± 0.24% of baseline at >15 min) (p < 0.05 between the groups at >15 min, Mann-Whitney test) (Fig 3C, S5 Data). We tested three of the recorded rat FSINs for pv immunoreaction and found them all positive. Thus, in both human and rat cortex, weak glutamatergic connections to L2–3 FSINs exhibit group I mGluR-dependent LTD if multiple glutamatergic inputs are activated simultaneously.
Given that interneuron–PC connections with VLEs have been proposed to be essential in generation of the neocortical complex events, we studied whether the LTD in these connections would selectively modify network activity. First, we confirmed that single PC AP-evoked VLEs in the FSIN as well as in the non-FSIN elicited firing of these interneurons from the resting membrane potential [11]. We found that VLE-evoked postsynaptic spikes in a FSIN (Em −69 mV) typically followed with a short 3–5 ms delay (Fig 4A) [10], whereas in a non-FSIN, (Em −69 mV) the spikes showed long delay with large jitter (S2 Fig). Similarly, whole-cell recordings between identified PCs revealed disynaptic GABAAR-mediated inhibitory currents (dIPSCs) in complex events elicited by a single AP (interval 10 s) (Fig 4B) [10,11]. The dIPSCs occurred with short delay (6.23 ± 0.72 ms, n = 16 pairs) and high probability (0.70 ± 0.05, n = 16 pairs in baseline conditions) (S6 Data). The dIPSCs showed longer and more variable delay to the presynaptic spike than monosynaptic GABAAR-mediated inhibitory currents (monIPSCs) from FSINs (0.96 ± 0.10 ms, n = 9, p < 0.001, t-test) (SW 0.48 ± 0.03 ms, n = 9) (S3 Fig, S9 Data). In addition, dIPSCs were blocked by the glutamate AMPAR blocker GYKI53655 (25 μM) (n = 3, p < 0.001, Chi-square test) (Fig 4C, S4 Fig, S6 Data). The evoked dIPSC amplitudes (averages excluding failures in all plasticity recordings 32.1 ± 3.7 pA, n = 12) were similar to monIPSCs (35.2 ± 5.6 pA, n = 9, t-test) (S5 Fig, S10 Data), indicating that these early complex event inhibitory currents (IPSCs) were generated by a single FSIN. The dIPSCs and the monIPSCs were recorded at −55 mV. The dIPSCs were detected in 3.0% of all potential connections tested (n = 1,056).
To determine whether plasticity modified this network, we performed long recordings from PC pairs showing that the probability and delay of the dIPSCs were stable for at least 30 min (n = 3) in normal conditions (Fig 4D, 4E and S6 Data). However, if the 40 Hz burst firing was delivered in the presynaptic PC (similar to the VLE LTD experiments in Fig 2), the dIPSC occurrence (total in 25 ms from PC spike) rapidly and permanently attenuated after a baseline showing strong LTD (from 0.71 ± 0.04 in baseline to 0.14 ± 0.09 at 15–20 min, n = 3, p < 0.05, Chi-square test) (Fig 4F, 4G and S6 Data). Interleaved experiments in the presence of group I mGluR antagonists LY367385 (100 μM) and MPEP (25 μM) showed that LTD of dIPSCs was blocked in the presence of the group I mGluR antagonists (0.81 ± 0.10 in baseline and 0.88 ± 0.12 at 15–20 min, n = 3, Chi-square test) (Fig 4H, 4I and S6 Data). In conclusion, the results demonstrate that a single PC burst firing at 40 Hz elicits robust LTD in VLE connections to FSINs and causes suppression of phase-locked early dIPSCs between PCs. These two LTDs both require group I mGluRs. Thus, the results show that plasticity of VLEs in FSINs changes the activation pattern of the neurons discharging in supragranular layers during complex events.
Almost a decade ago, Molnar et al. [10] first reported that a subset of excitatory PC synapses in the human neocortex form VLEs in local GABAergic interneurons in supragranular layers. These strong connections specifically from PCs to inhibitory interneurons have since been reported in the frontal, prefrontal, parietal, and temporal cortices, where the VLEs often are suprathreshold, driving assemblies of inhibitory interneurons to fire after a single PC spike [11,37]. Similar strong connections between two individual neocortical neurons have not been found in nonhuman species [37–39]. Analyses of large datasets from rodent visual and somatosensory cortices have revealed that neurons in local networks are not randomly connected, but specific local connectivity patterns exist between neuron types, and the strongest excitatory synapses control local network activity [40–42]. This suggests that there is a skeleton of strong connections in the network that dominates the activity [40]. Therefore, it has been proposed that the VLE connections in the human might be important in generating neocortical cell assemblies and be involved in higher cognitive functions [11]. However, until now it had remained unknown how neuronal plasticity regulates these connections and whether their selective modulation indeed alters discharge of neuronal assemblies in the human neocortical network activity.
Our result that the VLEs occur specifically in human interneurons, and not in between PCs, is consistent with previous studies [10,11]. In addition, we demonstrate that VLEs are generated in different interneuron subpopulations, including fast-spiking (such as basket cells) and non–fast-spiking supragranular vgat+ neurons. Furthermore, single GABAergic interneurons receive both VLEs and more common small EPSP connections from layer 2–3 PCs. Indeed, anatomically identified L2–3 basket cells show huge variability in strength of PC inputs. It is likely, although we do not directly demonstrate it here, that a single presynaptic L2–3 PC cell evokes VLEs and small EPSPs in different L2–3 postsynaptic interneurons showing synapse specificity. A comprehensive recent study in rodent demonstrated that certain connectivity patterns between neurons are repeated in the neocortex across different regions [39]. The strong VLE connection, occurring between PCs and L2–3 inhibitory interneurons in various neocortical regions is one such feature, and probably specific to the human neocortical microcircuits.
We show that strength of VLE connections is controlled by their activity history. A common form of mGluR-dependent LTD characterized in the extracellular stimulation experiments of glutamatergic synapses in the rodent cortex [24] converts the VLEs to small amplitude EPSPs. This LTD suppresses VLE synapses through a presynaptic mechanism most likely controlling the vesicular transmitter release as indicated by the changes in the EPSP amplitude PPR and the coefficient of variation in the depressed synapses [37]. The LTD of VLEs in FSINs, many of which are also pv+, involves the group I mGluRs. In the rodent cortex, group I mGluRs have been shown to play a central role in long-term plasticity processes including presynaptic LTD and LTP [24,29–31,34,43–46]. Our results from the group I mGluR-dependent LTD in FSINs in the human and the rodent neocortex indicate that this is an evolutionarily conserved plasticity mechanism for controlling the fast-spiking interneuron activity in the mammalian brain.
Interestingly, we report that in a non–fast-spiking cell, LTD was not blocked by the group I mGluR antagonists (see S2 Fig). In addition, we found significant LTD in two of three PC–non-FSIN connections with small EPSP when the postsynaptic cell was depolarized during PC bursts. These suggest that there may be various LTD forms in human cortical interneuron synapses [24,25], and that non-FSINs may exhibit different LTD mechanism than the FSINs, possibly depending on postsynaptic depolarization. Indeed, many synapse-specific properties including long-term plasticity have been reported in glutamatergic fibers in the rodent cortex [21,32,47–50]. Correspondingly, synapses originating from the same PC in the human neocortex may exhibit distinct long-term plasticity, depending upon the postsynaptic target cell; and different activity patterns may be required for plasticity in the synapses [24,50]. In this study, we have used elevated extracellular calcium (3 mM) to increase the stability of disynaptic IPSCs in baseline conditions. However, compared to recordings with 2 mM extracellular calcium, this modification is unlikely to strongly affect the high-frequency firing-evoked synaptic release and the long-term plasticity in FSIN synapses: the probability of synaptic release in VLEs in FSINs is already very high at 2 mM Ca2+, and only slightly modulated by further increase of calcium [37]. Yet, as demonstrated in the recent study by Molnar et al. [37], the VLE release probability can markedly decrease when extracellular calcium is reduced from 3 mM to 1.5 mM, which is considered lower range of the cerebrospinal fluid total calcium level in physiological conditions [51]. Therefore, it is also possible that in calcium concentrations close to 1.5 mM, the PC firing pattern used in this study may not be sufficient for such a robust LTD as reported here in the FSINs. Importantly, some non-FSINs show VLEs with low presynaptic release probability even in 3 mM calcium, as indicated by the large EPSP amplitude coefficient of variation. In these synapses, extracellular calcium modulations may have even stronger effects on the short- and long-term plasticities than in the FSINs.
Strikingly, in the human neocortex, the activity of a single PC is sufficient to trigger mGluR-dependent LTD in the VLE connections, but not in weak glutamatergic pathways. The level of the postsynaptic depolarization does not explain the failure of LTD in small EPSP connections to fast-spiking cells, and a potential explanation is that there is insufficient activation of the postsynaptic group I mGluRs [34,43]. The connections with VLEs are likely to release more glutamate and activate the critical mGluRs [52–54]. The hypothesis on strong glutamate release is supported by our finding that the small EPSP connections were unable to generate LTD by single fiber activity, but they showed the mGluR-dependent plasticity when multiple fibers were activated simultaneously with local extracellular stimulation [55]. Indeed, a recent study revealed that human neocortical PC–FSIN synapses with VLEs have more transmitter vesicle release sites, although the glutamate release quantal size is similar compared to synapses in rat neocortex [37]. This indicates multivesicular release in synapses with VLEs, and it is interesting to speculate that the conversion to common small EPSPs via the presynaptic LTD might reflect their transformation from a multivesicular release site to a single vesicle-releasing synapse.
Although a link between very large excitatory synapses and human cortical complex events has been suggested earlier [10,11], a relation between their selective modulation and complex events had not been directly demonstrated until this study. The LTD triggered by a single PC firing in the current conditions is specific to connections with VLEs and therefore provides a useful tool to test the relation between VLEs and neocortical network activity. Results here show that the VLE connections indeed trigger the complex events, and the LTD changes their temporal structure. The activation of individual fast-spiking interneurons was commonly observed in dual PC recording as disynaptic GABAergic currents with timing corresponding to the APs in fast spiking interneurons [11]. LTD of the VLE in FSINs and the suppression of dIPSCs in complex events were both induced by single PC burst firing, and they both required group I mGluRs. Some non-FSINs also exhibit VLEs and can show LTD, although these cells were unlikely to contribute to the GABAergic disynaptic currents investigated here: the dIPSCs occurred with short delay and high precision, whereas the non-FSINs show long and variable delay in their response to fire APs (see S2 Fig). However, the results of non-FSINs are based on small sample sizes, and should therefore be interpreted with caution.
In conclusion, the human neocortex is unique in many aspects, since its microcircuits show differences at the molecular, ultrastructural, and physiological levels compared to other mammalian species [1,3,56]. The capacity of the human neocortex to perform extraordinary and highly complex tasks may at least partly result from these microcircuit level specializations. We propose that VLEs with robust activity-induced plasticity and their contribution to neocortical cell assemblies may be crucial for higher cognitive functions and abstract mental abilities of the human brain. In addition, evidence in animal models suggests the involvement of group I mGluR-mediated plasticity in neocortical learning processes, and perturbation of the mGluR-dependent plasticity has been reported with mental decline [43]. Therefore, the human-specific microcircuit features may also be substrates for pathological processes resulting in cognitive decline and other neurological and neuropsychiatric dysfunctions that we as a species are vulnerable to [57–61].
All procedures were performed according to the Declaration of Helsinki with the approval of the University of Szeged Ethical Committee and Regional Human Investigation Review Board (ref. 75/2014).
Human neocortical slices were derived from material that had to be removed to gain access to the surgical treatment of deep-brain tumors from the left and right frontal, temporal, and parietal regions with written informed consent of the patients prior to surgery. The patients were 10–85 y of age (mean ± SD = 50 ± 4 y), including 17 males and 14 females. The tissue obtained from underage patients was provided with agreement from a parent or guardian. The resected samples were cut from the frontal and temporal lobes of left or right hemisphere. Anesthesia was induced with intravenous midazolam and fentanyl (0.03 mg/kg, 1–2 lg/kg, respectively). A bolus dose of propofol (1–2 mg/kg) was administered intravenously. The patients received 0.5 mg/kg rocuronium to facilitate endotracheal intubation. After 2 min, the trachea was intubated and the patient was ventilated with O2/N2O mixture (a ratio of 1:2). Anesthesia was maintained with sevoflurane at monitored anesthesia care volume of 1.2–1.5. After surgical removal, the resected tissue blocks were immediately immersed in ice-cold standard solution containing (in mM): 130 NaCl, 3.5 KCl, 1 NaH2PO4, 24 NaHCO3, 1 CaCl2, 3 MgSO4, 10 D(+)-glucose, and saturated with 95% O2 and 5% CO2. Slices were cut perpendicular to cortical layers at a thickness of 350 μm with a microtome (Microm HM 650 V) and were incubated at room temperature (20–24°C) for 1 h in the same solution. Rat neocortical slices were prepared as described before [62]. Male Wistar rats were anaesthetized using halothane, and following decapitation (320 μm thick), coronal slices were prepared from the somatosensory cortex. The solution used during electrophysiology experiments was identical to the slicing solution, except it contained 3 mM CaCl2 and 1.5 mM MgSO4. Recordings were performed in a submerged chamber (perfused 8 ml/min) at approximately 36–37°C. Cells were patched using water-immersion 20× objective with additional zoom (up to 4x) and infrared differential interference contrast video microscopy. Micropipettes (5–8 MOhm) were filled with intracellular solution for whole-cell patch-clamp recording (in mM): 126 K-gluconate, 8 KCl, 4 ATP-Mg, 0.3 Na2–GTP, 10 HEPES, 10 phosphocreatine (pH 7.20; 300 mOsm) with 0.3% (w/v) biocytin. Current and voltage clamp recordings were performed with Mutliclamp 2B amplifier (Axon Instruments), low-pass filtered at 6 kHz (Bessel filter). Series resistance (Rs) and pipette capacitance were compensated in current clamp mode and pipette capacitance in voltage clamp mode. Cell capacitance compensation was not applied. Rs was monitored and recorded continuously during the experiments. The recording in voltage clamp mode was discarded if the Rs was higher than 25 ΩM or changed more than 20%.
In paired cell recordings, APs were generated in the presynaptic cell with brief (2–3 ms) suprathreshold depolarizing (60–70 mV) paired pulses (50 ms interval) in voltage clamp delivered every 10 s from −60 mV. Postsynaptic cells were at resting membrane potential in current clamp mode. In some cells with VLEs, the postsynaptic cell was hyperpolarized (up to −10 mV) with constant current to prevent the VLE from triggering an AP. The 40 Hz firing protocol was similarly applied in voltage clamp mode with series of 2–3 ms depolarizing pulses (5 pulses at 40 Hz, delivered every 0.5 sec 40 times), while the postsynaptic cell was held in current clamp resting membrane potential. In some experiments (Fig 3A, 4–5), the postsynaptic cell was depolarized in voltage clamp during presynaptic 40 Hz firing with a continuous step (20–30 mV, 250 ms). This elicited on average 2.2 postsynaptic spikes for 1st presynaptic spike (in following 25 ms, n = 200 in 5 cells) of the 40 Hz train, and on average 0.80 postsynaptic spike probability for the 2nd–5th PC AP.
Extracellular stimulation was applied with a concentric bipolar electrode (125 μm tip diameter, FHC Inc., US) positioned on L2–3. Paired pulse stimuli (50 μs, with 50 ms interval, intensity range from 20 to 300 μA) were delivered every 15 s with current isolator stimulator (Model DS3, Digitimer, UK). Compound EPSCs in Fig 3 were confirmed by observing less than 100 pA increases in the evoked EPSC amplitude when gradually increasing stimulation intensity.
Data were acquired with Clampex software (Axon Instruments, US) at 20 kHz. EPSC/P, IPSC, action current duration, and the cell input resistance were analyzed off-line with p-Clamp software (Axon Instruments, US) and Spike2 (version 7.0, Cambridge Electronic Design, UK). Liquid junction potential was not corrected. EPSC amplitude and kinetics analysis in voltage clamp mode (time-to-peak from onset) was omitted when access resistance was higher than 25 MΩ. SW was calculated from the onset of inward action current till recovery to baseline holding level. Data for SWs were collected in the beginning of experiments when synaptic connections from all cells were briefly tested in voltage clamp mode. All data are presented as mean ± s.e.m. and when showing baseline-normalized EPSPs of many cells, the values were calculated from binned (30 s bin) data. In rare cases when cell-spiked and accurate EPSP amplitude data was not available, bin includes two instead of three data points. For statistical analysis, ANOVA with posthoc Tukey’s test and t-test were used for data with normal distribution (Shapiro-Wilk test) and sample sizes larger than n = 6. Chi-square test was used for categorical variables (occurrence of dIPSC in 25 ms time window from PC spike). Otherwise, Mann-Whitney U-test (unpaired) and Wilcoxon Signed Rank Test (paired) were used. EPSP amplitude in individual plasticity experiments was tested with paired t-test comparing data points in 5 min baseline and an equal time window at 20–25 min following the presynaptic bursts, unless stated otherwise (in some shorter experiments in the last 5 min of the recording). Correlation was determined with Pearson’s r-test. Differences were accepted as significant if p < 0.05. Failures were included in the EPSP mean values (in binned data) in plasticity analysis.
DL-APV, GYKI53655, LY367385, MPEP, and picrotoxin (PiTX) were applied via bath and purchased from Sigma Aldrich (Hungary).
After electrophysiological recording, slices were immediately fixed in a fixative containing 4% paraformaldehyde and 15% picric acid in 0.1 M phosphate buffer (PB; pH = 7.4) at 4°C for at least 12 h, then stored in 0.1 M PB with 0.05% sodium azide as a preservative at 4°C. Slices were embedded in 10% gelatin and further sectioned into slices of 50 μm thickness in cold PB using a vibratome VT1000S (Leica Microsystems, UK). After sectioning, the slices were rinsed in 0.1 M PB (3 x 10 min) and cryoprotected: first step in 10% (30 min), and later in 20% sucrose (1 h) dissolved in PB and then permeabilized using a freeze and thaw procedure. Finally, they were incubated in fluorophore (Cy3)-conjugated streptavidin (1:400, Jackson ImmunoResearch Lab.Inc. US) in 0.1 M Tris-buffered saline (TBS, pH 7.4) for 2.5 h (at 22–24°C). After washing with 0.1 M PB (3 x 10 min), the sections were covered in Vectashield mounting medium (Vector Laboratories Inc, US), put under cover slips, and examined under epifluorescence microscope (Leica DM 5000 B, UK). Sections selected for immunohistochemistry and cell reconstruction were dismounted and processed as explained below.
Some sections for cell structure illustrations were further incubated in a solution of conjugated avidin-biotin horseradish peroxidase (ABC; 1:300; Vector Labs, UK) in Tris-buffered saline (TBS, pH = 7.4) at 4°C overnight. The enzyme reaction was revealed by the glucose oxidase-DAB-nickel method using 3’3-diaminobenzidine tetrahydrochloride (0.05%) as chromogen and 0.01% H2O2 as oxidant. Sections were postfixed with 1% OsO4 in 0.1M PB. After several washes in distilled water, sections were stained in 1% uranyl acetate and dehydrated in ascending series of ethanol. Sections were infiltrated with epoxy resin (Durcupan) overnight and embedded on glass slices. Three-dimensional light microscopic reconstructions from sections were carried out using the Neurolucida system with 100 x objective (Olympus BX51, Olympus UPlanFI, Hungary). Images were collapsed in z-axis for illustration. Cells in Fig 1 were reconstructed from confocal microscope z-stack images of streptavidin fluorophore signal using Image-J software as described previously [32]. Vgat immunoreaction analysis was used in parallel to confirm the interneuron axon.
For immunohistological reactions, free-floating sections were washed 3 times in TBS-TX 0.3% (15 min) at 22–24°C, then moved in 20% blocking solution with horse serum in TBS-TX 0.3%. The sections were incubated in primary antibodies diluted in 1% serum in TBS-TX 0.3% over three nights at 4°C, then put in relevant fluorochrome-conjugated secondary antibodies in 1% of blocking serum in TBS-TX 0.3% overnight at 4°C. Sections were washed at first step in TBS-TX 0.3% (3 x 20 min) and later in 0.1 M PB (3 x 20 min) and mounted on glass slides with Vectashield mounting medium (Vector Lab.Inc., UK). The characterizations of antibodies: pv (goat anti-pv, 1:500, Swant, Switzerland, www.swant.com), sst (rat anti-sst, 1:50, Merck Millipore, Germany, www.merckmillipore.com) and vgat (rabbit anti-vgat, 1:500, Synaptic Systems, Germany, www.sysy.com). Fluorophore-labelled secondary antibodies were: DyLight 488 (Donkey anti goat, 1:400, Jackson ImmunoResearch Lab. Inc., www.jacksonimmuno.com, US), Alexa488 (Donkey anti rat, 1:400, Jackson ImmunoResearch Lab. Inc.) and Cy5 (Donkey anti rabbit, 1:500, Jackson ImmunoResearch Lab. Inc.). Labelling of neurons by neurobiotin and immunoreactions were evaluated using first epifluorescence (Leica DM 5000 B, UK) and then laser scanning confocal microscopy (Olympus FV1000, Hungary). Immunoreaction was considered to be negative when fluorescence was not detected in relevant neurobiotin-labelled cell, but immunopositivity was detected in the same area in unlabelled cells. Immunoreactions were studied in axon boutons (vgat and pv) and soma and dendrites (pv, sst).
|
10.1371/journal.pntd.0005760 | Zika virus alters the microRNA expression profile and elicits an RNAi response in Aedes aegypti mosquitoes | Zika virus (ZIKV), a flavivirus transmitted primarily by Aedes aegypti, has recently spread globally in an unprecedented fashion, yet we have a poor understanding of host-microbe interactions in this system. To gain insights into the interplay between ZIKV and the mosquito, we sequenced the small RNA profiles in ZIKV-infected and non-infected Ae. aegypti mosquitoes at 2, 7 and 14 days post-infection. ZIKA induced an RNAi response in the mosquito with virus-derived short interfering RNAs and PIWI-interacting RNAs dramatically increased in abundance post-infection. Further, we found 17 host microRNAs (miRNAs) that were modulated by ZIKV infection at all time points. Strikingly, many of these regulated miRNAs have been reported to have their expression altered by dengue and West Nile viruses, while the response was divergent from that induced by the alphavirus Chikungunya virus in mosquitoes. This suggests that conserved miRNA responses occur within mosquitoes in response to flavivirus infection. This study expands our understanding of ZIKV-vector interactions and provides potential avenues to be further investigated to target ZIKV in the mosquito host.
| Vector-borne viruses have immense impacts on human health by causing mortality and morbidity. Control of diseases caused by these viruses have mostly concentrated on vector control or inhibition of virus transmission by the vectors. This requires a thorough understanding of vector-virus interactions. In this study, we investigated the RNA interference (RNAi) response in Aedes aegypti mosquitoes infected with the Zika virus (ZIKV) strain isolated from the current pandemic using deep sequencing technologies. We found that infection alters the microRNA (miRNA) profile between infected and uninfected mosquitoes and that changes in miRNA expression occur over time. The short interfering RNA pathway, which is the main mosquito defense as part of the RNAi pathway, was also induced by ZIKV infection with the number of short interfering RNAs increasing significantly as infection progressed. Our results indicate that ZIKV induces the mosquito host defense similar to infection with other flaviviruses.
| Zika virus (ZIKV) is a flavivirus related to dengue virus (DENV), West Nile virus (WNV) and Yellow fever virus (YFV) that is transmitted to humans by Aedes mosquitoes. In the urban transmission cycle, Aedes aegypti is thought to be the dominant vector, while several Aedes species are implicated in transmission in the sylvatic cycle [1,2]. The virus was originally discovered in the Ziika forest in Uganda [3] and has likely been circulating in monkey and human populations in Africa and Asia. In the last 10 years, an Asian virus lineage has rapidly spread on an unprecedented timescale around the pacific and the Americas. In humans, the neurotropic virus causes microcephaly in newborns and has been implicated in other neurological disorders such as Guillain-Barre syndrome [4]. The explosive spread of the virus and its effect on infants created a public health emergency and stimulated research efforts to investigate new treatments and vaccines to reduce these conditions. Although significant progress has been achieved concerning the interaction of ZIKV with the mammalian host since the outbreak, we still have a poor understanding of the molecular interplay between the virus and the mosquito host. As vector control is the only viable option for alleviating the diseases caused by ZIKV, a more thorough understanding on these interactions is critical.
Arbovirus infection of mosquitoes elicits complex interactions between the host and the virus. In some cases, the mosquito’s innate immune pathways, which can be antagonistic to viral infection, are provoked by arboviruses. However, these immune pathways appear to be virus specific as the Toll and JAK-STAT pathways are antagonistic to DENV yet do not appear to influence other arboviruses such as Chikungunya virus (CHIKV) or ZIKV [5–8]. In addition to these classical immune pathways, RNA interference (RNAi) and microRNAs (miRNAs) are important components that dictate host-microbe interactions for arboviruses and their mosquito vectors [9–11]. PIWI-interacting RNAs (piRNAs), another group of noncoding small RNAs of 25–30 nt, could also potentially be involved in arbovirus-mosquito interactions [12].
miRNAs are small non-coding RNAs (~22 nt) that regulate gene expression post transcriptionally. In mosquitoes, miRNAs are important in many developmental processes and nutrition [13,14] and it is becoming clear that these molecules are critical in host-pathogen interactions [9,10,15]. Several studies have shown that pathogen infection alters the miRNA expression profile in mosquitoes (reviewed in [11]). This alteration could be due to the host responding to the pathogen or by the pathogen attempting to alter gene expression in the host to make its environment more suitable. For example, the mosquito-borne alphavirus North American eastern equine encephalitis virus (EEEV) alters a host miRNA to avoid the host’s immune response [16]. In Ae. aegypti, infection with DENV alters the miRNA profile [17], with temporal variation in miRNA expression observed with 23 miRNAs altered at 9 day post infection (dpi) compared to five or less at 2 and 4 dpi. In the Asian tiger mosquito, Aedes albopictus, the miRNA, miR-252, increased after a DENV infected blood meal, and inhibition of this miRNA resulted in increased viral copies while overexpression of the miRNA suppressed virus [18]. Taken together, these studies demonstrate that miRNAs can contribute to the complex interactions occurring between invading arboviral pathogens and their mosquito host, and that this interplay likely dictates vector competence.
While our understanding of these pathways on arbovirus vector competence is expanding, there is a dearth of knowledge related to how ZIKV may alter the miRNA profile in the vector or the human host. To address this issue, here we used high throughput sequencing to examine the small RNA profiles after viral infection of the primary ZIKV vector Ae. aegypti. We examined host miRNA, virus-derived short interfering RNA (viRNA) and piRNA profiles at various time points post-infection. Our results provide the first molecular evidence that infection of ZIKV alters the miRNA profile of a host and the mosquito host mounts an RNAi response against the virus.
The ZIKV strain was acquired from the World Reference Center for Emerging Viruses and Arboviruses at the University of Texas Medical Branch (Galveston, TX, USA). The virus was originally isolated from an Ae. aegypti mosquito (Chiapas State, Mexico). ZIKV protocols were approved by the University of Texas Medical Branch Institutional Biosafety Committee (Reference number: 2016055).
Four-six day old female Ae. aegypti (Galveston strain) mosquitoes were orally infected with ZIKV (Mex 1–7 strain) at 2 x 105 focus forming units (FFU)/ml) in a sheep blood meal (Colorado Serum Company). At 2, 7 and 14 days post-infection (dpi) RNA was extracted from whole mosquitoes using the mirVana RNA extraction kit (Life Technologies) following the protocol for extraction of total RNA. Viral infection in mosquitoes was confirmed by Taqman qPCR on ABI StepOnePlus machine (Applied Biosystems) using a ZIKV-specific probe and primers (S4 Table). RNA from ZIKV positive samples was pooled (N = 5) for time points 7 and 14. Limited ZIKV positive samples were detected at day 2, likely due to the virus titer being at the limits of detection for qPCR. For this time point, at least 1 qPCR positive individual was included in each pool. For all time points, three independent pools were used to create libraries for infected and uninfected samples. Control mosquitoes were fed with blood devoid of ZIKV and collected at the same time points and processed in the same way as infected ones.
Small RNA libraries were created using the New England Biolabs small RNA library protocol (New England Biolabs). Library construction used a two-step ligation process to create templates compatible with Illumina based next generation sequence (NGS) analysis. Where appropriate, RNA samples were quantified using a Qubit fluorometric assay (Thermo Fisher Scientific). RNA quality was assessed using a pico-RNA chip on an Agilent 2100 Bioanalyzer (Agilent Technologies). Library creation uses a sequential addition of first a 3’ adapter sequence followed by a 5’ adapter sequence. A cDNA copy was then synthesized using ProtoScript reverse transcriptase (New England Biolabs) and a primer complimentary to a segment of the 3’ adapter. Amplification of the template population was performed in 15 cycles (94°C for 30 sec; 62°C for 30 sec; 70°C for 30 sec) and the amplified templates were PAGE (polyacrylamide gel electrophoresis) purified (147 bp DNA) prior to sequencing. All NGS libraries were indexed. The final concentration of all NGS libraries was determined using a Qubit fluorometric assay and the DNA fragment size of each library was assessed using a DNA 1000 high sensitivity chip and an Agilent 2100 Bioanalyzer. Sequence analysis was performed using the rapid run platform and single end 50 base sequencing by synthesis on an Illumina Hi-Seq 1500 using the TruSeq SBS kit v3.
CLC Genomic Workbench (version 7.5.1) was used to remove adapter sequences and reads with low quality scores from datasets. We applied the quality score of 0.05 as cut off for trimming. As described in CLC Genomic Workbench manual the program uses the modified-Mott trimming algorithm for this purpose. The Phred quality scores (Q), defined as: Q = -10log10(P), where P is the base-calling error probability, can then be used to calculate the error probabilities, which in turn can be used to set the limit for which bases should be trimmed. Reads without 3’ adapters or with less than 16 nt were also discarded from the libraries. Clean data were considered as mappable reads for further analysis. We used small RNA tool in CLC Genomic Workbench to extract and count unique small RNA reads with minimum five sampling count. Tab separated files with the read sequences and their counts were used as input file for novel and homologous miRNA analysis using sRNAtoolbox [19]. All known Ae. aegypti precursor miRNAs reported in miRBase 21 were used as reference for miRNA annotation [20]. The ultrafast short read aligner Bowtie was used to align the reads to the Ae. aegypti genome and the miRNA database. The alignment type “n” was selected and we allowed a maximum of one mismatch in the Bowtie seed region for genome, and known and homologous miRNA database in our mapping parameters. The seed alignment length for Bowtie was 20 for all the analyses. Differential expression of miRNAs between two conditions was calculated and normalized based on the DESeq package with EdgeR [21] on sRNAtoolbox server, and final fold change values were given in log2 scale.
To understand the RNAi activity against ZIKV, we mapped all the small RNAs to the viral genome (Accession No. KX247632). We implemented strict mapping criteria (mismatch, insertion and deletion costs: 2: 3: 3, respectively). The minimum similarity and length fraction of 0.9 between a mapped segment and the reference were allowed in mapping criteria. We ignored reads with more than one match to viral genome in mapping parameters. Mappable reads in all libraries were filtered and only reads with 21 nt in length were selected to check their mapping pattern to negative and positive strands of the virus genome. We also sorted all mappable reads between 25–30 nt to the viral genome for checking any potential piRNA signature.
We used three different algorithms including RNA22 [22], miRanda [23] and RNAhybrid [24] to predict potential miRNA binding sites in all the Ae. aegypti annotated genes (GCF_000004015.3_AaegL2) and ZIKV genome (KX247632). The small RNA sequence was hybridized to the best fitting portion of the mRNA or viral genome by RNAhybrid. We did not allow G:U pairing in the seed region (nucleotides 2–8 from the 5’ end of the miRNA) and forced miRNA-target duplexes to have a helix in this region. Maximum 5 nt were approved as unpaired nucleotides in either side of an internal loop. miRanda also considers matching along the entire miRNA sequence but we ran the program in strict mode which demands strict 5’ seed region (nucleotide 2–8 from the 5’ end) pairing. It takes the seed region into account by adding more value to matches in the seed region. RNA22 v2 is a pattern based target prediction program which first searches for reverse complement sites of patterns within a given mRNA sequence and identifies the hot spots. In the next step, the algorithm is searched for miRNAs that are likely to bind to these sites. We allowed maximum 1 mismatch in the seed region and minimum 12 nt matches in the entire binding site. We set the sensitivity and specificity thresholds to 63% and 61%, respectively. miRNA binding sites on Ae. aegypti mRNAs, which were predicted by all the three algorithms are considered as highly confident miRNA binding sites.
RNA samples were converted to cDNA using a miSCRIPT II RT kit (Qiagen) using the HiSpec buffer to assure that the cDNA produced was derived only from mature miRNA molecules. 5μL of RNA was used per reaction with an average 605ng per sample. One additional reaction was prepared with no RNA template. The reaction was heated on a Mastercycler-Pro thermal cycler (Eppendorf). Real-time PCR was performed using an IQ5 cycler (BioRad) and with Quantitech SYBR master mix (Qiagen). The process was performed using the proprietary-sequence universal primer provided with the kit as the reverse primer and 10 μM of one of nine miRNA-specific forward primers (IDT), the sequence of which is listed in S4 Table. The cDNA was diluted with 60 μL of nuclease-free water per 30 μL of RT product solution, and 2 μL of diluted cDNA was used per reaction. The volumes of the master mix and primers used were those recommended by their manufacturer. Each sample was run in duplicate and the Ct values averaged for further mathematical processing. The amplification program began with 95°C for 15min, followed by forty cycles of 94°C for 15s, 55°C for 30s, and 70°C for 30s. Gene expression analysis was performed using the ΔΔCt (Livak) method [25]. The miRNA expression in each sample was normalized to the expression of U6B small nuclear RNA (RNU6B). Our RT-qPCR results confirmed that U6B remained quite stable across infected and non-infected samples (S1 Fig). For each day, six RNA samples were used: three from mock-infected mosquitoes, and three from ZIKV-infected mosquitoes. For each day post-infection, individual ΔCt values for both mock and ZIKV samples were used to calculate relative difference of expression. “No-template” controls were included on each plate run.
The accession number for the raw and trimmed sequencing data reported in this paper is GEO: GSE97523.
Illumina small RNA deep sequencing platform was used to produce small RNA profiles in ZIKV-infected and non-infected Ae. aegypti mosquitoes. RNA samples were extracted from whole mosquitoes collected at 2, 7 and 14 days post-infection (dpi) to explore host miRNA and RNAi responses to ZIKV infection. ZIKV infection was confirmed in individual mosquitoes by RT-qPCR, which indicated increases in viral load as infection progressed (S2 Fig). We obtained 59.5–61.8 million combined raw reads from the non-infected libraries in day 2, 7 and 14 samples, respectively (S1 Table). From ZIKV-infected libraries, 54.7–84.8 million reads were acquired after combining all the three biological replicates in day 2, 7 and 14 post-infection, respectively (S1 Table). 15–25% of reads were discarded in different libraries due to their low-quality score or lack of adapter sequence. We detected most of the annotated Ae. aegypti miRNAs present on miRBase in our data representing 10–17% of clean reads in different libraries. In all libraries, total read numbers over different lengths showed a peak at 21–22 nucleotides (nt) representing the typical length of miRNAs and short interfering RNAs (siRNAs) (Fig 1). Another smaller peak at 27–29 nt was obtained probably pertaining to PIWI-interacting RNAs (piRNAs), which are common in most insect small RNA libraries.
Small RNA libraries from ZIKV-infected Ae. aegypti mosquitoes showed alteration of miRNA profiles compared with non-infected controls at 2, 7 and 14 dpi. However, only 17 miRNAs were identified as differentially modulated at all the time points, with the majority of them significantly depleted in response to ZIKV infection (Table 1). At day 2, 10 Ae. aegypti miRNAs showed significant changes in their abundance in response to infection. The maximum fold change (FC) was found in aae-miR-286a, aae-miR-2944b-3p and aae-miR-980-3p with log2 FC of -1.82, -1.54 and -1.43, respectively (Table 1). Among all the differentially regulated miRNAs, aae-miR-308-3p showed the most considerable depletion (-3.78) at 7 dpi. These values are comparable with miRNA changes seen after DENV infection [17]. However, our study and the DENV study [17], sequenced miRNAs using RNA extracted from whole mosquitoes. More pronounced changes are likely to be observed when using specific tissues that are infected with virus. Furthermore, comparison of infected and uninfected tissues may be useful in determining tissue-specific versus systemic changes in miRNAs. Only miRNAs aae-miR-2940-3p, which is mosquito specific, and aae-miR-1-5p were significantly enriched in ZIKV-infected libraries at this time point. We spotted less alteration in miRNA profile at 14 dpi libraries despite mosquitoes at this time point having the highest viral load (S2 Fig). Overall, among all the differentially expressed miRNAs due to ZIKV infection, significant declines in miRNA abundances are more pronounced than enrichment. A similar observation was also reported in a previous study with DENV2, where only 4 miRNAs out of 35 modulated miRNAs during the course of infection were enriched in response to DENV infection [17]. Further studies investigating the effect of distinct flaviviruses on miRNA expression in Aedes mosquitoes are required to confirm if depletion is a general response to infection. The abundance of a few miRNAs was altered in more than one time point after ZIKV infection including, aae-miR-309a, aae-miR-308, aae-miR-286b, aae-miR-2941 and aae-miR-989.
To validate the differentially expressed miRNAs, nine miRNAs were selected. For this, RNA samples extracted from non-infected and ZIKV-infected whole mosquitoes at 2, 7 and 14 dpi were subjected to miRNA-specific RT-qPCR. Our results showed broad agreement between qPCR and NGS values. While it is not uncommon to find inconsistences between these two quantification approaches [26,27], in 18 out of 27 cases, the direction of gene expression was the same (i.e. both enriched or both depleted) (Fig 2). Where discrepancies were observed, the trend was for NGS data to indicate depletion of the miRNA, while the qPCR suggested no significant changes. A notable inconsistency was seen with the miRNA miR-308-3p that was seen to be enriched by qPCR but depleted by deep sequencing at 7dpi.
A cell line study using Ae. aegypti Aag2 cells found miRNAs were only mildly affected by DENV infection [28], but in contrast a number of mosquito studies, reported differentially abundant miRNAs in response to a number of arboviruses. However, in most cases, follow up studies to explore the functional significance of those changes and effects on host target genes and virus replication are lacking. Therefore, below we mainly compare the miRNA changes identified in our study with those in previous ones. In Ae. aegypti mosquitoes infected with DENV2, five, three and 23 miRNAs were differentially expressed at 2, 4 and 9 dpi, respectively [17]. Among those, miR-308-3p and miR-305-5p (9dpi) overlap with those in ZIKV-infected mosquitoes at 7 and 14 dpi; in both host-virus systems both miRNAs showed depletion. In Ae. albopictus DENV2-infected mosquitoes, overlapping differentially abundant miRNAs with ZIKV-infected mosquitoes from this study are miR-2940-3p (depleted in DENV, but enriched in ZIKV), miR-263a-5p (depleted in both), miR-308-5p (enriched in both), miR-989 (depleted in DENV, but enriched in ZIKV), and miR-2941 (depleted in both) [27]. In another study from the same group with Ae. albopictus and DENV2 infection specifically in the midgut tissue, three miRNAs (miR-2941, miR-989, miR-2943) were differentially expressed [29], the first two also with change in abundance upon ZIKV infection in this study. Furthermore, miR-989 was found to be depleted in Culex quinquefasciatus mosquitoes by 2.8-fold when infected with WNV [30]; although this miRNA was enriched by about 1.8-fold at 2 and 14 dpi with ZIKV in the present study. miR-980 was also differentially expressed in the Cx. quinquefasciatus-WNV interaction [22].
It appears that the identified differentially expressed miRNAs in different host mosquitoes upon flavivirus infections overlap more with each other than infections with other viruses, such as alphaviruses. For example, none of the major Ae. albopictus miRNAs that were differentially abundant after CHIKV infection (miR-100, miR-283, miR-305-3p, miR-927) [31] were found among the list of differentially expressed miRNAs from this study; although some of the differentially expressed miRNAs as a result of ZIKV infection could be found among miRNAs showing low levels of differential expression in the CHIKV-mosquito interaction. The similarities in miRNA changes in mosquitoes when infected with flaviviruses as compared to alphavirus infections could be due to (1) antigenic differences between flaviviruses and alphaviruses that may elicit slightly different host responses, or (2) differences in replication strategies; for example, production of subgenomic flavivirus RNA (sfRNA) by flaviviruses, which could function as decoys or sponges against host derived miRNA, suppress the RNAi response, and play other important roles in mosquito-virus interaction [32–34]. Interestingly, sfRNA from WNV has been shown to efficiently suppress siRNA and miRNA-induced RNAi pathways in mosquito cells and its engineering into a Semliki Forest virus (SFV, an alphavirus) replicon led to enhanced replication of SFV in RNAi-competent mosquito cells [32]. While alphaviruses do not produce such RNAs and must rely on other mechanisms to deregulate the host RNAi response.
The hypothetical binding sites for all the differentially abundant miRNAs upon ZIKV infection were predicted by command line tools miRanda, RNAhybrid and RNA22 v2 using their default parameters. High confidence potential targets were defined as those containing a unique binding site for each miRNA in all the algorithms, with a maximum of 10 nucleotides shifting. We predicted 898 mRNAs, which can potentially be regulated by the differentially abundant miRNAs upon ZIKV infection (S2 Table). Among these predicted target genes, 247 binding sites were identified for aae-miRNA-980-3p while only six predicted binding sites were detected for aae-miR-308-3p. Although this miRNA showed more profound regulation in response to viral infection (day 7), we only identified Rho GTPase as its predicted target gene (S2 Table). Other predicted binding sites for this miRNA are located on coding regions of some hypothetical proteins. Rho proteins are small signaling G proteins, which are involved in a wide range of cellular functions such as cell polarity, vesicular trafficking, the cell cycle and transcriptome dynamics [35]. Among the predicted targets, a number of immune-related genes were found, such as leucine-rich immune protein and Toll-like receptor, possibly indicating the ability of ZIKV to modulate mosquito immunity. While the list of targets provides a catalogue of high confidence targets of Ae. aegypti differentially abundant miRNAs upon ZIKV infection, further investigations are required to experimentally establish miRNA-target interactions.
Whilst miRNA-target studies have not been carried out on any of the miRNAs reported to be differentially abundant following viral infection in mosquitoes (previous section), except aae-miR-2940-5p, the role of some of these miRNAs are known in other aspects of mosquito or Drosophila biology. For example, a number of the differentially expressed miRNAs upon ZIKV infection were also found differentially expressed upon blood feeding in the fat body tissue [36]. These include, miR-308-5p, miR-263a-5p, miR-305-5p, miR-989, miR-2941, miR-286b, miR-2946. miR-309a, specifically was shown to control ovarian development by targeting the Homeobox gene SIX4 [36], and miR-375 was found highly induced in blood fed mosquitoes regulating a number of mosquito genes, including upregulating cactus and downregulating Rel1 [37]. Application of miR-375 mimic in Aag2 cells led to enhanced DENV replication. While this miRNA was found to be mostly depleted after ZIKV infection (Fig 2), it will be interesting to experimentally test if manipulation of this miRNA could have any effect on ZIKV infection by regulating the Toll pathway. In D. melanogaster, the role of miR-308 in development [38], miR-980 in memory [39], and miR-305 in homeostasis [40] have been reported.
We also screened the ZIKV genome for potential miRNA binding sites of all the 17 modulated miRNAs. Eighty-five possible interactions were identified by three different target predicting algorithms (miRanda, RNAhybrid and RNA22). S3 Table summarizes highly confident binding sites that were predicted by more than one tool. Some miRNAs such as aae-miR263a-5p, aae-miR-286, aae-miR-305-5p, aae-miR308-5p, aae-miR-989 and aae-miR-980-3p can potentially bind to more than one place in the viral genome. Previously, targeting of genomes of RNA viruses by host miRNAs have been reported in mammalian cells [41]. In particular, a number of human miRNAs (hsa-miR-133a, hsa-miR-548g-3p, hsa-miR-223) with potential binding sites in the 5’ and 3’UTRs of different DENV serotypes have been shown to negatively affect replication of the viruses when overexpressed in mammalian cells [42,43]. In mosquitoes, a midgut-specific alb-miR-281 from Ae. albopictus was shown to target the 5’UTR of DENV2 thereby enhancing replication of the virus [44].
Flaviviruses generally produce dsRNA intermediates during their replication, which are the target of their invertebrate host RNAi machinery [10]. The long dsRNAs are recognised and subsequently diced by the ribonuclease Dicer-2 into 21 nt virus-derived short interfering RNAs (viRNAs) that are double stranded and induce the formation of the RNA induced silencing complex (RISC). One of the strands of the duplex is degraded and the other one guides the RISC complex to viral target sequences with complete complementarity. This binding results in the cleavage and degradation of viral RNAs produced during replication of the virus.
To investigate potential RNAi activity against ZIKV, we mapped all the small RNAs to the viral genome (accession no. KX247632). In total, 3,288, 20,360 and 57,867 reads mapped to the viral genome at 2, 7 and 14 dpi, respectively, ranging in size from 15–35 nt. The total number of reads at 14 dpi that mapped to the virus genome accounted for 0.16% of the total small RNA reads at this time point after infection (36,115,068; S1 Table), which is close to the percentage (0.05%) found in DENV2-infected Ae. aegypti whole mosquitoes at 9 dpi [45]. The number could possibly be higher if small RNAs are analysed in specific tissues where virus infection primarily occurs. Using whole mosquitoes, which is a mixture of infected and non-infected tissues, may result in dampening of the percentage of virus-specific small RNAs. While at 2 dpi the distribution of small RNAs was across different sizes, at 7 and 14 dpi the majority of the mapped reads were at 21 nt, typical of viRNA size in mosquitoes (Fig 3A). When only the 21 nt reads were mapped to the viral genome, the number of viRNAs increased dramatically during the course of infection; 201 (2 days), 6,250 (7 days), and 20,732 (14 days). This also confirmed successful replication of the virus in the mosquitoes. In addition, the viRNAs mapped across the entire length of the viral genome, on both positive and negative strands of the viral genome (Fig 3B). The pattern of mapped reads indicated a bias towards the positive strand; 62% to the positive strand and 38% to the negative strand–the percentages were very similar both at 7 and 14 dpi. We did not find distinct hot-spots (large number of viRNA production) across the viral genome, except one towards the end of the NS5 region at both 2 dpi and 7 dpi, which is also present at 14 dpi but not as a pronounced peak among others (Fig 3B). These results confirm that ZIKV is exposed to the mosquito host RNAi response, with the replicative dsRNA intermediates being the major substrate for Dicer-2. These findings are consistent with other examples of flaviviruses [28,45–50].
Virus-derived piRNA-like small RNAs (25–30 nt), which are also referred to as viral-derived piRNAs (vpiRNAs), have been identified in insects infected with flavivirues, bunyaviruses and alphaviruses [45,51–54]. It has been shown that knockdown of the piRNA pathway proteins leads to enhanced replication of arboviruses in mosquito cells, suggesting their potential antiviral properties in mosquitoes. For example, knockdown of Piwi-4 in Ae. aegypti Aag2 cell line increased replication of the mosquito-borne alphavirus, SFV [51]. In another study in the same cell line, specifically silencing Ago3 and Piwi-5 led to significantly reduced production of vpiRNAs against another alphavirus, Sindbis virus (SINV) [55].
To find out if any virus-derived piRNA-like small RNAs are produced in Ae. aegypti mosquitoes infected with ZIKV, we mapped 25–30 nt small RNA reads from the three time points post-infection to the viral genome. The number of reads increased as infection progressed, and they mapped to the entire ZIKV genome with no particular hot spots identified (Fig 4). However, we found a significant bias for reads mapped to the positive strand; for example, in 14 dpi samples 5,300 of 25–30 nt reads mapped to the positive stand and only 60 reads mapped to the negative strand (Fig 4). In DENV2 infected Ae. aegypti mosquitoes, the number of 25–30 nt reads that mapped to the negative strand of the virus were also extremely low. Further, no bias for a specific base or sequence-specific piRNA signature (U1 and A10 bias) was observed in this study, as would normally be expected for ping-pong derived piRNAs [56].
Similar observations were reported in other flavivirus-infected mosquitoes or mosquito cell lines. We recently demonstrated that in Ae. aegypti mosquitoes infected with an insect-specific flavivirus (Palm Creek virus), small RNA reads in the range of 25–30 nt do not harbor any of the classical sequence-specific piRNA features [57]. Hess et al. (2011) also showed that DENV2 piRNA-like sequences do not display any bias for U in the first position and only a slight bias for A10 [50]. However, in mosquito cells infected with alphaviruses SFV [51] and SINV [52], and bunyaviruses La Crosse virus [52] and Rift Valley Fever virus [58] clear U1 and A10 ping-pong piRNA signature was observed. Hence, currently we do not have enough evidence to classify the 25–30 nt reads that mapped to the ZIKV genome as vpiRNA since they might be artefacts of viral genome degradation.
In summary, we found that ZIKV infection in Ae. aegypti altered the small RNA profile of mosquitoes with peaks seen at 21–22 and 27–29 nt. Overall, ZIKV infection modulated 17 miRNAs with the majority of these small RNAs being depleted. Several immune related transcripts were the predicted targets of differentially abundant miRNAs suggesting that ZIKV may interact with mosquito immunity. At 7 and 14 dpi, viral infection initiated an RNAi response indicated by the presence of viRNAs. At these times points, virus-derived small RNAs in the size range of piRNAs were also found in infected mosquitoes, although they lacked the typical piRNA signature. This study increases our understanding of ZIKV-mosquito interactions and broadens our comprehension of the Aedes miRNA response to flavivirus infection.
|
10.1371/journal.ppat.1003570 | Influenza A Virus Migration and Persistence in North American Wild Birds | Wild birds have been implicated in the emergence of human and livestock influenza. The successful prediction of viral spread and disease emergence, as well as formulation of preparedness plans have been hampered by a critical lack of knowledge of viral movements between different host populations. The patterns of viral spread and subsequent risk posed by wild bird viruses therefore remain unpredictable. Here we analyze genomic data, including 287 newly sequenced avian influenza A virus (AIV) samples isolated over a 34-year period of continuous systematic surveillance of North American migratory birds. We use a Bayesian statistical framework to test hypotheses of viral migration, population structure and patterns of genetic reassortment. Our results reveal that despite the high prevalence of Charadriiformes infected in Delaware Bay this host population does not appear to significantly contribute to the North American AIV diversity sampled in Anseriformes. In contrast, influenza viruses sampled from Anseriformes in Alberta are representative of the AIV diversity circulating in North American Anseriformes. While AIV may be restricted to specific migratory flyways over short time frames, our large-scale analysis showed that the long-term persistence of AIV was independent of bird flyways with migration between populations throughout North America. Analysis of long-term surveillance data provides vital insights to develop appropriately informed predictive models critical for pandemic preparedness and livestock protection.
| Despite continuous virological surveillance (1976–2009) in wild waterfowl (Anseriformes) and shorebirds (Charadriiformes), the ecological and evolutionary dynamics of avian influenza A virus (AIV) in these hosts is poorly understood. Comparative genomic analysis of AIV data revealed that the high prevalence of Charadriiformes infected in Delaware Bay is a reservoir of AIV that is phylogenetically distinct from AIV sampled from most North American Anseriformes. In contrast, influenza viruses sampled from Anseriformes in Alberta are representative of the remaining AIV diversity sampled across North America. While AIV may be restricted to specific migratory flyways over short time frames, our large-scale analysis showed that this population genetic structure was transient and the long-term persistence of AIV was independent of bird flyways. These results suggest an introduced virus lineage may initially be restricted to one flyway, but migration to a major congregation site such as Alberta could occur followed by subsequent spread across flyways. These generalized predictions for virus movement will be critical to assess the associated risk for widespread diffusion and inform surveillance for pandemic preparedness.
| Migrating wild birds have been implicated in the spread and emergence of human and livestock influenza, including pandemic influenza and highly pathogenic H5N1 avian influenza [1]–[3]. Viral transmission between wild birds and domestic poultry has contributed to genomic reassortment and confounded disease control efforts [2], [4]. Subsequently, with the reintroduction of H5N1 to wild birds the virus has spread throughout Eurasia and Africa [5]–[9]. While it is contentious as to whether wild birds are the primary vectors spreading H5N1 viruses over long distances, there is little doubt that these animals play a role in confounding disease surveillance and control efforts.
It is estimated worldwide that over 50 billion birds migrate annually between breeding and non-breeding areas [10]. Even though there is evidence that Anseriformes infected with influenza A virus have hampered migration, these hosts vector influenza viruses vast distances [11]–[12]. Disease transmissions between the millions of conspecific birds at congregating sites throughout the world contribute to the genetic variability and reassortment of influenza A viruses [13], [14]. It is not coincidental that these major breeding, feeding, and staging sites are also regions of high viral prevalence [14]–[21].
Recent efforts to assess invasive virological threats have focused on increased surveillance and early detection of introduced viral strains [22]–[24]. Influenza A viruses have transmitted between the Eurasian and North American wild Anseriformes and Charadriformes gene pools where birds from both continental regions commingle and therefore the threat posed by introduction of H5N1 to North America remains. However, once a virological threat has entered the North American bird population there is little information regarding how that virus may behave or diffuse between spatially distant migratory bird populations.
The prediction of viral spread and disease emergence, as well as formulation of preparedness plans has generally been based on ad hoc approaches. This is largely due to a critical lack of knowledge of viral movements between different host populations [13]–[17]. The patterns of viral spread and subsequent risk posed by wild bird viruses therefore remain unpredictable. Methodological advances present an opportunity for large-scale assessment of spatiotemporal patterns of viral movement between migrating bird populations.
In this study we identified 20 discrete regions in North America where influenza viruses have been systematically collected from wild birds to determine whether the viral population was structured according to host migratory flyways, and rates of gene flow between these populations. Avian influenza viruses were isolated annually throughout our surveillance in Alberta, Canada and Delaware Bay, USA and an additional 287 genomes were sequenced. Using full genome data we characterize the reassortment dynamics, spatial diffusion patterns and evolutionary genomics of influenza A viruses in North America collected over a 25-year period from migratory birds.
Avian influenza H3 viruses were among the most frequently isolated influenza subtype from our surveillance in Alberta, Canada and Delaware Bay, USA [17]. We therefore randomly selected 200 H3 subtype isolates collected from 1976 to 2009 – plus an additional 100 influenza isolates of multiple subtypes – for full genome sequencing. Thirteen isolates could not be sequenced and a number of additional isolates were mixed samples containing multiple subtypes. As a result, 163 H3 subtype viruses and 124 isolates of other subtypes were sequenced. The newly sequenced H3-HA genes were analyzed with publically available H3-HA data to estimate the phylogenetic history (number of taxa (ntax) = 531). This large scale phylogeny of globally sampled H3 viruses from wild birds revealed three major lineages, two circulating in North America (Lineages I and II) and a third lineage that is a mix of North American and Eurasian isolates (Figure S1). All gene sequences that were of Eurasian origin were excluded from all further analysis in this study, including those that belonged to the mixed Eurasian/North American lineage.
Comparative genomic analysis of H3 subtype viruses isolated from the Alberta and Delaware Bay sites was conducted to test AIV evolutionary dynamics in different hosts. In Alberta, where birds sampled were primarily juvenile Anseriformes [20] the H3-HA phylogeny showed that H3 viruses were recovered in almost every year (ntax = 94), with both Lineage I and II viruses present (Figure 1A). In contrast, in Delaware Bay, where only Charadriiformes were sampled, H3 viruses were detected in only 7 years (ntax = 69) from 24 years of surveillance (Figure 1B). In those years when H3 viruses were isolated in Delaware Bay, only a single clade was detected each sampling season and no co-circulation of these clades was apparent. While viral prevalence in Delaware Bay and Alberta are similar [17], Anseriformes host a representative diversity of AIV in North America. In contrast, Charadriiformes host limited viral diversity exhibiting local epidemic-like dynamics [25] suggesting Charadriformes in Delaware Bay are being infected from a currently undetected AIV population.
We used multidimensional scaling of times of most recent common ancestor (tMRCAs) and patristic distances for each gene segment (excluding NA) to test differences in reassortment between populations (Figure 1C, D). In this analysis, the spread of each point cloud represents the statistical uncertainty in the phylogenetic history of each gene and we expect non-reassortant genes will have overlapping point clouds [26]. For both Alberta and Delaware Bay these analyses clearly indicate high levels of reassortment and that the evolutionary histories of the HA and internal genes are therefore partially independent, although the HA and PB1 from Delaware Bay show a higher level of similarity.
To evaluate evolutionary dynamics and migration patterns of H3 subtype viruses throughout North America we identified viruses from avian hosts sampled in 20 defined discrete geographic regions excluding those sequences with recently introduced from Eurasia as described above (ntax = 437). The tMRCA of Lineages I and II was estimated to be ∼1942 (95% Bayesian Credibility Interval 1926–1962). The mechanism for maintenance of this deep divergence remains unknown, as viruses from both lineages have co-circulated in geographically overlapping host populations, primarily Anseriformes, throughout the entire surveillance period. One possibility is that this deep divergence is the product of (i) a very large host meta-population and (ii) relatively rare cross-species transmission rate when compared to annual seasonal epidemic dynamics leading to a lack of synchronicity of partial immunity across host species so that more than one lineage can effectively survive long periods of time. Although there was little evidence for geographic structuring of the virus population over extended periods, an obvious exception is a single lineage that has circulated for more than 10 years in birds sampled from Delaware Bay (Figure 2).
Ancestral state reconstruction of virus geographic location suggests that the population of Lineage II was localized in southeast Alberta prior to migrating to other locations across all North American flyways (Figure 2). However, the apparent geographic isolation of viruses from Alberta may be an artifact as sampling in this location began 12 years before other sites. Furthermore, in Lineage I, where sampling was temporally and spatially more consistent, we found no evidence of localized ancestral populations.
We next estimated rates of viral migration between discrete geographic locations treating each gene as an independent dataset to capitalize on the extra historical information generated by genetic reassortment. While each gene segment analysed supported lateral diffusion between migratory flyways over time, analysis of migration paths using single gene segments yielded contradictory answers (Figure S2, S3, S4, S5, S6, S7, S8). For example, the PB1 gene analysis highly supported migration events within the Pacific flyways, although none of the other gene segment analyses did (Figure S4). This is probably a reflection of the high rates of reassortment unlinking the evolutionary history of individual gene segments between subtypes.
We further analyzed all publically available PA, PB1, PB2, NP and M sequence data from wild aquatic birds isolated between 1985–2009 in North America. The HA, NA and NS gene segments were not included in this analysis due to the deep divergence between the subtypes [16]. In this analysis we defined 16 geographic states and a 17th state termed “Other”, that maintained phylogenetic tree structure. The “Other” state included taxa isolated prior to 1998 where few geographic locations were sampled and locations where few isolates were encountered over the surveillance period [27]. This analysis included more than 1300 sequences for each gene. The migration pattern was jointly estimated from all gene datasets in a single analysis even though the taxon number and subtype between each gene dataset was not identical. The phylogenetic tree space was sampled independently for each dataset, but we assumed the migration parameters were linked. These parameters were estimated across all gene trees to elucidate the migration history of the avian influenza population in North American wild birds and showed similar levels of within versus between flyway migration rates (Figure 3). This was confirmed by statistical comparison of these rates, which showed no significant difference in diffusion patterns (mean within flyway rate>mean between flyway rate, Bayes factor (BF) = 0.968; mean between flyway rate>mean within flyway rate, BF = 1.033).
Table 1 shows the mean migration rates for all statistically supported state transitions recovered from our analysis. The diffusion patterns recovered from this analysis show that when all subtypes, hosts and locations are considered there is extensive mixing of influenza A virus between populations (Figure 4). However, it is unlikely that this pattern can be generalized for individual subtypes. For example, analysis of H3-HA gene segments with the six other internal gene segments (excluding NA) showed greater within flyway migration compared to between flyway migration (Figure S2, S3, S4, S5, S6, S7, S8, S10). Surprisingly, we could not reject the null hypothesis that migration rates are unrelated to the distance between locations (Pearson correlation coefficient = −0.037; Mantel test of rates vs distance, p = 0.317, Figure S10). However, the large-scale spatial diffusion and persistence of AIV is facilitated by comingling of birds in congregation sites located where multiple flyways overlap, such as Alberta (Figure 4). Taken together these results suggest that the AIV population mixes extensively and rapidly despite large geographic separation between sampling locations.
Our goal was to understand the migration dynamics and diffusion patterns of influenza virus in their natural hosts by utilizing over 30 years of continuous systematic surveillance data. We show that our surveillance within Alberta, which includes convergence points for all four migratory flyways [28], [29], is capturing the majority of genetic diversity of the North American influenza gene pool. Breeding birds converging in this region facilitate the spread and generation of influenza virus genetic diversity indicating the importance of Anseriformes' social behavior in persistence of the virus population.
The site at Delaware Bay has been identified as a hotspot for avian influenza A viruses [30], where hundreds of thousands of migrating Charadriiformes stopover annually to feed in highly dense congregations. Our results showed limited genetic diversity coupled with high prevalence of infection indicating an epizootic in Charadriiformes that does not play a significant role in the shaping the sampled AIV diversity within North American Anseriformes. Even though this hotspot is not representative of gene pool diversity, these viruses are ultimately derived from the same population of viruses common throughout North America. The transmission of viruses between populations of birds is most likely occurring where migratory Anseriformes and Charadriiformes commingle, possibly in South and Central America or Arctic breeding grounds. The role of Charadriiformes in the persistence and transmission of influenza A viruses therefore warrants further study, especially on a more comprehensive spatial scale.
We show that the long-term persistence of the influenza A virus gene pool in North American wild birds may be independent of migratory flyways. Although virus migration could be restricted within a flyway over short time periods, our results show strong support for longer-term lateral diffusion of viral lineages between host populations. In our study, data points were not assigned to a flyway but discrete sites were assigned and used to inform within and between flyway migration rates using tip-dated time-dependent phylogenetic reconstructions. While this does contradict previous work by Lam et al [27], which suggested that migratory flyways and distance might represent a barrier for migration, both studies show that migration between flyways does occur [27]. Our study shows that the short-term evolutionary consequences of these ecological barriers may be rapidly erased by East-West virus migration, and that such diffusion may be critical for the survival and persistence of novel virus lineages introduced to North American wild birds.
Subtype specific host distribution, geographic state definition and host ecology may also be a source for the differences observed between the two studies [27]. While we found no correlation between distance migrated and rate of migration, analysis of the H3-HA indicated that subtype specific diffusion patterns might be different. In turn this may be related to host specificity of H3 viruses. Furthermore, in our study we cannot detect migration events where the distance migrated is less than 400 km due to the definition we used for geographic states (5′×5′ latitude-longitude square).
The data used in our analysis included collections from resident and short distance migratory birds [31]. This data was unavailable to Lam et al [27], and may further account for the observed differences. In our study we assume that virus migration was the same regardless of host. This assumption may be valid when analyzing viruses from all hosts in a single analysis, it is unlikely to be justified when considering specific hosts. Flyways are often applied universally to all hosts, whereas there are clear differences in the behavior and ecological habits of different hosts (see supporting information Text S1).
Using our model for virus transmission generalized predictions for movement of an introduced Eurasian virus and the associated risk for widespread diffusion can be inferred. An introduced virus lineage to Alaska might initially be restricted to the Pacific Flyway, but migration to a major congregation site such as Alberta could occur with subsequent spread across flyways occurring shortly after. While the establishment of introduced lineages into North America may be rare, introduction and reassortment events with Eurasian and North American strains probably occur more frequently than detected [16], [17], [32].
The development of fully resolved ecological and viral risk models depend upon the continued long-term active surveillance in major bird congregation zones. While the resolution and detection of migration events has been enhanced with increased surveillance in recent years, critical information for wild bird surveillance remains sparse. This is especially evident as no sampling in Central and South America was available for this study. A comprehensive understanding of spatial diffusion patterns of viruses introduced to wild animal populations is critical for the development of preparedness plans in response to emerging viral threats.
Systematic influenza surveillance has been conducted in ducks in Alberta, Canada since 1976, and in shorebirds and gulls at Delaware Bay (Delaware and New Jersey) since 1985. Ducks were sampled post-breeding and prior to southern migration during July through early September at various wetlands in the following regions of Alberta: Vermilion (1976–1978), Grand Prairie/Fairview (1979–1984, 1992–2011), Edmonton/Stettler (1979, 1981, 1983–2009), Brooks (1992–1995), and High River (1993–2000, 2002–2003, 2005–2007). Sampling occurred during duck banding operations conducted by the Canadian Wildlife Service after ducks were captured in swim-in bait traps. Birds banded in Alberta have been recovered in all four North American flyways but most mallards are recovered in the Central and pacific flyways. In 1984 samples were also collected from ducks captured in decoy traps during late April to early May in the Vermilion area. Overall, the majority of samples were obtained as cloacal swabs (n = 18,057) and tracheal/oropharyngeal specimens accounted for most of the remaining samples (n = 1,641; 1,293 of the oral swabs being collected since 2007). Hatch-year ducks were sampled more frequently than after-hatch-year ducks (n = 11,923 versus 7,559, respectively). A variety of duck species were sampled – primarily dabbling ducks. The most abundantly sampled species are mallard (Anas platyrhynchos), northern pintail (Anas acuta), and blue-winged teal (Anas discors) with these three species accounting for 93% of the total specimens. Other species (listed in decreasing rank order of samples obtained) include redhead (Aythya americana), green-winged teal (Anas crecca), american wigeon (Anas americana), gadwall (Anas strepera), canvasback (Aythya valisineria), lesser scaup (Aythya affinis), american coot (Fulica americana), northern shoveler (Anas clypeata), bufflehead (Bucephala albeola), cinnamon teal (Anas cyanoptera), common goldeneye (Bucephala clangula), ruddy duck (Oxyura jamaicensis), greater scaup (Aythya marila), hooded merganser (Lophodytes cucullatus), and wood duck (Aix sponsa).
Fecal samples from Charadriiformes – shorebirds and gulls - were collected in May at Delaware Bay from ruddy turnstone (Arenaria interpres), red knot (Calidiris canutus), semipalmated sandpiper (Calidris pusilla), sanderling (Calidiris alba), and dunlin (Calidris alpina) starting in 1985 and continuing to the present. Samples were also obtained from breeding colonies of gulls – primarily laughing gull (Larus atricilla) and herring gull (Larus argentatus). It is during this period in May that shorebirds (waders) are migrating north from South America to their breeding grounds in the Canadian Arctic. Delaware Bay serves as a stopover point where the birds can re-fuel on the abundance of eggs deposited by the coincident spawning of horseshoe crabs (Limulus polyphemus).
Although most of the 10,350 samples obtained were from freshly deposited feces on beaches we also collected 213 cloacal swabs from captured birds spanning the years 1986–1989 and 2000. A subset of 440 samples was collected outside of the May surveillance period at the following times; September 1985, September and November 1986, and June-September 1988. It should be noted that from 1988 through 2002 multiple swabs (usually 3) were combined to constitute a single sample vial. In the years prior to 1988 most sample vials contained an individual swab, and all samples since 2003 have been from single fecal deposits.
Approximately 19 sample sites were established around Delaware Bay and varied from year-to-year. Six sites were used on the west side of Delaware Bay in Maryland and Delaware from 1985 through 1989. Sampling was performed at 13 sites on the east side of the bay in New Jersey in all years. Table S1 summarizes prevalence and bird population estimates from Delaware Bay, the Prairie pothole region and the central flyway [33]–[37].
The majority of the swabs were derived from fecal deposits and therefore it was not possible to identify the species that served as the source of the sample in over half of the specimens. However, the birds tend to congregate in groups of like species, and gull feces were easily discriminated from other bird droppings, therefore in many instances we could attribute the source of the sample to a particular species. Otherwise the sample was considered “shorebird” or “gull”.
Swabs were collected using a dacron tipped applicator and placed in transport medium containing 50% phosphate buffered saline and 50% glycerol adjusted to pH 7.2 and supplemented with penicillin G, streptomycin, polymyxin B, gentamycin, and nystatin. In Alberta the duck swabs were placed immediately in liquid nitrogen and returned to the laboratory. Shorebird samples from Delaware Bay were immediately placed on ice and shipped to the laboratory within 6 days of collection. Storage of the specimens prior to testing was at −70°C.
Viruses were isolated in 10-day-old embryonated chicken eggs as previously described [38], [39]. Virus subtypes were determined by antigenic analysis in hemagglutination inhibition tests [38], neuraminidase inhibition tests, and/or by RT-PCR [40] and sequence analysis.
Through exploratory examination of surveillance records from Alberta and Delaware Bay we determined that H3 subtype viruses have been most frequently isolated throughout the time period 1985–2009. We therefore focused our sequencing efforts on this time period and randomly selected 200 viruses for full genome sequencing. This data was further supplemented with an additional 100 viruses randomly selected for genomic sequencing of various subtypes.
All samples were sequenced using a high-throughput Next-Generation sequencing pipeline at the JCVI that includes the 454/Roche GS-FLX and the Illumina HiSeq 2000. Viral RNA was first reverse transcribed and amplified by multi-segment RT-PCR (M-RTPCR) [41], which simultaneously and specifically amplifies all influenza A virus segments in a single reaction, irrespective of the virus subtype. The amplicons were barcoded and amplified using an optimized SISPA protocol [42]. Barcoded amplicons were quantitated, pooled and size selected (∼800 bp or ∼200 bp) and the pools were used for Next Generation library construction (50–100 viruses/library).
One library was prepared for sequencing on the 454/Roche GS-FLX platform using Titanium chemistry while the other was made into a library for sequencing on the Illumina HiSeq 2000. The sequence reads from the 454/Roche GS-FLX data were sorted by barcode, binned by sample, trimmed, searched by TBLASTX against custom nucleotide databases of full-length influenza A segments downloaded from GenBank to filter out both chimeric influenza sequences and non-influenza sequences amplified during the random hexamer-primed amplification. For each sample, the filtered 454/Roche GS-FLX reads were then binned by segment, and de novo assembled using CLC Bio's clc_novo_assemble program. The resulting contigs were searched against the corresponding custom full-length influenza segment nucleotide database to find the closest reference sequence for each segment. Because of the short read length of the sequences obtained from the barcode-trimmed Illumina, HiSeq 2000 these were not subjected to the TBLASTX filtering step. Both 454/Roche GS-FLX and Illumina HiSeq 2000 reads were then mapped to the selected reference influenza A virus segments using the clc_ref_assemble_long program.
At loci where both GS-FLX and Illumina sequence data agreed on a variation (as compared to the reference sequence), the reference sequence was updated to reflect the difference. A final mapping of all next-generation sequences to the updated reference sequences was then performed. Any regions of the viral genomes that were poorly covered or ambiguous after Next Generation sequencing were PCR amplified and sequenced using standard Sanger sequencing approach.
Through sequencing, some of these selected viruses have been identified as more than one isolate (“Mixed” in table S3). The direct sequencing method does not allow us to determine which internal gene segments are associated with which subtype. Furthermore, some variants could not yield unique gene sequences for each potential virus identified. Hence, some mixed variants contain more than 8 associated sequences, but fewer than 16. As such, these were not included in the analysis of genomic reassortment patterns. Other variants could not be completely sequenced and have subsequently been submitted as “Draft.” Out of the 300 variants submitted for sequencing, 287 full genomes have been completed. All data generated for this study has been made publicly available via the Influenza Virus Resource at NCBI [43] (Accession numbers CY101081to CY103740).
We analyzed 1441 genomic sequences of influenza A viruses in wild birds (Table S2 shows NCBI accession numbers). For each dataset prepared we removed all recent introductions from Eurasia and focused this study solely on viral gene segments that have been circulating in North America for the last 25 years. Each internal gene dataset contained >1300 sequences. While no whole genomes with Eurasian origins were evident in the datasets examined, numerous reassortant genes with recent Eurasian ancestry were detected. The neuraminidase (NA) gene was not included in the analysis due to the deep divergence between NA subtypes, while distribution of locations and time was sparse or inconsistent for individual NA genes. However, H3-HA gene sequences were sampled throughout North America and we therefore analyzed all H3-HA gene sequences isolated from wild aquatic birds (ntax = 437).
We used time-stamped sequence data with a relaxed-clock Bayesian Markov chain Monte Carlo method as implemented in BEAST v1.6.2 and BEAST 2 for phylogenetic analysis [44], [45]. For all analyses we used the uncorrelated lognormal relaxed molecular clock to accommodate variation in molecular evolutionary rate amongst lineages, the SRD06 codon position model, with a different rate of nucleotide substitution for the 1st plus 2nd versus the 3rd codon position, and the HKY85 substitution model then applied to these codon divisions [46]. This analysis was conducted with a time-aware linear Bayesian skyride coalescent tree prior over the unknown tree space with relatively uninformative priors on all model parameters a normal prior on the mean skyride size (log units) of 11.0 (standard deviation 1.8) [47]. We performed three independent analyses of 50 million generations. These analyses were combined after the removal of an appropriate burn-in (10%–20% of the samples in most cases) with 5000 generations sampled from each run for a total of 15,000 trees and parameter estimates.
We further compared relative genetic diversity and reassortment patterns of viral isolates from Alberta and Delaware Bay by estimating phylogenies as described above for these populations independently.
Analysis of migration paths using single gene segments yields answers that do not have to agree with each other, due to multiple factors such as sampling bias and/or reassortment. Therefore, we implemented one inclusive analysis of all genes in which each gene is treated as an independent dataset, but shares the migration parameters with all other genes. In order to estimate migration patterns for a single subtype as well as an average migration pattern of the entire AIV gene pool we devised two datasets. The first dataset focused on seven gene segments from H3 influenza A (excluding NA) as this was the most commonly isolated subtype throughout the surveillance period in both Alberta and Delaware Bay. Secondly, we analyzed all publically available PB1, PB2, PA, NP, M gene segments (excluding recent introductions from Eurasia) to estimate the viral migration patterns across the entire population of birds regardless of subtype. HA, NA and NS genes were not included due to the deep divergence between subtypes. This latter analysis resulted in a dataset of more than 1300 sequences for each of the five genes included.
While the phylogeny and substitution rates were separate for each gene, based on a joint migration process a single migration matrix was estimated. We used a reversible continuous-time Markov chain model to estimate the migration rates between geographical regions and the general patterns of avian influenza A virus circulation in different populations [48]. In these analyses we used a constant-population coalescent process prior over the phylogenies and uncorrelated lognormal relaxed molecular clocks. Here we identified 16 discrete geographic regions, based on observed sampling locations, estimated from a 5′×5′ latitude-longitude square (Supporting Data Files; File S1, Table S2, S3, Figure S12), plus an additional character state containing taxa isolated prior to 1998 and locations with fewer than four sequences isolated. We selected discrete geographic sites based on the grid instead of assigning taxa to discrete flyways as these vary to a large degree between potential host populations and overlap between geographic zones. By defining the discrete characters in such a manner we were able to group a number of sampling sites and establish a parameter limit that could be addressed by the data available. A limitation of this approach is that migration rates between locations less than 400 km could not be detected. The ancestral states were mapped onto the internal nodes of phylogenetic trees sampled during the Bayesian analysis (Supporting Data Files; Figures S2, S3, S4, S5, S6, S7, S8). Given the large number of states, a Bayesian stochastic search variable selection (BSSVS) was employed to reduce the number of parameters to those with significantly non-zero transition rates [48]. The BSSVS explores and efficiently reduces the state space by employing a binary indicator (I) [48]. From the BSSVS results, a Bayes factor (BF) test can be applied to assess the support for individual transitions between discrete geographic states. The BF was deemed statistically significant where I>0.5 and the BF>6 from the combined independent analyses. Therefore our minimal critical cutoff for statistical supports were 6≤BF< 10 indicating substantial support, 10≤BF<30 indicating strong support, 30≤BF<100 indicates very strong support and BF>100 indicating decisive support [48]–[50]. Within flyway rate estimates were compared with between flyway rate estimates to determine if migration of the viral population was structured by flyway. The Pearson correlation coefficient and the Mantel statistical test of correlation (100000 permutations) were conducted to test correlation between migration rate and distance between sites.
We used multidimensional scaling plots to visually assess the strength of reassortment in Alberta and Delaware Bay. In this analysis the tree-to-tree variation in branch lengths is visualized as a cloud of points where the centroid of the cloud represents the mean from the 500 trees used in the analysis. Here we assume that gene segments with similar evolutionary histories will occupy the similar locations in the 2-dimensional Euclidean space where the cloud of points should overlap. We used two metrics to assess the degree of reassortment of the influenza A virus populations in the two discrete sampling regions: the time to the most recent common ancestor (tMRCA) or patristic distances calculated from a posterior distribution of trees. From a posterior distribution of phylogenetic trees we estimated the tMRCA for influenza A viruses sampled in each location from each gene during each year and computed the correlation coefficient of the tMRCAs between each pair of trees. This method of tree to tree comparisons has been applied to seasonal influenza A viruses [26] where the uncertainty of the phylogenetic history in the Bayesian posterior sampling of trees for each influenza A gene segments was compared using the tMRCA estimated for annual seasonal influenza A virus outbreaks in two geographic locations.
In our data sets there was a sparseness of sampling through time, especially in Delaware Bay. Therefore we encountered high levels of uncertainty where no clear pattern was discernable and zero distances between trees resulted in computational errors by using the tMRCA to estimate phylogenetic uncertainty between gene trees. To overcome this we computed the correlation matrix of the pairwise tree distances. Here we calculated the correlation coefficient for each pair of trees using the patristic distances between every taxon, where the patristic distance is the sum of branch lengths between two nodes. The dissimilarity matrix was obtained by calculating one minus the correlation matrix.
All animal experiments were performed following Protocol Number 081 approved on August 19, 2011 by the St. Jude Children's Research Hospital Institutional Animal Care and Use Committee in compliance with the Guide for the Care and Use of Laboratory Animals, 8th Ed. These guidelines were established by the Institute of Laboratory Animal Resources and approved by the Governing Board of the U.S. National Research Council.
|
10.1371/journal.pntd.0006455 | Molecular detection of Mycobacterium ulcerans in the environment and its relationship with Buruli ulcer occurrence in Zio and Yoto districts of maritime region in Togo | Buruli Ulcer (BU) is a neglected tropical skin infection caused by Mycobacterium ulcerans. Residence near aquatic areas has been identified as an important source of transmission of M. ulcerans with increased risk of contracting Buruli ulcer. However, the reservoir and the mode of transmission are not yet well known. The aim of this study was to identify the presence of M. ulcerans in the environment and its relationship with Buruli ulcer occurrence in Zio and Yoto districts of the maritime region in south Togo.
A total of 219 environmental samples including soil (n = 119), water (n = 65), biofilms/plants (n = 29) and animals’ feces (n = 6) were collected in 17 villages of Zio and Yoto districts of the maritime region in Togo. DNA of M. ulcerans including IS2404 and IS2606 insertions sequences and mycolactone ketoreductase-B gene (KR-B) was detected using real time PCR amplification (qPCR) technique. In parallel, clinical samples of patients were tested to establish a comparison of the genetic profile of M. ulcerans between the two types of samples. A calibration curve was generated for IS2404 from a synthetic gene of M. ulcerans Transposase pMUM001, the plasmid of virulence.
In the absence of inhibition of the qPCR, 6/219 (2.7%) samples were tested positive for M. ulcerans DNA containing three sequences (IS2404/IS2606/KR-B). Positive samples of M. ulcerans were consisting of biofilms/plants (3/29; 10.3%), water (1/65; 1.7%) and soil (2/119; 1.5%). Comparative analysis between DNA detected in environmental and clinical samples from BU patients showed the same genetic profile of M. ulcerans in the same environment. All these samples were collected in the environment of Haho and Zio rivers in the maritime region.
This study confirms the presence of M. ulcerans in the environment of the Zio and Yoto districts of the maritime region of Togo. This may explain partially, the high rates of Buruli ulcer patients in this region. Also, water, plants and soil along the rivers could be possible reservoirs of the bacterium. Therefore, Haho and Zio rivers could be potential sources of infection with M. ulcerans in humans in these districts.
| Buruli ulcer is a skin disease caused by Mycobacterium ulcerans. Although residence near aquatic areas has been identified as an important source of increased risk of contracting Buruli ulcer, the reservoir and the mode of transmission are not well known. To improve the understanding of the mode of transmission of M. ulcerans in human, we report here the first study of detection of M. ulcerans in the environment and its relationship with BU occurrence in Togo. This study confirms the presence of M. ulcerans DNA in the environment of Zio and Yoto districts of the maritime region. This study provides information on some possible reservoirs of the bacterium such as water, plants/biofilms and soil. Also, in this article we showed that the rivers of Haho and Zio could be potential sources of infection of M. ulcerans in human. Finally, data obtained from this study could explain partially the high rate of Buruli ulcer cases in these two districts of the maritime region in south Togo.
| Buruli ulcer (BU) is an infectious skin disease caused by the Mycobacterium ulcerans [1–3]. BU is the third most common mycobacterial disease after tuberculosis and leprosy in immunocompetent hosts. Although the rate of mortality of Buruli ulcer is low, the serious morbidity caused by the disease includes functional disabilities that may result in permanent social, economic and developmental problems. At least 50% of those affected by BU are children aged less than 15 years. Rates of infections among males and females are equivalent [1–4]. Infection with M. ulcerans often leads to extensive destruction of skin and soft tissue with the formation of large ulcers, commonly on limbs [4]. Necrosis and ulceration are induced by a diffuse cytotoxic macrolide lipid called mycolactone, which represents the key of the pathogenesis of the disease. Mycolactone is the product of three major complex enzymes called polyketide synthases which are coded by mlsA1 (51Kb), mlsA2 (7 Kb) and mlsB (42 Kb) genes. These genes are located on the plasmid of virulence of the mycobacterium known as pMUM001[5–6].
To date, BU cases have been reported in over 30 countries, particularly in tropical and subtropical climate regions but also in temperate climate zones such as Japan and southern Australia [1–4]. BU is a neglected tropical disease (NTD) with a poorly known global prevalence and mainly affects remote rural African communities [7]. According to the WHO (2016), from an estimated 7,000 BU cases reported annually worldwide and more than 4,000 cases occurred in Sub-Saharan Africa. The largest numbers of reported BU cases were from West African countries, particularly from Ivory Coast (about 2,000 cases annually), Benin and Ghana, each of which reported about 1,000 cases a year (WHO, 2016) [1–4].
All BU cases reported must be confirmed by laboratory techniques as recommended by WHO such as direct smears for detection of acid fast bacilli (AFB), in vitro culture and PCR amplification targeting IS2404 sequence [8–9]. However, given the fact that M. ulcerans is a slow-growing bacterium, it may take 8 to 12 weeks to confirm a case by culture and this could delay the implementation of the treatment [8–9]. The development of the conventional PCR technique targeting IS2404 sequence, an insertion sequence present in more than 200 copies per M. ulcerans genome is therefore considered a more sensitive and faster technique to confirm BU cases [10–11]. This method has been also used for testing environmental samples and allowed to detect IS2404 insertion, suggesting a probable presence of M. ulcerans in water samples [12–14], aquatic insects [15], plants [16] and fish [17]. Although this technique is highly sensitive and specific for M. ulcerans detection in clinical samples, its application on environmental samples remains difficult and non-specific due to the presence of PCR inhibitors and the existence of other environmental mycobacterial species harbouring IS2404 such as Mycobacterium lifandii, Mycobacterium pseudoshottsii and Mycobacterium marinum [18–20]. In order to increase the level of specificity and reliability of PCR results as well as increasing testing speediness of clinical and environmental samples, Fyfe et al. [21] have developed two multiplex real time PCR (qPCR) targeting two insertion sequences (IS2404 and IS2606) and Ketoreductase-B domain gene. This new method allowed to distinguish M. ulcerans from other Mycobacterium species that also contain IS2404 sequence [21].
Several epidemiologic studies in Africa [12,21–23] and Australia [13–24] have identified aquatic sources as important sources of M. ulcerans transmission with a high risk of contracting Buruli ulcer. However, the exact mechanism of the transmission of the bacterium is still unknown. The absence of evidence for human-to-human transmission suggests that M. ulcerans is an environment microorganism [25]. Human-linked changes in the aquatic environment such as dam constructions on rivers, deforestation, agriculture and mining have led to environmental disturbance and may contribute to the spread of M. ulcerans [26–27]. This could increase the prevalence of Buruli ulcer cases in endemic areas and lead to the emergence of the bacterium in areas where the pathogen was previously absent [26].
In order to improve the understanding of the mode of transmission of M. ulcerans, it is important at a first stage, to determine the ecology of the bacterium. Thus, some studies used the method described by Fyfe et al. and have identified M. ulcerans reservoirs by detecting DNA in multiples environment samples [12–13,22,28,29].
In Togo, the first cases of Buruli ulcer have been described in 1996 by Portaels et al [30]. Since 2007, several collaborations with German leprosy and tuberculosis relief association in Togo (DAHWT) and the department of infectious and tropical medicine (DITM) of the University of Munich like the BuruliVac project between 2011 and 2013 have proved that Buruli ulcer is endemic in the maritime region in south Togo [31–32]. The availability of a national reference laboratory using PCR technique allows to confirm every year about 30 to 65 new cases of which 85% are from Zio and Yoto districts [32]. Some studies [31–37], mainly clinical were carried out in Togo on BU. However, no environmental data exist on this disease and the risk of infection in human in this country. The present study aims to determine the presence of M. ulcerans in the environment and its relationship with the Buruli ulcer occurrence in Zio and Yoto districts of the maritime region of Togo.
We conducted a cross-sectional study in two districts of maritime region in south Togo. The sample collection method was based on a non-standardised sampling. Then environmental samples were collected from May 19 to 30, 2015 in 17 villages of Zio and Yoto districts where more than 85% of confirmed BU patients originated.
The sampling sites are in maritime region of south Togo which covers the entire coastal part with an area of 6,359 km2. The population is estimated at 1,762,518 inhabitants in 2012. The climate is tropical and humid with two rainy seasons and two dry seasons. The region has a flat topography, with a low contrast characterized by a sedimentary basin that covers 4/5 of the region, a low altitude (50-80m on average) and crossed by the depression of the Lama. The soil, mainly clay remains soggy and muddy in the rainy season with stagnant water for several months. The hydrographic network comprises 3 large rivers which are on the one hand the Mono in the east and other hand, in the center, the Zio and the Haho. Both Zio and Haho have several small tributaries and flow into the “lac Togo” (Fig 1). All these streams have a low flow, closely linked to seasonal variations of precipitations [38].
Samples were collected by three people consisting of two laboratory technicians and a health community volunteer (CHV). The volunteer has served as guide to find various collection sites in the villages. The samples consisted in water, plants, soil and animal feces.
Water samples were collected from ponds, open borehole, cisterns, pumps and borehole. At each point, 50 ml of water were taken and put in the Falcon tubes (greiner bio-one). In the rivers, samples were taken in the middle and at the edges upstream and downstream.
Two to three most frequent plants or herbs (Nymphea lotus, polygonome senegalensis, Ludwigia erecta, Pistia stratiotes, panicum maximum) were collected from inside and along the edges upstream and downstream of the rivers. Each sample consisting of roots, stems and leaves was put in a same plastic sealable bag. To build up the biofilm, 50 ml of sterile water were added in the bag. Biofilms made were taken after 24 to 48 hours and put in 2 ml tube (Eppendorf).
One gram of soil was taken at the surface and put in a 2 ml tube from houses as well as around of ponds, open cisterns, wells and pumps. For rivers, samples were taken along, upstream and downstream and at 5 m of the edge.
Feces samples were taken from chicken, goats, sheep or cattle from house, henhouse and livestock farms.
All samples collected were stored in refrigerator at 4°C until laboratory analyses.
The detection of M. ulcerans in environmental samples was performed using the following protocol (http://dx.doi.org/10.17504/protocols.io.pb7dirn).
Prior to DNA extraction, plant, soil and fecal samples were homogenized using the FastPrep-24 instrument (ver. 6004.2) at the laboratory of the Togolese agricultural research institute (ITRA, Lomé, Togo). Briefly, 200μl of each sample of soil, plant, and animals’ feces was transferred to a lysing matrix E tube in the presence of MT Buffer and sodium phosphate buffer After a rapid homogenization on the FastPrep-24 instrument for 40s at a speed setting at 6.0, the Matrix E tubes were directly centrifuged at 14000xg for 10 minutes to pellet debris.
For the biofilm and water samples, 200μl were transferred into the Matrix E tube and directly centrifuged at 14000xg for 10 minutes. After completion of the centrifugation, the supernatant was collected for DNA extraction.
The DNA was extracted at the molecular laboratory of the national institute of hygiene (INH, Lomé, Togo) using FastDNA Spin Kit for Soil for the 4 types of specimens following the recommendations of the manufacturer. One negative extraction control (sterile water) for each extraction batch has been added to check for a possible cross-contamination during the extraction process.
Real time PCR (qPCR) was performed as previously described by Fyfe et al. [21]. Three primers pairs with probes targeting sequences of two insertions sequences (IS2404/IS2606) and Ketereductase B-domain gene (KR-B) present on the M. ulcerans virulence plasmid pMUM001(GenBank accession no. BX649209) were used (Table 1). These targets were chosen because they were reported to be present in multiple copies in the M. ulcerans genome and are absent in the closely related species Mycobacterium marinum [19, 21].
To confirm the presence of M. ulcerans in an environmental sample, three consecutives qPCR runs were realized.
The first real time IS2404-qPCR run was quantitative using a Taqman probe targeting IS2404 with three controls. An internal positive control (IPC,) to determine the level of inhibition, a no template control (NTC) and a positive control included in quadruplicate. A calibration curve was generated based on a serial dilution of known copies of IS2404 from a synthetic gene of M. ulcerans Transposase pMUM001 (58 bp), the plasmid of virulence. All the dilutions of IS2404-DNA were tested both with samples to determine the sensitivity of the qPCR.
The amplification reaction was obtained from a mixture of 3μl of DNA extracted and 22μl of master mix which contained 0.5μl IPC DNA, 2.5μl IPC master mix, 1.25μl IS2404 forward primer, 1.25μl IS2404 reverse primer, 1.25 IS2404 probe, 2.75μl water and 12.5 μl TaqMan Environmental Master mix 2.0. The reaction was run on ABI 7300 machine in the following conditions: 50°C for 2 minutes, 95°C for 15 minutes and 40 cycles of 95°C for 15 seconds and 60°C for 60 seconds.
All samples for which the internal control IPC and the IS2404 insertion did not show the amplification curve were considered as inhibited samples. All inhibited samples were tested again after 10-fold dilution in DNase/RNase free water.
Samples found positive for IS2404 have consecutively been tested in a semi-quantitative IS2606-qPCR run.
The internal control IPC was no longer used, and the master mix prepared including 1.25μl of the IS2606 forward primer, 1.25μl IS2606 reverse primer, 1.25 μl IS2606 probe, 5.75μl water and 12.5 μl du TaqMan Environmental Master mix 2.0. The amplification reaction was carried out from a mixture of 3μl DNA extract and 22μl master mix with the amplification conditions as described in the IS2404-PCR.
After the second run, all samples that were positive for both insertions sequences (IS2404/IS2606) were analyzed for detection of the mycolactone gene, KR-B.
In this semi-quantitative qPCR, the KR gene was amplified from samples positive for the two insertions sequences (IS2404 and IS2606). The amplification reaction was composed of a mixture of 5μl DNA extract and 20μl of master mix in the same conditions as the IS2606-qPCR.
All samples were tested in duplicate. An environmental sample was considered positive for M. ulcerans if the qPCR was positive for the two insertions sequences (IS2404/IS2606) and KR-B gene (with a threshold cycle, Ct<40) in replicates and if the difference (ΔCt) between (IS2404-IS2606) was < 7.
To compare the genetic profile of the M. ulcerans strains detected in the clinical and environmental samples, 50 DNA extracts from clinical samples of BU patients were tested for the two insertions sequences (IS2404/IS2606) and KR-B gene in the same conditions of qPCR as for environmental samples. The clinical DNA was obtained from extraction of 31 liquid of fine needle aspiration (FNA) from nodule and 19 swabs collected from ulcers. The DNA extraction was performed with the Gentra Purgene DNA extraction kit as previously described [32].
Statistical analysis was carried out by SPSS software (Statistical Package for Social Science, Version 16.0, SPSS Inc. and Chicago, IL). Student t-test was used for comparison of proportion of IS2404 and IS2606 insertions sequences and KR-B gene between different matrices with significant level set at p≤0.05.
The study protocol was approved by the National Program for Buruli Ulcer Control, (Authorization No.006/2014/MS/DGS/DSSP/PNLUB-LP) and the Ministry of Health as an integral part of the surveillance of the disease. However, this study did not require a review of the ethics committee.
Accession number of sequences on M. ulcerans gene:
Whatever the qPCR run, all the negative extraction control and the no template control, NTC did not show an amplification curve. This means the absence of any contamination during the extraction process and from the water used to prepare the master mix. All samples that did not generated an amplification curve were considered negative to the targeted sequences.
The internal control (IPC) and the positive control showed exponential amplification curves. All sample having an exponential curve like the positive control with the Ct<40, was considered as positive to the targeted sequence.
M. ulcerans genetic profile detected in environmental samples.
A total of 219 samples were analysed using real time PCR (qPCR) technique to determine IS2404 and IS2606 insertions and KR-B gene sequences.
for the IS2404-qPCR, 10 (5%) samples did not show an amplification curve neither for the internal control IPC nor the IS2404 sequence indicating an inhibition reaction. After 10-fold dilution and retesting, these samples were negative for IS2404 sequence. Overall, 37 (17%) out of 219 samples analysed were tested positive for IS2404 insertion with Ct values ranging from 26.6 to 38.3, suggesting a probable presence of M. ulcerans in environmental samples (Table 2). The calibration curve generated by IS2404-qPCR results of plasmid standards has shown high detection sensitivity up to 0.01 copies of the M. ulcerans genome in a sample (S1 Table).
To confirm the presence of M. ulcerans in environmental samples, all the above mentioned IS2404 positive samples were also tested for IS2606 insertion sequence and KR-B gene. Thus, out of 37 IS2404-positive samples, 14 (38%) samples were tested positive for IS2606 with Ct values varying from 33.3 to 38.3. The difference (ΔCt) between IS2404 and IS2606 (IS2606-IS2404) was analysed and presented in Table 2. The mean Ct difference between IS2404 and IS2606 was 1.1 (interval from 0.3 to 1.9) (Table 2). This ΔCt value was < 7 indicating that DNA amplified belongs to M. ulcerans strains, lineage 3 which are found in human lesions and contain high copies of IS2606 per genome and not for other mycobacteria or non-virulent mycobacteria (lineage 1) which are fish and frog pathogens, or lineage 2 M. ulcerans, both of which harbor only few copies of IS2606 [19–21].
Finally, 6 (43%) out of 14 samples that containing both IS2404 and IS2606 insertions sequences were tested positive for mycolactone KR-B gene. In conclusion, we detected 6 (2.7%) samples positive for M. ulcerans (IS2404/IS2606/KR) out of 219 analysed (Table 2).
To identify possible reservoirs of M. ulcerans in the environment of BU patients, samples from different sources were tested and consisted of water (n = 65), plants (n = 29), soil (n = 119) and animal faeces (n = 6) (Table 3). The distribution of the number of samples tested overall and tested positive according the source is presented in Table 3.
Overall, IS2404 positivity rate was not significantly different (p = 0.79) between water source (15%), plants (17%), soil (17%) and animals’ feces (17%) (Table 3). For IS2606 sequence, a high positivity rate was observed for water (40%) and biofilms/plants (50%) but were not significantly different from soil samples (35%; p = 0.84) (Table 3). The KR gene was found at 100% in the biofilm/plants samples compared to water (25%) and soil (29%). However, there was no significant difference of KR-B proportion between these three matrices (p = 0.99) (Table 3).
In Table 3, we observed that 1/10 water sample from open borehole/cisterns was tested positive for IS2404 but did not reveal amplification of IS2606 and KR gene sequences. From water of pump/borehole, no sample was tested positive for both insertions sequences and the KR gene. Stagnant water samples (5/14) and water from rivers (4/30) were tested positive for IS2404. The two matrices had high positivity rates for IS2606 (≥40%). However, the KR-B gene was only detected in water samples (50%) collected form rivers (Table 3).
Concerning biofilms/plants, samples collected from ponds did show amplification for the three sequences. However, samples collected from rivers, at the surface and at along edges were positive for IS2404 in 25% (5/20) and IS2606 in 40% (2/5) insertions sequences. The positivity rate of KR-B gene was found in the 2 samples (100%) tested IS2606-positive (Table 3).
For soil samples, IS2404 insertion was positive in samples collected along river banks (8/20), around stagnant water (2/20) and in homes and other locations (10/74). However, IS2606 sequence was detected in 37% (3/8) of samples collected along river banks and in samples from homes and other locations in 40% (4/10). The KR-B gene was detected in these two sources in more than 25% (Table 3).
Only one sample of animal feces was tested positive for IS2404 but negative for other sequences (IS2606/KR-B gene) (Table 3).
In summary, M. ulcerans including three sequences (IS2404/IS2606/KR) were detected in matrices consisting of water in 1.5% (1/65); biofilms/plants in 10.3% (3/29) and soil in 1.7% (2/119) (Table 3). However, there was no significant predominance (p = 0.96) between all sources of matrixes tested.
This comparison aims to establish a link between the disease and M. ulcerans in the living milieu of Buruli ulcer patients. Then, 50 clinical samples of BU patients have been tested using qPCR in the same conditions as describe for environmental samples. The results of qPCR for clinical samples with comparison with environmental samples are presented in Table 4. The real time PCR had showed that all DNA tested from clinical samples was positive for both the two insertions (IS2404/IS2606) and the KR-B gene sequences (Table 4).
The analysis of the mean ΔCt of the difference between IS2404 and IS2606 had showed that the ΔCt was < 7 for the swabs and FNA (Table 4). This mean ΔCt confirms that DNA detected in clinical samples is belonging to M. ulcerans and not for any other mycolactone- producing mycobacteria [19, 21]. The comparison of this result obtained in clinical samples (ΔCt < 7) to the one found in environmental samples had led to conclude that the M. ulcerans could have the same genetic profile in the two types of samples.
The M. ulcerans detected had a distribution limited to 4 villages including Fongbé Apédomé (2 cases), Yobo Sedjro (1 case), Tchékpo Dévé (1 case) and Tchékpo Dedekpoe (1 case) around the Haho river. Only one case was detected in the village of Gapé Kpodji near the Zio River (Fig 2). This distribution of environmental samples is like the confirmed Buruli ulcer patients in the same villages of residence (Fig 1). However, there was no significant correlation (p = 0.59) between the number of BU patients and the presence of M. ulcerans in the environment.
Based on epidemiological evidences [12–13, 21–24], it has been suggested that M. ulcerans is an environmental organism that sometimes infects humans [25]. The mode of transmission is still poorly understood although primary contact of the skin with contaminated aquatic environment is a possible route of infection [25]. In order to improve the understanding of the mode of transmission of M. ulcerans in human, it is important at first stage to determine the ecology of the M. ulcerans [39]. For this purpose, this study aimed to determine the presence of M. ulcerans in the environment and its relationship with BU disease.
The real time PCR analysis of environmental samples in our study has shown that M. ulcerans including three sequences (IS2404/IS2606/KR) was detected in 2.7% (6/219) of samples tested. This is the first evidence that M. ulcerans is present in the environment of the Zio and Yoto districts of the maritime region in Togo. This percentage is higher than what was found in Ghana [40]. However this frequency remained lower than the positivity rate found in other studies conducted in Africa [12,22–23,29], South America [39] and Australia [13]. The difference between frequencies observed may be explained by the criteria of M. ulcerans detection. In our study, the confirmation method of the presence of the M. ulcerans was based on the detection of three sequences including two insertions (IS2404 and IS2606) and the mycolactone KR-B gene. This method has been used in Cameroon [12] and in Ghana [40]. However, in other studies from Benin [22], Côte d’Ivoire [29] and South America [39], the authors have detected two sequences consisting of IS2404 insertion and KR-B gene. On the other hand, Stinear [13] had identified M. ulcerans in Australia, basing only on the two insertions (IS2404/IS2606) without detecting the KR-B gene.
Indeed, although IS2404 sequence is considered a specific marker of M. ulcerans detection in clinical samples [19], the existence of M. ulcerans ecotypes (lineage 1) positive for this sequence which are largely non-virulent for humans complicates the interpretation of real-time PCR results in environmental samples [19, 21]. This requires that samples should be also tested for detection of IS2606 sequence and the difference ΔCt between IS2606 and IS2404 will be analysed [21]. The M. ulcerans (lineage 3) ecotypes that cause human diseases in Africa and Australia incorporate a higher number of IS2606 sequences than the lineage 1 [21]. Thus, such ecotypes can be differentiated based on the ΔCt value of the difference between IS2606 to IS2404. Indeed, a ΔCt <7 allow to identify the ecotype of M. ulcerans, (subspecies) ulcerans, which are virulent strains of lineage 3 compared to other non-ulcerans mycolactone- producing mycobacteria (MPM) or non-M. ulcerans strains virulent (lineage 1) [19–21].
To determine probable reservoirs or habitats of M. ulcerans, we analysed environmental samples. The qPCR results showed that samples of biofilms/plants (10.3%), soil (1.7%) and river water (1.5%) were positive to M. ulcerans DNA including three sequences (IS2404/IS2606/KR). However, the positivity rate of M. ulcerans was not significantly different between these three sources. In contrast to this observation, other studies [12,22,29,39] had often identified M. ulcerans DNA in a large proportion of water samples. Whilst it could be concluded that mycobacteria are more frequent in water, the difference between these results may be due to the DNA extraction method and the sample collection site. In general, water samples are collected in a large volume with possibility of concentration on filters, whereas samples of biofilms/plants and soil are used in small quantities (0.25 g) according to the kits available for DNA extraction [39]. The site collection of samples could explain the prevalence of M. ulcerans DNA as some authors have especially collected samples from aquatic areas [13, 23, 39] which are at high risk of M. ulcerans infection while other performed sampling in both aquatic and dry areas [12, 22, 29].
Due to the abundance of other and faster growing microorganisms in the environment, routine cultivation of M. ulcerans from environmental samples has mostly failed [8–9]. Because of theses relative difficulties, the real-time PCR is commonly used to confirm the presence of M. ulcerans in the environment by amplifying DNA of the bacteria [21]. Although this method does not allow to prove the presence of viable mycobacteria in a sample, in the absence of a culture isolate, concurrent detection of IS2404 and IS2606, can be used to provide convincing evidence of the presence of M. ulcerans [41].
The relationship between the presence of M. ulcerans in environment and Buruli ulcer disease was analysed. Thus, using qPCR to detect three markers in the M. ulcerans genome, we found that the genetic profile detected was similar between environment and clinical samples. In addition, this mycobacterium identified in environmental samples in our study had similar distribution to BU patients in the same villages where they resided (Fig 2). This suggests a co-existence between patients and the pathogen in the same environment. Also, Possible reservoirs identified in environmental samples were water, biofilms/plants and soil (mud) which was collected at the surface or the edges of Haho and Zio rivers. The environment of these rivers could be potential source M. ulcerans infection in human. This observation confirms results from other studies [15–16,22–23,26,29,39] that aquatic sources are important source of risk of contracting Buruli ulcer. Furthermore, our study identifies the presence of M. ulcerans in living houses of BU patients (1.3%; Table 3) in dry areas. This result could explain that this bacterium was widely distributed in both aquatic and dry zones.
This study was limited by the lack of culture isolates of M. ulcerans from environmental samples. This should provide an undeniable proof of the presence of the viable mycobacteria and its association with Buruli ulcer disease. In the other hand, the absence of data from non-endemic areas has reduced the knowledge about the real distribution of M. ulcerans in the environment in surveyed districts.
This study confirms the presence of M. ulcerans in the environment of the districts of Zio and Yoto in the maritime region of south Togo. This may explain partially, the high rates of Buruli ulcer patients in this region. Possible reservoirs of M. ulcerans identified were water, biofilms/plants and soil in the neighbourhood of rivers. Haho and Zio rivers could be potential sources of M. ulcerans infection in Human in these districts of the maritime region of south Togo.
|
10.1371/journal.pgen.1006807 | Dynamics of DNA methylomes underlie oyster development | DNA methylation is a critical epigenetic regulator of development in mammals and social insects, but its significance in development outside these groups is not understood. Here we investigated the genome-wide dynamics of DNA methylation in a mollusc model, the oyster Crassostrea gigas, from the egg to the completion of organogenesis. Large-scale methylation maps reveal that the oyster genome displays a succession of methylated and non methylated regions, which persist throughout development. Differentially methylated regions (DMRs) are strongly regulated during cleavage and metamorphosis. The distribution and levels of methylated DNA within genomic features (exons, introns, promoters, repeats and transposons) show different developmental lansdscapes marked by a strong increase in the methylation of exons against introns after metamorphosis. Kinetics of methylation in gene-bodies correlate to their transcription regulation and to distinct functional gene clusters, and DMRs at cleavage and metamorphosis bear the genes functionally related to these steps, respectively. This study shows that DNA methylome dynamics underlie development through transcription regulation in the oyster, a lophotrochozoan species. To our knowledge, this is the first demonstration of such epigenetic regulation outside vertebrates and ecdysozoan models, bringing new insights into the evolution and the epigenetic regulation of developmental processes.
| Elucidating the mechanisms which govern the development of multicellular animals and their evolution is a fundamental task. Epigenetic mechanisms like DNA methylation have recently emerged as critical regulators of mammalian development through the control of genes that determine the identity of cells and the transmission of parental imprints. In invertebrates however, DNA is mostly unmethylated and does not play a role in development except in the peculiar case of social insects. Therefore the significance of DNA methylation in development is thought to be restricted to vertebrates, and thereby considered a recent evolutionary acquisition, and the situation in more distant organisms is unknown. Here we investigated the dynamics of genome-wide DNA methylation patterns in a mollusc, the oyster C. gigas, throughout its development. We found that the dynamics of DNA methylation correspond to the expression dynamics of distinct functional gene clusters that control two critical development steps, cleavage and metamorphosis, and we provide insights into the underlying molecular mechanisms in a non-vertebrate species. These findings challenge the present considerations on the evolution of developmental processes and their epigenetic regulation, and open a new area of research in molecular and developmental biology in invertebrates.
| The methylation of DNA is a prevalent epigenetic mark that is deeply rooted in evolution and found from bacteria to mammals. Despite that metazoan organisms display methylation on cytosines, great variations exist in the amount and distribution of methylcytosines (meCs) across taxa. DNA methylation is an essential feature of mammalian development because meC patterns are associated with a wide range of cell processes whose subtle combination is required for the embryo to develop into a complex adult organism exhibiting differentiated cell types. In mammals, ca. 60 to 80% of CpG cytosines are methylated and exhibit mostly stable patterns across tissues. CpG rich regions (CpG islands) are prevalent at transcription start sites, and the methylation of promoters correlates to gene silencing during development [1]. CpG dinucleotides are overrepresented in promoters of development and housekeeping genes which are protected from methylation by transcription factor binding and subsequent DNA methyltransferase exclusion [2], reflecting poor methylation in the germline over evolutionary time. However, DNA methylation can be highly dynamic at precise locations during development, as illustrated by the demethylation wave observed in parental pronuclei, the epigenetic reprogramming of the germline or the differences between the epigenomes of germ and somatic cells [1, 3]. Consistently, DNA methylation shapes cell differentiation (reviewed in [4]) notably through silencing of pluripotency factors [5, 6] and of germline specific genes in somatic cells [7] at lineage commitment by de novo methylation. DNA methylation is also implicated in genome defence against transposable element activity [8], maintenance of parental imprints [9, 10], and X chromosome inactivation (review in [11]). Developmental processes are not only triggered by DNA methylation, whose causal role remains debated [12, 13], but by networks of epigenetic regulators including histone modifiers [14], non coding RNAs [15], transcription factors [16] and DNA methyltransferases [17, 18]. DNA methylation stabilizes the chromatin context underlying cell fate decisions that are propagated through cell generations by maintenance of the meC landscapes (review in [4]).
In invertebrates, DNA is much less methylated and meCs are not evenly distributed but exhibit mosaic patterns [19, 20]. DNA methylation in insect models is rare and mostly confined to gene bodies (gene body methylation, GBM) [20]. In hymenopterans, GBM controls exon selection[21] and governs important developmental outcomes such as caste differentiation in the honeybee [22, 23] and in ants [24, 25], as well as developmental gene expression in the wasp N. vitripennis [26, 27]. However, DNA methylation and its developmental significance seem essentially restricted to a peculiar evolutionary acquisition in hymenopterans. Indeed, in Drosophila, early genes are controlled by cis-regulatory elements, non-coding lnc- and miRNAs, and transcription factors including polycomb and trithorax complexes (review in [28]) but not by DNA methylation. The actual presence and function of meCs in the fruitfly genome have been under discussion, and the nematode C. elegans even lacks conserved DNA methylation machinery. Therefore DNA methylation is considered absent in the ecdysozoan common ancestor, in line with animal genomes evolving towards an overall loss of DNA methylation in protostomes [20] such as insects, compared to deuterostomes [29] such as mammals. As a consequence of this basic divergence between ‘methylated vertebrates’ and ‘unmethylated invertebrates’, and in spite of the tremendous variability of organisms and life traits within protostomes, DNA methylation is largely neglected outside insects.
However, recent studies in lophotrochozoans (that include molluscs and annelids), the sister group of ecdysozoans (that include insects and nematodes), suggest a more complex situation. Although meCs similarly exhibit a mosaic distribution, mollusc genomes are far more methylated than insects’, where methylated genomes display ca. 0.15% of meCs [30], whereas this value reaches ca. 2% in the snail Biomphalaria glabrata [31] and in the gills [32] and mantle [33] of the oyster Crassostrea gigas. In this bivalve of greatest ecological and economical importance, GBM is predominantand associated to mRNA content [32, 34]. Surprisingly, exposure to a DNMT inhibitor disrupts the oyster embryogenesis [35], and meCs are present in the promoter of some development genes with a direct influence on their expression [36]. These data point to developmental significance for DNA methylation in a lophotrochozoan species [37], challenging the current view on the evolution of epigenetic regulation of developmental processes.
Here, to shed light on this point, we provide the first characterisation, to our knowledge, of genome-wide DNA methylation dynamics covering the development of a lophotrochozoan species. Using a development stage-wise MeDIP-seq approach, we characterized the methylome dynamics from the egg to the completion of organogenesis in the oyster C. gigas. Epigenetic landscapes were analysed at both a global, physical and more local, feature-related scales, together with mRNA expression and functional annotation, and indicate a dynamic regulation of DNA methylation at critical developmental steps.
The methylated DNA immunoprecipitation followed by high throughput sequencing (MeDIP-seq) approach enabled genome-wide assessment of methylation and its variations during oyster development. Large-scale genome-wide methylome dynamics were investigated by analyses of differentially methylated regions (DMRs) and physical maps. DMRs highlighted 4 main developmental phases (oocytes, 2–8 cells, mid-larval, spat) separated by 3 main developmental steps: cleavage (C step), gastrulation and organogenesis (I step, intermediate) and metamorphosis (M step), respectively (S1A Fig).The morula, blastula, gastrula, trochophore and D-larva stages were grouped into an intermediate mid-larval stage, because DMRs and individual feature methylation profiles (see below) showed only minor differences. Physical methylation maps of genomic scaffolds confirm the mosaic characteristic of oyster DNA methylomes, which display a succession of methylated and non methylated regions bearing gene clusters of variable length with no obvious organisation or relationship to CpG content (Fig 1). Developmental methylation dynamics mostly affect regions that were already methylated in oocytes, with hypermethylation prevailing during development (Fig 1). Regions not methylated in oocytes mostly remain unmethylated and only little de novo methylation of previously unmethylated regions is observed that lie almost exclusively between two adjacent previously methylated regions or at their direct proximity (Fig 1). Indeed, 98.7% of the genes that are methylated at the spat stage were already methylated in oocytes. The mean distance from genomic features to the nearest DMR was drastically shorter at the M step (ca.5 kb for CDS to 20 kb for TE) than at other steps (ca. 100 to 200 kb at C and I), indicating that DNA methylation is more evenly regulated throughout the genome at the M step than in the C step (S1B Fig). Although DMR length did not exhibit marked variations, they were not equally distributed regarding genome features amongdevelopment steps. Many more DMRs were found at the C (n = 1043) and M (n = 2230) steps than at the I step (n = 14), with methylation being preferentially regulated in exons (CDS), repeats (REP) and transposable elements (TEs) (S1C Fig), and in class I TEs (i.e. retrotransposons) compared to class II TEs (i.e. DNA transposons) (Pearson’s χ2: p<0.0001***; C step: 122 DMRs in class I TEs vs. 81 in class II (60.1% vs. 39.9%); M step: 121 in class I TEs vs. 63 in class II (65.7% vs. 34.2%); genome, 64150 class I TEs vs. 21263 class II (75.1 vs.24.9%)).
In parallel to genome-scale investigations, methylation landscapes were examined at the level of individual genomic features (i.e. exons (CDS), introns (INT), promoters (PRO), repeats (REP) and transposable elements (TE)). Most of the reads (81.5 ± 0.95%) mapped to the considered features and a great majority (ca. 90%) of methylation was found within gene bodies (CDS and INT, S4A Fig). Overall, the distribution of methylation depends on the development stage (Pearson’s χ2: p<2.10−16***, S2A Fig). The relative methylation of CDS is strongly increased after metamorphosis at the expense of the other features, especially introns (correlation between methylation in CDS and INT: Pearson R = -0.985, p<0.0001***), but not TEs (S2A Fig). This is because the intermediately methylated genes (ca. 2 to 6 log counts per million (CPM)) have their CDS markedly hypermethylated at the spat stage (Fig 2). The individual methylation level of an important number of TEs increases in 2–8 cell embryos,and the stability observed thereafter is due to compensation between individual TE hypermethylation and hypomethylation in spats (Fig 2). The biological coefficient of variation (BCV) analyses of feature methylation between development stages clearly discriminate the 2–8 cells and spat stages from one another and apart from the other stages, which are grouped and may display a gradual distribution regarding embryogenesis chronology (S4B Fig). The methylation profiles depend on the feature considered (S2C Fig) thereby confirming both the feature- and development stage-specificity of the dynamics of oyster DNA methylomes, especially marked at the 2–8 cells (cleavage) and spat (post metamorphosis) stages.
DMR proximity is associated to gene expression variability, and whether the DMR lies upstream or downstream has no influence (S3A Fig). Consistently, DMR-associated genes have their expression level more regulated than genes not associated to a DMR at each developmental step, although DMR and mRNA level variations were not correlated (S3B Fig). Compared to moderate changes, extreme methylation variations tend to hinder mRNA level regulation (S3B Fig). At a finer scale, most genes display a detectable methylation (20704 methylated genes vs. 7197 non methylated genes) during oyster development. The non methylated genes are mostly silent whereas the methylated genes are dramatically more expressed (Fig 3A). These genes have their mRNA level positively associated to their CDS methylation level, with a slight drop for genes within the 10th expression decile. Conversely, the methylation level decreases with expression variability (Fig 3A). These results indicate that methylation marks highly and stably expressed genes. Although the exact localisation of methylcytosines is hampered by the resolution of MeDIPseq (ca. 250 bp), gene expression decreases with the hypermethylation of the INT or CDS feature over the other (ANOVA p<0.001), and the expression variability is correlated to the methylation pattern variability (p<2.10−16***) suggesting an optimal in-gene methylation pattern for maximum transcription (S4 Fig).A large set (26%) of oyster genes exhibits a dynamic CDS methylation during development (ANOVA across stages p<0.01**), and gene clusters can be discriminated based on their distinct CDS methylation kinetics (Fig 3B). However, methylation and mRNA level kinetics are correlated for only ca. 10% of these genes (r2>0, p<0.05*) (Fig 3B). All together, these results indicate that during oyster development, unmethylated DNA is associated to transcription repression whereas methylated DNA corresponds to gene expression, with the dynamics of gene-body methylation being associated with transcription regulation.
There was no association between PRO methylation and gene expression. The methylation level or coefficient of variation of gene features (PRO, CDS or INT) was not found to be correlated to the number of transcript variants during oyster development.
Gene ontology (GO) annotation of DMRs depends on the development step considered. Cleavage DMRs bear genes with functional annotation (Biological Process ontology) related to egg vitellogenic resource consumption, mRNA metabolism and nuclear genome processes, whereas metamorphosis DMR genes are enriched in terms related to transport within the cell and protein degradation (Table 1). The gene clusters based on methylation kinetics exhibit specific GO term distribution in ontologies (Biological Process, Molecular Function and Cell Component) and little GO terms in common (S5A Fig). Conversely, selected Biological Process ontology terms display specific methylation level and developmental dynamics (S5B Fig), indicating that methylation dynamics correspond to the distinct functional pathways related to specific steps of oyster development.
The present work constitutes, to our knowledge, the first description of genome-wide methylome dynamics in a lophotrochozoan model. Their functional significance in oyster development brings new insights into the transcriptional regulation mediated by DNA methylation and the evolution of the epigenetic mechanisms underlying ontogenic processes in a lophotrochozoan species.
DNA methylation landscapes are highly dynamic during oyster ontogenesis and depend on the development stage (Fig 1), as suspected from variations of the global DNA methylation levels [35],. Furthermore, meC patterns also depend on genome features (CDS, INT, TE). DNA is mostly methylated within genes, and particularly in exons, in line with previous reports on oyster gills [32], mantle [33] or gametes [34]. Methylated coding sequences get hypermethylated during development at the expense of other genomic features, especially introns but not TEs. However, not all genes get hypermethylated (Fig 2), unravelling a precise and individual regulation of their DNA methylation (Fig 2 and S2C Fig). Such regulation is marked at cleavage and metamorphosis, assuming a participation of methylome dynamics at these two precise and critical development steps, as suggested by the altered phenotypes observed upon DNMT inhibition during oyster development [35].
The relationship between gene body methylation (GBM) and mRNA levels clearly indicates a biological significance of DNA methylation dynamics in gene expression during oyster development. Indeed, exon methylation is almost always required for transcription and marks stable and high expression (Fig 3A), in line with single time-point methylomes of adult tissues [32–34]. In addition, our study suggests that regardless of the level, the pattern of methylation within genes may be associated to transcriptional regulation. Indeed, a skewed GBM pattern corresponds to a diminished transcription, and the variability of methylation patterns and of mRNA levels are positively correlated (S4 Fig), in spite of the fact that the limited resolution of MeDIP-seq does not allow a more precise localisation of meCs within exons and introns. Such an influence is reported here for the first time to our knowledge and supports previous hypotheses of GBM increasing ‘transcriptional opportunities’ of methylated genes [38]. However the methylation of gene features was not associated to the number of transcript variants. This unexpected finding does not substantiate a role for methylation in exon selection in the oyster, in contrast to insects [21] and to previous hypotheses [33], although higher resolution methylomes might be needed to clarify this point.
In addition to transcriptional regulation, this work shows that methylation dynamics are associated with developmental stage-specific functional pathways in the oyster. Indeed, GBM dynamics define gene clusters with specific functional annotation (Fig 3 and S5 Fig). Consistently, DMR annotations at cleavage are clearly relevant to egg mRNA and vitellogenic resource consumption as well as nuclear genome processes (Table 1), bringing epigenetic indications of early zygotic genome activation in the oyster. The requirement of transcription for the cleavage in the distant annelid, the leech Helobdella [39] suggests that this situation could be general in lophotrochozoans. It implies stabilisation of open chromatin states at loci of cleavage genes that lie within DMRs during oyster development. Conversely, persistent unmethylated regions could be of functional significance for developmental gene silencing [40] as reflected by the strong repression of unmethylated genes (Fig 3A). The primary resemblance with oocyte DNA methylation patterns (Fig 1) suggests that these loci could be inherited, their transcription being further regulated by local methylation dynamics in the embryo, reminiscent of recent findings in the mouse [41]. Oysters exhibit a mosaic development and the fate of their blastomeres is determined during early ontogenesis notably through specific transcriptomes [42]. Besides, the oyster genes with maximum sequence similarity to vertebrate pluripotency factors Pou5f1 and Sox2 (GenBank accession number CGI_10005968 and CGI_10010085, respectively) display their highest embryolarval methylation at cleavage (S1 File). This suggests that DNA methylation could link zygotic genome activation and cell differentiation [43], consistent with the hypothesis that methylome dynamics participate in both cell differentiation and epigenetic memory in the oyster. Surprisingly, gastrulas and later mid-larval stages display little methylation difference despite dramatic morpho-physiological changes such as organ and shell formation, larval growth and the onset of digestion. Such processes are likely governed by other factors, such as Hox genes and Polycomb repressive complex orthologues [36], although target genes should lie in chromatin competent for transcription. Long intergenic non-coding RNAs (lincRNAs) were recently hypothesised to play a role during oyster development [44], but they are mostly expressed late after metamorphosis and therefore unlikely contribute to mid-larval state epigenetic control. The ‘M’ step DMRs are more abundant and widespread across the genome (Fig 1 and S1 Fig), further supporting the association of distinct methylation landscapes with the transcriptomes of the various cell types within post-metamorphosis oyster larvae. Their functional annotation related to transport and targeted protein degradation evokes the importance of cell morphology changes related to metamorphosis.
Transposable elements can be either hypermethylated or hypomethylated after metamorphosis (Fig 2) independently of being retro- or DNA transposons. Some young repetitive elements such as SINEs are preferentially methylated [33] and may still be active in the oyster genome [42]. Therefore DNA methylation may be important for TE control and genome plasticity during oyster development, unlike most ecdyzosoans [23, 27] but like vertebrates [45]. Whether DNA methylation facilitates SINE transcription is unknown, but retrotransposons are more associated to DMRs, raising the possibility that methylation influences their transcription, which may be silenced by other regulators such as piRNAs [46].
The precise molecular pathways associating transcription regulation with DNA methylation dynamics in the oyster remain unknown. Nevertheless, taken altogether our results raise the assumption that DNA methylation locally impairs chromatin compaction, and propagates transcription-competent states outwards to flanking unmethylated sequences. In this context, initiation may be hampered by promoter methylation [36], and elongation by too high and/or inconsistent meC density within gene bodies. This hypothesis does not contradict the situation in vertebrates where high meC levels inhibit transcription initiation [47]. Indeed, vertebrate genomes are highly methylated whereas methylation of oyster DNA is scarce. Then, both high meC levels in the oyster and meC depletion in vertebrates would correspond to moderate methylation levels allowing transcription. It remains to be determined whether DNA methylation is a cause or a stabiliser of transcriptional regulation, or both, which is under debate [12, 13]. The association between GBM and expression could also be explained by increased accessibility of transcribed DNA to methyltransferases [47, 48].
This study brings new insights into the evolution of the epigenetic regulation of developmental processes. Regarding the DNA methylation in development, oysters resemble deuterostomes more than other protostomes. Therefore such epigenetic regulation could be present in the bilaterian common ancestor and might be an ancient trait of metazoan organisms that would have been generally lost in ecdysozoans. Besides, there is poor consistency over evolutionary time in the methylation level of both oyster development genes and environment-response genes [49]. Such resemblance may suggest that development is an inducible biological process rather than a fixed program. Stochastics of ontogeny may thus be under environmental inputs implicating epigenetic mediation, and their reproducibility could result from probability canalisation by biological systems over evolutionary time. Future work towards the understanding of the interactions between epigenetic marks and chromatin dynamics across evolutionary lineages, as well as insights into the causal and/or stabilising role of DNA methylation in this context and upon environmental inputs, is required to decipher this issue.
The present work constitutes, to our knowledge, the first description of genome-wide methylome dynamics in a lophotrochozoan model. Dynamics of DNA methylation in gene bodies are associated with transcriptional regulation, and the control of transposable elements may imply DNA methylation. The shifts in methyl DNA profiles and their functional outcomes are prevalent at cleavage and metamorphosis, and suggest the importance of inherited methylomes. These results demonstrate that DNA methylation dynamics underlie Crassostrea gigas development. The developmental significance of gene body methylation in the oyster brings new insights into the epigenetic regulation of developmental processes and its evolution.
Oyster embryos were obtained as previously described [50]. Wild individuals were collected in Marennes- Oléron, France in August 2008 then transferred in mesh bags in February 2009 to Paimpol (northern Brittany, France, 48°48’ 24.49”N, 3° 0’ 22.84”W) until February 2010 and then to the Ifremer grow-out farm located at Aber-Benoît (northern Brittany, France, 48° 34’ 29.976”N, 4°36’ 18.378”W). These animals were exposed to disease during the spring of 2009 and suffered ca. 75% mortality. In April 2010 (Experiment 1 and 2) and February 2011 (Experiment 3), 60 individuals were transferred to the Ifremer marine station located at Argenton (Brittany, France, 48° 31’ 16.320”N, 4°46’ 01.998”W) for broodstock conditioning (6 weeks in 500 L flow-through tanks with UV-treated and 1 μm filtered seawater (TSW) at 19°C, enriched with a 1:1 in dry weight mixture of Isochrysis affinis galbana and Chaetoceros gracilis corresponding to a daily diet of a ration equivalent to 6% of the oyster dry weight). Diploidy of oysters was confirmed by flow cytometry of gill cells from randomly sampled animals as previously described [51]. Gametes from mature specimen (13♂ and 27♀, 10♂ and 24♀, 13♂ and 21♀, Experiment 1, 2 and 3 respectively) were obtained by stripping and filtered on a 100 μm mesh for the removal of large debris. For females, oocytes were harvested as the remaining fraction on a 30 μm mesh; for males, spermatozoa were harvested as the passing fraction through a 30 μm mesh. Oocytes were pre-incubated in TSW then mixed in a 5 L jar at 50–100 spermatozoids per oocyte (22 November 2010, 5 January 2011 and 12 April 2011 for experiment 1, 2 and 3 respectively). The embryonic development was completed in TSW in oxygenated 150 L tanks at 21°C for 48 h. The D-larvae were then collected and reared in flow- through rearing systems at 25°C. At the end of the pelagic phase (16 d), competent larvae were collected on a 225 μm sieve and allowed to settle on cultch. Post-larvae were maintained in downwelling systems where they were continuously supplied with enriched seawater. After 10 d, the spat were collected on 400 μm mesh. In the larval and post-larval stages, the oysters were fed the same diet as the broodstock. Throughout this time, the oysters were free of any abnormal mortality and OsHV-1 virus. Embryos were left unattended until sampling, i.e. before fertilization for control oocytes, and ca. 1 hour post-fertilization (hpf) for 2–8 cells stage, ca. 3 hpf for morulae, ca. 6 hpf for blastulae, ca. 9 hpf for gastrulae, ca.16 hpf for trochophore larvae, and ca. 24 hpf for D larvae. Spat was collected at 26 days, after settlement and metamorphosis. Developmental stages were assayed by microscopic observation based on morphological and motility criteria before and after fixation using 70% ethanol. Samples were split in aliquots of 2 million larvae, stored dry at -80°C and thawed only once before use. Each development stage was sampled from three distinct fertilization experiments (experiments 1, 2 and 3).
Genomic DNA from ca. 2 million larvae per sample was purified by affinity chromatography (Macherey Nagel) following the manufacturer’s instructions. Degradation of contaminating RNA was realized using RNAse. DNA purity and concentration were assayed by spectrometry (Nanodrop, Thermo) and on-chip gel electrophoresis (Tape Station 2200, Agilent). DNA was sheared in ca. 250 bp fragments using a Covaris S2 sonicator (duty cycle: 10%, intensity: 4, cycles: 200, time: 80s).
Twenty samples (n = 2 to 3 biological replicates per development stages) were processed for MeDIP-seq library preparation following the protocol of Taiwo et al. [52]. Briefly, 5μg DNA from each sample were used for DNA end-repair and dA-tailing (NEBNext reagents, New England Biolabs). Immunoprecipitation of methylated DNA (MeDIP) was realised on 1 μg DNA after end-repair, dA-tailing and purification using the MagMeDIP kit (Diagenode) using the manufacturer’s instructions. All DNA purifications were carried out using Ampure XP magnetic beads (Beckman Coulter) according to the recommended procedure. Ten randomly chosen samples were assayed following the manufacturer’s recommendations for MeDIP specificity with a mean value of 96.9%. Immunoprecipitated DNA was then amplified using Phusion DNA polymerase (New England Biolabs) and purified. After size selection and quality control, libraries were submitted to 2x76 bp paired-end sequencing using a GAIIx sequence analyzer (Illumina). This strategy produced ca. 80 million paired end reads i.e. 3.6±0.25x106 reads per sample among which 80.7±0.1% were aligned to the genome, giving a 24.8±3.3-fold mean coverage.
Data were analysed using a using a combination of dedicated R (bioconductor.org) and bash script as well as in-house R, bash, TiCL and PERL scripts. Data source files (NCBI project PRJNA324546) and scripts used for analyses are publicly available (github.com/BOREA-UNICAEN/MeDIPSeq-Dev-Gigas). Primary analysis was performed with RTA (Illumina) with default parameters and reads were demultiplexed using CASAVA v.1.8. Bases with a QC>30 were retained for further analyses. Paired-end reads were mapped to the oyster genome (assembly v.9) using BWA with default parameters and pair-sorted. Paired reads mapping to the following genomic features: exons (CDS), introns (INT), promoters (PRO), repeats (REP) and transposable elements (TE) [34] were counted using HTseq-count [53]. Only promoter sequences longer than 100 bp were retained for further analyses and ambiguous read pairs were discarded.
|
10.1371/journal.pcbi.0030109 | Modeling Systems-Level Regulation of Host Immune Responses | Many pathogens are able to manipulate the signaling pathways responsible for the generation of host immune responses. Here we examine and model a respiratory infection system in which disruption of host immune functions or of bacterial factors changes the dynamics of the infection. We synthesize the network of interactions between host immune components and two closely related bacteria in the genus Bordetellae. We incorporate existing experimental information on the timing of immune regulatory events into a discrete dynamic model, and verify the model by comparing the effects of simulated disruptions to the experimental outcome of knockout mutations. Our model indicates that the infection time course of both Bordetellae can be separated into three distinct phases based on the most active immune processes. We compare and discuss the effect of the species-specific virulence factors on disrupting the immune response during their infection of naive, antibody-treated, diseased, or convalescent hosts. Our model offers predictions regarding cytokine regulation, key immune components, and clearance of secondary infections; we experimentally validate two of these predictions. This type of modeling provides new insights into the virulence, pathogenesis, and host adaptation of disease-causing microorganisms and allows systems-level analysis that is not always possible using traditional methods.
| The immune response is a complex network of processes activated in a host upon infection. Pathogens seek to disrupt or evade these processes to ensure their own survival and proliferation. This article provides a systems-level analysis of the immune response against two related bacterial species in the Bordetella genus, B. bronchiseptica and B. pertussis. B. pertussis, the causative agent of whooping cough, has lost many of the virulence factors of its B. bronchiseptica–like progenitor, and is using different strategies for the modulation of the immune system. We have synthesized two separate network models for the interaction of these pathogens with their hosts. Each network is then translated into a predictive dynamic model and is validated by comparison with available experimental data. The model offers predictions regarding cytokine regulation and the effects of perturbations of the immune system, as well as the time course of infections in hosts that had previously encountered the pathogens. We experimentally validate the prediction that convalescent hosts can rapidly clear both pathogens, while antibody transfer cannot substantially reduce the duration of a B. pertussis infection. This type of modeling provides new insights into the virulence, pathogenesis, and host adaptation of disease-causing microorganisms and can be readily extended to other pathogens.
| Bacteria persist within their hosts by subverting phagocytosis by immune cells, interfering with antigen processing or presentation [1], or by promoting anti-inflammatory or immunosuppressive responses that normally function to terminate the protective effector immune responses of the host [2]. The dynamic interplay between pathogen and host can have one of three outcomes: death of the host, persistent disease, or recovery. To understand and influence this complex system, it is imperative that we identify the subset of key components and regulatory interactions whose perturbation or tuning leads to significant functional changes. Mathematical modeling can assist in this process by integrating the behavior of multiple components into a comprehensive network model, and by addressing questions that are not yet accessible to experimental analysis.
We used two species of the genus Bordetellae as model organisms because (1) they are examples of pathogens that successfully overcome the defenses of their mammalian hosts, (2) their genomes are completely sequenced, and (3) two closely related species of Bordetellae provide a comparative model to understand how virulence factors modulate immune responses. The Bordetellae are small, Gram-negative coccobacilli, some of which colonize the respiratory tracts of their hosts, adhering to ciliated epithelia and spreading via respiratory droplets. B. bronchiseptica and B. pertussis are two very closely related species that have different host ranges and cause different diseases in their hosts. B. bronchiseptica naturally infects wild and domesticated animals, including leopards, koala bears, cows, dogs, rabbits, and mice [3–5], and causes a persistent disease typified by atrophic rhinitis in pigs and by kennel cough in dogs. B. pertussis, which evolved from a B. bronchiseptica–like progenitor, causes whooping cough (pertussis) in humans and is endemic in much of the world. Whooping cough is an acute illness characterized by severe coughing that can become spasmodic, and in some cases leads to death. Although the human pathogen does not cause persistent infection, its rapid spread within relatively dense and mobile human populations is apparently sufficient to allow transient infections to circulate on an ongoing basis, allowing the bacteria to survive within a population.
The different persistence strategies of B. bronchiseptica and B. pertussis are surprising in light of their high genetic relatedness. The Bordetella strains mainly evolve through loss of genes and acquisition of insertion sequences. The two strains of Bordetellae studied in this paper share 3,394 genes with a synonymous substitution rate of 0.021 [6]. The majority of known virulence factors, including adhesins (filamentous hemagglutinin [FHA], pertactin, and fimbriae) and toxins (adenylate cyclase toxin [ACT] and dermonecrotic toxin) are expressed by both B. bronchiseptica and B. pertussis. Despite this, the genome of B. pertussis is 30% smaller than that of B. bronchiseptica, due in part to the loss of numerous sizable multigenic regions (e.g., the 22-kb genomic region required for the assembly of a predominant antigen, O-antigen). Interestingly, there also appear to be a number of genes present but not expressed by one pathogen or the other (e.g., the genes encoding pertussis toxin [PTX] are only expressed by B. pertussis; see Table 1 [7–12]). Though limited, the genetic variation between B. bronchiseptica and B. pertussis allows substantial differences in their pathogenesis mechanisms.
Mouse infection models with the Bordetellae provide an excellent experimental setup in which specific interactions between the host and pathogen can be discovered and manipulated. Both B. bronchiseptica and B. pertussis efficiently colonize the upper and lower respiratory tracts of their hosts and increase in numbers rapidly in the first few days after inoculation. The inflammatory infiltrate, leukocytosis, gradual generation of antibody and T cell responses, and the delayed bacterial clearance from the lower respiratory tract are qualitatively similar to aspects of the clinical pertussis disease. The major aspects of Bordetellae virulence and host response have been identified and quantified in the past 20 years, and a wealth of data is available in the literature.
The immune response to a pathogen includes a sequence of processes that are activated by immune cells after sensing bacteria. Here, we construct a network model synthesizing these processes activated in response to the sequenced strains of B. bronchiseptica and B. pertussis. We analyze the differences in host immune responses due to the exclusive virulence factors present in the two species by developing separate dynamic models for the infection of the lower respiratory tract by these two species. Discrete dynamic simulation based on the available time course data allows us to monitor the progression of infection in time and to determine the dynamic outcome of Bordetellae lung infections. We use the model to predict the outcome of infection scenarios not yet studied and experimentally verify two such predictions.
We started by synthesizing the available data from the literature and from our own experiments (Table S1) into an interaction network (Figure 1). Bacteria and the components of the immune system (i.e., immune cells and cytokines) were represented as network nodes; and interactions, regulatory relationships, and transformations among components were represented as directed edges starting from the source (regulator) node and ending in the target node. We incorporated regulatory relationships that modulate a process (or an unspecified mediator of a process) as edges directed toward another edge. The regulatory effect of each edge was classified into activation or inhibition, and is represented by an incoming black arrow or an incoming red blunt segment, respectively, in the figures. Since not all processes involved in natural B. pertussis clearance are known or addressable through the mouse infection model, we extended the set of known interactions by putative interactions based on general immunological knowledge.
We assumed that the bacteria activating the chain of immune responses shown in Figure 1 express the generic adhesins and toxins (Table 1) of the sequenced Bordetella strains; we did not assign independent nodes for these virulence factors. However, we did separately include the set of virulence factors identified in the literature as modulating immune responses in a species-specific manner. The resulting network had 18 nodes common in B. bronchiseptica and B. pertussis; two species-specific nodes in B. bronchiseptica (O-antigen and the type III secretion system [TTSS]), two species-specific nodes in B. pertussis (PTX, and a combined node for FHA and ACT). Here, we first describe the network common to both species, illustrating the sequence of processes activated after bacterial invasion (Figure 1), followed by a description of the species-specific nodes in the two bacteria.
The first immune mechanisms that respond to Gram-negative bacterial pathogens are Toll-like receptor 4 (TLR4)–mediated recognition of bacteria and the alternative complement pathway. Both species express lipopolyssacharide (LPS) that is recognized by TLR4 receptors on respiratory epithelial cells and dendritic cells (DCs). TLR4-mediated signaling in response to pathogen-associated molecular patterns such as LPS activates the production of cytokines and chemokines. Many of these, including IL-1, IL-6, TNF-α, and TNF-β, are proinflammatory and recruit polymorphonuclear leukocytes (PMNs) to the site of infection [13]. Complement-activated PMNs produce cytokines, which in turn recruit more phagocytes. Following the commencement of phagocytosis, DCs (the main antigen-presenting cells included in the network) present antigens to T0 cells (naive T cells). Signal transduction networks activated during this interaction induce cytokines, leading to the differentiation of T0 cells into either T helper type 1 (Th1) cells or Th2 cells. Th1/Th2 cells also produce the cytokines required for T cell differentiation, leading to a positive feedback in the differentiation process. We denote the common cytokines inducing differentiation of T0 cells to Th1 (or Th2) cells and produced in turn by Th1 (or Th2) cells as Th1 (or Th2)–related cytokines (Th1RCs or Th2RCs). These cytokines are mutually inhibitory in that Th1RCs (such as IFN-γ and IL-12) inhibit the production and function of Th2RCs (for example, IL-4 and IL-10), and vice versa. Some Th2RCs, such as IL-10, are anti-inflammatory, and inhibit the production of proinflammatory cytokines (PICs), reducing the recruitment of PMNs. The balance between the production of Th1RCs and Th2RCs plays an important role in the time course of the immune responses.
Th2 cells activate the clonal expansion of B cells, which in turn produce antibodies against bacterial antigens. Opsonization of bacteria by complement-fixing antibodies leads to the activation of the classical complement pathway. PMNs express complement receptors that recognize complement-coated bacteria as well as Fc receptors that recognize the Fc region of antibodies bound to bacterial antigens; both mechanisms result in the activation of PMNs. In general, the classical complement pathway can directly lyse bacteria; however, this mechanism is found to be weak for Bordetellae [14,15]. Therefore, we did not include bacterial lysis in the network. Th1 cells produce a set of cytokines such as IFN-γ, TNF-β, and IL-2, which activate phagocytes to ingest and kill bacteria. PICs and Th1RCs attract more phagocytes to the site of infection, where they are activated and eliminate antibody-bound bacteria. Thus, the Th2- and Th1-mediated adaptive responses constitute a positive feedback to phagocytes for antigen-specific elimination of the pathogen. We assume that the main mechanism of natural clearance of the Bordetellae is via phagocytosis by activated phagocytes [14,15].
Bacterial virulence factors such as LPS, PTX, TTSS, FHA, and ACT modulate these host immune processes at several levels. The B. bronchiseptica LPS contains long repeats of O-antigen that inhibit the activation of the alternative complement pathway [16]. B. bronchiseptica expresses a TTSS that induces the necrosis of PMNs [17,18]. The TTSS is known to inhibit the activation of Th1RCs [17]; specifically, the TTSS in association with ACT [19] inhibits IFN-γ production during the first week of the infection, thus inhibiting the differentiation of T0 cells into Th1 cells [20,21]. Consequently, IL-10 (a Th2RC) is produced, facilitating the differentiation of T0 cells into Th2 cells. In a B. pertussis infection, the alternative complement pathway is active [22]; however, early recruitment of PMNs is inhibited by the activity of PTX, and thus PMN activation and phagocytosis are delayed [23]. Like B. bronchiseptica, B. pertussis has also developed mechanisms for the suppression of Th1-related responses. FHA [24] and ACT [25], in association with LPS, stimulate IL-10 production by DCs, transiently inhibiting IL-12 and Th1 responses [26]. ACT also enhances DC activation and maturation, promoting the early differentiation of T0 cells into T regulatory 1 (Tr1) and Th2 cells [25,27]. As Tr1 and Th2 cells have a functionally similar role, we represent them as a single node. Note that although the immunomodulation due to the synergistic activity of TTSS and ACT (in B. bronchiseptica) and FHA and ACT (in B. pertussis) appear similar, their exact target of action is different [20,21,27]. Thus, both pathogens have evolved strategies to suppress antigen-specific Th1 responses during the acute phase of infection [28], modulating the balance between Th1 and Th2 responses to their favor.
During the course of infection, the time-dependent expression of specific virulence factors and/or immune components results in a differential infection time course in the two species. We integrated the known temporal information with the interaction network and developed dynamic models for B. bronchiseptica and B. pertussis interactions with their hosts. The models incorporate (1) the interactions and regulatory relationships between components (i.e., the interaction network of Figure 1, augmented with the relevant virulence factors), (2) how the strength of the interactions depends on the state of the interacting components (i.e., the transfer functions, given in Table 2), and (3) the initial state of each component in the system. Given the above inputs, the model generates the time evolution of the states of the components of the network (e.g., the time course of bacterial presence, of cytokine concentration, or of immune cell activity).
Given the scarcity of kinetic and quantitative characterizations of the processes involved in the bacteria–immune system interaction network (Figure 1), we used a discrete dynamic modeling approach. The network's nodes were assumed to have two qualitative states: 0 (off) and 1 (on), corresponding to a baseline (below-threshold) and high (above-threshold) concentration or activity, respectively. The state change of each node was described by a Boolean transfer function F that depends on the state of the nodes connected to it by directed edges and on its own state. The transfer functions were developed from the knowledge of the nodes directly upstream of each target node in the network, and augmented with dynamic information from the literature and basic immunology when available. The state of target nodes having a single activator and no inhibitors follows the state of the activator with a delay. Often, the target node is regulated by more than one pathway. We used the AND operator whenever synergy between two (or more) nodes is absolutely necessary to activate the target node. When either of nodes connected to the target node could activate it, the OR operator was used. For inhibition, we used the AND NOT operator, requiring a low level or inactivity for the inhibitor in order for the activation of the target node. Table 2 lists the transfer functions of each node, and a detailed justification of each transfer function is available in Text S1.
The transfer functions define a discrete dynamic system in which iteration determines the evolution of the state of nodes. We employed both the frequently used method of synchronous update [29], using the hypothesis that all regulatory processes have the same duration, and as a more realistic alternative, we used random asynchronous update, where the time scales of each regulatory process are randomly chosen [30–32]. Both methods assume that time is quantized into regular intervals (time steps). The synchronous update can be described as
= Fi(
…), where
represents the state of node i at time step t, and Fi is the Boolean function associated with the node i and its upstream regulators a, b, c, … . The asynchronous method entails updating the nodes in a randomly selected order during each time step, and interprets the time step as the longest duration required for a node to respond to a change in the state of its regulator(s) (also called a round of update) [30]. In the asynchronous algorithm, the Boolean updating rules are written as
= Fi(
…), where ta, tb, and tc are the time points corresponding to the last change in the state of the input nodes a, b, and c and could be either in a previous or the current time step. Asynchronicity does not change the steady states of the dynamic system, but induces a variability (stochasticity) in the steady states reachable from an initial condition and in the time courses necessary to reach these steady states [31,32].
We extended the general Boolean framework to incorporate known quantitative information by introducing decay rates for cytokines (Th1RCs and Th2RCs) and virulence factors (FHA/ACT and TTSS) and a threshold for the duration of DC activity required for T0 cell differentiation and the duration of phagocytosis necessary for clearance (see Materials and Methods). We performed a systematic search in parameter space to determine the parameter regions that satisfy the following two criteria derived from the literature: (1) reaching bacterial clearance under normal conditions and (2) association of bacterial clearance from the lower respiratory tract with Th1-related activity (see Materials and Methods and Text S2 for a detailed parameter analysis). The two methods of update gave remarkably similar results; thus, in the following we will describe the results obtained by the asynchronous algorithm and refer to the synchronous algorithm only to describe divergent behaviors.
We started with a completely asynchronous algorithm and studied the outcome by monitoring bacterial clearance and the order of activation of immune processes. We performed 1,000 runs with different orders of update selected randomly in each time step. The disease was allowed to evolve for 70 time steps in each simulation. We obtained a distribution of the time steps necessary for bacterial clearance with mean μ = 24 and standard deviation σ = 0.97 in B. bronchiseptica, and mean μ = 22.5 and standard deviation σ = 1.2 in B. pertussis (Figure 2).
Next, we aimed to incorporate existing knowledge on the relative duration of the processes represented by single edges in the network. These processes include ligand–receptor binding, signal transduction, cytokine production, cell differentiation, and cellular chemotaxis, and span a range of time scales. We incorporated inequalities between two process durations as updating one node before the other in each round of update. Candidate inequalities were accepted only if the resulting dynamics of the model did not contradict experimentally known temporal information (e.g., that B. bronchiseptica and B. pertussis infections of mice with an initial dose of ~5 × 105 colony-forming units [CFUs] are reproducibly cleared from the lung by days 70 and 50, respectively; see Figure S1 and Tables S2 and S3 for a compilation of known temporal information). First, to incorporate the fact that the residential epithelial cells are first to recognize bacteria, epithelial cells were always updated before DCs in both species. The implementation of this criterion decreased the standard deviation of the clearance time distribution to 0.88 in B. bronchiseptica (μ = 24) and 1.07 in B. pertussis (μ = 22) (see Figure S2A).
Second, because cytokine regulation is important in directing the course of infection and because only limited information was available on the timing of cytokine production and decay, we explored all possible relative durations for the synthesis of the three cytokine classes (PICs, Th1RCs, and Th2RCs), and found that updating Th1RCs before Th2RCs and Th2RCs before PICs led to a significant narrowing of the clearance time distributions (μ = 27, σ = 0.62 in B. bronchiseptica and μ = 22, σ = 0.64 in B. pertussis; see Figure S2B). The first part of this condition implies that either the Th1RCs are produced faster than the Th2RCs or Th1RCs are sensed faster than Th2RCs during the interaction between T0 cells and DCs. The validity of this assumption is supported by the experimental observation that in the absence of the immunomodulating virulence factors TTSS/ACT and FHA/ACT, IFN-γ (a Th1RC) is produced before IL-10 (a Th2RC) [20,21,27]. The condition on PICs was necessary only for their second, adaptive immunity–related activation, and implies that the inhibition of proinflammatory cytokines by Th2-related cytokines is released later than their inhibition of Th1RCs (Th1-related cytokines) [22]. The cytokine timing criterion described above was also necessary to reproduce the known earlier clearance of certain mutant bacteria (described in Table S3; see Systemic Effects of Deletions and Comparison with Experimental Results for more information).
Third, to incorporate the fact that the activation of the recruited phagocytes and consequent phagocytosis of bacteria are complex multistep processes, the corresponding nodes were updated penultimately and last, respectively, in each step. After the implementation of these conditions, B. bronchiseptica and B. pertussis were cleared on a single time step, the 26th and 21st, respectively, in all 1,000 simulations. Thus, the sequence of update orders incorporated in the model offers predictions on the population level differences in the relative time scale of the immune processes. Experimental testing of the timing inequalities expressed by our update criteria can potentially elucidate the possibilities for variation in the clearance time scale of natural infections. The next step was to analyze the activity of each node during our in silico pathogenesis.
To represent a previously uninfected host encountering bacteria, we started the dynamic simulation with initial state (at time step t = 0) of all nodes at 0 (off), except for bacteria that were assumed to be at 1 (on). We observed three distinct temporal patterns common to all 1,000 simulations with asynchronous update, allowing us to identify three phases in the course of infection: innate responses, including PIC production and the recruitment of PMNs and DCs (phase I); B cell– and antibody-mediated responses (phase II); and Th1-related responses leading to a significant activation of phagocyte recruitment and activation, and ultimately leading to bacterial clearance (phase III). To illustrate these phases of pathogenesis, in Figures 3–5 we present subsets of the network of Figure 1, additionally including species-specific virulence factors; these figures delineate the processes that are most active in the three infection stages within each of the two organisms and indicate timing information whenever an estimate was available (see also Table S4 for a compilation of active nodes in each phase). The experimental data support the existence of distinct responses in these three phases. For example, in the bacterial growth curves shown in Figure 6, the bacterial numbers increase exponentially in the first 7 d of infection (our phase I), after which an immune-mediated decline is observed. While the second and third phases do not clearly separate on the bacterial growth curves, there is ample evidence for the existence of both humoral (Th2-related) adaptive immunity (our phase II) and cellular (Th1-related) adaptive immunity (our phase III) in in vivo infections by wild-type strains. Moreover, both for B. bronchiseptica and B. pertussis, recovery from infection was associated with the development of pathogen-specific Th1 cells [33,34]. The long-term steady state of our model after the clearance of bacteria indicates all immune components in an unperturbed (subthreshold) state, with the exception of the two antibody nodes. This steady state is a simplified representation of immunological memory. In the following, we present the in silico time course by focusing on the activity of seven selected nodes (see Figure 7, wild-type [WT]), and we describe the experimental observations supporting our results.
We tested the contribution of individual components/nodes on bacterial clearance by simulating their knockout mutants. For each deletion, we set the affected node's state to off (0) and kept it in this state during the simulation. The asynchronous algorithm was run 1,000 times, in each time step generating a new order of update satisfying the three conditions described in Relative Duration of Regulatory Immune Processes. We determined the effect of the node's knockout by monitoring the time step when bacteria are cleared (if at all) and the behavior of other nodes. Figure 7 represents the pattern of active nodes in one representative simulation. The constraints on the relative duration of immune processes led to bacterial clearance during the same time step in all simulated disruptions as well; however, the activation pattern of individual nodes varied slightly depending on the randomly selected order of update.
To analyze the efficiency of immune responses in detecting and clearing a second challenge with pathogens, we performed simulations of secondary infections. The secondary infection with the same or different species (cross-infections) allowed us to better understand the complex pathogen–host interplay and to generate testable predictions. The secondary infection was modeled by a “secondary” initial state comprising the state of all nodes in the stage of the primary infection when the host encountered the second challenge. This representation of the secondary challenge is not simply a continuation of the primary infection. Although the state of the node “bacteria” does not change at the time of the second challenge (it already is on), the second challenge can lead to the reactivation of certain cytokines and virulence factors that have previously decayed during the primary infection. We modeled three scenarios: first, reinfection of diseased hosts, by using a secondary initial state corresponding to time step 9, in phase II, for both B. bronchiseptica and B. pertussis. Second, we modeled the infection of convalescent hosts by a secondary initial state corresponding to time step 21 for B. bronchiseptica and time step 18 for B. pertussis, in phase III (refer to Figure 7 for the state of key nodes at these time steps). Third, we modeled the reinfection of hosts with immunological memory by a secondary initial state that has antibodies on. In summary, the “secondary” initial state indicates the state of the host when the second bacterial invasion takes place (see Materials and Methods), and is different from the initial state of the primary infection (where only bacteria are on) and from a continuation of the primary infection. Correspondingly, the in silico pathogenesis of the secondary challenge showed a different pattern of node activation than the time course of a primary infection (compare Figure 8 with Figure 7).
We first studied secondary infections by the same species. We simulated the secondary initial state in phase III by preserving the states of TTSS and FHA/ACT off, assuming that the host recognized those virulence factors of the new bacteria and removed them (or became desensitized). Secondary dosage given in a particular phase generally reset the pathogenesis to the beginning of the same phase of the combined infection, with a few notable exceptions. When the secondary dosage was given in phase II of the primary infection, the antigen–antibody complex was active starting with the first time step of the secondary infection, and phase II of the combined infection was one time step shorter than phase II of the primary infection for both species. When the dosage was given in phase III, the recruited PMN peak was delayed compared with the primary infection in B. pertussis, but phagocytosis was immediately activated. We found that the secondary infection was cleared faster than a single (primary) infection in both species and both initial conditions. Secondary infection in phase II (i.e., of diseased hosts) took longer to clear (21 time steps in B. bronchiseptica and 18 time steps in B. pertussis; Figure 8, third column), whereas secondary infection in phase III (i.e., of convalescent hosts) was cleared faster (six time steps in B. bronchiseptica and in B. pertussis; Figure 8, fourth column). The clearance was always associated with Th1-related responses (phase III). Thus, immune components active in phase III led to early clearance of a secondary infection. In conclusion, the phase of the primary infection when the host encountered a second dose of Bordetellae had an important effect on the progression of the second infection.
We simulated B. pertussis after B. bronchiseptica and B. bronchiseptica after B. pertussis cross-infections exploring scenarios with or without antibody cross-reactivity (see Text S3 and Table S5). We found that the only case of faster clearance is secondary infection of convalescent hosts in the presence of antibody cross-reactivity, a result confirmed by preliminary experiments (Wolfe DN and ETH, unpublished data).
We also studied secondary infections of immune hosts by using an initial condition where antibodies are active, a state that approximates immunological memory. It is known that memory T cells and B cells are also generated after an infection; however, these cells require reactivation, and the mechanisms involved are not understood. The capacity of memory B cells to proliferate faster after priming and to produce antibodies with increased affinity could not be incorporated in our current qualitative model. Our approximation thus serves as an upper limit of immunological memory. We found similar results as in the case of prior treatment with antibodies (i.e., B. bronchiseptica was cleared by the sixth time step, but there was no early clearance of B. pertussis). This latter result suggests the possibility of reinfection of previously vaccinated individuals, probably associated with a subclinical disease but a nonzero transmission probability [42].
Immunological responses have been traditionally studied by biochemical and molecular biology techniques. These approaches allow us to manipulate components that are experimentally detectable, and they increase our knowledge about individual immune mechanisms and related responses. In the present study, we synthesized these separate sets of information into an integrated network (Figure 1) that gives a comprehensive view of the system. Immunological studies often focus on model organisms; however, the results ultimately need to be applicable to the natural host. We overcame such limitations by constructing unbiased (consensus) networks such as that represented in Figure 1 and adding on pathogen-specific mechanisms as specific nodes and edges. Although the edges of the network are based on information from experimental observations, the network is more than the sum of its parts because it enables the evaluation of the direct and indirect effects of perturbing each node. Constructing such consensus models has the potential for accelerating new discoveries in a field; such advances are sorely needed for B. pertussis, which is still persistent in human populations in part because the information about human-specific immune responses is limited. Though our network analysis is dependent on the definition of the nodes and edges, it is flexible enough to describe the system under study accurately.
The network-based dynamic model enabled us to analyze the time course of the immunological responses and bacterial clearance. Dynamic modeling usually employs continuous or discrete methods. For continuous models, detailed information about the interaction kinetics, rate constants, and component concentrations is necessary [43,44]. Previous models of immunological response to pathogen invasion, mostly based on ordinary differential equations [45,46] or cellular automata models [47], have focused on the interaction between a few cell types, cytokines, and pathogens. The kinetic parameters of these models were estimated by comparison of pathogen and/or cytokine concentration time courses from experiments and models. By focusing on a small subset of the immune response, these models do not reflect the diversity, complexity, and long-range feedbacks present in pathogen–host interactions, and thus may lead to unrealistic results. In our comprehensive network model, several nodes represent populations of cells or families of cytokines, and edges represent whole signal transduction pathways; thus, the molecular-level description of each node would need quantitative knowledge of complex subnetworks, knowledge that is currently missing. The fact that we have limited knowledge, even at the coarse-grained level presented here, does not allow us to use continuous methods.
The qualitative dynamic descriptions we use, in addition to being the practical choice, are well suited for networks that need to function robustly despite changes in external and internal parameters [48,49]. Qualitative discrete modeling such as ours has been previously successfully implemented in gene regulatory networks and signal transduction networks for predicting the dynamic trajectory of biological circuits and for accessing the reliability of gene regulatory networks in signal processing [31,32,50]. In the study of immunological responses, this approach has been implemented in small networks for the analysis of T cell activation and anergy [51] and for the analysis of lymphocyte subsets [52]. Here, a comprehensive network was constructed to study the immunological responses at the systems level, and the dynamic model of this network was successfully validated.
The ingredients (node states, transfer functions) of our dynamic model refer to the node (component) level, and there is no explicit control over pathway-level effects. Moreover, the combinatorial transfer functions we used are, to varying extents, conjectures, informed by the best available experimental information. For example, FcγR, C3, and Th1-mediated activation of phagocytes might have a complex, partly redundant, and partly synergistic relationship, as there is a requirement for both the recruitment and activation (by at least one mechanism) of phagocytes in the natural clearance of both B. bronchiseptica and B. pertussis. Our assumption for the node “activated phagocytes” allowed all three processes mentioned above to contribute towards bacterial phagocytosis, and the dynamic model allowed us to analyze the temporal separation between contributions to phagocytosis by these three processes. As the model's dynamics is an emergent, systems-level property, and our choice of parameters was based on the normal time course, an agreement between experimental and theoretical results of node disruptions is not inherent, and provides a validation of the model. Indeed, our model agrees with experimental results on numerous negative or positive perturbations in immune mechanisms (such as the deletion of B cells or adoptive transfer of antibodies).
We incorporated possible stochastic differences within individual host responses by randomly sampling update orders, and incorporated known relative temporal information on interactions and processes as restrictions on the order of update; for example, epithelial cells were always updated before dendritic cells. At a molecular level, we can interpret the restrictions on the order of updates either as restrictions on the relative duration of processes or on the time required to activate their target nodes. For example, updating Th1RCs before Th2RCs either means that the signal transduction pathway activating Th1RCs is faster or that Th1RCs once produced can activate downstream processes more quickly than can Th2RCs. Moreover, the orders of updates also gave an insight into the regulatory processes; for example, we find that a slow phagocytosis process ensures better, more reproducible bacterial clearance. The model identified two possible oscillatory behaviors of Th1RCs and Th2RCs: short oscillations in the absence of modulating factors, and the switch-like behavior in the presence of those modulating factors.
Our network model gives insight into how virulence factors change the course of innate and adaptive immune responses, and allows the comparison of species-specific virulence factors. For example, both the B. bronchiseptica O-antigen and the B. pertussis PTX inhibit early immune responses, assisting bacterial multiplication and survival. Our directed network representation allows for tracing the sequence of immune responses following recognition of the pathogen. The O-antigen interacts with host components earlier in this sequence (in the pseudodynamic sense introduced in [53]) than does PTX, as it inhibits bacterial recognition, while PTX disables the recruitment of immune cells that recognize bacteria. However, the inhibitory effect of PTX is ultimately more effective since it gives B. pertussis a means to resist the effect of serum antibodies. Comparing the way in which the B. bronchiseptica TTSS and the B. pertussis FHA/ACT modulate T helper cell balance gives insight into why the TTSS gene is not expressed in B. pertussis. The TTSS causes necrosis of PMNs [17], which is possibly a side effect of the evolution of the TTSS's primary role to modulate cytokines, and which ultimately leads to a stronger immune response. Unlike B. bronchiseptica, the milder pathology during B. pertussis infection allows it to modulate immunity silently.
The model also provides insight into regulatory mechanisms of antigens and suggests differential regulation of virulence factors by two distinct mechanisms: (1) immune-mediated neutralization/elimination, or (2) decay due to bacterial gene regulation or to a reduction in bacterial numbers. The conclusion that the effects of the secreted factors ACT and TTSS are longer lasting than the effects of PTX is supported by the observation that PTX does not modulate cytokines or chemokines [23]. The model also predicts that the decay rates of the (effect of) Th2RCs, TTSS, and FHA/ACT are similar. Results of new experiments will lead to further refinement of the model; for example, the speculations that the LPS containing O-antigen decreases host cells' access to other factors such as adhesins, or that it facilitates secretion of toxins, can be incorporated if supported by dynamic evidence.
One of the most important contributions of our model is the identification of the pathogenesis phases. The definition of three infection phases is preferable to using exact timing because specific processes can have different durations in different hosts, and timing results obtained in model organisms might not be directly applicable to the pathogens' natural hosts. The time scale of processes in our dynamic model does not correspond exactly to the experimentally observed timing in mice infections, but the transition from one phase to another is captured. Comparing the model and experimental time scales, we estimate that one time step corresponds to 1–2 d of infection. The time span of the third phase is shorter in the model than in experiments because quantitative differences cannot be reproduced in the present qualitative model; or equivalently, the implicit bacterial concentration threshold of our model is higher than the experimental detection threshold.
Models such as this allow the efficient use and logical representation of available information. We extracted host- and pathogen-specific processes from the experimental literature, and used overlapping information, such as the modulation of Th1 and Th2 cell types, from other host–pathogen systems. Our model allows the identification of new relationships and the making of new predictions that would be difficult to derive from less formal analysis. First, the logical network representation of the pathogen- and host-related information will allow microbiologists and immunologists to see the knowledge gaps in the results from different laboratories and appreciate the synergy between the pathogen and the host. For example, the role of FHA in cytokine balance and its possible synergy with TTSS and ACT has not been studied in the ancestral pathogen B. bronchiseptica. Second, the timing constraints and parameters derived from our dynamic model give predictions regarding the time scales of several of the processes. Experimental testing of our results on the degradation of toxins and cytokines (or their effects) will be able to establish the mechanism responsible for this degradation. Third, and perhaps most important, the model is able to predict the outcome of perturbations not yet explored experimentally and to direct future experimental efforts. A considerable amount of sustained manual effort is necessary to study immunological processes by traditional techniques. Modeling leads to efficient ways for analyzing the effect of knockout mutations, as it is straightforward to delete certain components in the model and observe the consequences on bacterial clearance and activation/inhibition of immunological components. While our model lacks quantitative kinetic and spatial detail, it can serve as a scaffold to which experimentalists and modelers can add future results on the regulation between immune components and bacterial mechanisms as they become available.
Bordetella evolution is leading to the frequent emergence of new strains and species. Our simulations of cross-infections with or without cross-immunity constitute the first step toward modeling the within-host effects of newly emerged Bordetella species. Such simulations can be extended to incorporate different combinations of known virulence factors, or to explore putative new virulence factors, providing reasoned expectations prior to costly and time-consuming animal experiments. As population-level dynamics are shaped by within-host interactions [42], models such as ours can increase our understanding of the population-level effects of specific molecular evolution. Infections by two different strains or species can result in a similar outcome (e.g., persistent disease) in terms of bacterial clearance, but have different pathological effects. These species-specific aspects can be readily studied in our model by systematically comparing the simulated dynamics of the immune system in the two species. Comparison between models of different strains and species will allow us to recognize crucial virulence mechanisms and can give us insight into the evolution of new virulence mechanisms.
The present study led us to the identification of the three phases in Bordetella infections. Putative medical treatments can thus be evaluated in silico by simulating them through adding/removing or activating/inhibiting certain nodes and studying their effect on these three phases. Qualitative conditions (e.g., for cytokine regulation) provided by such models can be used in the future to detect the phase of infection in patients, and ultimately to predict recovery and design medication. Some of the manipulations of the model (e.g., the deletion of FHA/ACT) result in early clearance of B. pertussis in the simulation. Our study of minimal components necessary for early clearance of a secondary infection suggests that Th1RCs along with antibodies can be used as a prophylactic measure. We believe that these and other such observations will be useful for the control of B. pertussis and could be applied to other diseases.
This study addresses the goals of systems biology by effectively encapsulating prior knowledge to generate a mechanistic and predictive understanding at the systems level. Understanding at the network level is necessary for formulating detailed quantitative models of within-host disease dynamics. Our methodology can be extended to other respiratory pathogens, and ultimately to pathogens in general. The outputs of within-host models serve as inputs to broader-scale epidemiological models; for example, the effect of host immune components (intrinsic factors) was shown to be important in cholera epidemics [54,55]. Thus, our study has implications in epidemiological models as well.
The networks in Figures 1, 3, 4, and 5 were drawn with the SmartDraw software (http://www.smartdraw.com). The dynamic model was implemented by custom Python code.
Here, we explain a few representative logical transfer functions used in our model; a detailed justification of all transfer functions is given in Text S1.
The Boolean rule for activated phagocytic cells is:
Here, the “AND” operator between phagocytes and antibody-mediated responses describes the fact that phagocytosis requires antigen-specific activation of phagocytes. The “AND” relationship between complement and complement-fixing antibodies denotes that only the classical complement pathway activated through antibodies could activate phagocytes above threshold levels. In case of other infections, the alternative complement pathway also activates phagocytes, but in Bordetella infections in B cell–deficient mice, the alternative complement pathway is not sufficient for significant decrease in bacterial numbers [39], supporting the above description. We used an OR relationship between the activation of phagocytes by complement + complement-fixing antibodies and antigen–antibody complexes because PMNs and macrophages are recruited and activated in C3-deficient mice [15,39]. The “OR” condition between PMNs and macrophages is justified by the fact that both of these major cell types contribute directly to the bacterial phagocytosis [56]. These contributions differ in their timing and extent because different activation signals exist for these cell types [57]. Hence, for a simplification in the model, we included separate nodes for PMNs and macrophages and not for each cell type.
The Boolean rules for T helper cells are:
Th cells are activated through the interaction of T0 cells with DCs in the model. Depending on the activation of either Th1RCs or Th2RCs, T0 cells differentiate into Th1 cells or Th2 cells, respectively. The presence of all three components is essential for the activation of a particular T helper cell type; hence, we used an “AND” relationship.
Our model has two types of parameters. First, the threshold parameters dcmax and pmax signify a condition for the continuous expression of the regulator nodes DCs and Phagocytosis to induce a state change in their respective targets, T0 cells and Bacteria. Second, the decay constants defined for certain antigens (τFHA/ACT/τTTSS) and cytokines (τTh1RC/τTh2RC) express the condition that these antigens and cytokines are degraded after being active for more than a given number of time steps, even if the conditions for their activation are still satisfied. Here, we give a biological justification of these parameters.
To incorporate the fact that bacteria are not cleared by innate or early adaptive responses, we assume that only a sustained process of phagocytosis is able to clear the infection (i.e., that phagocytosis needs to be continuously on for pmax time steps).
The notation
was introduced to indicate that the present (i = 0) as well as past (i = 1
⋯ pmax) states of the node Phagocytosis compound synergistically to determine the state of the node Bacteria. This condition signifies that only sustained above-threshold levels of phagocytosis lead to bacterial clearance.
Secreted bacterial toxins and bacterial adhesins such as LPS follow different time courses during pathogenesis. Toxin levels can be reduced by interactions with host immune components (e.g., antigen-specific antibodies) or free decay. Adhesin levels decrease due to a reduction in bacterial numbers or a regulation of the genes encoding the antigens. Though both secreted factors and adhesins can activate antibodies, the removal (neutralization) of the former is not associated with bacterial clearance as they are secreted and released in the host, unlike the latter. We assumed that the levels of PTX are determined by the presence of bacteria and of pertussis-specific antibodies, while O-antigen is expressed constitutively in B. bronchiseptica.
The effect of other virulence factors such as TTSS and FHA/ACT is more complex, possibly due to the cooperation between different virulence factors. The node TTSS expresses the cooperation between TTSS and ACT in B. bronchiseptica, and the node FHA/ACT expresses the cooperation between the adhesins FHA, LPS, and the secreted factor ACT in B. pertussis. The effects of the TTSS and FHA/ACT are shown to be removed or decreased below threshold levels during late pathogenesis [21,23,27,58]. This reduction could not be reproduced simply by the neutralization of TTSS-secreted factors or of ACT by antibodies. In the absence of detailed information, we assumed that the activity or effect of these nodes decays after a continuous presence for a prespecified number of time steps τTTSS and τFHA/ACT, even if the conditions that activated them persist. Thus, the transfer function for TTSS is TTSS* = Bacteria AND NOT
; similar transfer function applies to FHA/ACT. Thus, the condition for an above-threshold concentration of TTSS or FHA/ACT nodes (state 1) is the presence of their activators at the previous time point, combined with the absence (subthreshold concentration) of TTSS or FHA/ACT τTTSS or τFHA/ACT time points ago, respectively. Note that it is possible that other virulence factors such as PTX also exhibit uncatalyzed decay/inactivation, but this decay rate is probably smaller than their removal rate by active immune mechanisms. The crucial difference between the effect of secreted factors PTX, on one hand, and ACT and TTSS, on the other hand, is that PTX does not modulate chemokine and cytokine production [23]. Hence, PTX is less likely to have long-lasting effects after its neutralization, unlike ACT and TTSS. Our description of the latter nodes allows us to capture such longer-lasting effects of the nodes by using a decay time longer than one time step.
The decision between generating Th1 cells or Th2 cells is determined by a group of cytokines produced during the interaction of antigen-presenting cells and T cells. How the balance between Th1 cells and Th2 cells maintained in the presence of double-negative feedback between Th1-related cytokines and Th2-related cytokines is still unknown. Our modeling of the double-negative feedback between Th1RCs and Th2RCs led to different dynamic outcomes in asynchronous and synchronous algorithms. Asynchronous algorithms gave advantage to the process that is activated earlier (an example of bistability), whereas in the synchronous algorithm synchronous oscillations of Th1RCs and Th2RCs were observed (refer to Table 2 and Text S1 for the Boolean transfer functions of these nodes and to Dynamic simulation for the algorithm). To remove the dependence on the order of update of the asynchronous method in the consensus Bordetellae model (without specific virulence factors), we assumed that Th1RCs/Th2RCs decay after a period τTh1RC/τTh2RC. We found that the value of these decay times is dependent on bacterial virulence factors (refer to Parameter analysis and Text S2 for details). We chose the decay times such that (1) in the species-specific models incorporating TTSS/FHA (in B. bronchiseptica) and FHA/ACT activity (in B. pertussis), a reproducible switch (i.e., activation of Th2RCs followed by Th1RC activation) is observed; and (2) in the consensus model (absent specific virulence factors), there is more than one such switching. Alternative activation of Th1RCs and Th2RCs is found in many disregulations, for example in allergies and allografts [59,60], when the balance between Th1RCs and Th2RCs is not directly modulated, and is also predicted by other models [61].
In this section we analyze the realistic range of the threshold parameter pmax and the decay constants τTTSS/τFHA/ACT. All parameters and their analysis are given in Text S2. We performed a systematic search in parameter space to determine the parameter regions that satisfy the following two criteria: (1) reaching bacterial clearance and (2) association of bacterial clearance with activation of phase III. We performed 1,000 simulations for each parameter value in a biologically realistic region. The disease was allowed to evolve for 70 time steps in each simulation. To test the first condition, we determined the distribution of the time steps at which bacteria were cleared; for the second criterion, we monitored the frequency of activation of the node Th1RC around the time step at which bacteria were cleared. Figure S4 shows the evaluation of condition 2 in the case of τTTSS, as an example.
We sampled the threshold for clearance-inducing phagocytosis, pmax, in the interval 0 ≤ pmax ≤ 10. For pmax < 2, B. bronchiseptica were cleared by innate immune responses, which is inconsistent with the literature [2,14,24,39]. For pmax > 1, the bacterial clearance was delayed by one time step for each unit increase in the pmax value in both species. Increasing pmax above 2 increased the length of phase III, which was shorter in the simulation than in experimental studies. Our perturbation analysis indicated that for pmax > 4, the perturbations in which early clearance was observed (TTSS deletion, antibody-mediated clearance in B. bronchiseptica; see Table S3) could not be reproduced, implying that the condition is too stringent. We chose pmax = 4; the success of the results obtained with this value suggest that in later phase III, bacterial concentration might decrease so much that phagocytosis is active at low (close to threshold or even subthreshold) levels, requiring a longer time for complete clearance. Experiments performed at shorter time intervals around the expected clearance time than the customary sampling at days 28, 50, and 70 will be necessary to better elucidate this parameter.
and
τFHA/ACT and τTTSS were varied in the intervals 1 ≤ τFHA/ACT ≤ 22, 1 ≤ τTTSS ≤ 22, and the distribution of the time steps of bacterial clearance was plotted (see Figure S4). Clearance was delayed by one time step with each unit increment. When τTTSS > 15 or τFHA/ACT > 18, bacteria were not cleared. For 3 ≤ τFHA/ACT ≤ 8 or 3 ≤ τTTSS ≤ 8, the bacterial clearance was not associated with Th1-related responses; for 0 ≤ τFHA/ACT or 0 ≤ τTTSS ≤ 3, there were spurious oscillations between Th2RCs and Th1RCs. As natural killer cell activation in B. pertussis infection activates IFN-γ (a Th1RC) production [38,62] earlier than in B. bronchiseptica infections, thus τFHA/ACT < τTTSS. Hence, we used τFHA/ACT = 12 and τTTSS = 15. The result that the composite action of the nodes FHA/ACT and TTSS decreases below threshold levels after 12 and 15 time steps, or after nine to 12 time steps of active antibody response, suggests that these nodes have a longer effect than other independently acting virulence factors such as PTX and O-antigen. A possible mechanism in support of this suggestion is damage to cellular systems; for example, TTSS permeabilizes cell membranes, which require a longer time to heal even after the toxin can be neutralized by antibodies.
Secondary infections were modeled by using initial conditions that represent the state of the host when a secondary bacterial invasion takes place. The secondary initial condition is defined by the active components of the immune system and bacteria. Three scenarios were simulated. First, an infection of a diseased host was represented using an initial condition where the nodes “Epithelial cells,” “Ag–Ab complex,” “Complement-fixing Ab,” “Other Ab,” “Th2RCs,” “T0 cells,” “Th2 cells,” and “DCs” were active. The node “Complement” was also active in the case of B. pertussis infection. Second, the infection of a convalescent host was represented by an initial condition in which the following nodes were on: “Epithelial cells,” “Complement,” “Ag–Ab complex,” “Complement-fixing Ab,” “Other Ab,” “B cells,” “Th1RCs,” “PICs,” “T0 cells,” “Th1 cells,” “Recruited PMNs,” “Macrophages,” and “DCs.” Last, the simulation of infection of immune hosts was performed by using the memory steady state of primary infections as initial condition (i.e., setting the two antibody nodes “Complement-fixing Ab” and “Other Ab” on, along with the node “Bacteria”).
We used the sequenced WT strains of B. bronchiseptica (RB50) and B. pertussis (BP536) as described previously [63,64]. Bacteria were maintained on Bordet-Gengou agar (Difco, http://www.bd.com), inoculated into Stainer-Scholte broth at optical densities of 0.1 or lower, and grown to mid-log phase at 37 °C on a roller drum.
C57BL/6 mice were obtained from The Jackson Laboratory (http://www.jax.org). Mice were maintained and treated at the Pennsylvania State University in accordance with approved institutional guidelines. Prior to inoculation, mice were lightly sedated with isoflurane (Abbott Laboratories, http://www.abbott.com) and were inoculated by pipetting 50 μl of phosphate-buffered saline containing the 5 × 105 CFU of B. bronchiseptica or B. pertussis onto the tip of the external nares. For bacterial enumeration, groups of four animals were killed at the indicated time point after inoculation. Colonization of respiratory organs was quantified by homogenization of each tissue in phosphate-buffered saline, plating onto Bordet-Gengou blood agar containing 20 μg/ml streptomycin, and colony counting. For reinfection experiments, mice were intranasally inoculated with 5 × 105 CFU of bacteria in 50 μL and reinfected with the same dose 28 d after the initial inoculation. Mice were dissected 3 d after the reinfection, and bacterial numbers were enumerated as described above.
The GeneDB (http://www.genedb.org) accession numbers for the bacterial genes discussed in this paper are: adenylate cyclase toxin (BB0324 in B. bronchiseptica, BP0760 in B. pertussis), filamentous hemagglutinin (BB2993 in B. bronchiseptica, BP1879 in B. pertussis), pertussis toxin (five subunits: BP3786, BP3785, BP3787, BP3783, BP3784), and type III secretion system (BB1628). The ExPASy database (http://ca.expasy.org) accession number for Toll-like receptor 4 is TLR4_HUMAN (O00206) and TLR4_MOUSE (Q9QUK6). |
10.1371/journal.pntd.0000886 | LXR Deficiency Confers Increased Protection against Visceral Leishmania Infection in Mice | The liver X receptors (LXRs) are a family of nuclear receptor transcription factors that are activated by oxysterols and have defined roles in both lipid metabolism and cholesterol regulation. LXRs also affect antimicrobial responses and have anti-inflammatory effects in macrophages. As mice lacking LXRs are more susceptible to infection by intracellular bacteria Listeria monocytogenes and Mycobacterium tuberculosis, we hypothesized that LXR might also influence macrophage responses to the intracellular protozoan parasite Leishmania chagasi/infantum, a causative agent of visceral leishmaniasis.
Surprisingly, both LXRα knock-out and LXRα/LXRβ double-knock-out (DKO) mice were markedly resistant to systemic L. chagasi/infantum infection compared to wild-type mice. Parasite loads in the livers and spleens of these animals were significantly lower than in wild-type mice 28 days after challenge. Bone marrow-derived macrophages from LXR-DKO mice infected with L. chagasi/infantum in vitro in the presence of IFN-γ were able to kill parasites more efficiently than wild-type macrophages. This enhanced killing by LXR-deficient macrophages correlated with higher levels of nitric oxide produced, as well as increased gene expression of IL-1β. Additionally, LXR ligands abrogated nitric oxide production in wild-type macrophages in response to infection.
These observations suggest that LXR-deficient mice and macrophages mount antimicrobial responses to Leishmania infection that are distinct from those mounted by wild-type mice and macrophages. Furthermore, comparison of these findings to other intracellular infection models suggests that LXR signaling pathways modulate host antimicrobial responses in a complex and pathogen-specific manner. The LXR pathway thus represents a potential therapeutic target for modulating immunity against Leishmania or other intracellular parasites.
| Leishmania spp. are protozoan single-cell parasites that are transmitted to humans by the bite of an infected sand fly, and can cause a wide spectrum of disease, ranging from self-healing skin lesions to potentially fatal systemic infections. Certain species of Leishmania that cause visceral (systemic) disease are a source of significant mortality worldwide. Here, we use a mouse model of visceral Leishmania infection to investigate the effect of a host gene called LXR. The LXRs have demonstrated important functions in both cholesterol regulation and inflammation. These processes, in turn, are closely related to lipid metabolism and the development of atherosclerosis. LXRs have also previously been shown to be involved in protection against other intracellular pathogens that infect macrophages, including certain bacteria. We demonstrate here that LXR is involved in susceptibility to Leishmania, as animals deficient in the LXR gene are much more resistant to infection with the parasite. We also demonstrate that macrophages lacking LXR kill parasites more readily, and make higher levels of nitric oxide (an antimicrobial mediator) and IL-1β (an inflammatory cytokine) in response to Leishmania infection. These results could have important implications in designing therapeutics against this deadly pathogen, as well as other intracellular microbial pathogens.
| Liver X receptors (LXRs) are a family of nuclear transcription factors that play an integral role in both lipid metabolism and the regulation of inflammation [1], [2]. Two isoforms of LXR exist in both mouse and human: LXRα is mainly expressed in metabolically active tissues such as liver, intestine, kidney and adipose tissue, in addition to macrophages and myeloid dendritic cells (DCs) [3]; LXRβ has a relatively ubiquitous expression pattern [4]. LXRs form functional heterodimers with retinoid X receptors (RXRs) that are activated upon binding to intracellular oxysterols, thus functioning as sensors of cellular cholesterol levels. Upon activation, LXR-RXR complexes promote expression of genes involved in cholesterol efflux, absorption, conversion to bile acids, and lipogenesis [5]. In addition to controlling these key elements of lipid homeostasis, activated LXR also inhibits the development of inflammatory pathways via repression of NF-κB signaling, particularly in macrophages [6]. This dual ability to promote cholesterol efflux and inhibit inflammation supports a protective function for the LXRs against diseases such as atherosclerosis, which are characterized by cholesterol-laden foam cells and chronic inflammation [1]. Consistent with this model, treatment with synthetic LXR agonist molecules reduces disease incidence in animal models of atherosclerosis [7]. LXR-deficient mice, conversely, develop enlarged, cholesterol-laden livers and elevated serum cholesterol upon exposure to high cholesterol diets, and are highly prone to atherosclerosis [8].
In addition to their involvement in the development of these chronic metabolic diseases, macrophages play a fundamental role in innate immune activity. The expression of LXR in macrophages, combined with the ability of these receptors to regulate inflammatory pathways, suggest a role for LXR in combating specific microbial pathogens. Previously it has been demonstrated that LXR-deficient mice are more susceptible than wild-type mice to infection with the intracellular bacterial pathogens Listeria monocytogenes [9] and Mycobacterium tuberculosis [10]. The increased susceptibility to Listeria was associated with an LXR-regulated gene expressed in macrophages called SPα, identified as having anti-apoptotic function. LXR-deficient mice displayed increased macrophage apoptosis during the course of Listeria infection, correlating with a decreased ability to clear the bacteria. Demonstration that LXR-associated pathways impact macrophage responses to an intracellular bacterium suggested that these receptors and their downstream metabolic networks might also influence immunity against other microbial pathogens.
The genus Leishmania, comprised of numerous distinct species of trypanosomatid protozoa, infects predominantly macrophage cells in their mammalian hosts. Leishmania species, particularly L. major, have been studied in detail over many decades, helping to define many aspects of host immunity, including initial characterizations of TH1- versus TH2-type, CD4+ T cell responses [11]. Invasion of phagocytic host cells occurs through complex interactions between Leishmania surface structures and phagocytic receptors in the host cell [12], which are organized within cholesterol-containing lipid rafts in the outer membrane. Upon cellular entry, promastigotes remain in endosomal structures, transform into amastigotes, multiply, and inhibit immune-mediated clearance by the host through various mechanisms of immunosuppression [13]. Infected macrophages and other immune cells express important proinflammatory cytokines and molecules that play roles in parasite resistance, including IL-1, IL-6, IL-12, TNF-α, CD40L, and IFN-γ [14]–[16]. Also critical to Leishmania immunity is the expression of inducible nitric oxide synthase (iNOS) by macrophages, which leads to microbicidal killing via NO production. This pathway is particularly essential in murine cells [17]–[19]. Experimental vaccination regimens in mice that augment immunity against Leishmania challenge often enhance these pathways, and correlate with increases in parasite killing.
In this study we examine the role of LXR activity in resistance to the visceralizing species L. chagasi/infantum, which replicates primarily in the liver and spleen following intravenous inoculation in the mouse. Given similarities in pathogenesis of infection between Leishmania and the bacteria Listeria (liver tropism, macrophage invasion, intracellular replication, and requirement for TH1 cellular responses for clearance), we hypothesized that LXR might also prove important for Leishmania resistance. In contrast to our published L. monocytogenes results, we demonstrate here that LXR-deficient mice are more resistant to L. chagasi/infantum infection compared to wild-type mice. This suggests that loss of LXR function does not simply lead to a global, general defect in immunity to intracellular pathogens, and broadens the implications of LXR modulation in medical therapies. Here we characterize the increased resistance of LXR-deficient mice to visceral Leishmania infection, and present data implicating the role of the macrophage in this phenotype.
Mice were housed in a temperature-controlled room under a 12-hour light/12-hour dark cycle and maintained on standard chow under specific-pathogen-free conditions. LXRαβ+/+(wild-type), LXRα−/−, LXRβ−/−, and LXRαβ −/− (DKO) mice were maintained in a pure C57Bl/6 background, following greater than 20 backcrosses. Knockout and wild-type mice in a mixed background (Sv129xC57Bl/6) were originally provided by David Mangelsdorf (University of Texas Southwestern Medical Center, Dallas, TX) and have been separately backcrossed with each other since their original creation in 1999. Mice were sex- and age-matched with littermates for all Leishmania challenge experiments. Mice ranged in age from 6–12 weeks at the beginning of experiments. Typical experiments included equal numbers of both male and female mice, but no significant differences in infection levels were observed between male and female mice in these experiments. All mouse experiments were approved by the Institutional Animal Care and Research Advisory Committee of UCLA, and the Institutional Animal Care and Use Committee at Los Angeles Biomedical Research Institute.
Leishmania chagasi/infantum (MHOM/BR/00/1669), originally isolated from a patient with visceral leishmaniasis in Northeast Brazil, was maintained by serial intracardiac injection in outbred male golden hamsters. Revirilized amastigotes were isolated from hamster spleens removed from animals infected approximately 3 months prior to euthanasia. Promastigote parasites were grown in hemoflagellate-modified minimal essential medium (HOMEM) [20] at 26°, and passaged for no longer than 6 weeks. To enrich for metacyclic promastigotes, parasites were initially seeded at 3×106 cells/mL and grown for 6 days until cultures reached stationary phase (5−7×107 cells/mL). Parasites grown in this way displayed similar infection kinetics, in both mice and macrophages, to parasites enriched for metacyclics by Ficoll gradient purification, in our hands. Parasites were washed in PBS, counted, and resuspended at 5×107/mL for tail vein injection of 1×107 parasites/200 µL. Following specified durations of infection, mice were euthanized, and parasite burdens were assessed microscopically by blinded counting of Giemsa-stained liver imprints. LDU (Leishman-Donovan units) equals the number of amastigotes per cell nuclei × liver weight (in mg).
Bone marrow-derived macrophages (BMDM) from C57Bl/6 wild-type or LXR-deficient mice were obtained by differentiating bone marrow cells in complete DMEM/20% FBS supplemented with M-CSF as described [21]. After 7 days, BMDM were collected following trypsinization and 1×106 cells were allowed to adhere overnight to coverslips in 12-well plates for infection assays. Primary peritoneal macrophages were harvested from euthanized animals following thioglycollate injection as described [22]. For CFSE labeling of parasites, L. chagasi/infantum promastigotes were incubated in 5 µM CFSE in PBS for 10 min. at 37°, and washed twice in PBS. Cells were infected with L. chagasi/infantum promastigotes (grown to stationary phase to enrich for metacyclics, as above) at multiplicity of infections (MOI) ranging from 1∶1 to 7∶1. Infections were performed in serum-free media, and plates centrifuged briefly to synchronize infection. After 1 hour of infection, extracellular parasites were removed by rinsing twice with PBS, and cells were incubated in fresh medium, at 37°, 5% CO2 for an additional hour. Macrophage infection levels were quantitated by microscopy by Giemsa staining cover slips and counting parasites. A minimum of either 500 amastigotes or mononuclear cell nuclei were counted per slide, and numbers reported as amastigotes/nuclei. Alternatively, infected macrophages were removed from plates, fixed in 2% paraformaldehyde/PBS, and analyzed by flow cytometry for CFSE (FL-1) positive events.
Primary macrophages were derived in culture as above. Nitric oxide production was measured following infection of 1×106 macrophages in 12-well plates with L. chagasi/infantum at an MOI of 5∶1 in complete DMEM medium. Designated wells were preincubated with 10 U/ml recombinant murine IFN-γ overnight for 16–24 hours prior to infection. At varying time points after initiating infection, supernatants were sampled and assayed for the accumulation of nitrite ion, which is proportionate to nitric oxide production, using a Griess Reagent System (Promega). Samples were analyzed by absorbance photometry at a wavelength of 535 nm and nitrite ion concentration calculated by comparison with a standard curve of serial dilutions of sodium nitrite.
Total genomic DNA was extracted from frozen liver and spleen samples (<25 mg) from infected animals, using an UltraClean Tissue DNA Isolation Kit (Mo Bio Laboratories, Inc., Carlsbad, CA) according to the manufacturer's instructions. Quantitative PCR assays were performed using TaqMan probes and TaqMan Universal PCR Master Mix (Applied Biosystems, Inc.), with typical 25 µL reactions containing 200 nM of each primer and probe and 2 µL of genomic DNA template diluted 1∶10 following column elution. L. chagasi/infantum parasite DNA was detected using primers and probe specific for GP63, a known virulence gene [23]. Quantitative values of Leishmania were normalized to levels of a mouse gene (TNF-α) to account for differences in amounts of tissue sampled. All primers and probes were synthesized by Integrated DNA Technologies (Coralville, Iowa); sequences available upon request. PCRs were performed on an Applied Biosystems Prism 7000 sequence detector, and data was analyzed using SDS1.2.3 software. GP63 Ct values were converted to absolute parasite counts using previously determined standard curves, and were normalized to the amount of tissue DNA in each sample as determined by TNF-α Ct values.
Total RNA was extracted from tissues or cells growing in vitro using TRIzol reagent (Invitrogen) and was reverse transcribed to obtain cDNA (Applied Biosystems). SYBR-based or Taqman-based real-time quantitative PCR assays were performed using an Applied Biosystems 7900HT sequence detector as previously described [24]. Data was normalized to housekeeping genes RPLP0 (36B4) or HPRT1, both of which were externally verified as appropriate reference genes under these experimental variables. Primer sequences are available upon request.
Differences between mean values of counted Giemsa-stained parasites, parasite loads measured by qPCR, and NO levels were analyzed by a two-tailed Student's t-test.
We have previously reported that LXRα−/− and LXRαβ−/− (LXR-DKO) mice are more susceptible to Listeria monocytogenes infection than wild-type control mice [9]. To determine the role of LXR in the pathogenesis of the macrophage-infecting Leishmania parasites, we challenged these same knockout mouse strains with the visceralizing species Leishmania chagasi/infantum. Because of the high expression levels of LXRα in the liver, we chose to study this visceral infection model of L. chagasi/infantum, as it is found predominantly in the liver and spleen of mice following intravenous injection. Wild-type C57Bl/6 mice display maximal numbers of L. chagasi/infantum in the liver four weeks after intravenous challenge infection with stationary-phase promastigotes, followed by subsequent parasite clearance associated with granuloma formation [25]. Parasite numbers in the spleen, in contrast, gradually increase over time and can persist chronically over the life of the animal [26]. We infected wild-type and LXRα−/− C57Bl/6 mice intravenously with virulent, stationary-phase L. chagasi/infantum promastigotes, and assessed parasite burden in the liver by microscopy four weeks later, when levels peak in wild-type mice. Mice deficient for LXRα had significantly decreased parasite burdens compared to wild-type animals 28 days following challenge (Fig. 1A, p = 0.007). LXRβ−/− mice, compared to wild-type mice, did not exhibit reproducible differences in parasite load in two experiments (data not shown), and were not examined further for this study. To determine if these results were mouse strain-dependent, we also challenged wild-type and LXR-deficient mice in a mixed strain background, Sv129xC57Bl/6. All mice, both wild-type and LXR-knockout, had been backcrossed within this mixed strain for many years and were therefore congenic. Although mixed background wild-type mice generally had lower parasite loads than C57Bl/6 wild-type mice, LXRα−/− and LXR-DKO mice in the mixed background also displayed significantly lower parasite loads than their wild-type littermates (Fig. 1B, LXRα−/−: p = 0.003, LXR-DKO: p = 0.003).
To assess the lower parasite levels in the spleen and at earlier time points after infection, we utilized quantitative PCR (qPCR). Liver and spleen genomic DNA was harvested and subjected to Taqman-based qPCR using Leishmania-specific primers and probes. In agreement with microscopy counts, parasite numbers were reduced in livers of LXRα−/− animals compared to wild-type mice one month after infection (Fig. 1C, p = 0.014). In addition, parasite numbers measured by qPCR were also reduced in LXRα−/− spleens (Fig. 1D, p = 0.06). To determine the level of Leishmania infection in LXR-KO mice at earlier time points, we measured parasite levels in these tissues at 3 and 7 days following intravenous challenge with L. chagasi/infantum. The amount of parasite DNA detected in LXRα−/− livers was significantly lower on both day 3 and day 7 than the amount detected in wild-type livers (Fig. 1E, d3: p = 0.002, d7: p = 0.0002). Spleens displayed relatively low overall levels of parasite DNA at early time points, but by day 7, parasite loads in wild-type spleens were significantly higher than in LXRα−/− spleens (Fig. 1F, d3: p = 0.04, d7: p = 0.02). Taken together, the differences in organ parasite loads in LXR-deficient mice suggest that LXR-regulated pathways can significantly influence the overall control of visceral leishmaniasis infections.
Given the high expression levels of LXRα in macrophages, and the central role these cells play in Leishmania pathogenesis, we examined the effect of macrophage LXR deficiency on the kinetics of parasite uptake and infection in vitro. We differentiated bone marrow-derived macrophages (BMDM) in vitro for 7 days, then infected these macrophages with stationary-phase L. chagasi/infantum promastigotes for 1 hour, before removing any extracellular parasites by washing. Parasites were labeled with the intracellular dye CFSE to allow for analysis of infected macrophages by flow cytometry. Two hours following infection, macrophages from wild-type, LXRα−/−, and DKO mice all displayed similar patterns of CFSE staining (Fig. 2A), suggesting that parasite uptake was similar in all cells. We also infected wild-type and LXR-deficient macrophages grown on cover slips with L. chagasi/infantum. Giemsa staining and microscopy also revealed identical percentages of infected cells, as well as identical numbers of parasites/cell in wild-type, LXRα−/−, and DKO macrophages (Fig. 2B). These results suggest that LXR does not affect the process of initial parasite internalization by macrophages in vitro.
We next measured the kinetics of infection and parasite clearance over time. When macrophages were infected with Leishmania alone, there was no difference between wild-type and DKO macrophages in the percentage of infected cells over 48 hours (Fig. 2C, left panel). During in vivo infections, however, macrophage killing of parasites is known to be dependent on endogenous IFN-γ produced by other cell types, including NK cells and T cells [16]. We thus repeated in vitro infections by pretreating macrophages with exogenous IFN-γ, in order to provide a relevant macrophage activating stimulus. In the presence of IFN-γ, DKO macrophages displayed more efficient clearance of internalized parasites over 48 hours than wild-type macrophages (Fig. 2C, right panel; p = 0.007 at 48 hr). Interestingly, IFN-γ did not appear to enhance parasite clearance in wild-type macrophages. These results suggest that LXR-dependent pathways in macrophages partially inhibit the IFN-γ-mediated clearance of Leishmania.
We next examined the effect that LXR deficiency has on specific parasite defense mechanisms of macrophages. Expression of inducible nitric oxide synthase (iNOS) by macrophages leads to nitric oxide (NO) production resulting in destruction of intracellular pathogens [27]. We measured NO production by macrophages following Leishmania infection in the presence or absence of IFN-γ in vitro. Without exogenous IFN-γ, levels of NO secreted by BMDM infected with Leishmania parasites over three days remained low (Fig. 3A, left panel). However, when BMDM were pretreated with IFN-γ before parasite infection, cells from LXR-DKO mice secreted enhanced levels of NO compared to wild-type cells (Fig. 3A, right panel). This finding was verified in both BMDM and peritoneal macrophages (Fig. 3B). The increased level of NO production by DKO macrophages in the presence of IFN-γ correlates with enhanced clearance of the parasite (Fig. 2C), suggesting that increased IFN-γ-induced NO production by macrophages may contribute to the enhanced resistance of LXR-deficient mice to Leishmania infection.
We hypothesized that if activated LXR complexes are functioning to downregulate NO production following induction with IFN-γ/Leishmania, then the addition of LXR-activating ligands to wild-type cells might result in decreased levels of NO in vitro. Since wild-type BMDM secrete measurable amounts of NO in response to IFN-γ/Leishmania, we tested the ability of LXR agonists to inhibit this production. We pre-exposed wild-type BMDM to the synthetic LXR/RXR ligands GW3965 (GW) and LG268 (LG) with or without IFN-γ overnight prior to incubation of cells with Leishmania for one hour. The NO produced in wild-type cells in response to IFN-γ/Leishmania was significantly reduced (by 75%) in cells pretreated with LXR/RXR ligands (Fig. 4). Pretreatment with L-NNMA, a broad NOS inhibitor, had similar effects on NO production as GW/LG pretreatment.
We assessed the effect of Leishmania infection on mouse tissue gene expression in vivo, and in macrophages in vitro. Wild-type and LXRα-KO mice were infected with L. chagasi/infantum 24 or 72 hours before sacrificing, and RNA was harvested from livers and spleens. LXRα expression in either tissue did not change upon infection of wild-type mice (as expected, LXRα was not expressed in LXRα-KO mice), and LXRβ was not induced by Leishmania infection in either wild-type or LXRα-KO mice (Fig. 5A). In accordance with previously published microarray data in human and mouse macrophages [14], [28], expression of both LXRα and LXRβ was unaltered by Leishmania infection of IFN-γ-pretreated BMDM (Fig. 5B). These results suggest that the Leishmania resistance observed in LXR-deficient mice cannot be attributed to a lack of Leishmania-induced upregulation of the LXR genes. We examined a prototypical LXR-regulated gene, apoE, which contains an LXR responsive element in its promoter, and is known to be upregulated upon LXR activation [29]. Baseline expression levels of apoE, as well as expression following Leishmania infection (which was decreased), were similar in wild-type and LXR-DKO macrophages (Fig. 5B). We also examined macrophage expression levels of several genes known to be involved in resistance to visceral Leishmania, including iNOS, IL-1β, IL-6, TNF-α, and IL-10. Wild-type and DKO BMDM expressed similar levels of iNOS RNA in response to IFN-γ and Leishmania exposure (Fig. 5B), suggesting that the observed differences in NO production by wild-type compared to LXR-deficient macrophages (Fig. 3) are not due to changes in iNOS transcription. Of the cytokine genes examined, IL-1β was significantly induced by IFN-γ and Leishmania infection in LXR-deficient macrophages, but not in wild-type macrophages. No other genes examined were differentially regulated in the absence of LXR, including IL-12 and IFN-γ, which were detected only at very low levels by RT-PCR (data not shown). These results suggest that IL-1β is one possible cytokine that might contribute to LXR control of Leishmania resistance during infection.
The LXR nuclear receptors regulate both lipid homeostasis and inflammation in mammalian cells. LXR has emerged as a potential drug target for chronic metabolic and inflammatory diseases such as atherosclerosis. Macrophages are central in these processes, and the expression and activation of both LXR isoforms in macrophages have been well established. Since macrophages are innate immune effectors, we and others have begun to investigate the relationship between these metabolic regulatory genes and the ability to mount responses against microbial pathogens. We demonstrate here an unexpected enhanced resistance of LXR-deficient mice to infection with the intracellular protozoan Leishmania chagasi/infantum, a parasite that resides and replicates predominantly in macrophage phagolysosomes. The in vivo resistance in mice lacking LXRα was evident in the liver as early as three days following inoculation, and continued over one month. In accordance with this resistance phenotype, we report that LXR-deficient macrophages pre-exposed to IFN-γ and infected with L. chagasi/infantum in vitro had fewer intracellular parasites after 48 hours and produced increased levels of NO compared to their wild-type counterparts. In addition, IFN-γ/Leishmania-induced NO production by wild-type macrophages could be inhibited by exposure to LXR ligands. Whereas the in vivo resistance phenotype was apparent in both LXRα−/− and LXR-DKO (LXRαβ −/−) mice, the in vitro phenotypes were only consistently observed in LXR-DKO macrophages, which may reflect different compensatory abilities of LXRβ in vitro versus in vivo. Overall, we hypothesize that the significantly decreased parasite burden in LXR-deficient mice might be broadly attributed to 1) changes in host cell susceptibility to parasite invasion, establishment, and/or persistence, and/or 2) differences in functional host immune responses following parasite establishment within macrophages.
The enhanced resistance of both LXRα−/− and DKO mice to L. chagasi/infantum was surprising in light of several recent findings documenting these genotypes' decreased immunity to various infections. Our group previously reported that LXR-deficient mice are acutely susceptible to infection with the bacterium Listeria monocytogenes [9]. Listeria, like Leishmania, resides intracellularly in mammalian host cells (including macrophages), and requires TH1-mediated immune responses for clearance. The absence of LXRα leads to increased apoptosis of macrophages in response to Listeria infection, a phenotype attributable in part to decreased expression of an LXR-regulated, anti-apoptotic factor called SPα [9]. We did not observe increased apoptosis levels in LXR-deficient BMDM compared to wild-type macrophages following Leishmania infection in vitro. One major difference between the two organisms is that Listeria, unlike Leishmania, escape the phagolysosomal compartment and enter the host cell cytoplasm. Both Listeria and Shigella flexneri, a gram-negative intracellular bacterium that also has a cytoplasmic niche, strongly induce LXRα expression in murine BMDM, in contrast to extracellular bacteria [9]. Addition of the NOD2 ligand muramyl dipeptide (MDP) to BMDM also induces expression of LXRα mRNA, suggesting that specific intracellular signaling pathways initiated by NOD-like receptors (NLRs) in the cytoplasm trigger upregulation of LXRα. These pathways are likely not directly activated by Leishmania in macrophages, as the parasite normally remains sequestered in phagosomal compartments. Our results (Figure 5), and microarray data of gene expression patterns in both human and mouse macrophages, suggest that Leishmania infection does not affect LXR expression levels or LXR-mediated gene expression [14], [28].
LXR deficiency in mice also results in increased susceptibility to the intracellular pathogen Mycobacterium tuberculosis [10]. Mice lacking both isoforms of LXR had decreased clearance of mycobacterial load from the lungs, spleen and liver over several weeks of infection. This defect correlated with a reduction of lung-infiltrating neutrophils in mice lacking LXRα, along with reduced neutrophil-attracting chemokines. Although typical pro-inflammatory/microbicidal mediators TNF-α and iNOS remained at basal levels following infection in both wt and KO mice, IL-12p40 levels were reduced in LXRα KO mice. Interestingly, a significant increase of arginase mRNA was observed in DKO mice, although it was unclear if this was specific for either isoform, ArgI or ArgII.
The metabolic state of macrophages can directly influence intracellular growth and/or survival of intracellular amastigotes. Arginine metabolism is regulated by LXR and has previously been implicated in Leishmania pathogenesis. The ArgII gene is induced greater than 50-fold upon ligand activation of LXRα, and its arginase enzyme product functions to hydrolyze the amino acid arginine to urea and ornithine [30]. Promotion of this reaction might influence Leishmania survival within host macrophages in two distinct ways. First, the products of the arginase pathway are precursors for polyamine synthesis that positively affect parasite growth and metabolism [31], [32]. Removal of LXRα might lead to a reduction in these small metabolites that are normally utilized by amastigotes, leading to inhibited growth and increased susceptibility to oxidative stress. Secondly, nitric oxide synthases (NOS) compete with arginase enzymes for the utilization of L-arginine as a substrate, alternatively converting it into NO and L-citrulline. Previous experiments in our group have demonstrated that overexpression of ArgII in RAW macrophages in vitro leads to an inhibition of nitrite production in response to inflammatory stimuli including LPS, poly I:C, PGN, LTA and CpG [30]. Decreased ArgII expression in LXR-deficient cells, conversely, might be predicted to lead to an increase in NO production, due to higher levels of arginine metabolite. We were unable to detect significant levels of ArgII in wild-type macrophages upon Leishmania infection, or in mice infected with Leishmania (data not shown), consistent with the absence of overt upregulation of LXR by infection. However, the exact role of arginase activity, and its impact on parasite survival in LXR-deficient animals, will require further investigation to define.
Specific genes known to be suppressed by activated LXR include pro-inflammatory cytokines induced by LPS and bacterial stimuli, including iNOS, cyclooxygenase (COX)-2, and IL-6 [6]. Early innate cytokine responses by macrophages and neutrophils determine in part whether Leishmania infections resolve, with Type I immune responses generally correlating with control of infections [25]. We demonstrate here that mRNA levels of IL-1β were higher in LXR-DKO macrophages than in wild-type macrophages following IFN-γ pretreatment and L. chagasi infection in vitro. However, we did not observe differences in macrophage expression levels of several other cytokines potentially involved in Leishmania resistance (including IL-6, TNF-α, and IL-10), nor in levels of iNOS. Enhanced expression of the pro-inflammatory cytokine IL-1β by macrophages might partially explain the early resistance of LXR-deficient mice, which displayed lower liver parasite load as early as three days following challenge. It is clear, however, that other cell types in addition to macrophages contribute to the global leishmanial resistance phenotype in vivo. Bensinger et. al. recently demonstrated a link between LXR signaling and lymphocyte proliferation, whereby loss of LXRβ led to enhanced T cell responses to antigenic challenge [33]. The effects of LXR deficiency on other cells, including TH1 CD4+ T cells, in terms of Leishmania immunity and cytokine expression, are currently being studied in our model. Additionally, LXR signaling was also recently implicated in the clearance of apoptotic cells by macrophages, and the maintenance of immune tolerance [34]. The relevance of this finding to the observed pathogen immune phenotypes remains to be defined.
The LXR-regulated network of genes affects many aspects of metabolism and inflammation, and has been extensively studied with hopes of positively influencing metabolic disorders such as hyperlipidemia and atherosclerosis. Mechanisms by which LXR pathways impact immune function will be important to consider when targeting these molecules for activation in cardiovascular disease. With regards to leishmaniasis, understanding how pathogenesis is impacted by LXR signaling may lead to novel interventions in this disease process as well. Since LXRs down-modulate inflammatory pathways, blocking LXR activation or expression might augment key cell-mediated immune responses necessary for Leishmania killing. The observation, however, that LXR-deficient mice, while more resistant to visceral Leishmania infection, are more susceptible to bacterial infections caused by Listeria monocytogenes and Mycobacterium tuberculosis, suggests a complex immune phenotype that is pathogen-dependent. Targeting LXR or LXR-mediated genes for antagonism may nonetheless represent a feasible strategy to modulate immunity against certain types of intracellular pathogens like Leishmania. Continued study into the role of LXR in resistance to specific microorganisms may also lead to enhanced understanding of complex host-pathogen relationships.
|
10.1371/journal.ppat.1003072 | Clostridium difficile Toxin B Causes Epithelial Cell Necrosis through an Autoprocessing-Independent Mechanism | Clostridium difficile is the most common cause of antibiotic-associated nosocomial infection in the United States. C. difficile secretes two homologous toxins, TcdA and TcdB, which are responsible for the symptoms of C. difficile associated disease. The mechanism of toxin action includes an autoprocessing event where a cysteine protease domain (CPD) releases a glucosyltransferase domain (GTD) into the cytosol. The GTD acts to modify and inactivate Rho-family GTPases. The presumed importance of autoprocessing in toxicity, and the apparent specificity of the CPD active site make it, potentially, an attractive target for small molecule drug discovery. In the course of exploring this potential, we have discovered that both wild-type TcdB and TcdB mutants with impaired autoprocessing or glucosyltransferase activities are able to induce rapid, necrotic cell death in HeLa and Caco-2 epithelial cell lines. The concentrations required to induce this phenotype correlate with pathology in a porcine colonic explant model of epithelial damage. We conclude that autoprocessing and GTD release is not required for epithelial cell necrosis and that targeting the autoprocessing activity of TcdB for the development of novel therapeutics will not prevent the colonic tissue damage that occurs in C. difficile – associated disease.
| Clostridium difficile is an anaerobic spore-forming bacterium that infects the human colon and causes diarrhea, pseudomembranous colitis, and toxic megacolon. Most people that develop disease symptoms have undergone antibiotic treatment, which alters the normal gut flora and allows C. difficile to flourish. C. difficile secretes two toxins, TcdA and TcdB, that are responsible for the fluid secretion, inflammation, and colonic tissue damage associated with disease. The emergence of hypervirulent strains of C. difficile that are linked to increased morbidity and mortality highlights the need for new therapeutic strategies. One strategy is to inhibit the function of the toxins, thereby decreasing damage to the colon while the patient clears the infection with antibiotics. Toxin function is thought to depend on an autoprocessing event that releases a catalytic ‘effector’ portion of the toxin into the host cell. In the course of trying to identify small molecules that would inhibit such a function, we found that TcdB induces a rapid necrosis in epithelial cells that is not dependent on autoprocessing. The physiological relevance of this observation is confirmed in colonic explants and suggests that inhibiting TcdB autoprocessing will not prevent the colonic tissue damage observed in C. difficile associated diseases.
| Clostridium difficile is a gram-positive, spore-forming anaerobe that infects the colon and causes a range of gastrointestinal disorders including diarrhea, pseudomembranous colitis, and toxic megacolon [1], [2]. This is a major healthcare concern as the number and severity of C. difficile-associated disease (CDAD) cases have increased dramatically in recent years [3]. Two large toxins, TcdA and TcdB (308 kDa and 270 kDa, respectively), are recognized as the main virulence factors of C. difficile [4], [5]. The C-terminal portion of these toxins is responsible for delivering an N-terminal glucosyltransferase domain (GTD) into the host cell [6], [7]. The GTD inactivates Rho family GTPases including Rho, Rac1, and Cdc42 [8], [9].
While there are numerous studies that report the effects of toxin-mediated glucosylation in cells, a consensus as to the conclusion of these reports, taken together, has been difficult due to differences in cell types, toxin concentrations, and assay methods. In addition, it appears that TcdA and TcdB can elicit different effects under similar conditions [10], [11]. In all reports, both toxins can induce a cytopathic effect characterized by cell rounding. In many reports, these cells go on to die by apoptotic mechanisms, but the time course can be up to 48 hours [12]–[19]. It has been noted, however, that apoptosis cannot be detected in cells treated with higher concentrations of TcdB [20]. In at least one study, the absence of apoptosis in cells treated with TcdB has led to suggestions of a necrotic mechanism of cell death [21].
The mechanism of GTD delivery for TcdA and TcdB involves binding a host cell receptor [22], [23], uptake by endocytosis [24], [25], pH-dependent pore formation [26]–[28], translocation across the endosomal membrane, host-factor dependent autoprocessing [29], and release of the GTD into the host cell cytosol [30]. Release is thought to allow the GTD access to the Rho-family GTPases tethered to the plasma membrane surface. An N-terminal sub-domain within the GTD is thought to serve as a membrane localization domain [31].
The autoprocessing function of the toxins is mediated by a cysteine protease domain (CPD) that follows the N-terminal GTD [32]. Inositolphosphates, predominantly inositol hexakisphosphate (InsP6), have been identified as the host factors responsible for inducing autoprocessing [29]. The InsP6-bound structures of the TcdA and TcdB CPDs reveal a positively charged InsP6-binding pocket that is distinct from the catalytic active site [33], [34]. InsP6 binding is thought to trigger conformational changes that permit the formation of the substrate-binding pocket and alignment of the catalytic residues [35]. The three catalytic amino acids Asp587, His653, and Cys698 (TcdB sequence) and the P1 substrate recognition site, Leu543, have been shown to be important for in vitro processing activity by genetic mutation [32]. Mutation and chemical modification of these residues has also been shown to prevent activity in various cell based assays [29], [32], [34], [36], [37]. For this reason, TcdB autoprocessing activity and GTD release have been considered important in the toxin mechanism, an idea which suggests that the CPD could serve as a useful target for novel small molecule inhibitor discovery.
The objective at the outset of this project was to conduct a high-throughput screen for small molecules that inhibit TcdB-mediated cell death. Our first step toward exploring this potential was to evaluate apoptotic and necrotic markers as cell death indicators. In observing a necrotic response to TcdB, we decided to specifically focus on the question of whether the assay would be able to detect inhibition of TcdB autoprocessing. We constructed mutant TcdB proteins with deficiencies in either the autoprocessing or glucosyltransferase activities and tested their effects on cell viability. Our unexpected observation that the mutants killed cells rapidly and at concentrations comparable to wild-type led us to investigate the role of autoprocessing and GTD release in cell death and cell rounding in greater detail. In this report, we provide evidence that epithelial cells and porcine colonic tissue challenged with TcdB undergo a rapid, necrotic cell death that is not dependent on autoprocessing and GTD release.
The objective at the outset of this project was to conduct a high-throughput screen for small molecules that inhibit TcdB-mediated cell death. Our first goal was, therefore, to establish conditions for an assay that was sensitive and homogeneous. HeLa cells were seeded into 384 well plates and treated with TcdB at multiple concentrations for varying lengths of time. Cells were then simultaneously assayed for caspase-3/7 activation and ATP levels using fluorescent and luminescent indicators, respectively. At all concentrations and time points tested, TcdB failed to activate caspase-3 and -7, central regulators in apoptotic cell death (Figure 1A). Conversely, staurosporine, a known inducer of apoptosis, triggered significant caspase-3/7 activation at a 5 hour time point. Since the result appeared to be in conflict with a previous report showing that TcdB-treatment of HeLa cells induced an increased rate of caspase-3 activity [18], we performed additional experiments using lower toxin concentrations, a 48 hour time point, and TcdA. We did not observe caspase-3/7 activation in any of the cells treated with TcdB and only saw TcdA-induced caspase-3/7 activation when the toxin was applied at concentrations of 100 nM (Figure S1A). While our initial experiments were performed with TcdB purified from a recombinant Bacillus megaterium expression system, we did not observe caspase-3/7 activation when we tested TcdB purified from C. difficile culture supernatants (Figure S1B).
Despite the lack of caspase-3/7 activation, the TcdB treatments had a significant impact on cellular ATP levels (Figure 1B). Decreases in ATP were observed after only 2.5 hours in cells treated with 1, 10, and 100 nM TcdB suggesting that these cells were no longer viable. The effect is specific to TcdB, as TcdA only impacted the viability at concentrations of 100 nM at 24 hours (Figure S2A). While lower concentrations of TcdB can induce cell death after a 48 hour application, the effect does not appear to be dose dependent at the 48 hour time point (Figure S2A).
In an attempt to correlate the viability indicators with cytopathic events, mock and TcdB treated cells were visualized by light microscopy. At concentrations of 10 pM, a characteristic cytopathic (cell rounding) effect was observed. In contrast, cells treated with 10 nM TcdB for 2.5 hours had completely lost their membrane integrity (Figure 1C). The rapid loss of ATP and membrane integrity suggested that cells treated with nM concentrations of TcdB were dying by necrosis. To further test this hypothesis, we assessed the effect of TcdB on LDH and HMGB1 release. LDH release was apparent 2.5 hours after intoxication and at an increased level after 8 hours (Figure 1D). Similar values for LDH release are observed when the cells are treated with TcdB from C. difficile supernatants (Figure S2B). Notably, LDH release is only detectable at toxin concentrations above 0.1 nM, consistent with the cell death data obtained with an ATP indicator (Figure 1B). HMGB1 is a nuclear protein that is released into the cytoplasm when the cell is dying by necrosis. We found that at 10 nM TcdB, HMGB1 was released into the cytoplasm after 1 hour (Figure 1E). As a result of these studies, CellTiterGlo, the luminescent indicator of cellular ATP levels, was deemed the best indicator of cell viability for high throughput screening. The rapid loss of ATP and membrane integrity, the release of LDH and HMGB1, and the lack of caspase-3/7 activation all suggest necrosis is the mechanism of TcdB-mediated death in HeLa cells.
We next generated autoprocessing-deficient mutants that could be used as negative controls in a secondary assay that would allow us to select for molecules that inhibit the autoprocessing activity of the toxin. Single amino acid point mutations were made in the TcdB autoprocessing active site (C698S, C698A, H653A, and D587N) and the cleavage site (L543A). Proteins were expressed in the B. megaterium expression system and purified to homogeneity. All mutants were tested for their in vitro autoprocessing activity (Figure 2A). TcdB autoprocessing can be induced with the addition of 1 uM InsP6, and the amount of processing increases as the concentration of InsP6 increases. At all concentrations of InsP6, TcdB C698S, TcdB C698A, and TcdB H653A were completely inactive in autoprocessing, as detected by Coomassie-stained SDS PAGE (Figure 2A) and densitometry (Figure 2B). TcdB D587N and TcdB L543A had residual cleavage activity, but were significantly cleavage-impaired. Cleavage of D587N was not induced until 100 uM InsP6 was added, and the amount of processed toxin was reduced.
We next wanted to confirm that the mutants were also defective for autoprocessing in the context of the cell. HeLa cells were treated with wild-type TcdB or autoprocessing deficient TcdB mutants for 50 min, lysed, and probed by Western blot using an anti-TcdBGTD antibody. Free GTD was detected in cells treated with wild-type TcdB but was not detected in cells intoxicated with TcdB mutants (Figure 2C). The same lysates were probed with an antibody specific for unglucosylated Rac1. Rac1 is glucosylated even when the cells have been treated with autoprocessing mutants. These data suggest that in cells treated with TcdB autoprocessing mutants, the GTDs are being translocated into the cytosol, but they remain tethered to the endosome where glucosylation of Rac1 can still occur.
To test the hypothesis that small molecule inhibitors of TcdB autoprocessing could be detected in a cell based screen, we assessed cell viability in response to three of the TcdB autoprocessing mutants: TcdB C698S, TcdB C698A, and TcdB L543A. HeLa cells were treated for 2.5 hours with multiple concentrations of TcdB and the TcdB mutants, and viability was assessed using CellTiterGlo. Unexpectedly, the autoprocessing deficient mutants were found to induce cell death at concentrations comparable to TcdB (Figure 3A). To test whether this response was unique to HeLa cells, we performed similar experiments with Caco2 cells, an epithelial cell line derived from human colon. As with the HeLa cells, wild-type and autoprocessing deficient TcdB mutants induced a decrease in cellular ATP at similar concentrations in Caco2 cells (Figure 3B). Caspase-3/7 activation was not detected in HeLa cells treated for 25 hours with autoprocessing deficient TcdB mutants (Figure 3C), and the amount of LDH released in HeLa cells treated with wild-type TcdB and the TcdB C698S, C698A, and L543A autoprocessing mutants was equivalent (Figure 3D). Finally, HeLa cells were treated with 10 nM wild-type and mutant TcdB proteins in the presence of a live/dead cell indicator and imaged every 10 minutes over a 2 hour time course. A representative movie of what we observed is included in the supplemental material (Video S1). The percentage of dead cells quantified over six fields suggests that the kinetics of cell death are identical for the four proteins (Figure S3). Collectively, these data suggest autoprocessing is not required for TcdB-mediated necrosis in epithelial cells.
The idea that TcdB-induced necrosis did not require autoproteolytic release of the GTD suggested that the TcdB glucosyltransferase activity would also not be required for cytotoxicity. To test this hypothesis, single amino acid point mutations were made in the glucosyltransferase active site (D270N, D270A, Y284A, W520A, and N384A) based on the crystal structure of the TcdB GTD bound to UDP-glucose [38]. Proteins were expressed in the B. megaterium expression system and purified to homogeneity. All mutants were tested for their in vitro glucosyltransferase activity in the presence of purified Rac1 and UDP[14C]glucose, and all were impaired relative to wild-type (Figure 4A). Of the five mutants, the TcdB D270N mutant showed the greatest defect in in vitro glucosyltransfer, with residual activity only evident in the highest concentrations of enzyme and substrate (Figure 4B). Even with differences in the amount of residual activity, all five mutants were defective in the modification of Rac1 in cells (Figure 4C). Furthermore, all 5 mutants were capable of inducing a cytotoxic effect similar to that of wild-type TcdB when applied to HeLa cells (Figure 4D) and Caco-2 cells (data not shown). We interpret these data to mean that the TcdB cytotoxic effect does not require the glucosyltransferase activity of the toxin.
The observation that TcdB autoprocessing mutants were able to glucosylate Rac1 in cells (Figure 2C) suggested that they would induce rearrangements in the actin cytoskeleton that result in the cytopathic ‘rounding’ phenotype. To investigate this, HeLa cells were treated with multiple concentrations of wild-type and mutant TcdB proteins and imaged every 10 minutes over a 2 hour time course. The percentage of round cells was quantified over six fields for each concentration and time point. At a 10 pM concentration, we observed similar rounding kinetics for TcdB and the three TcdB autoprocessing-deficient mutants (Figure 5A). Differences in the kinetics of rounding began to appear at a concentration of 100 fM (Figure 5B) but were not fully evident until the concentration of toxins was dropped to 1 fM (Figure 5C). The full dataset collected at concentrations spanning 8 orders of magnitude and a movie of what we observed with 10 fM wild-type TcdB is included in the supplemental material (Figure S4 and Video S2). While not required for cytotoxicity, autoprocessing and GTD release appear to be important for cytopathic processes that occur at very low concentrations. In HeLa cells, we see that at concentrations where cytopathic effects can be observed (1 fM–10 pM, Figure 5), the cells are not dead (Figure 3A). These data provide a clear distinction between the cytotoxic and cytopathic effects induced by TcdB.
The distinction between cytopathic and cytotoxic events in cell culture led us to question if either event might correlate with disease pathology. Since the formation of necrotic lesions in the colon is a hallmark of CDAD pathology, we sought to determine the concentration of toxin required to induce these effects and whether autoprocessing was required. Porcine colonic explants were incubated with multiple concentrations of toxin for 5 hours. The tissue was fixed with formalin, embedded in paraffin, and sections were stained with H&E (Figure 6A). The slides were scored in a blinded fashion and given a score (0–3) to reflect the level of epithelial damage (Figure 6B). Damage ranged from a mostly intact surface epithelium to mucosal loss of 50% or greater in the depth of colonic crypts. The scores indicated a loss of surface epithelium in tissue treated for 5 hours with 10 nM TcdB and TcdB C698A. There was little damage in tissues treated with a buffer control or in tissues treated with wild-type TcdB and TcdB C698A at a concentration of 10 pM. Statistical analysis by two-way ANOVA revealed a significant difference in scores for tissues treated with the toxins over the range of concentrations (p<0.001), while there was no statistical difference between tissues treated with wild-type TcdB and TcdB C698A. A subsequent Bonferroni's test revealed that scores given to tissue treated with 10 nM TcdB and 10 nM TcdB C698A were significantly different from scores given to tissue treated with 10 pM TcdB and 10 pM TcdB C698A (p<0.001). The tissues were stained with an anti-pan cytokeratin antibody to confirm the keratin positive cells at the luminal surface of the colon were disrupted (Figure 6C) and an anti-activated caspase-3 antibody to confirm that the toxin treatment did not induce an apoptotic response (Figure 6D). The data reveal a correlation between the concentration of toxin required to kill epithelial cells in culture with the concentration required to disrupt epithelial integrity in colonic tissue and indicate that autoprocessing is not required for tissue damage.
TcdB is a multi-functional protein with a central role in CDAD pathogenesis. Our goal at the outset of this study was to conduct a screen for small molecule inhibitors that could aid in the dissection of the TcdB mechanism and the generation of new leads for therapeutic intervention. Our strategy was to combine a cell-based phenotypic screen with target-specific secondary assays. In the course of setting up our screening assays, we made two unexpected observations that warranted further investigation.
First, in contrast to a previous report [18], TcdB did not trigger the induction of apoptosis in cultured epithelial cells as measured by caspase-3/7 activation (Figure 1A, S1). Since there was an overlap in the cells, concentration of toxin, and timepoints used for analysis, we are left to speculate that the difference stems from advances in the detection reagent. The newer reagent for detecting caspase-3/7 activation allows one to directly quantitate the relative quantity of activated caspase-3/7 as opposed to the overall rate of caspase activity.
While TcdB-treatment did not induce the activation of caspase-3/7, the rapid ATP depletion observed in both HeLa (Figure 1B, 3A, S2A) and Caco2 (Figure 3B) cells suggested that the mechanism of TcdB-induced cell death was likely necrosis. The observed loss of membrane integrity (Figure 1C), rapid LDH (Figure 1D, S2B), and HMGB1 release (Figure 1E) support this conclusion.
We next questioned whether a cell-based assay for small molecule inhibitors of TcdB-induced necrosis would allow us to detect molecules that interfered with autoprocessing. We were particularly interested in targeting the autoprocessing activity of the toxin since, in theory, one could identify molecules that either activate (e.g. InsP6) or inhibit the function of the cysteine protease domain. We generated five TcdB point mutants in which key residues of the cysteine protease active site or cleavage site were mutated. Three of these mutations, C698S, C698A, and L543A, rendered TcdB non-functional for InsP6-induced autoprocessing in an in vitro assay, even when InsP6 was added at a 1 mM concentration (Figure 2A, 2B). The mutants were also defective for autoprocessing in the context of cells since free GTD could be detected in cells treated with wild-type TcdB but not in cells treated with the autoprocessing mutants (Figure 2C). While we cannot rule out the possibility of an alternate cleavage mechanism that results in a quantity of free GTD that is less than the detection limit of the assay, the free GTD concentration generated from such a mechanism would be too small to account for the identical cytotoxicity profiles observed in Figures 3A and 3B.
The unexpected observation that cytotoxicity does not require autoproteolytic release of the GTD led us to directly test whether the glucosyltransferase activity of the toxin was required (Figure 4). We generated five single amino acid point mutants of TcdB that differed in their residual glucosyltransferase activities in vitro (Figure 4A, 4B). Despite the different enzyme activity levels, all were significantly impaired relative to wild-type TcdB in their capacity to modify Rac1 in cells (Figure 4C), and all were comparable to wild-type TcdB in their cytotoxic effects (Figure 4D). These data are consistent with the observation that autoprocessing is not required and suggest that the cytotoxic response to TcdB is triggered by an event upstream of GTD release.
While not required for cytotoxicity, autoprocessing and GTD release are important for cytopathic processes that occur at low concentrations [29], [32], [34], [36], [37]. Our data are consistent with these previous reports and indicate differences in rounding kinetics emerging at concentrations of 100 fM (Figure 5C and SF4). While our Western experiment indicated TcdB autoprocessing mutants were still able to modify Rac1 in cells (Figure 2C), a similar observation has been made for a non-cleavable form of TcdA and is thought to reflect continuous vesicle trafficking and an exchange of membranous compartments that allow the uncleaved toxin to come into contact with the membrane-bound GTPases [39]. This capacity to modify Rac1 while still tethered to the endosomal membrane presumably accounts for the similar rounding kinetics that we observed when the TcdB autoprocessing mutants were applied to HeLa cells at concentrations of 1 pM and higher (Figure 5A, S4).
The concentrations of TcdB needed to induce cytopathic effects (≤1 fM, Figure S4) are significantly lower than what is required to induce the cytotoxic effect (1 nM, Figure 3). At a concentration of 10 pM TcdB, the cells are clearly round (Figure 5A) but not dead (Figure 3). The distinction between cytopathic and cytotoxic events in cell culture raises the question of whether either process correlates with mechanisms of pathology observed in the host. To address this question, we decided to test what concentration of toxin was required to induce epithelial cell damage in colonic tissue explants. Visual assessment of H&E stained colonic tissue integrity in a blinded fashion indicated damage with treatments of 10 nM TcdB but not with 10 pM TcdB (Figure 6). Similar observations were made with the TcdB C698A mutant suggesting that the damage that occurs to colonic tissue in response to TcdB does not depend on the autoprocessing activity. Pan-cytokeratin staining confirmed that the cells on the luminal surface of the tissue remained intact in the presence of 10 pM TcdB or TcdB C698A but were being disrupted in samples treated with 10 nM TcdB, 10 nM TcdB C698A, or 100 uM staurosporine. The staurosporine control revealed strong caspase-3 activation into the crypts (Figure 6D). The untreated control tissue demonstrated a low level of caspase-3 activation in the cells on the luminal surface and strong activation in single cells coming off the surface of the tissue. Tissues treated with 10 pM TcdB and TcdB C698A showed caspase-3 activation levels similar to those of the untreated tissue. Tissue treated with 10 nM TcdB or TcdB C698A demonstrated even lower levels of caspase-3 activation, presumably because the cells on the luminal surface have been shed. Unlike the untreated, staurosporine-treated, and 10 pM TcdB-treated tissues, caspase-3 activation was generally not observed in the cells that were in the process of being shed in tissues treated with 10 nM TcdB or TcdB C698A (Figure 6D). This suggests that tissue damage is not only independent of autoprocessing activity, but also not likely due to apoptosis.
The phenotypic differences with concentration led us to wonder what concentration of toxin is present in the colons of individuals experiencing the symptoms of CDAD. We found only one published report, where TcdB was quantitated using a real-time cell analysis system [40]. In this report, the TcdB concentrations in stool samples from 10 patients experiencing mild to severe symptoms of CDAD ranged from 4.9 pM to 413 pM with a mean concentration of 146 pM. Presumably, the concentration of TcdB would be much higher at the colonic epithelium prior to dilution by diarrhea. Of note, the average TcdB concentration in samples from 9 individuals who were not experiencing CDAD symptoms was 1 pM, with a range of 0.1 pM to 3.3 pM. This analysis suggests that the cytotoxic effects observed in cells and tissues treated with 1 to 10 nM TcdB are better correlated with pathology than the cytopathic effects that are induced at 1 fM concentrations.
Our data suggest that inhibiting TcdB autoprocessing will not prevent the colonic tissue damage observed in C. difficile associated diseases. However, while the colonic epithelium is the primary barrier separating C. difficile from the host, it is possible that the autoprocessing function of TcdB is important in another setting relevant to pathogenesis. For example, the colonic explant model used in this study does not account for the impact of the toxins on inflammation or the potential impact of an anaerobic environment. Evaluating the effect of autoprocessing- and glucosyltransferase-deficient toxins in an animal model of C. difficile infection therefore represents a priority for future studies. In addition, it will be important to define the mechanism of TcdB-mediated necrosis in cells and tissue. Relevant comparisons may come from the study of other toxins. For example, the Bordetella pertussis adenylate cyclase (AC) toxin is known to have multiple mechanisms that contribute to cytotoxicity [41]. Identifying the autoprocessing- and glucosyltransferase-dependent and –independent aspects of TcdB-mediated pathology represents an exciting path for future study.
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. Animal husbandry and experimental procedures related to the porcine colonic explants were performed in accordance with the Vanderbilt University Institutional Animal Care and Use Committee (IACUC) policy. Discarded colon tissues were obtained from pigs following euthanization at the end of IACUC-approved animal use protocols. Animal husbandry and experimental procedures related to the generation of the anti-TcdBGTD monoclonal antibody were performed in accordance with the Washington University Animal Studies Committee policy, approval number 20100113.
Single amino acid point mutations were made in the TcdB autoprocessing active site (C698S, C698A, H653A, and D587N), the cleavage site (L543A), and the glucosyltransferase domain (D270N, D270A, Y284A, N384A, and W520A) using the QuickChange mutagenesis protocol (Stratagene). The template for mutagenesis and clone for the production of wild-type TcdB was a B. megaterium expression vector encoding the strain 10643 of TcdB [42]. A similar clone was used for expression of recombinant TcdA [42]. Plasmids for expressing TcdA, TcdB, and TcdB point mutants were transformed into B. megaterium according to the manufacturer's instructions (MoBiTec, Göttingen, Germany). 1 L of LB was inoculated with 35 mL overnight culture and 10 mg/L tetracycline and grown at 37°C and 230 rpm. At an OD600 of 0.3, expression was induced with 5 g of D-xylose. Cells were harvested after 4 h by centrifugation and resuspended in 20 mM Tris, pH 8.0, 500 mM NaCl and protease inhibitors. Cells were lysed by French press, and lysates were centrifuged at 48,000 g for 25 min. The proteins were purified by Ni-affinity chromatography, Q-sepharose anion exchange chromatography, and gel filtration chromatography in 20 mM HEPES, pH 6.9, 50 mM NaCl.
Proteins were expressed and purified as previously described [1].
HeLa and Caco2 cells (cultured in DMEM, 10% FBS, 5% CO2 and MEM, 10% FBS, 5% CO2, respectively) were seeded in a black 384-well plate at a concentration of 3,000 or 1,000 cells/well, respectively. HeLa cells were intoxicated the next day, and Caco2 cells were intoxicated 36 h later. After intoxication, the cells were incubated at 37°C, 5% CO2 for either 2.5 h (HeLa) or 18 h (Caco2). The amount of ATP (cell viability) was assessed with a luminescence-based indicator, CellTiterGlo (Promega). LDH release was assessed with a luminescence-based indicator, CytoToxGlo (Promega). Caspase-3/7 activation was determined using a fluorescent indicator, Apo-One (Promega). Staurosporine (Sigma, 1 mM) was used as a positive control for caspase-3/7 activation. Plates were read in a Biotek Synergy 4 plate reader.
HeLa cells were seeded into a tissue culture treated chamber slide at 2×104 cells per well and incubated overnight. Cells were synchronized at 4°C and intoxicated with 10 nM TcdB for 1 h. Cells were then shifted to 37°C for 1 h. Media was removed from the cells, and the cells were washed with PBS. They were fixed with 4% paraformaldehyde at room temperature for 10 minutes and quenched with 1 mM glycine. Cells were permeated with 0.2% Triton X-100 in PBS for 5 minutes, washed in PBS, and blocked for 30 minutes in PBS, 2% BSA, 0.1% Tween 20. Cells were stained with a monoclonal antibody against HMGBI (Abcam, ab77302), and an Alexa Fluor 488 anti-mouse antibody (Invitrogen, A11001). Cells were visualized with an LSM 510 Confocal microscope.
1 uL InsP6 stock solution (100×) or buffer was added to 200 nM TcdB or TcdB autoprocessing mutant and incubated for 2 h at 37°C. The reactions were stopped with the addition of loading buffer and boiling and analyzed by Coomassie stained SDS PAGE.
Genomic DNA of C. difficile clinical isolate 630 was obtained from American Type Culture Collection, and the region encoding residues 1 to 549 of TcdB, which is known to encode the substrate binding and enzymatic domains of the toxin, was amplified in frame with a carboxy-terminal (His)6-tag using upstream primer:5′- CCGGATGTACAGTTGAGGGGGTAAAATGAGTTTAGTTAATAGAAAACAGTTAG -3′ and downstream primer 5′- GGTCCTCAATGATGGTGATGGTGATGAAGATTATCATCTTCACCAAGAGAACC -3′. The resulting product was cloned into plasmid pcDNA3.1 (Invitrogen, Carlsbad CA) and sequenced to ensure fidelity of the amplified product. The gene was then released with restriction enzymes BsrG1 and AgeI and cloned into similarly digested vector pHIS1525 (MoBiTec), placing the gene under control of a xylose-inducible promoter. Recombinant protein was expressed in B. megaterium and purified by sequential nickel affinity and gel filtration chromatography. Two mice were immunized bi-weekly by intraperitoneal injection with 100 µg purified TcdB-GTD. Three days after the third vaccination, splenocytes were harvested and fused to P3X63Ag8.6.5.3 myeloma cells using polyethylene glycol 1500 [43]. Hybridomas producing anti-TcdB-GTD MAbs were identified by ELISA, subcloned by limiting dilution, and purified by protein G immunoaffinity chromatography.
HeLa cells were synchronized by cooling to 4°C and then intoxicated with 10 nM TcdB, autoprocessing mutant, or buffer. The cells were returned to 4°C for 1 h, and then shifted to 37°C for 50 min. The cells were harvested and lysed, samples were boiled, and proteins were separated by SDS PAGE. Samples were analyzed by Western with primary antibodies specific for the TcdB GTD, unglucosylated Rac1 (BD, 610650), total Rac1 (Millipore, clone 23A8), and GAPDH (Santa Cruz Biotechnology, sc-25778). Binding of an anti-mouse, HRP-conjugated secondary antibody (Jackson ImmunoResearch Laboratories, 115-035-174) was detected with a LumiGLO kit (Cell Signaling) according to manufacturer's instructions.
Unless otherwise noted, 100 nM TcdB or TcdB glucosyltransferase mutants and 2 uM Rac1 were mixed with 20 mM UDP-[14C]glucose (250 mCi/mmol, Perkin Elmer) in a total reaction volume of 10 uL. The buffer contained 50 mM HEPES pH 7.5, 100 mM KCl, 1 mM MnCl2, 2 mM MgCl2, and 0.1 mg/mL BSA. Reactions were incubated at 37°C for 1 h and stopped with the addition of loading buffer and boiling. Proteins were separated by SDS PAGE, and glucosylation of Rac1 was detected by phosphorimaging.
HeLa cells were seeded in a black 96-well imaging plate (PerkinElmer) and incubated overnight. Cells were pretreated with live/dead cell imaging dyes (Molecular Probes, R37601) and then treated with multiple concentrations of wild-type and mutant TcdB proteins. Cells were imaged in an environment-controlled chamber (37°C, 5% CO2) every 10 minutes over a 2 hour timecourse using an Opera High-Throughput Confocal Screening Microscope and Peltier-cooled, confocal CCD cameras. The percentage of dead cells and round cells was quantified over six fields for each concentration and time point using the Columbus Analysis software. Dead cells were defined as red cells with an intensity greater than 450 relative units, and round cells were defined as having an area less than 500 um2 and a width-to-length ratio of less than 0.4.
Colonic tissue was harvested from purpose-bred 25–35 kg, male or female, York-Landrace crossbred pigs. Following an overnight fast and immediately after euthanasia, a midline incision was performed and 15 cm of distal colon proximal to the rectum was excised and placed in PBS. The colon was opened, the luminal side was washed 3×5 min in 1 mM DTT to remove the mucus, and 3×5 min in PBS prior to dissection. Individual tissue sections were placed in wells of a 24-well plate. A nutrient buffer [44] containing (mM/liter): 122.0 NaCl, 2.0 CaCl2, 1.3 MgSO4, 5.0 KCl, 20.0 glucose, 25.0 NaHCO3 (pH 7.5) was pre-conditioned with HeLa cells overnight at 37°C and used to dilute the toxins. Explants were treated with wild-type TcdB, mutant TcdB, staurosporine (100 uM, Enzo Life Sciences, ALX-380-014-C250) or nutrient buffer for 5 hours at 37°C. The tissues were fixed with formalin for 56 h, washed in PBS, and transferred to cassettes. The tissue blocks were then embedded in paraffin, and 4 µm sections were cut and stained with hematoxylin and eosin (H&E) by the Vanderbilt University Translational Pathology Shared Resource core. Stained sections were coded and evaluated by six individuals, using a semi-quantitative injury scale: 0- no damage; 1-superficial damage, damage limited to intact surface epithelial cells; 2-loss of up to 50% of surface epithelial cells or gland length, crypts intact; 3-loss of over 50% of surface epithelial cells and damage in greater than 50% of gland length. An injury score was calculated as the mean score for sections evaluated seven times by six individuals. Statistical analysis was performed using a two-way ANOVA and Bonferroni's test. For keratin and caspase staining, sections were de-paraffinized with Histo-clear (National Diagnostics) and antigens were retrieved by citric acid. The sections were blocked with Serum-free protein block (Dako), stained with a rabbit anti-pan cytokeratin or anti-active caspase-3 antibody (Santa Cruz Biotechnology, sc-15367; Abcam, ab13847), and diluted in Dako's antigen diluent with background reducing components overnight at 4°C. The sections were washed with PBS and incubated for 1 hr at RT with an AlexaFluor 546 donkey anti-rabbit antibody (Invitrogen A10040). The sections were washed with PBS and mounted with Prolong Gold with DAPI (Invitrogen). H&E, pan-cytokeratin, and caspase-3 stained sections were imaged using an Ariol SL-50 (Epithelial Biology Center Imaging Core).
|
10.1371/journal.pcbi.0030151 | Helicobacter pylori Evolution: Lineage- Specific Adaptations in Homologs of Eukaryotic Sel1-Like Genes | Geographic partitioning is postulated to foster divergence of Helicobacter pylori populations as an adaptive response to local differences in predominant host physiology. H. pylori's ability to establish persistent infection despite host inflammatory responses likely involves active management of host defenses using bacterial proteins that may themselves be targets for adaptive evolution. Sequenced H. pylori genomes encode a family of eight or nine secreted proteins containing repeat motifs that are characteristic of the eukaryotic Sel1 regulatory protein, whereas the related Campylobacter and Wolinella genomes each contain only one or two such “Sel1-like repeat” (SLR) genes (“slr genes”). Signatures of positive selection (ratio of nonsynonymous to synonymous mutations, dN/dS = ω > 1) were evident in the evolutionary history of H. pylori slr gene family expansion. Sequence analysis of six of these slr genes (hp0160, hp0211, hp0235, hp0519, hp0628, and hp1117) from representative East Asian, European, and African H. pylori strains revealed that all but hp0628 had undergone positive selection, with different amino acids often selected in different regions. Most striking was a divergence of Japanese and Korean alleles of hp0519, with Japanese alleles having undergone particularly strong positive selection (ωJ > 25), whereas alleles of other genes from these populations were intermingled. Homology-based structural modeling localized most residues under positive selection to SLR protein surfaces. Rapid evolution of certain slr genes in specific H. pylori lineages suggests a model of adaptive change driven by selection for fine-tuning of host responses, and facilitated by geographic isolation. Characterization of such local adaptations should help elucidate how H. pylori manages persistent infection, and potentially lead to interventions tailored to diverse human populations.
| Helicobacter pylori is a genetically diverse bacterial species that infects billions of people worldwide, typically for decades. Long-term infection is a major risk factor for stomach ulcers and cancer, although most infections are benign, and the risks of various disease outcomes vary markedly among human populations. Analyses of housekeeping genes, whose encoded proteins perform normal cellular metabolic functions, had established that H. pylori strains from different geographic regions differed in their DNA sequences. Here, we analyzed the H. pylori slr multigene family that encodes up to nine secreted proteins (called SLR proteins) quite similar to the human protein Sel1. We showed that most members of the H. pylori slr gene family evolved significantly more rapidly than normal housekeeping genes. Different amino acids were selected in different H. pylori lineages, often on exposed surfaces of SLR proteins where they were potentially positioned to interact with host components. We propose that these amino acid differences affect the SLR protein function, likely contributing to H. pylori's adaptation to local differences in human stomach physiology. Further characterization of H. pylori proteins with lineage-specific differences in amino acids should improve understanding of geographic differences in H. pylori–host interactions and human disease, and of the interplay between different evolutionary forces in natural populations of any species.
| Helicobacter pylori chronically infects billions of people worldwide, typically for decades. Most H. pylori reside on gastric epithelial cell surfaces and in the overlying mucin layer, a tissue that turns over rapidly, is infiltrated by inflammatory cells after infection, and is buffeted by gastric acidity on its luminal side. The gastric mucosa is hostile to most bacterial species, and constitutes an unstable niche to which only the Helicobacters among bacterial taxa have become well adapted, in part perhaps through effective management of host responses to infection [1–4].
Great genetic diversity, geographic differences in predominant genotypes, and rapid evolvability are hallmarks of H. pylori populations [5–9]. Independent isolates generally differ by some 2% or more in DNA sequence of any metabolic (housekeeping) gene, with most such differences being synonymous (protein sequences unchanged). Phylogenetic analyses of H. pylori housekeeping gene sequences revealed differences in predominant genotypes between East Asian, European, and African populations that are far greater than those seen with most other pathogens. Such patterns reflect a combination of mutation, recombination, selection to retain gene function, and random genetic drift, which itself likely stems from H. pylori's highly localized (preferentially intrafamilial; nonepidemic) mode of transmission, and a resulting relative lack of H. pylori gene flow between well-separated human populations [6,7,10–12]. Correlated with this geographic partitioning of H. pylori populations are striking differences in predominant clinical consequences of infection. To illustrate, duodenal ulcer, which is typically associated with excess gastric acidity (hyperchlorhydria), is far more common in India than in Japan, whereas gastric cancer, which is typically associated with hypochlorhydria, is far more common in Japan than in India [13,14]. The near universality of H. pylori infection until very recently, the extraordinary chronicity of infection, and geographic differences among H. pylori populations all have contributed to an idea that H. pylori may have co-evolved with its human hosts [15]. Geographically isolated populations are also more likely to adapt to differences in local environment [16,17]. Human genetic or physiological traits that diverged during our evolution, that differ geographically, and that are important to H. pylori could have selected for adaptive changes in cognate H. pylori genes.
The virulence-associated cagA, and vacA genes provide examples in which evolutionary dynamics are likely to have been shaped by local differences in host physiology. CagA and VacA proteins each enter target cells and affect several normal cellular signal transduction pathways, with strengths and specificities that vary geographically. For example, East Asian and Western type CagA proteins differ most in sequences of domains responsible for phosphorylation and in resulting interactions with host SHP-2 phosphatase, an intracellular regulator of various cell proliferative, morphogenetic, and motility signaling pathways [18,19]. Similarly, highly active “s1,m1”–type alleles of the vacA toxin gene predominate in Japan and Korea, whereas nontoxigenic s2,m2–type alleles are common in the West [20,21], and a recombinant s1,m2 form predominates in coastal China [22]; the “m” region of VacA determines the cell type specificity of toxin action. We suggest that these geographic differences reflect types of selection pressures that predominate(d) in the various human populations, either currently or in centuries past, superimposed on the random genetic drift that figures so importantly in geographic partitioning of housekeeping genes; and that such patterns may be common among genes whose products interact with host components. Furthermore, the extraordinary chronicity of H. pylori infection suggests a possible need for effective management and potentially even exploitation of host responses. For example, although inflammatory responses help protect potential hosts against casual pathogen encounters, H. pylori is thought also to use metabolites leached from inflammation-damaged host tissues for its nutrition [23]. In addition, many strains use host sialylated glycolipids, synthesized during the inflammatory response, as receptors for adherence [24]. In this framework, much of H. pylori-induced gastric pathology might reflect how host signaling pathways are modulated by contact with the bacterium or its secreted products.
Sequenced H. pylori genomes contain a gene-family whose encoded proteins are likely secreted, and contain two or more copies of a degenerate 34–36 amino acid repeat motif that is characteristic of eukaryotic “Sel1” proteins, which themselves help regulate diverse signal transduction pathways [25]. The proteins bearing this repeat are typically built of several consecutive α/α motifs, the antiparallel α-helices of the motifs being connected by a short loop [26]. Five of the nine members of this “SLR” (for Sel1-like repeat) protein family are rich in cysteine residues and had been designated as “Helicobacter cysteine rich proteins” (Hcp) [27,28]; The α-helices of each SLR repeat are bridged by a disulfide bond, which is a unique feature of Hcps [26]. Although the in vivo function of H. pylori SLR proteins is not known, some Hcps bind β-lactam compounds [27,29], which suggests possible interactions with immunomodulatory peptidoglycan fragments, that could affect the innate immune response [30]. High antibody titres against four SLR proteins [Hp0211 (HcpA), Hp0235 (HcpE), Hp0336 (HcpB), and Hp1098 (HcpC)] were found in H. pylori–infected people [28], indicating in vivo expression and immune recognition. Furthermore, recombinant HcpA elicited IL12-dependent IFN-γ secretion in a naïve mouse splenocyte model [30], and HP1117 elicited protective antibodies during mouse infection [31]. Only one or two slr homologs are found in members of the closely related Campylobacter and Wolinella genera, whereas strains of H. acinonychis (from big cats) and of the nongastric mouse pathogen H. hepaticus (implicated in liver cancer) contain seven and six slr homologs, respectively (Figure 1A). It is appealing to imagine that this expanded family of secreted proteins affects bacterial–host interactions during chronic Helicobacter infection of mucosal tissues.
Here, we posited that if geographically isolated H. pylori populations had adapted to local differences in host physiology caused by factors such as host nutrition, genotype, or infection by other pathogens that in turn impacted on responses to H. pylori, these adaptations would leave an imprint of natural selection on the affected genes, superimposed on H. pylori's overall population genetic structure. For protein-coding genes, selection pressures and adaptive evolution can be detected and examined by comparing rates of fixation of synonymous (silent; dS) versus nonsynonymous (amino acid altering; dN) mutations in the population [32,33]. The ratio dN/dS (= ω) indicates whether amino acid change is unaffected, inhibited, or promoted by natural selection. Considering that most synonymous substitutions have little if any effect on fitness, dS is often equated to the rate of neutral nucleotide substitution. Under neutral evolution, one would expect dN and dS to be equal (ω = 1), and functionally critical genes (e.g., housekeeping genes, responsible for intracellular metabolic functions) to show very low dN (ω < 1) [32]. In certain genes, however, nonsynonymous substitutions are in excess (ω > 1) because changes in their encoded proteins are advantageous and thus have been selected in particular environmental contexts. This is often termed positive selection (sometimes also called diversifying or Darwinian selection). Many such substitutions are likely to have been selected specifically to change the activity or structure of encoded proteins [34]. Since differences in local conditions (e.g., host features) can lead to geographic differences in patterns of selection [16], evolutionary rates of amino acid substitutions are likely to vary among H. pylori lineages; and an elevated dN (ω >>1) in any specific lineage would indicate adaptive evolution. Such adaptive evolution tends to be episodic, in that it operates sporadically, and affects only a few amino acid positions in the protein [34]. Consequently, methods that estimate average dN and dS (summed over all codons and all lineages) often fail to detect adaptive evolution [33]. Here, we applied codon-based models of sequence evolution in conjunction with maximum-likelihood (ML) computational methods [35,36] that are particularly useful for detecting adaptive changes at specific sites in proteins, to study the evolutionary dynamics of the H. pylori slr gene family. We found that most slr family members had experienced positive selection, and accumulated adaptive mutations in specific H. pylori lineages, preferentially affecting specific surface-exposed sites in encoded proteins. These outcomes suggested selection for management and fine-tuning of host responses during chronic infection. Our results illustrate the utility of population-based phylogenetic strategies for identifying human population-specific adaptive determinants of H. pylori.
This study began with subtractive hybridization (as in [37,38]) to find genes or alleles that differed markedly between representative Japanese versus Western strains. One recovered clone contained a fragment of gene hp0519. Subsequent sequencing of hp0519 alleles from representative strains identified two in-frame deletions: a 24-bp segment that was absent from the Japanese strain (Δ24) but present in US reference strain J99 (nt 133–156), and a 15-bp segment in the Japanese strain that was absent from J99 (Δ15) (nt 640–654). PCR tests indicated that 70 of 87 Japanese strains carried Δ24 15+ type alleles, and only 14 carried the reciprocal 24+ Δ15 type allele, whereas 45 of 47 Spanish strains tested carried 24+ Δ15, the allele type that was uncommon in Japanese strains. Remarkably, 24 of 28 Korean strains tested also carried 24+ Δ15 type alleles, not the Δ24 15+ type that predominated in Japan. This difference in hp0519 pattern seemed extraordinary because Japanese and Korean strains were closely related in sequences of all other slr genes tested (see below); they were also closely related in sequences of housekeeping genes (Figure S2B; and Dailide and Berg, unpublished), which should be subject only to purifying selection to maintain function within the bacterial cell. Such relatedness was expected given the proximity of Japan and Korea and the shared history of their peoples.
Inspection of the Hp0519 protein sequences using the SignalP (http://www.cbs.dtu.dk/services/SignalP) and SMART (simple modular research tool) (http://smart.embl-heidelberg.de/) programs identified an N-terminal signal sequence and three SLRs with 40% sequence similarity to the human Sel1L protein (Tables S1 and S2; Figure S1A). Hp0519 belongs to the cluster of orthologous group (COG0790), which contains eight other family members in the genome of H. pylori strain 26695 ([39], and seven in the genomes of strains J99 [40] and HPAG1[41]) (http://www.ncbi.nlm.nih.gov/COG) (Figure 1A). Each COG0790 family member is predicted to encode a secreted protein with two or more SLRs (Table S2). Seven of these slr genes are present in each of the three sequenced genomes—hp0160 (jhp0148/hpag0158), hp0211 (jhp0197/hpag0212), hp0235 (jhp0220/hpag0238), hp0519 (jhp0468/hpag0493), hp0628 (jhp0571/hpag0610), hp1098 (jhp1024/hpag1036), hp1117 (jhp1045/hpag1055)—(prefixes “hp” in strain 26695; “jhp” in strain J99; “hpag” in strain HPAG1) (Figure 1A). Each genome also contains strain-specific slr genes: hp0336 in 26695, and jhp0318 (similar to hpag0339) and jhp1437 in J99. Reciprocal BLAST analysis revealed sequence similarities with domains of human Sel1L, ranging from 38% with jhp1437 to 51% with hp1117 (Figure S1), an intriguing pattern, even though such homologies do not by themselves demonstrate functional equivalence.
A strain of the related H. acinonychis encodes seven SLR homologs [42], six of which are nearly identical in amino acid sequence to SLRs of H. pylori (Figure 1A); this suggests either equivalent selection of slr gene–related function in these two species or recent interspecies transfer between H. pylori and H. acinonychis, which occurs readily in culture or in vivo [43]. A strain of the related nongastric pathogen H. hepaticus encodes six SLR homologs [44], which are relatively less related to those of H. pylori. In contrast, strains from genera most closely related to Helicobacter, Campylobacter, and Wolinella [45] each contain only one or two slr genes (Figure 1A). The slr genes are organized into discrete repeating units, which should be prone to duplication events [32]. Amino acid identities of 25%–70% are seen among SLR motifs in different members of this protein family in H. pylori (Table S3). Three of the six H. hepaticus slr genes have close homologs in H. pylori (hh1827 and hp0235; hh0718 and hp0628; and hh0816 and hp0519). These features suggest a possible Helicobacter-lineage–specific slr gene family expansion (duplication) after Helicobacters diverged from Campylobacter and Wolinella, and near the time of H. pylori and H. acinonychis versus H. hepaticus divergence. Also tenable is a model of separate gene family expansions in H. pylori and H. hepaticus, and even in H. acinonychis, since the corresponding slr homologs have different chromosomal locations (flanking genes) in these three species (unpublished data).
When a gene family's expansion is adaptive, the sequences of individual members should reflect selective forces that operated during and after this expansion. In general, paralogs that subsequently suffer inactivating mutations tend to be lost from the population over time [32,46]. More important evolutionarily are the paralogs that diverged, and acquired new functions, or optimized or subdivided complex ancestral functions. Purifying selection (ω < 1) predominates in the evolution of genes whose roles remained constant, whereas positive selection (ω > 1) predominates in cases of genes whose functions have diverged [32,46]. Accordingly, we determined ω values to test if slr gene family expansion was accompanied by functional divergence of paralogs, and, more generally, examined selection pressures that operated on this gene family.
We applied two codon-based models of sequence evolution to obtain ML estimates of selective pressures during slr gene family expansion, starting with sequences of the eight and nine slr genes in the three sequenced H. pylori genomes (Figure 1B). The simplest one-ratio model assumes the same ω for all branches, whereas the free-ratio (FR) model allows ω to vary among branches [33]. These models are nested, and hence can be compared using a standard likelihood ratio test (LRT). ML estimates were computed under varying conditions using different sets of initial values for ω and κ to confirm optimal algorithm convergence. Regardless of underlying assumptions, the FR model fit the data significantly better than the M0 model (−InL(FR) = 11286.639; −InL(M0) = 11356.182; χ2 = 140.062, degrees of freedom = 47; p < 0.00001; initial ω = 2, κ = 2; equilibrium codon frequencies estimated as free parameters). This suggested that ω varied significantly among individual branches of the slr gene-family phylogeny (Figure 1B; Table S4). Strong positive selection was evident in several branches, again indicating that slr gene family expansion was driven by selection for functional divergence among paralogs.
Given the multiple slr genes in these three Helicobacter species, slr family expansion might have occurred well before H. pylori became widespread in humans: that is, in ancient non-human hosts, possibly reflecting generalized selection pressures during mucosal colonization. Alternatively, because distributions of slr genes in the three sequenced H. pylori genomes vary, these expansions could have been more recent, especially given the ease with which gene duplications can arise [47,48], possibly facilitated in H. pylori by its lack of a MutHSL DNA repair system [49], and/or induced by reactive metabolites generated during infection [50]. In either case, evidence of functional divergence among H. pylori slr homologs driven by positive selection makes it appealing to imagine their products affecting traits important for H. pylori mucosal colonization. With this perspective, and prompted by the nonrandom geographic distribution of hp0519 indels, we determined DNA sequences of six slr genes present in most East Asian, Western European, and African H. pylori strains using isolates from a representative strain collection.
Sequences of Japanese and Korean alleles of housekeeping genes are typically intermingled in the same clusters (Figure S2B; and Dailide and Berg, unpublished). Therefore, we asked if hp0519 single nucleotide polymorphisms (SNPs) also differed geographically by sequencing a 322-bp hp0519 segment internal to these “24” and “15” indels in 78 strains. These internal hp0519 sequences contained many SNPs, which fell into separate Japanese (n =20) and Korean (n = 16) allele clusters (Figure 2A), in accord with the PCR-based indel results. Permutation–randomization tests of this 322-bp internal segment suggested great genetic differentiation, perhaps reflecting separation of Japanese island and Korean mainland populations (FST = 0.5, p < 0.001; Table S5), and a critical difference in the forces that had operated in these two regions.
To better understand the evolutionary forces driving this divergence, we next determined full-length hp0519 sequences (approximately 873 bp) from African, European, and East Asian strains (n = 27), chosen randomly from the larger dataset shown in Figure 2A, and reconstructed an ML phylogeny with these data. This revealed Japanese–Korean allele separation (bootstrap support = 100), as expected (Figure 3A). Further pairwise permutation-randomization tests of all 27 full-length hp0519 sequences also showed genetic differentiation among H. pylori subpopulations in various geographic regions (FST > 0.5, all comparisons; Table S5). Additional pairwise comparisons of five Korean and five Japanese full-length hp0519 sequences revealed 56 fixed differences (sites at which all sequences in one population differed from all sequences in a second population), versus 33 polymorphisms shared between them. Both sets showed unique polymorphisms (i.e., sites polymorphic in one set, monomorphic in the other; 68 in Japanese and 42 in Korean, respectively). This inverse relationship between fixed and shared polymorphisms suggests either ancient separation of Korean and Japanese hp0519 alleles or selection for accelerated accumulation of SNPs in at least one population. Application of the McDonald–Kreitman test for adaptive evolution showed that the ratio of nonsynonymous changes to synonymous changes among fixed differences (48/8) was significantly higher than that among polymorphic differences (68/42) (p < 0.001, Fisher's exact test and G-test) (Figure 2B). This outcome suggested that the accumulation of nonsynonymous substitutions in Japanese versus Korean hp0519 alleles had been driven by positive selection. In contrast, equivalent tests of hp0519 sequences from other populations suggested divergence between them due mostly to random genetic drift. ML phylogenies of the hp0518 and hp0520 genes that flank hp0519 from six representative Korean and six representative Japanese strains did not show any distinction between these two East Asian populations (Figure S2A). This indicated that divergence of Japanese versus Korean hp0519 alleles was due to selection on hp0519 itself, not due to linkage to an even more highly selected gene.
Selection pressures on hp0519′s individual codons and branches of its phylogenetic tree were next studied in detail using three groups of codon-based models of sequence evolution and ML-based LRTs: 1) site-specific models (SSMs), which examine variation in selection pressures across codons and assume a single ω across the phylogeny [35]; 2) lineage-specific models (LSMs), which allow ω to vary among lineages, while assuming a single rate across all codons [51]; and 3) lineage-site–specific models (LSSMs), which allow ω to vary both among codons and across the phylogeny [36]. SSMs confidently identified 18 sites under positive selection (ω2 = 3.515; Bayesian probability > 0.99) (Figure 2C; Table S6A), which suggested different selective pressures at different sites in hp0519. [Equivalent site-specific positive selection was also detected in hp0519 codons by the single-likelihood ancestor counting (SLAC) and fixed-effects likelihood (FEL) methods hosted at http://www.datamonkey.org (unpublished data)]. Previous work has shown that codon-based models implemented, in M7 and M8 models, in particular, are usually not adversely affected by recombination in bacterial datasets [52–54]. However, because the ML approach used here explicitly assumes a phylogenetic tree when estimating selection pressures, we also assessed if the extensive recombination typically seen in H. pylori populations could have produced a false-positive signal for positive selection. This entailed repeating the analysis assuming that sequences were linked by a “star” phylogeny, where lineages diverge simultaneously from a single root node; this removes the effect of phylogenetic history, including recombination events from the outcome. This analysis again indicated positive selection (p < 0.0001 for M3 versus M2 and M8 versus M7), with higher ω values under the M3 and M8 models, and with the same sites usually in the positively selected class as in the original analysis (Table S6C).
Next, a two-ratio LSM model, M2J (ωJ for Japanese foreground branch; ωR for all other branches (background lineages)) was constructed, to test a priori whether ωJ was significantly different from ωR. M2J fit the data significantly better than M0 (p < 0.0001; Table S6B) and suggested that divergence of Japanese hp0519 lineage was driven by positive selection (ωJ = 1.6). In accord with this, the FR model, which assigned an independent ω for each lineage, also fits the data significantly better than M0 (p < 0.0001; Table 6B). This suggested that hp0519 alleles had been subject to significantly different selective pressures in different lineages (Figure 3A). To identify rapidly evolving codons in the Japanese lineage, we constructed two LSSMs, M2JM2 and M2JM3, which assigned a different value to the Japanese lineage (ωJ) and compared them for fit against M1 and M3 SSMs, respectively. Both LSSMs confidently identified 26 sites that had been strongly selected (ωJ >> 20; Table S6B) in the ancestral Japanese lineage. This confirmed that divergence of Korean and Japanese hp0519 alleles was driven by strong positive selection that had favored specific adaptive changes in Japan. These differences are not explained by models invoking founder effects or random genetic drift alone. We suggest, rather, that this divergence reflects a condition unique to H. pylori in the Japanese islands at some recent evolutionary time. Possibilities include differences between the islands and the East Asian mainland in prevalence of other pathogens or parasites that affect host responses to H. pylori [55,56] and how H. pylori can best manage them; or host genotype, diet or nutrition, or sociocultural features that could also affect host responses to infection. The divergent Japanese-type hp0519 alleles might have existed at low frequency before being strongly selected in Japan. Alternatively, they might have arisen by more recent stepwise mutation and selection, perhaps only starting when rice-based agriculture was brought to Japan some 2,300 years ago, along with changes in diet, lifestyle, risk of infection, etc.
To examine adaptive evolution in the Japanese hp0519 lineage in a protein structure–function context, we applied several methods for comparative secondary structure analysis and modeled Hp0519 three-dimensional structure on the experimentally determined crystal structures of its homologs HcpB (Hp0336) [57] and HcpC (Hp1098) [58]. Twenty four of 26 (92%) sites under positive selection were in regions near but not overlapping the predicted SLRs, with 19 of 26 (73%) between the second and third SLR. Two of these adaptive residues (44 and 216) were located within the Δ24 or Δ15 indels, suggesting that these regions were also important functionally, whereas the other adaptive residues were separate from these indels. Strong conservation of Hp0519 SLR motifs suggests that they are critical for protein function. Secondary structure comparisons revealed striking differences between Japanese-type Hp0519 proteins and all others in amino acid charge distribution, hydrophilicity, surface probability, and antigenic index (Figure 3B). Although modelling of Hp0519 structure was complicated by significantly lower sequence conservation and the presence of indels, threading analysis suggested that Hp0519 folds into a three-dimensional structure similar to that of HcpB and HcpC (z-score = 31.19; Table S7). Secondary structure analysis localized most sites under positive selection to loop regions, which are generally surface-exposed and potentially positioned to contact cognate host proteins. Data supporting the computational prediction that Hp0519 protein is secreted was obtained by generating plasmids that encode recombinant Hp0519-FLAG fusion proteins with and without predicted signal peptides, and expressing them in BL21 DE3 Escherichia coli cells. Tests using the α-FLAG M2 antibody showed that most Hp0519 encoded by full-length hp0519 (predicted signal peptide intact) was secreted into the culture supernatant, whereas that encoded by an engineered hp0519 variant that lacked the signal peptide coding sequence remained with the cell pellet (Figure S3). Expecting equivalent signal peptide–dependent secretion of Hp0519 protein in H. pylori, we propose that many of the surface-exposed adaptive changes identified computationally may affect the strength or specificity of Hp0519′s interaction with cognate receptors or other host components. Other surface-exposed Hp0519 residues might have evolved under host immune selection, in particular if certain Hp0519-specific immune responses could inhibit H. pylori growth or persistence. In this last scenario, the directional nature of positive selection, specifically in the Japanese lineage, would suggest unique immunological pressures in Japan at some point in H. pylori's evolution. Finally, some of the observed adaptive changes might have been context-dependent, compensating for deleterious mutations elsewhere in the same gene [59–61], which themselves could have accumulated by chance (drift), or been specifically selected at an earlier time, if different selective pressures operated then.
To learn if selection for amino acid sequence change in different populations was common to the slr gene family, in general, versus specific to hp0519, we sequenced five additional slr family members present in nearly all H. pylori strains (hp0160, hp0211, hp0235, hp0628, and hp1117) from ≥32 isolates variously from East Asia (Japan, Korea), West Europe (Spain, England), and Africa (South Africa, The Gambia). The slr gene phylogenies revealed separate clustering of African, European, and East Asian alleles, as is typical of housekeeping genes. In striking contrast to hp0519, there was no distinction between Korean and Japanese alleles of these other slr genes (Figures S4A–S7A and S8). Analysis of selective pressures by SSMs suggested that each slr gene, except hp0628, that was analyzed here had experienced different selective pressures at different amino acid sites during their evolution (Figure 4A–4D; Tables S8–S12). Homology-based structural modeling suggested that most sites under positive selection in these slr genes also encoded surface-localized amino acids (Figure 4E– 4F). This outcome supports the idea of adaptive sites in SLR proteins potentially affecting their interactions with cognate host components.
LSMs suggested that the different hp0160, hp0211, and hp1117 lineages also evolved at different rates (Figure S4B–S7B). For example, positive selection was more evident in the East Asian and The Gambian hp0160 lineages (ω = 2.26 and 2.32, respectively), than in South African and European lineages (ω = 0.29 and 0.001, respectively). Similar lineage-specific positive selection was also seen in phylogenies of hp1117 and of hp0211; Hp0211 (HcpA) protein induces a cytokine IL12–mediated pro-inflammatory response [30]. Although SSMs suggested heterogeneous selective pressures and rapid evolution of certain codons in hp0235, LSMs indicated that ω did not vary significantly among the lineages studied (Table S11); a sampling of additional populations, however, might identify hp0235 lineages that had evolved more rapidly. The hp0628 gene was exceptional, in that SSMs and LSMs indicated evolution dominated by purifying selection (no codon class under positive selection), suggesting that it is functionally constrained, although ω did vary among codon sites (Table S12). Taken together, these outcomes illustrate that H. pylori evolution has not been strictly neutral genome-wide: that amino acid level polymorphisms fixed by positive selection are particularly abundant in certain key genes, perhaps often selected by features of the host milieu that vary geographically. Different sets of selective pressures operated in different members of the slr gene family and only on particular sites in any given gene.
The many decades during which H. pylori can persist in the gastric mucosa despite inflammation and other host defenses, the differences among individuals and geographic regions in the intensity and specificity of host responses, and the changes in responses with age and as infection progresses, all coupled with the possibility of H. pylori exploiting these responses while avoiding clearance by them [23], suggests a need for active response management by the bacterium. It is especially in this framework that we studied the evolutionary dynamics of the H. pylori slr gene family. The members of this family encode secreted proteins with homology to the Sel1 group of eukaryotic regulatory proteins that, through their interaction with other eukaryotic proteins, affect cell proliferation, apoptosis, immune response, and intracellular trafficking [62,63]. We found that positive selection played a dominant role in slr gene evolution: that different amino acids were selected at particular sites in a given protein in different geographic areas, and that effects were more extreme for some slr genes than for others in the populations examined. We suggest that these findings be interpreted in terms of within- and between-host pathogen dynamics: that differences among hosts in physiologic traits had selected for changes in cognate microbial proteins (here H. pylori SLR proteins). Also to be included in this category, we suggest, should be selection for altered interaction with or recognition by critical components of the host immune system, which can differ among humans genetically or physiologically, reflecting factors such as infectious disease history, nutrition, stress, etc. Successful adaptation to these forces should also contribute to the extraordinary chronicity of H. pylori infection. Geographic differences in predominant H. pylori–associated diseases noted above [13,14,64] are likely due to multiple factors that may include infections by other pathogens that affect responses to H. pylori infection, diet, and human genotype [3,15]. It is tempting to consider these trends also being affected by several aspects of H. pylori genotype, including predominant allele types of slr genes.
H. pylori isolates were obtained from phylogenetically distinct West European (Spain, England), African (The Gambia, South Africa), and East Asian (Japan, Korea) populations and have been described earlier [8,52]. All isolates were obtained from patients with gastric complaints who had undergone diagnostic endoscopy with informed consent. Standard methods were used for H. pylori propagation and storage. Primers used for PCR and sequencing are listed in Table S13. Standard methods were used for genomic DNA preparations, PCRs, and sequencing.
Nucleotide diversities within and between population, FST and permutation tests, and McDonald–Kreitman tests were done with DNASP version 4.1 (http://www.ub.es./dnasp). Phylogenetic reconstruction using slr gene sequences was performed with the ML approach implemented in PAUP*4b10 (http://paup.csit.fsu.edu/). An ML phylogeny was reconstructed under the best-fit model [determined with MODELTEST, version 3.7 (http://darwin.uvigo.es/software/modeltest.html)] by using a combination of heuristic searches and branch swapping to further optimize the likelihood score and substitution parameters. The significance of observed phylogenetic groupings was assessed by a bootstrap analysis performed with 1,000 replicates under the distance optimality criterion, while incorporating ML-optimized model and parameters. Phylogenetic trees were visualized with TreeView version 1.6.6 (http://taxonomy.zoology.gla.ac.uk/rod/treeview.html). The selective pressures operating on H. pylori slr genes were measured using an ML method that takes into account the sequence phylogeny and assesses the fit to the data of various models of codon evolution that differ in how ω varies across the sequence or across the phylogeny [33]. Three classes of codon-based analysis were used in this study: (a) SSM analysis, whereby models (M0, M1, M2, M3, M7, and M8) assume a single ω for all branches of the tree, but allow ω to vary among individual codon sites, thereby providing a measure of heterogeneity in selection pressures acting across the gene sequence [35]; M7 and M8 perform robustly even when recombination has occurred; (b) LSM analysis, wherein models (FR, two-ratio, etc.) assume that ω varies among individual branches of the phylogeny, but that all codon sites are under the same selective pressure, thereby providing a measure of selective pressures acting on the gene in different lineages [51]; and (c) LSSM analysis, which allows ω to vary simultaneously among sites and lineages [36]. Positive selection was inferred when codons with ω of >1 were identified and the likelihood score (−InL) of the codon substitution model in question was significantly higher than the likelihood of a nested model that did not take positive selection into account. The probability that a specific codon belonged to the neutral, negative, or positively selected class was calculated by using Bayesian methods implemented in PAML version 3.14 (http://abacus.gene.ucl.ac.uk/software/paml.html). Multiple runs, assuming different initial ω and κ values, and different models for estimating equilibrium codon frequencies (calculated from the average nucleotide frequencies at the three codon positions tables (F3X4) or used as free parameters) were analyzed for each gene to verify the convergence optima for each model.
Homology modeling of Hp0160, Hp0211, Hp0235, Hp0519, and Hp1117 started with a multiple sequence alignment including the protein sequences of the template structures HcpB (1KLX) [57] and HcpC (1OUV) [58] using program CLUSTALW. Due to the modular architecture of Hcps, different superpositions of HcpB and HcpC are possible and meaningful. This molecular feature increased conformational space and allowed modeling of protein sequences that were significantly longer than the template structures. The resulting structure-based sequence alignment was merged with the multiple sequence alignment. The alignment was manually curated taking into account the predicted secondary structure [program JPRED (http://www.compbio.dundee.ac.uk/∼www-jpred)]. Homology models were generated using program MODELLER (http://salilab.org/modeller/modeller.html). To identify distant structural homologues of Hp0519, its protein sequence was threaded against databases of protein structures using the programs FUGUE (http://www-cryst.bioc.cam.ac.uk/fugue) and LOOPP (http://cbsu.tc.cornell.edu/software/loopp/). Figures were generated with PYMOL (http://www.pymol.org).
GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession numbers for sequences generated in this study are from EF372636 to EF372923. |
10.1371/journal.pgen.1008123 | Mouse genome-wide association and systems genetics identifies Lhfp as a regulator of bone mass | Bone mineral density (BMD) is a strong predictor of osteoporotic fracture. It is also one of the most heritable disease-associated quantitative traits. As a result, there has been considerable effort focused on dissecting its genetic basis. Here, we performed a genome-wide association study (GWAS) in a panel of inbred strains to identify associations influencing BMD. This analysis identified a significant (P = 3.1 x 10−12) BMD locus on Chromosome [email protected] Mbp that replicated in two separate inbred strain panels and overlapped a BMD quantitative trait locus (QTL) previously identified in a F2 intercross. The association mapped to a 300 Kbp region containing four genes; Gm2447, Gm20750, Cog6, and Lhfp. Further analysis found that Lipoma HMGIC Fusion Partner (Lhfp) was highly expressed in bone and osteoblasts. Furthermore, its expression was regulated by a local expression QTL (eQTL), which overlapped the BMD association. A co-expression network analysis revealed that Lhfp was strongly connected to genes involved in osteoblast differentiation. To directly evaluate its role in bone, Lhfp deficient mice (Lhfp-/-) were created using CRISPR/Cas9. Consistent with genetic and network predictions, bone marrow stromal cells (BMSCs) from Lhfp-/- mice displayed increased osteogenic differentiation. Lhfp-/- mice also had elevated BMD due to increased cortical bone mass. Lastly, we identified SNPs in human LHFP that were associated (P = 1.2 x 10−5) with heel BMD. In conclusion, we used GWAS and systems genetics to identify Lhfp as a regulator of osteoblast activity and bone mass.
| Osteoporosis is a common, chronic disease characterized by low bone mineral density (BMD) that puts millions of Americans at high risk of fracture. Variation in BMD in the general population is, in large part, determined by genetic factors. To identify novel genes influencing BMD, we performed a genome-wide association study in a panel of inbred mouse strains. We identified a locus on Chromosome 3 strongly associated with BMD. Using a combination of systems genetics approaches, we connected the expression of the Lhfp gene with BMD-associated genetic variants and predicted it influenced BMD by altering the activity of bone-forming osteoblasts. Using mice deficient in Lhfp, we demonstrated that Lhfp negatively regulates bone formation and BMD. These data suggest that inhibiting Lhfp may represent a novel therapeutic strategy to increase BMD and decrease the risk of fracture.
| It is currently estimated that half of all Americans over the age of 50 already have or are at high risk of developing osteoporosis [1]. Bone mineral density (BMD) is used clinically to diagnose osteoporosis and beyond age, it is the single strongest predictor of the risk of fracture [2]. BMD is also one of the most heritable disease-associated quantitative traits with studies demonstrating that up to 80% of the variance in peak bone mass is heritable [3–6]. Consistent with its high heritability, genome-wide association studies (GWASs) in humans have identified hundreds of loci for BMD [7–9]. However, only a small fraction of the variance in BMD can be collectively explained by these loci, suggesting that BMD is influenced by a large number of small effect size loci [10]. As a result, there remains much to be discovered regarding the genetics of bone mass and genetic mapping efforts using mouse models is a complementary approach to identify novel regulators of bone mass [11–13].
Historically, linkage analyses in intercrosses, backcrosses, and recombinant inbred strain panels were the mainstay of mouse genetics [14]. These approaches were used to identify dozens of quantitative trait loci (QTL) for BMD and other bone traits [15,16]. However, identifying causative genes underlying QTL proved challenging [17]. Over the last decade, gene mapping approaches have transitioned from low-resolution linkage mapping to high-resolution GWASs [11]. The first GWASs in mice used panels of inbred mouse strains [18–21] and by leveraging accumulated recombinations, this approach significantly increased mapping resolution [19]. However, the approach was limited by population structure and low statistical power, due to the complicated breeding histories of inbred mouse strains and the small number of easily accessible and appropriate inbred strains (N typically < 30), respectively. Later studies demonstrated that these issues could be partly addressed by accounting for population structure and leveraging information from linkage-based QTL studies [22,23]. Given the significant amount of existing phenotypic and genotypic data on inbred strain panels [24], this approach is potentially a cost-effective strategy to identify novel regulators of complex traits.
High-resolution mapping approaches have significantly increased our ability to identify narrow regions of the genome harboring trait associated genetic variants. It is still, however, a challenge to identify causal genes and several approaches have been developed that can assist in bridging this gap. Specifically, systems genetics approaches involving the integration of other types of “-omics” data have proven useful [25]. Two systems genetics approaches for informing GWAS are expression quantitative trait loci (eQTL) discovery and co-expression network analysis [26]. EQTL discovery allows one to link variants associated with a trait, such as BMD, to changes in gene expression which leads to the hypothesis that the change in gene expression is causal for the change in phenotype. EQTL studies have been tremendously successful in identifying target genes downstream of genome-wide significant variants (as examples; [27,28]). However, in many cases the identified target genes have no known connection to the phenotype under investigation. It has been shown that co-expressed genes often operate in the same pathway or are functionally related [29]. Therefore, by using co-expression networks, which cluster genes based on patterns of co-expression across a series of perturbations [30], it is possible to develop hypotheses as to the function of a novel gene. When a locus has been resolved down to a small number of genes using genetic methods, unknown or poorly characterized genes can be ranked as the most likely candidate based on their function predicted from a co-expression network generated in a disease relevant tissue or cell-type [12,31].
Here, we used GWAS in an inbred strain panel to identify two chromosomal regions harboring variants influencing BMD. One of the associations, located on Chromosome (Chr.) 3, affected BMD in both sexes and was replicated in two separate inbred strain panels and an F2 intercross. This locus mapped to a 300 Kbp interval (NCBI37/mm9; Chr3:52.5–52.8 Mbp) encompassing four genes, Gm2447, Gm20750, Cog6, and Lhfp. An eQTL analysis and examination of a bone co-expression network suggested that Lhfp was a causal gene at this locus. The analysis of BMD, and other bone parameters, in Lhfp mutant mice supported this hypothesis. Additionally, SNPs within human LHFP were associated with heel BMD. Thus, we have used GWAS and systems genetics to identify Lhfp as a novel regulator of bone mass.
We performed a GWAS for total body BMD in 26 classical (non wild-derived) inbred strains at 12 months of age fed a chow diet. Genome scans were performed separately for each sex using the Efficient Mixed Model Algorithm (EMMA) to account for population stratification (S1 and S2 Files). In female mice, a significant (permutation determined threshold of -log10(P)>6) association was identified on Chromosome (Chr.) 3 and, in males, significant (permutation determined threshold of -log10(P)>5.9) loci were identified on Chrs. 2 and 3 (Fig 1).
Given our goal of identifying novel genes influencing BMD, we selected the Chr. 3 locus for further investigation. This locus was chosen because it was the most significant and the only one identified in both sexes (Fig 1). However, upon closer inspection, Chr. 3 harbored two associations, with peaks at 52.5 and 63.3 Mbp. In males, the 52.5 Mbp peak was the most significant (-log10(P) = 11.5), whereas in females the 63.3 Mbp peak was the most significant (-log10(P) = 5.9). The lead SNPs at both peaks were in moderate linkage disequilibrium (r2 = 0.46), making it unclear if they represented independent loci. We performed conditional analyses in males and in both cases each peak still exceeded chromosome-wise significance (-log10(P)>2.9) after controlling for the other, suggesting they represent independent loci.
We next sought to identify independent datasets supporting the validity of the Chr. 3 associations. There were 17 lead SNPs (the B6 reference allele was the minor allele at all SNPs with a frequency of 0.42), with the exact same strain distribution pattern (SDP) (S3 File), at the [email protected] Mbp association. All 16 were polymorphic between B6 and C3H. We previously identified a QTL, Bmd40, affecting femoral BMD on Chr. 3 in 32 week-old mice from a C57BL/6J (B6) x C3H/HeJ (C3H) (BXH) F2 intercross fed a high-fat diet [16]. The peak of Bmd40 overlaps both associations (Fig 2A). We also identified two sets of inbred strains with BMD measurements (the “Naggert” and “Tordoff” studies; data available from the Mouse Phenome Database [32] (https://phenome.jax.org/)) that were large enough (N strains > 25) to attempt to replicate the associations. In both “Naggert” and “Tordoff” panels, the strains used largely overlapped, but they did represent independent measures of BMD at different ages and conditions (Naggert—15–17 wks old, high-fat diet; Tordoff—14–18 weeks, chow diet). In both strain sets the exact same sets of SNPs at 52.5 Mbp reached chromosome-wide significance (-log10(P)>2.9) in both sexes (Naggert males -logP = 3.03, Naggert females -logP = 3.09, Tordoff males -log10P = 7.73, and Tordoff females -log10P = 5.17) (Fig 2B–2E). The association at 63.3 Mbp replicated in the Tordoff cohort (male -log10P = 3.18 and female -log10P = 2.92), but not in the Naggert cohort (male -log10P = 1.15 and female -log10P = 1.09) (Fig 2B–2E). These data provide additional support for the BMD association at 52.5 Mbp. Importantly, in all three inbred strain panels (“Ackert”, “Naggert” and “Tordoff”) and the BXH F2 intercross, reference (B6) alleles were associated with increased BMD relative to non-reference (C3H) alleles (Fig 2F–2I). Together, these data, from independent sources, are consistent with the hypothesis that a variant(s) in proximity of 52.5 Mbp on Chr. 3 influences BMD.
The set of SNPs that were the most significantly associated with BMD spanned a 300 Kbp interval from 52.5 to 52.8 Mbp (Fig 3A and 3B). This region contained four RefSeq transcripts: Gm2447, Gm20750, Cog6, and the 5’ end of Lhfp. Gm2447 and Gm20750 were listed as “predicted” RefSeq transcripts and annotated as long non-coding RNAs (lncRNAs). The evidence for these transcripts was based on prediction models and a small number of expressed sequence tag (EST) sequences. Neither of these transcripts have homologs in humans, rats, or any other mammalian species. To determine if Gm2447 and Gm20750 were expressed in mouse bone or bone cells, we performed total RNA-seq (poly A+ and poly A-) on three bone and three marrow-derived osteoblasts samples. Gm2447 and Gm20750 were not expressed, whereas the other two transcripts, Cog6 and Lhfp, which are well-annotated protein-coding sequences, were highly expressed in both bone (Fig 3C) and osteoblasts (Fig 3D). We also analyzed the expression profiles of Cog6 and Lhfp in 96 mouse tissues and cell lines using data available from BioGPS (http://biogps.org/) [33]. Cog6 was highly expressed in all tissues profiled (Fig 3E). Lhfp showed a more restrictive expression profile (Fig 3E). Importantly, Lhfp expression in primary calvarial osteoblasts was among the highest of any of the 96 samples surveyed (Fig 3E). Cog6 is part of the conserved oligomeric Golgi complex required for maintaining normal structure and activity of the Golgi apparatus [34]. Lhfp is a member of the lipoma HMGIC fusion partner (LHFP) gene family with no known function [35]. All other transcripts on either side of the region were >200 Kbp away.
We cannot exclude Gm2447 and Gm20750 (or for that matter other genes flanking the association); however, based on the data above we focused on interrogating Cog6 or Lhfp as potential causal genes. First, we evaluated Cog6 and Lhfp for coding polymorphisms among inbred strains. Based on whole genome-sequence data from C57BL6/J and C3H/HeJ (which carry alternative alleles at the association) there are no coding variants between the strains for Lhfp [36]. In contrast, there were three non-synonymous SNPs in Cog6 between B6 and C3H. These SNPs resulted in (rs30302002) I461V, (rs30323949) V620I and (rs30323946) S643N amino acid substitutions. However, using PolyPhen2, SIFT, and PROVEAN all three substitutions were predicted to be benign/tolerated and not impact Cog6 function [37–39].
We next determined if the same SNPs associated with BMD regulated the expression of Lhfp or Cog6 (or any gene ± 1Mbp of the association). We searched for local expression quantitative trait loci (eQTL) using expression data in liver, brain, adipose and muscle tissues in the BXH F2 intercross. Although expression data on bone or bone cells would have been ideal for this analysis, these data were not available. This did, however, allow us to identify local eQTL that might also be operative in bone. We observed a highly significant local eQTL for Lhfp (LOD = 19.9) in liver (Fig 4A). Cog6 and all other genes in proximity of the region were not regulated by a local eQTL in any tissue (max cis eQTL LOD = 1.8 across all four tissues). The lead Lhfp eQTL SNP (rs3665395) was located in the first intron of Lhfp and B6 alleles of rs3665395 were associated with increased expression of Lhfp relative to C3H alleles (Fig 4B). In liver, we observed a negative correlation (r = -0.29, P = 1.5 x 10−4) between Lhfp and BMD, as would be expected given that B6 alleles of rs3665395 were associated with decreased BMD and increased Lhfp (Fig 4C). We also searched for local eQTL using expression data from bone in the Hybrid Mouse Diversity Panel (HMDP) panel, but did not identify local eQTL for either Cog6 or Lhfp.
Our group and others [12,27,31,40] have shown that co-expression network analysis can identify interactions among genes and knowledge of these interactions can assist in predicting gene function and/or the cell type in which a gene is operative. Therefore, we next used a bone co-expression network to further evaluate Lhfp and Cog6. For the analysis we used a previously generated whole bone (femur with marrow removed) co-expression network from the Hybrid Mouse Diversity Panel (HMDP) that consisted of 13,759 genes partitioned into 21 co-expression modules [41,42]. In this network, Lhfp was a member of module 9 and Cog6 was a member of module 2. Module 2 was enriched in a large number of gene ontology terms including “mitochondrion”, “oxidative phosphorylation” and “actin cytoskeleton”; all of which are important to bone. However, module 2 did not have a signature of a particular bone cell-type, nor was it enriched for genes known to influence BMD. In contrast, we have previously demonstrated that module 9 is enriched for genes 1) directly involved in osteoblast differentiation, 2) implicated by BMD GWAS, and 3) when knocked-out in mice impact BMD [31,42].
To investigate specific network connections for Lhfp and Cog6, we identified the 150 genesmost strongly connected to each gene in their respective module (S4 and S5 Files). The genes with the strongest connections to Cog6 were enriched for genes involved in “muscle structure development” (FDR = 2.9 x 10−10), “muscle cell development” (FDR = 4.4 x 10−10), among many other similar muscle-related categories (S6 File). In contrast the genes with the strongest connections to Lhfp were, similar to module 9, enriched for genes involved in “ossification” (FDR = 1.5 x 10−7), “osteoblast differentiation” (FDR = 8.0 x 10−4), “skeletal system development” (FDR = 6.4 x 10−3), “bone development” (FDR = 3.3 x 10−2), among many other related bone-related functional categories (S7 File and Fig 5A). The Lhfp-centric network contained a number of genes with key roles in osteoblast differentiation and activity, including Sp7, Pthr1, Akp2, Tmem119, and Bmp3 (Fig 5B). Together, these data suggest that Lhfp is involved in the activity of osteoblasts, a process of direct relevance to the regulation of bone mass.
Bone marrow stromal cells (BMSCs) are adherent marrow cells that contain the mesenchymal progenitors of osteoblasts [43]. To test the role of Lhfp in osteoblast function, we quantified the number of BMSCs and their ability to form osteoblasts from mice lacking Lhfp. Using CRISPR/Cas9, we created five small deletions (ranging from 4–16 bps) in exon 2 (ATG start codon is in exon 2) of Lhfp (Table 1). All five were frameshift mutations resulting in a truncated LHFP protein (S1 Fig). As expected, we observed significantly decreased Lhfp transcript levels in heterozygotes (Lhfp+/-) and mutants (Lhfp-/-) from all five lines (Fig 6A). Since all five mutations impacted Lhfp expression in the same manner, we grouped littermate mice by genotype from all lines for all downstream experiments.
Next, we performed colony-forming unit-fibroblast (CFU-F) assays, a direct measure of BMSCs, in 16 week-old Lhfp-/- and littermate Lhfp+/+ mice. We observed similar trends in both sexes; therefore, all data were combined and adjusted for the effects of sex to increase power. In Lhfp-/- mice, we observed a significant (P = 0.02) increase in CFU-F number (Fig 6B). We next evaluated the ability of BMSCs from Lhfp+/+ and Lhfp-/- mice to differentiate into mineralizing osteoblasts. Consistent with network predictions, Lhfp-/- BMSCs exhibited increased mineralization as measured by bound alizarin red (P = 0.02; Fig 6C).
We next determined if bone mass was altered in Lhfp mutant lines. To replicate the conditions of the Ackert inbred strain panel and the BXH F2, we generated two cohorts of mice. The first was fed a chow diet for 12 months, while the second was fed a high-fat diet from 8 to 32 weeks of age. Based on the negative correlation between BMD and Lhfp expression, the direction of the genetic effects on expression in liver, and increased osteoblast activity observed above, we predicted increased BMD in Lhfp-/- mice. In both cohorts, BMD was measured in mice of all three genotypes and cortical and trabecular microarchitecture was measured by microCT only in Lhfp+/+ and Lhfp-/- mice. At 32 weeks of age in mice on a high-fat diet we observed significantly (P = 2.6 x 10−3) increased femoral BMD as a function of mutant Lhfp alleles in females, but not males (Fig 6D).
BMD is an inherently noisy phenotype; therefore, to generate a more detailed understanding of the effects of Lhfp in bone we used microCT to investigate the amount of bone in both the femoral trabecular and cortical compartments. We did not observe effects on trabecular bone mass at the distal femur in either male or female mice. However, Lhfp-/- mice of both sexes had significantly (P<0.05) increased femoral cortical bone area fraction (BA/TA) and cortical thickness (Ct.Th) as compared to Lhfp+/+ littermates (Fig 6E and 6F). We also observed a significant (P = 0.03) increase in tissue mineral density (TMD) in male Lhfp-/- mice (Fig 6G). In general, we observed the same trends of increased cortical bone mass in Lhfp-/- mice at 52 weeks of age; however, only Ct.Th in females was significant (P = 0.03) (Fig 6H–6K). These data indicate that Lhfp is a negative regulator of cortical bone mass in both male and female mice. They are also consistent with Lhfp underlying, at least in part, the BMD association on Chr. [email protected] Mbp.
We next determined if the human region syntenic with the mouse BMD locus harbored SNPs associated with variation in BMD. For this analysis we utilized data from the largest GWAS performed to date (N~426K) for heel BMD [9]. Heel BMD has been demonstrated to be highly genetically correlated with BMD at more clinically relevant sites such as the spine and femoral neck [8,9]. The human region syntenic with the mouse Chr. [email protected] Mbp spanned from 39.9 to 40.6 Mbp on Chr. 13. This region harbored 4055 SNPs. A set of 14 SNPs were significantly associated (P = 1.2 x 10−5) after adjusting for the total number of SNPs in the region (P<1.23 x 10−5) (Fig 7). These SNPs were located in intron 3 of LHFP. We queried eQTL for the Gene Tissue Expression (GTEx) project [44], but there were no eQTL for any genes ± 1 Mbp of the association that colocalized with the heel BMD association. Though these data do not directly implicate LHFP, they do support its potential involvement in the regulation of human BMD.
In this study, we used GWAS in a mouse inbred strain panel and a multifaceted systems genetics approach to identify and validate a high-resolution association for BMD on Chr. 3. The association directly implicated four genes: Gm2447, Gm20750, Cog6 and Lhfp. Of these, Lhfp expression was regulated by a local eQTL in liver and was predicted, based on a bone gene co-expression network, to be involved in osteoblast-mediated bone formation. We demonstrated that mice deficient in Lhfp displayed increased BMSC number and increased BMSC osteogenic differentiation. Furthermore, Lhfp-/- had increased BMD due to increased cortical bone mass. Together these data strongly suggest that Lhfp is responsible, at least in part, for the BMD association we identified on Chr. [email protected] Mbp. This work defines Lhfp as a negative regulator of the pool of osteoprogenitor cells, osteoblast activity, and cortical bone mass.
GWAS in mice has proven to be a powerful approach for the identification of genomic regions harboring trait-associated genetic variation [11]. The earliest applications of GWAS in mice used panels of readily accessible inbred strains [18–20]. However, such approaches were plagued by false positives due to population stratification [22]. Aware of this limitation, we first performed GWAS for BMD after correcting for population structure in inbred strains and then replicated the analysis in two separate strain panels (containing many of the same strains, but representing independent measures of BMD in different environments) and an F2 intercross. Of the multiple loci identified, the association on Chr. 3 at 52.5 Mbp was identified in all datasets, strongly suggesting it represents a bona-fide genetic association.
The Chr. 3 locus, as defined by the interval harboring the most significant SNPs, contained four genes; Gm2447, Gm20750, Cog6 and Lhfp. Gm2447 and Gm20750 were both predicted lncRNAs. This prediction is based on limited data and the fact that we did not observe their expression in bone tissue or osteoblasts (though we only measured their expression in one inbred strain), suggest they are not likely causal for the locus; though, this alone is not enough to definitely exclude their involvement. For Cog6 and Lhfp we used eQTL data and a bone co-expression network to assist in evaluating their potential causality. Both analyses supported a role for Lhfp. Using eQTL data from liver tissue in the BXH F2 intercross, we observed that variants associated with decreased BMD were associated with increased expression of Lhfp. We did not observe an association between the BMD-associated variants and Cog6 expression. Furthermore, Lhfp was a member of a well-studied module of co-expressed genes in mouse bone. This module is highly enriched for genes that play a role in osteoblast function, which provides a direct explanation as to how Lhfp may be impacting BMD. In contrast, Cog6 was a member of a module enriched for genes involved in a wide range of “energy-generating” functions. Importantly, all of our experimental results confirmed that Lhfp is a negative regulator of osteoblast activity and BMD. While these data support a role for Lhfp in the effects of the Chr. 3 locus, they do not exclude any of the other genes in, as well as flanking the locus.
In all four genetic populations used to identify the association on Chr. [email protected] Mbp, the strength of the association differed by sex. For example, in the “Ackert” population the association was stronger in males relative to females. In the “Naggert” strain set the strength of the association was similar in both sexes, albeit both were lower than seen in the other three populations. Similar to the “Ackert” strains, the association was stronger in males than females in the “Tordoff” strain set. In the BXH F2, the Chr. 3 QTL was male-specific, with little to no signal in females. The increase in cortical bone mass in Lhfp-/- mice was also sexually dimorphic. Although Lhfp deficiency increased cortical bone mass in both sexes in general, the effects were slightly more pronounced in females than males. This discrepancy could be the result of inaccuracies in estimating genetic effect sizes in the relatively small strain sets, the extent of linkage in the F2 confounding the sex effects, or influences from the different genetic backgrounds of the populations studied (strains sets vs. F2 vs. knockout).
Little is known regarding the molecular function of Lhfp. Lhfp is a member of the Lhfp-like gene family, which is a subset of the superfamily of tetraspan transmembrane protein encoding genes. It was first identified as a translocation partner with the HMGIC gene in benign lipomas [35]. The human LHFP/COG6 locus was also identified by GWAS as harboring variants associated with hippocampal volume [45]. However, prior to this study Lhfp had not been connected to the regulation of osteoblast function or BMD. Based on our experimental results, we hypothesize that Lhfp regulates bone mass through a role in cells of the osteoblast lineage. This does fit with prior work implicating Lhfp in the mesenchymal differentiation of gliosarcoma [46]. It is possible that Lhfp serves as a “brake” regulating the number of osteogenic precursor cells in the bone marrow microenvironment as well as their differentiation potential. However, further work will be required to elucidate its precise molecular role in osteoblasts and bone.
In summary, we have used GWAS in a set of inbred strains to identify an association impacting femoral BMD on Chr. 3 at 52.5 Mbp. We show using a variety of approaches that Lhfp is likely responsible for most, if not all, of the effects of this locus. Our results identify Lhfp as a novel negative regulator of osteoblast function and BMD and increase our understanding of the genetics of BMD.
The animal protocol for the generation and characterization of Lhfp mutant mice was approved by the Institutional Care and Use Committee (IACUC) at the University of Virginia.
The “Ackert” strain set contained BMD data on 32 inbred strains at three time points (6, 12 and 18 months). These data were collected by The Jackson Laboratory Nathan Shock Center of Excellence in the Basic Biology of Aging. Cohorts of males and female mice of 32 inbred strains of mice were aged to 6, 12 and 18 months of age. The number of mice per sex and per strain for each age point ranged from 1 mouse to 9 mice per group, with the majority of groups containing 6–7 mice. For the 12 month data set, the focus of this paper, the group size ranged from 3 to 9 mice per strain, per sex. At 6, 12 and 18 months of age, a varity of phenotypes were measured using a cross sectional study design with the hopes of capturing the main definers of Healthspan. This study, and the phenotypes available, is described in detail elsewhere [47]. Whole body BMD, sans the head, was measured by Dual X-ray Absorptiometry as previously described [48]. The complete dataset is available from the Mouse Phenome Database (MPD) (https://phenome.jax.org/projects/Ackert1). After removing wild-derived strains, and C57BLKS/J (due to inclusion of this strain producing spurious results) we were left with data on 26 strains. To identify loci influencing BMD, we used the Efficient Mixed Model Association (EMMA) algorithm [23]. For the analysis BMD was rankZ transformed. SNPs were obtained from strains genotyped on the Mouse Diversity Array (http://churchill-lab.jax.org/website/MDA) [49]. SNPs with a minor allele frequency < 0.05 were removed, leaving 228,085 SNPs. These SNPs were used to generate a kinship using the ‘emma.kinship’ R script available in the EMMA R package (available at http://mouse.cs.ucla.edu/emma/) [23]. The emma.REML.t function of EMMA was used to perform all mapping analyses. The significance of the maximum association peak was assessed by performing 1,000 permutations of the data. In each permutation, the minimum p-value was recorded to produce an empirical distribution of minimum permutation p-values. The quantiles of this distribution were used to assign adjusted p-values. P-values exceeding a genome-wide significant of P<0.05 were used as thresholds to identify associated loci. GWAS resulted were visualized using the “qqman” R package [50]. Replication of the association on Chr. [email protected] Mbp was performed using “Tordoff” and “Naggert” inbred strains sets. These data are available from MPD (https://phenome.jax.org/projects/Naggert1 and https://phenome.jax.org/projects/Tordoff3). Replication analyses were restricted to Chr. 3 and otherwise performed as described above.
Femora were isolated from an inbred Collaborative Cross strain (CC016/GeniUnc; Jackson Lab Stock #024684) (N = 3 mice). Marrow was isolated and bone marrow stromal cells (BMSCs) were differentiated as described below. Total RNA was then isolated from bone and BMSC-derived osteoblasts using RNeasy Plus Mini Kit (Qiagen). RNA-Seq libraries were constructed using TruSeq RNA Library Prep Kit v2 sample prep kits (Illumina). Samples were sequenced to an average depth of 24.6 million 2 x 75 bp paired-end reads on an Illumina NextSeq500 sequencer. Fastq files were aligned to the mouse reference (GRCm38) using HISAT2 v 2.0.5 (https://ccb.jhu.edu/software/hisat2/index.shtml) with a SNP aware reference index (genome_snp) [51]. Expression levels in Fragments Per Kilobase of transcript per Million mapped reads (FPKM) were generated using Stringtie [51]. The data are available from GEO (GSE121887). Microarray profiles for Cog6 and Lhfp in 96 tissues/cell-types were downloaded from BioGPS (http://biogps.org).
The generation of microarray expression data and eQTL analyses on bone from the 96 strains of the Hybrid Mouse Diversity Panel (HMDP) has been previously described [41,42]. These data are available from NCBI Gene Expression Omnibus (GEO) (GSE27483). A t-test was used to test for differences in Cog6 and Lhfp expression in strains stratified by genotype at rs3691451 (of the 17 peak BMD SNPs). Liver, brain, muscle and adipose eQTL in the BXH F2 were identified using R/qtl [52]. The expression data is available from GEO (GSE11338, GSE11065, GSE12798, and GSE12795). The genotypes and expression data are also available from GeneNetwork (“BH/HB F2 UCLA”, http://www.genenetwork.org/webqtl/main.py).
The generation of a bone co-expression network and characterization of the module 9 (M9) is described in [42]. We identified genes with the strongest connections to Cog6 and Lhfp based on Topological Overlap Measures (TOMs), calculated as described in [30]. Network depictions were constructed using Cytoscape [53]. Gene Ontology (GO) analysis was performed using the PANTHER database statistical overrepresentation test (http://www.pantherdb.org/) [54]. The analysis was restricted to the “GO biological process complete” annotation data set.
The Lhfp knockout mice used in this study were generated using the CRISPR/Cas9 genome editing technique. Cas9 mRNA that was injected into C57BL6/N embryos was synthesized exactly as outlined in [55] while the guide RNA (sgRNA) was generated with some modifications. Briefly, the 20 nucleotide (nt) sequence that would be used to generate the sgRNA was chosen using the CRISPR design tool developed by the Zhang lab (crispr.mit.edu). The chosen sequence and its genome map position is homologous to a region in Exon 2 that is approximately 300 bp, 3’ of the start codon (the ‘ATG’ is located in Exon 2 of Lhfp) (S1 Table). To generate the sgRNA that would be used for injections, oligonucleotides of the chosen sequence, as well as the reverse complement (S1 Table, primer 1 and 2, respectively), were synthesized such that an additional 4 nts (CTTC and AAAC) were added at the 5’ ends of the oligonucleotides for cloning purposes. These oligonucleotides were annealed to each other by combining equal molar amounts, heating to 90°C for 5 min. and allowing the mixture to passively cool to room temperature. The annealed oligonucleotides were combined with BbsI digested pX330 plasmid vector (provided by the Zhang lab through Addgene; https://www.addgene.org/) and T4 DNA ligase (NEB) and subsequently used to transform Stbl3 competent bacteria (Thermo Fisher) following the manufacturer's’ protocols. Plasmid DNAs from selected clones were sequenced from primer 4 (S1 Table) and DNA that demonstrated accurate sequence and position of the guide were used for all downstream applications. The DNA template used in the synthesis of the sgRNA was the product of a PCR using the verified plasmid DNA and primers 3 and 5 (S1 Table). The sgRNA was synthesized via in vitro transcription (IVT) by way of the MAXIscript T7 kit (Thermo Fisher) following the manufacturer's protocol. sgRNAs were purified and concentrated using the RNeasy Plus Micro kit (Qiagen) following the manufacturer's protocol.
C57BL/6N female mice (Envigo) were super-ovulated and mated with C57BL/6N males. The females were sacrificed and the fertilized eggs were isolated from the oviducts. The fertilized eggs were co-injected with the purified Cas9 mRNA (100 ng/μl) and sgRNA (30 ng/μl) under a Leica inverted microscope equipped with Leitz micromanipulators (Leica Microsystems). Injected eggs were incubated overnight in KSOM-AA medium (Millipore Sigma). Two-cell stage embryos were implanted on the following day into the oviducts of pseudo pregnant ICR female mice (Taconic or Envigo). Pups were screened by PCR of tail DNA using primers 6 and 7 with subsequent sequencing of the resultant product from primer 8 (S1 Table).
Two sets of injections (of ~100 eggs each) were performed resulting in 2 mice possessing mutations from each set of injections (A,B and C,D, respectively; Table 1). All 4 mice possessed out of frame bi-allelic deletions ranging from 1–16 bp; progeny from only 3 of the founders (mice A, B, C, Table 1)) were used in this study. Note that an identical 11bp deletion was found in two mice from two separate injections. qPCR with primers 9 and 10 (S1 Table) was used to assess Lhfp expression as outlined in [31].
Isolation [56] and differentiation of mesenchymal stromal cells (MSC) from the bone marrow of mouse femurs was performed as described for osteoblasts [57] with minor modifications. Briefly, one or both femurs from a given mouse were aseptically isolated, denuded of soft tissue and the marrow extracted by removing the proximal end of each bone and centrifuging at 2000 xg for 30 s such that the marrow collects into 25 μl of fetal bovine serum (FBS). Exudates from a single femur were dispersed, via trituration, in 5ml complete media (MEM-alpha, 10% FBS, 100U penicillin/100ug Streptomycin per ml, 2 mM glutamine). Cells were manually counted where upon 4 million were used to seed a 10 cm dish for CFU-F determination and the remainder were applied to a 60mm dish. Media on the 10 cm dishes was changed on days 2, 4 and 8; on day 14, cells were fixed (NBF), stained (Coomassie, BioRad #161–0436) and the number of CFU-Fs determined via Image J analysis and the manual counting of colonies. Media on the 60 mm dishes was changed on day 2 with cells removed via trypsin/EDTA (Gibco) digestion on day 4. Detached cells were triturated in 5ml complete media, pelleted at 1000 xg for 5 min., re-suspended in 1 ml complete media and counted where upon 150,000 cells were used to seed a well of a 12 well dish. A minimum of 2 wells per sample were obtained for all samples reported here in. Osteoblast differentiation was initiated 3 days after plating (7 days after bone marrow isolation) by replacing the media with complete media supplemented with 50 μg/ml ascorbic acid, 10 mM beta-glycerophosphate and 10 nM dexamethasone. Media was changed every other day for 8 days at which time cells were either used as a source for RNA (mirVana, Thermo Fisher) or used to determine the amount of hydroxyapatite formed during differentiation [31]. Briefly, cells were washed with PBS, fixed with neutral buffer formalin (NBF) for 15 min. and subsequently stained with 40 mM Alizarin Red (AR), pH 5.6 for 20 min and washed extensively with H20. The amount of AR bound to mineral was quantitated by Image J analysis of scanned images as well as the 5% Perchloric Acid eluate absorbance at 405 nm.
Femoral BMD was measured ex vivo using a Lunar PIXImus II Mouse Densitometer (GE Medical Systems Model 51045; Madison, WI, USA). Morphologies of the trabecular bone of the distal femur and cortical bone of the femoral midshaft were measured using micro-focus X-ray computed tomography (vivaCT 40, Scanco Medical AG, Bassersdorf, Switzerland) following guidelines for assessment of bone microstructure [58]. Tomographic volumes were acquired at 55 kV and 145 μA, collecting 2000 projections per rotation at 300 millisecond integration time. Three-dimensional 16-bit grayscale images were reconstructed using Scanco Evaluation software, Version 6.5–3. Threshold values were adjusted to best match the silhouette of features of interest in the threshold-subtracted image compared to the grey-scale image. The resulting threshold for hydroxyapatite-equivalent density was 370 mg/cm3 for compact or cortical bone and 270 mg/cm3 for the trabecular bone region; these values were applied to subsequent samples. Volumetric analysis was confined to the trabecular region for the distal femur by manual exclusion of the cortical bone. A 1.03 mm high region of interest was analyzed beginning at 1 mm proximal to the growth plate. For the cortical bone, a 0.3 mm high region was analyzed at the mid-diaphysis. Measures analyzing the distal femur trabecular site included total volume, bone volume, trabecular bone volume fraction (BV/TV), thickness, number, connectivity density and structure model index (SMI). Cross-sectional measurements of the cortical bone included bone volume, total volume, marrow area and polar moment of inertia.
All statistical analyses were conducted using the R language and environment for statistical computing [59]. The Lhfp qPCR data was analyzed using a t-test. Data are presented as means ± 1.5 times the interquartile range. CFU-F and osteogenic differentiation data was analyzed using the “lsmeans” R package [60]. The data were fit to a linear model including the effects of genotype and sex. P-values were adjusted using the “tukey” method. Data are presented as lsmeans ± s.e.m. BMD and microarchitectural bone data were analyzed using ANOVA with a linear model including the effects of genotype, body weight, and any other phenotype-specific covariates. Data are presented as means ± 1.5 times the interquartile range.
Data from the largest heel BMD GWAS performed to date were downloaded from http://www.gefos.org/?q=content/data-release-2018 [9]. LocusZoom was used to create a regional association plot [61]. The GTEx database V7 was queried for colocalizing eQTL(https://gtexportal.org/home/) [44].
|
10.1371/journal.pgen.1002788 | Parallel Evolution of Auditory Genes for Echolocation in Bats and Toothed Whales | The ability of bats and toothed whales to echolocate is a remarkable case of convergent evolution. Previous genetic studies have documented parallel evolution of nucleotide sequences in Prestin and KCNQ4, both of which are associated with voltage motility during the cochlear amplification of signals. Echolocation involves complex mechanisms. The most important factors include cochlear amplification, nerve transmission, and signal re-coding. Herein, we screen three genes that play different roles in this auditory system. Cadherin 23 (Cdh23) and its ligand, protocadherin 15 (Pcdh15), are essential for bundling motility in the sensory hair. Otoferlin (Otof) responds to nerve signal transmission in the auditory inner hair cell. Signals of parallel evolution occur in all three genes in the three groups of echolocators—two groups of bats (Yangochiroptera and Rhinolophoidea) plus the dolphin. Significant signals of positive selection also occur in Cdh23 in the Rhinolophoidea and dolphin, and Pcdh15 in Yangochiroptera. In addition, adult echolocating bats have higher levels of Otof expression in the auditory cortex than do their embryos and non-echolocation bats. Cdh23 and Pcdh15 encode the upper and lower parts of tip-links, and both genes show signals of convergent evolution and positive selection in echolocators, implying that they may co-evolve to optimize cochlear amplification. Convergent evolution and expression patterns of Otof suggest the potential role of nerve and brain in echolocation. Our synthesis of gene sequence and gene expression analyses reveals that positive selection, parallel evolution, and perhaps co-evolution and gene expression affect multiple hearing genes that play different roles in audition, including voltage and bundle motility in cochlear amplification, nerve transmission, and brain function.
| The convergent origins of laryngeal echolocation in two groups of bats (Yangochiroptera and Rhinolophoidea) and toothed whales have long been a focus of interest for biologists. We screened three candidate genes—Cdh23, Pcdh15, and Otof—involved in different steps in the echolocation system. Signals of parallel evolution occurred in all three genes in the three groups of echolocators. Cdh23 and Pcdh15 constitute part of the mechanical link within the hair bundle of the ear. Both genes showed signals of both convergent evolution and positive selection, which implied they may have co-evolved to optimize cochlear amplification. Further, three lines of evidence suggest that Otof plays an important role in the transmission of signals in the brain during echolocation. First, the gene is more highly expressed in the auditory cortex of the brain in echolocating adult bats than in other cerebral cortexes. Second, this expression is higher in adult echolocating female bats than in their embryos, which do not use echolocation. Third, echolocators also have a higher level of expression in their cerebral cortexes than do non-echolocating bats. Taken with other evidence, the independent origins of echolocation involve the same genes that have evolved in precisely identical ways.
| The ability of echolocation using ultrahigh frequency sounds occurs in two groups of bats (Yangochiroptera and Rhinolophoidea) and in toothed whales including dolphins [1]–[3]. These mammals use this complex bio-sonar system to assist with orientation and feeding [4], [5]. Echolocation by bats and dolphins provides an iconic example of either parallel or convergent evolution via natural selection.
Previous molecular studies on echolocation have mainly focused on the Organ of Corti. In this organ, the motor protein prestin plays a key role in voltage motility [6]–[8]. It appears to have undergone sequence convergence between bats and dolphins [1], [2], as well as within laryngeal echolocating bats [9]. Further, the voltage-gated potassium channel gene KCNQ4 underwent parallel evolution in echolocating bats [3], [10]. Mammalian audition requires not only voltage motility, but also hair bundle motility, which is executed by outer hair cells in the cochlea [11]. Proteins encoded by the genes Cdh23 and Pcdh15 are essential to hair bundle motility [12]–[14], and their malfunctions in humans cause deafness in newborns and progressive retinitis pigmentosa (Usher syndrome type I) [15]. Homodimers of Cdh23 and Pcdh15 directly link to each other via their amino termini; they form the upper and lower part of tip-links, respectively (Figure S1), which lie between the stereocilia within the hair bundle [14], [16], [17]. The auditory system involves the perception and enhancement of sound signals, as well as transformation of the mechanical signals to ion fluxes in inner hair cells. Management of the electric signals to the brain involves a series of nerve channel openings [18]. Genetic mutations in the gene encoding otoferlin (Otof) cause a clinical, autosomal recessive nonsyndromic form of prelingual and sensorineural deafness [19]–[21]. This protein that may act as the major Ca2+ sensor that triggers membrane fusion at the ribbon synapse of the auditory inner hair cell [22]. Although the above functions are involved in the conversion of sound signals into electrical impulses in the inner ear, the expression of Otof also occurs in neurons and nerve fibers in the brain [23].
The molecular mechanism of voltage motility in echolocation is widely studied. Echolocation is a complex system that includes signal reception by hair cells in the Organ of Corti, nerve transmission, and signal processing in the brain [24]. Therefore, herein we investigate the gene sequence evolution of Cdh23, Pcdh15, and Otof. These proteins function in different steps during echolocation. Because the brain modulates sensory information from peripheral sensory organs [25], and because Otof is involved in transferring sound signal by electrical impulses, we also examine the expression patterns of Otof in the cerebral cortexes of different species. We synthesize evidence from sequences and expressions to study the convergent evolution of echolocation in bats and dolphins.
We built ML, BI, and NJ trees based on both nucleotide and amino acid sequences. The length of aligned nucleotides for Cdh23 was 9657 base pairs (bp). The gene trees for Cdh23 based on nucleotide sequences (Figure 1A) were basically the same as the well-accepted species tree (Figure 1C) [26]–[28] in all methods of tree-building. Echolocating Hipposideros clustered with Old World (OW) fruit bats, which represented the Yinpterochiroptera. The genera Taphozous, Chaerephon, Miniopterus, and Myotis clustered together, forming the Yangochiroptera, the sister group of Yinpterochiroptera. However, the topologies based on amino acid sequences (Figure 1B) differed substantially from those based on nucleotide sequences. All echolocators incorrectly grouped together and then became the sister-group of the OW fruit bats, which do not possess the ability of laryngeal echolocation. The topology of the tree based on synonymous sites of Cdh23 was nearly the same as that based on nucleotide sequences, as well as the species-tree. The NJ tree using nonsynonymous changes was congruent with the amino acid tree (Figure 1B).
For Pcdh15, the aligned length was 5835 bp. The tree based on its nucleotide sequences (Figure 2A) depicted the well-accepted species tree (Figure 2C) and the nodes received high bootstrap values. In contrast, the amino acid trees (Figure 2B) clustered all echolocating bats together, and this arrangement differed from the nucleotide trees. As with Cdh23, the topology of the tree based on synonymous sites was virtually identical to that based on nucleotide sequences, while the tree based on the nonsynonymous changes was congruent with the amino acid tree (Figure 2B).
The sequenced coding region of Otof varied from 5363 to 5645 bp. As common in the Yinpterochiroptera, a 20 bp deletion occurred from bp site 1248 to 1267. The trees for Otof showed patterns similar to those of the previous two genes; the amino acid trees and NJ topology for nonsynonymous sites incorrectly clustered all echolocators (Figure 3B), and this association conflicted with the nucleotide trees (Figure 3A). Again, the nucleotide trees were consistent with the species tree (Figure 3C).
Branches b, d, and g in the species tree (Figure 1C, Figure 2C, Figure 3C) lead to mammals that have the ability to echolocate. Branch b represented the common ancestor of the Rhinolophoidea, d of the Yangochiroptera, and g of the dolphin. We reconstructed ancestral nodes and mapped amino acid changes along these branches. For Cdh23, branches b, d, and g shared one amino acid change (R204Q). Branches b and g shared the following 21 parallel mutations: R204Q, D517N, P518A, S639N, N737S, S747T, A1080S, K1141T, S1314T, A1382S, I1673V, N1697D, L1960F, L1974I, A2146V, G2229S, V2427I, T2439R, R2639K, Q2725L, and N3180S. Parallel evolution of these two branches statistically differed from random expectations (P<0.001). Branches b and d had four parallel mutations (R204Q, R535K, T904I, and V1691I). Again, parallel evolution was statistically significant (P<0.001). All changed sites were mapped in Figure 1C. The positions of these sites in the domain structure of Cdh23 were mapped in Figure S2. We constructed a BI tree from the aligned amino acids of Cdh23 while excluding all parallel-evolved sites. The BI tree agreed with the species tree (Figure S3).
Two parallel mutations in Pcdh15 (I946M and E1278D) were shared by branches b, d, and g. Parallel evolution was statistically significant (P<0.001) between branches b and g for the following 22 parallel mutations: N218D, Q310E, E393V, T427S, A433V, I438V, T490I, V546F, I643V, N666K, K820R, I853V, K856T, M946I, V952A, R999L, T1139R, F1160L, A1173S, K1275R, D1278E, and I1404V. Parallel evolution was also statistically significant (P<0.001) between branches d and g for three parallel mutations: A726D, M946I, and D1278E. Finally, eight parallel mutations occurred between branches b and d, including L423V, Q468P, H765Y, F876L, M946I, F984S, V1019I, and D1278E (Figure 2C). Parallel evolution between these two branches was statistically significant (P<0.001). The positions of these sites in the domain structure of Pcdh15 were mapped in Figure S4. The BI tree based on amino acids excluding all parallel-evolved sites differed somewhat with the species tree, but it did not group echolocators together (Figure S5).
Otof had one amino acid change shared among echolocators (D396E). The four parallel-evolved sites along branches b and g were P191H, G213A, D396E, and V440M (Figure 3C), and parallel evolution was statistically significant (P<0.001) between these two branches. Parallel evolution of branch d and g was also significant (P<0.001). The conserved sites were shown in Figure S6. The BI tree based on amino acids while excluding all parallel-evolved sites agreed with the species tree (Figure S7).
Selective pressure was evaluated using the PAML package and test results were presented in Table 1 and Table S1. The one-ratio model obtained an average ω (Ka/Ks ratio) of 0.0546 (lnL = −35796.7768) for Cdh23. For specific branches, we alternatively set the echolocating bats (Rhinolophoidea and Yangochiroptera; branches b and d in Figure 1C, respectively) or dolphin (branch g in Figure 1C) as foreground branch. In both conditions, although the ratios of ω for the foreground branches were greater than background branches, they were less than 1 (ωbranch (b+d) = 0.1107 while ω0 = 0.0500; ωbranch g = 0.1145 while ω0 = 0.0516). When considering all echolocators together, ωecholocators was 0.1126 for the foreground branch compared with 0.0465 for background branch (2Δl = 660.6044, df = 1, P<0.001). For the branch-site models, test 2 (Model A vs. null model) was used to control the false positive signals. Branch b was detected to have undergone significant positive selection (P<0.01), and eight sites with BEB values >0.90 (213 A 0.984, 692 F 0.984, 1165 N 0.989, 1171 S 0.958, 1256 D 0.989, 1356 I 0.990, 1687 T 0.925, and 2492 L 0.970). When we set the dolphin as the foreground branch (branch g), significantly positive selection was also detected (2Δl = 10.7224, df = 1, P<0.01), and six sites (580 S 0.974, 840 H 0.973, 1011 H 0.973, 1014 T 0.978, 2133 T 0.952, and 2224 T 0.974) with BEB values >0.90. When all echolocating species were combined as the foreground branch, again significant positive selection signals were obtained (2Δl = 8.6179, df = 1, P<0.01). The positions of these positively selected sites were mapped in Figure S2.
For Pcdh15, we implemented the same series of analysis as for Cdh23. All branch models were significantly better than the null model that fixed ω of the foreground branch to 1, although the value of ω never exceeded 1 (ωbranch (b+d) = 0.3211, ω0 = 0.0998, 2Δl = 83.0381, df = 1, P<0.001; ωbranch g = 0.4469, ω0 = 0.1109, 2Δl = 16.6620, df = 1, P<0.001; and ωecholocators = 0.3561, ω0 = 0.0851, 2Δl = 96.3532, df = 1, P<0.001). For the branch-site models, a significant signal of positive selection was detected on branch d and one site had a BEB value >0.90 (P<0.05). The position of this positively selected site was mapped in Figure S4.
For Otof, the ω ratios in branch models were greater than ratios of the background branches, but less than 1 (ωbranch (b+d) = 0.0572, ω0 = 0.0364; ωbranch g = 0.0886, ω0 = 0.0352; and ωecholocators = 0.0732 with ω0 = 0.0340). Branch-site models did not detect any signals of positive selection in the echolocators (Table 1).
Real-Time PCR was used to assess expression patterns of Otof in the brain. We set the mean value of gene expression in the cerebellum of the adult Common Bent-wing Bat (Miniopterus schreibersii) as the baseline unit (marked * in Figure 4), and then compared expression patterns of Otof in different cortexes of the brain (Figure 4 and Table S2). The level of expression in the auditory cortex was more than 70-fold that of the baseline value. The levels of expression from the visual cortex, and motor and sensory cortex were more than 40-fold and 30-fold greater than the baseline value, respectively, whereas the expression in the olfactory bulb was nearly 17-fold greater. We compared this expression pattern with that of embryos. Expression levels of Otof in the auditory cortexes of three embryonic Common Bent-wing Bats were nearly 13-fold the baseline value and the visual cortex was over 3-fold. In contrast, expression levels in motor and sensory cortex, olfactory bulb, and cerebellum were similar to the baseline values. In adult Old World Fruit Bats (Rousettus leschenaultii), which do not echolocate, expression levels in the auditory cortex, visual cortex, and motor and sensory cortex were all around 3-fold greater than the baseline value, and expression levels in the olfactory bulb and cerebellum were less than the baseline value.
Morphological development is very complex and involves a suite of genes. Some mammals have independently developed similar features, such as the ability to echolocate objects. Echolocation by bats and dolphins provides an extreme example of parallel or convergent evolution. Although the morphologies of sending and receiving sonic signals differ greatly, the hearing of ultrasonic sounds and the mechanisms of decoding signals are shared [29], [30]. Thus, the genes coding for the auditory system become ideal candidates for adaptive evolution at the molecular level during echolocation.
The genes Cdh23 and Pcdh15 are essential to hair bundle motility [12]–[14], [31], [32]. Otof encodes a protein that may act as the major Ca2+ sensor to trigger membrane fusion at the auditory inner hair cell ribbon synapse [22]. The three genes are involved in different steps of the hearing system. Their nucleotide gene trees are largely congruent with the species tree and the nodes enjoy high support. In contrast, the amino acid trees conflict with the species tree; all unite the echolocators. Further, trees based on nonsynonymous sites have topologies similar to those based on corresponding amino acids, and yet the trees based on synonymous mutations do not show this pattern. Clearly, the difference in branching order is the result of amino acid changes (nonsynonymous mutations).
Three independent lineages of mammals can echolocate and it is important to identify whether this involves convergent or parallel evolution. If echolocation involves convergent evolution, then similar traits or functions independently emerge in two or more lineages from different ancestral states. Parallel evolution differs in that similar ancestral traits descend into similar extant states in different lineages [33].
Reconstructions of the ancestral sequences of the internal nodes can detect which amino acid changes cause the incongruence between the amino acid and species trees. A suite of sites show parallel changes in these echolocators, and these changes have a statistically significant signal of parallel evolution. Twenty-one amino acid sites converge on the same residue in branch g and branch b and with a highly significant probability (P<0.001). One convergent site appears in branches g and d (P<0.001), and four parallel sites occur in branches b and d (P<0.001) in Cdh23. In Pcdh15, 22 and three convergent sites occur between branch g and branches b and d, respectively (P<0.001 in both cases) and eight parallel sites have been identified between branches b and d (P<0.001). Four parallel sites in Otof occur on branches g and b (P<0.001). Convergent evolution is not accommodated by current phylogenetic methods and it can strongly mislead phylogenetic inference [34]. Upon excluding the parallel-evolved amino acid sites, the reconstructed amino acid trees did not incorrectly unite all echolocators (Figures S3, S5, S7). Thus, the discovery of multiple parallel-evolving amino acid sites explains the unnatural uniting of echolocators in the amino acid trees.
In addition to the parallel evolution of Prestin [1], [2], [9] and KCNQ4 [3], our analyses document a high level of complexity in a large-scale, multigene adaptation. Functional assays have proven that parallel and convergent amino acid changes are responsible for parallel and convergent functional changes [35], [36]. Thus, parallel evolution of gene sequences may have driven phenotypic and functional convergence in echolocating bats and dolphins. Further functional assays are needed to affirm this association.
Our analyses of selection pressure detect signals of positive selection in Cdh23 on the branches leading to the Rhinolophoidea and dolphin. The same occurs in Pcdh15 along the branch leading to Yinpterochiroptera (see Table 1 and Table S1). Positive selection appears to have acted on these genes to fit the requirements for echolocation. Whereas outer hair cells amplify sound by somatic [37], [38] or ciliary [39], [40] mechanisms, inner hair cells are passive detectors of the amplified vibratory signal. The signals are converted into electrical impulses by activating fibers of the cochlear (auditory) nerve [41], which then sends the signals to higher auditory processing centers in the brain. Prestin mediates the voltage somatic motility unique to mammals [42]–[44]. Part of the cochlear amplification, Cdh23 and Pcdh15 participate in hair bundle motility [39], [45], [46]. Otof participates by releasing neurotransmitter to nerves [22]. Convergent evolution and positive selection on these genes reflect the pathway from receipt of signal to signal amplification, and then to neural transduction. All parts of the pathway are involved in the auditory system, and thus may play important roles in echolocation.
Homodimers of Cdh23 and Pcdh15 directly link to each other via their amino termini, and they constitute the upper and lower part of tip-links, respectively, that lie between the stereocilia within the hair bundle [14], [16], [17]. Sequence alignments suggest that Cdh23 and Pcdh15 have 27 and 11 extracellular cadherin repeats, respectively [47]. Functional evidence encompassing a classical genetic approach shows that mutations at these two cadherin proteins can interact to cause hearing loss in digenic heterozygotes of both mice and humans [48]. Cdh23 and Pcdh15 appear to have undergone both convergent functional evolution and positive selection in echolocators. This finding suggests a strong interaction between the proteins and co-evolution is likely to have optimized their function in cochlear amplification for the development of echolocation.
Dolphins and bats employ different tools to echolocate. Whereas the larynx generates sound in echolocating bats [29], the monkey lip/dorsal bursa complex does the same in dolphins [30]. In bats, the sound involves a constant frequency (CF) and frequency modulation (FM), but dolphins use FM and amplitude modulation (AM) [49]. Convergence occurs via both auditory systems being adapted to receiving and processing ultrahigh frequency sounds. Our study discovers evidence of parallel sequence evolution in three genes involved in hearing and this may indicate a genetic basis for echolocation.
Sequence evolution only tells a part of the story. Expression pattern is usually a strong indicator of protein-demand and function [50], [51]. The central auditory system plays the crucial role of receiving impulses from auditory nerves and sending messages back to the cochlea [52]. Because Otof plays a role in neural signal transmission, we have evaluated its patterns of expression in different cerebral cortexes. Otof is most highly expressed in the auditory cortex of echolocating adult female Common Bent-Wing Bats. Adult females, which frequently use ultrasonic sounds to explore their environments, have higher levels of expression than their embryos, which do not use echolocation. Further, a comparison of expression of Otof between the brains of an adult echolocating bat (Common Bent-Wing Bat, Miniopterus schreibersii) and a non-echolocating bat (Old World Fruit Bat, Rousettus leschenaultii) reveals a higher level of expression in the former. Indeed, the gene exhibits great differences in levels of expression in both different cerebral cortexes and in species with or without the ability to echolocate.
Echolocation signals begin at the hair cells in the Organ of Corti, continue along the auditory nerve, and terminate in the auditory cortex of the brain [24]. Combined with sequence data and expression data, we conclude that multiple instances of parallel sequence evolution are involved in genes in different parts of auditory system between the three groups of echolocators. This occurs not only in well-studied voltage motility, but also bundle motility, and not only in cochlear amplification, but also in neural transduction. Further, co-evolution optimizes the function of homodimers of Cdh23 and Pcdh15 in cochlear amplification. The expression pattern of Otof in different cerebral cortexes implies that the evolution of gene expression might be required for echolocation. In conclusion, we synthesize gene sequence and gene expression analyses and conclude that positive selection, convergent evolution, and perhaps co-evolution and gene expression evolution play roles in audition (voltage motility and bundle motility in cochlear amplification, nerve transmission, and brain) during the independent origins of echolocation in bats and dolphins.
All research involving animals used in this study followed the guidelines and bylaws on animal experimentation. Bats were anesthetized using an intraperitoneal injection of sodium pentobarbital (C11H17N2NaO3) at a dosage of 100 mg per kg body weight. Following anesthesia, bats were euthanized and then their brains were sampled. Protocols were approved by the Ethics and Experimental Animal Committee of the Kunming Institute of Zoology, Chinese Academy of Sciences.
Sixteen species of bats were used in our analysis (Table S3). Total RNA was isolated from the brain using a RNAiso Plus Kit (Takara, China), and RT-PCR was performed on 2 ug of RNA using the PrimeScript RT-PCR Kit (Takara, China) to obtain cDNA. Subsequently, genes were amplified from cDNA using gene-specific primers (Table S4). PCR products were purified on a 1% agarose gel and a Watson Gel Purification Kit (Watson BioTechnology, Shanghai), and finally transformed into the pMD18-T vector (Takara, China). Each strand was sequenced in both directions with an ABI 3730 sequencer. RNA samples of brain were not available for the dolphin (Tursiops truncatus), so genomic sequences were amplified from total genomic DNA, which was extracted from muscle tissue using a standard 3-step phenol/chloroform extraction method [53].
Raw nucleotide sequences were edited using Lasergene SeqMan software (DNASTAR Inc., Madison, WI, USA). Newly determined sequences were deposited in GenBank (Accession numbers JF808081–JF808094, JQ284400–JQ284430). The sequences of background species came from the Ensembl database (Release 66) and those of high quality were used (Table S3). All the sequences were aligned using ClustalX 1.81 [54] and then visually checked for accuracy (the aligned sequences are available by request).
The best-fit models were selected by jModeltest v0.1.1 [55], [56] for nucleotide sequences and ProtTest 3.0 beta [57] for amino acid sequences. Maximum likelihood (ML) trees were reconstructed by PAUP* [58] with 1,000 replications, and Bayesian inference (BI) trees were reconstructed by MrBayes 3.1.2 with 1,000,000 replications [59], [60].
Neighbor-joining (NJ) phenograms were based on Kimura 2-parameter corrected distances for nucleotide sequences and uncorrected P-distances for amino acid sequences, each with 1,000 bootstrap replications. We implemented the Li-Wu-Luo method [61] to reconstruct NJ trees based on both synonymous and nonsynonymous sites.
The sequences of the internal nodes were reconstructed using distance-based Bayesian methods, which included the branch lengths estimated by the least squares method and the ancestral amino acids inferred by the Bayesian approach. These data were used to obtain an unbiased estimate of the true probability [62]. Convergent and parallel amino acid substitutions along each lineage were detected. The statistical significance of these amino acid changes was tested with the method developed by Zhang and Kumar [63].
The CODEML program in PAML 4 [64] was used to detect selective pressure. The species tree [26]–[28] was used as guide tree for analysis. Four models of evolution were used: one-ratio model, free-ratio model, branch models, and branch-site models. For the last two models, three groups of echolocators were set as the foreground branch to detect whether or not they had undergone positive selection.
We sampled auditory cortex, visual cortex, motor and sensory cortex, olfactory bulb, and cerebellum from euthanized adult and embryonic Miniopterus schreibersii, and adult Rousettus leschenaultii, which represented the adult echolocating bats, embryonic echolocating bats, and adult non-echolocating bats, respectively. Tissue was selected according the human brain atlas [65]. These tissues were stored in liquid nitrogen. Total RNA was isolated using a RNAiso Plus kit (Takara, China). Next, DNA-free RNA samples were condensed using a RNAqueous-4PCR Kit (Applied Biosystems, US). We synthesized cDNA using a PrimeScript RT-PCR Kit (Takara, China), and then used it in Real-Time PCR. TaqMan Gene Expression Assays were custom designed by Applied Biosystems based on our sequencing data of Otof and Actb in bats. The assay details were listed in Table S5. Sequences of Otof and Actb were amplified and detected using an ABI PRISM 7000 Sequence Detection System with a PCR profile as follows: 50°C for 2 min, 95°C for 10 min, followed by 40 cycles at 95°C for 15 s, and 60°C for 1 min. The products were purified and sequenced in both directions with an ABI 3730 sequencer to insure the assays' specificity. Real-Time PCR was performed on 96-well reaction plates in a 20 µl reaction volume containing 100 ng of cDNA per reaction with TaqMan Universal Master Mix II (Applied Biosystems, US). For each group, three individuals were used and each sample was performed three times. Expression data for the target gene Otof were normalized relative to the housekeeping gene Actb. Raw data were obtained and analyzed using the 7000 SDS 1.1 software (Applied Biosystems, US). The comparative CT method (ΔΔCt) was chosen to further calculate the relative expressions between different groups.
|
10.1371/journal.pgen.1000949 | Transposed Genes in Arabidopsis Are Often Associated with Flanking Repeats | Much of the eukaryotic genome is known to be mobile, largely due to the movement of transposons and other parasitic elements. Recent work in plants and Drosophila suggests that mobility is also a feature of many nontransposon genes and gene families. Indeed, analysis of the Arabidopsis genome suggested that as many as half of all genes had moved to unlinked positions since Arabidopsis diverged from papaya roughly 72 million years ago, and that these mobile genes tend to fall into distinct gene families. However, the mechanism by which single gene transposition occurred was not deduced. By comparing two closely related species, Arabidopsis thaliana and Arabidopsis lyrata, we sought to determine the nature of gene transposition in Arabidopsis. We found that certain categories of genes are much more likely to have transposed than others, and that many of these transposed genes are flanked by direct repeat sequence that was homologous to sequence within the orthologous target site in A. lyrata and which was predominantly genic in identity. We suggest that intrachromosomal recombination between tandemly duplicated sequences, and subsequent insertion of the circular product, is the predominant mechanism of gene transposition.
| Repetitive DNA, such as satellite repeats and transposons, is ubiquitous throughout the genome. Such repeats have been associated with DNA loss, circle formation, and gene transposition in plants and Drosophila. In this work we suggest that, in plants, one mechanism of gene mobility is intrachromosomal recombination via tandem repeats. In addition, we have demonstrated that the classes of genes that tend to form tandem duplications are more likely to have transposed than other gene classes. We conclude that tandem duplications may particularly facilitate gene excision and may also provide targets for gene insertion.
| Much of the eukaryotic genome is known to be mobile. This is a characteristic feature of transposable elements, and a large proportion of many eukaryotic genomes are composed of these parasitic elements. However, recent work in plants [1], [2] and Drosophila [3] demonstrates that mobility is also a feature of many non-transposon genes and gene families. An analysis of the Arabidopsis genome suggests that as many as half of all genes had moved to unlinked positions since Arabidopsis diverged from papaya roughly 72 million years ago [4], and that these mobile genes tended to fall into distinct gene families [2]. With the exceptions of unannotated transposons and retroposed genes, the exact mechanism by which single-gene transposition occurred was not deduced, though potential mechanisms include transposon-mediated transduplication or “highjacking” [5], recombination between repeated sequences [6], or nonhomologous end-joining of double-stranded breaks [7]. Unfortunately, ancient gene transposition events, such as those detected in Arabidopsis to date, are an unlikely source of clues because random mutation would be expected to erode all evidence of the mechanism of transposition. In order to detect such evidence, we examined more recent gene transposition events by comparing two relatively closely related (∼5MYA, [8]) Arabidopsis species, A. thaliana and A. lyrata. In this way we were able to identify a large number of recently transposed A. thaliana genes. We found that flanking direct repeats were associated with nearly half of these transposed genes, indicating that these repeats have a role in the process of gene transposition.
In order to detect recently transposed nontransposon genes in A. thaliana, we used a semi-automated procedure (Methods). Briefly, we automated a procedure we call the flanking gene method, which compares the location of two sequential genes in a region orthologous between two species such that, given genes A and C, if gene B is present between the two genes in one species but not the other, gene B is denoted as a possible transposed gene (as previously described in [2]). This method can identify genes that are present at a given position in A. thaliana, but absent at the orthologous position in A. lyrata. In order to distinguish between a gain in A. thaliana and a loss in A. lyrata, each region was compared to the orthologous regions in Carica papaya (papaya) and Vitis vinefera (grape), two more distantly related species in the rosid clade (Figure 1) (Methods). The absence of a given gene at the syntenous position in both of the outgroups and in A. lyrata was interpreted as evidence for insertion in A. thaliana; if it was not possible to substantiate the status of the candidate in the outgroup—for example because its expected position in the outgroup was in an unsequenced region—that particular candidate was excluded. We found a total of 420 genes that were present at a given position in A. thaliana and absent at the expected position in A. lyrata. Analysis using the two outgroup species suggested that that 226 of these genes were new insertions in A. thaliana (Table S1). This figure is most conservative, as our methods purposefully discarded questionable data.
As had been observed previously [2], we found that certain gene families are much more likely to have transposed than others (Table 1). Other than genes that encode unknown proteins, F-box genes were the most common class of transposed genes (6.2%), followed by MADS/AGL genes and LRR-type disease resistance genes (3.5% each), then defensins (2.7%). Defensins, due to their small size and rapidly changing sequences, were mostly undetected in C. papaya. Similarly, the same transcription factor gene families that, according to [2], tend not to transpose in the period following the divergence of A. thaliana and C. papaya (e.g. TF-GRAS genes, WRKY genes, WD40 genes, and GERMINS), are not found at all within our list of 226 genes that had transposed following the divergence of the A. thaliana and A. lyrata lineages. Our data suggests that, correcting for divergence time, the rate (in duplication-transpositions/MY) of gene transposition detected when comparing A. thaliana and A. lyrata (5 MYA) is roughly the same as the rate observed when comparing A. thaliana and C. papaya (72 MYA) [2]. We can calculate that, since the divergence of A. thaliana and A. lyrata, gene transposition has occurred approximately once every 22,000 years (226 transposed genes/5 MYA). Altogether, these data suggest that the more recent gene transposition events we detect in A. thaliana are representative of transpositions that have been occurring over the past 72 million years in the Arabidopsis lineage.
To examine the nature of these gene transpositions in closer detail, we looked for evidence of recent transposition events from a parental or source position. We focused on transposition events that were not the result of retroposition, but rather DNA-level recombination. These are distinguished as transposed sequences that contain intron and/or non-coding flanking sequences that exist in their parental copy. Assuming that gene transposition happens continuously over time, we expect that a recently transposed gene would retain noncoding sequence similarity to its parent if the parent still existed in the genome. We would also expect that only half of all transposed genes would have remaining donor sites within a given genome if the donor site and the transposed gene were unlinked, as, once transposition occurs, the donor site is then heterozygous for gene in question, and the donor site may be lost via segregation. However, the ratio of identifiable donor sites may be even less than half if the transposition event in question had been relatively ancient, and the noncoding sequences for the donor site and transposed gene had significantly diverged. We felt that the most recent transposition events would also retain evidence of their mechanism of transposition, and we focused on those. To do this, we looked for an unlinked paralog that had sequence similarity with the transposed genes higher than 75% identity across at least 50 base pairs of noncoding sequence. These criteria were used in order to restrict the level of detection to genes that had only recently transposed. Of the 126 transposed genes we examined manually (excluding the “unknown” genes), 106 of our transposed genes did not have a best hit with noncoding sequence similarity greater than or equal to 75%/50 bp. However, 25 (19.8%) had a best hit whose noncoding sequence fit the above criteria, consistent with being relatively recent transposition events, and making them candidate source genes (Table 2, Table S2). In 60% (15/25) of such cases, the parental gene was in a position syntenous in A. lyrata, suggesting that the parental gene itself had not transposed from another position within the last 5 MY. As a control, we also examined 102 genes that had not transposed since the divergence of A. thaliana and A. lyrata (Methods). None of these genes had a best hit whose noncoding sequence similarity was above 75% identity over 50 bp (Table 2, Table S3).
Next, we examined our transposed genes for signatures of the transposition mechanism. In particular, we searched for the presence of direct repeats flanking the transposed sequence because such repeats have already been shown to be associated with indels in Arabidopsis (though the absence of an outgroup prevented the distinction between insertion and deletions) [9]. In addition, whole-gene transposition in Drosophila [3] has also been associated with direct repeats of highly repetitive DNA. To look for flanking direct repeats around our transposed genes, we used the genome visualization platform GEvo to visually compare the 5′ region ∼500 bp upstream of our target sequence to the sequence ∼500 bp downstream of its 3′ region (Methods). We limited our search to BLAST hits that were greater than or equal to the e-value of a 15/15 bp exact match, and excluded simple sequences. Using these criteria, 17% (22/126) of our total transposed genes had flanking repeats greater than 15 bp (Table 3). However, when we enriched for transposed genes that had an identifiable parental site, 44% (11/25) had flanking direct repeats equal to or greater than 15 bp in length. In contrast, only 5.9% (6/102) of the control, nontransposed genes had flanking repeats greater than 15 bp (Table 3) (p-value 0.00025). The difference was even more dramatic for longer repeats: 36% (9/25) of parental-site transposed genes had flanking repeat sequence over 30 bp in length, versus only 2% (2/102) of the control genes (p-value 0.000051).
Upon closer examination of the nine transposed genes whose flanking repeats were greater than 30 bp, we found that in six cases, the flanking repeat sequence was detected at the parent site at least once (Table 4, Figure 2, Figure S1). Repeat carryover from a parent sequence had been associated with transposed genes in Drosophila [3], suggesting a flanking-repeat excision model. Two of our nine long-direct-repeat genes had inverted repeats associated with them as well as direct repeats (AT5G10330 and AT1G49715, Table 4). Inverted repeats are a hallmark of DNA transposons that are known to “highjack” foreign DNA sequence [5]. In fact, one of our transposed genes in this study, AT1G49715 was a PACK MULE, with characteristically long terminal inverted repeats flanked by 10-bp target-site duplications. Of the unknown genes excluded from this study, we found at least five genes that had clearly been captured by a transposon-like mechanism, based on the fact that were not transcribed, only a portion of them transposed from their putative donor site, and they were flanked by either terminal inverted repeats (TIRs) or long terminal repeats (LTRs) (data not shown). In short, we were able to detect transposons and transposons with genic insertions, so these did not confuse our study of the transposition of more typical genes.
Unexpectedly, in five of the transposed genes with long flanking repeats, the sequence of the repeat could be found at the orthologous site in A. lyrata (AT2G34290, AT2G26010, AT3G10845, AT4G01640, and AT2G16930, Table 4). This A. lyrata sequence could well correspond to the recombination site for the insertion in A. thaliana. In addition, of these five genes, three (AT2G34290, AT2G26010, and AT5G01080) were flanked by repeat sequence that could also be found at the parental site at least once, as though this repeat sequence were a source for intrachromosomal recombination out of the site of origin as well as recombination into the target site orthologous to the A. lyrata homologous sequence (as illustrated in Figure 3). Notably, the repeats surrounding two of these genes (AT2G26010 and AT5G01080) were chimeric between the original flanking sequence and the A. lyrata homeologous target sequence, as demonstrated in Figure 3B. Additionally, among the transposed genes in our study that did not have an identifiable parent source or donor site, nine of these unpaired, transposed genes also had flanking repeat sequence greater than 30 bp in length within 500 bp of the coding region. In eight of these cases, the flanking repeat sequence was found at the orthologous target region in A. lyrata (Table 5). Therefore, among the eighteen total transposed genes that had flanking repeats over 30 bp in length, thirteen (72%) of them had repeat sequence that corresponded to the orthologous site in A. lyrata.
Previous work in Drosophila found that the sequence identity of repeats surrounding transposed genes usually comprised transposon sequence [3], and an argument was made suggesting that transposon sequence distributed randomly within the genome could facilitate gene transposition by ectopic recombination. However, when we examined the sequences of the repeats flanking our transposed genes, we generally found these repeats to be made up of host non-transposon sequence of varying kinds, including what appear to be the remnants of fractionated genes (Table 4, Table 5).
We decided to examine all genes in the A. thaliana genome to determine whether the gene classes that tended to transpose were more likely to be flanked by direct repeats, and if so, what the identity of the sequence of those direct repeats were. We made a prediction that the gene classes that tended to transpose would also tend to be flanked by direct repeats; such flanking repeats would endow these gene classes with a propensity for excision and insertion via ectopic recombination.
Using an automated blast search with the parameters described in Methods, we retrieved 1088 genes, including pseudogenes, that were identified as having flanking direct repeats according to the algorithm. These included genes that had not transposed in the 5 MY since the A. thaliana/A. lyrata split. When we examined the classes of genes that tended to have these flanking repeats, we found that, aside from unknown genes, F-box genes were the highest represented as being flanked by direct repeats (4.5%,) closely followed by pseudogenes, then defensins and LRR genes (2.8% and 2.2%, respectively) (Table 6). These percentages are proportional to the percentages that these gene families make up in transposed genes (Table 1). Most notably, these three gene classes are also those that tend to either form, or insert into (interrupt), tandem duplications. Indeed, in 80% of the cases examined, the sequence identity of the flanking repeats is genic, and in the cases of F-box genes and LRR genes, the sequence is similar to the gene itself 40% of the time, as though the flanking repeats are remnants of a tandem duplication.
Repeat sequences in general are unstable regions of the genome due to their potential for unequal crossing over, intrachromosomal crossing over, and ectopic recombination [10], [11]. Intrachromosomal recombination between flanking repeats—the sort of event that generates a circular fragment—has been associated in plants and Drosophila with small deletions within transposons. Flanking repeats have also been associated with indels in Arabidopsis, but insertions were not distinguished from deletions [9]. Yang et al. have shown that whole-gene transpositions in Drosophila are primarily associated with 44- to 433-bp flanking repeats of transposable element sequences [3]. Our results in Arabidopsis are similar to those in Drosophila, but our flanking repeats generally derive from host non-transposon sequence.
Forty-four percent of our transposed genes that had an identifiable donor site had (detectable by our Methods) flanking repeats at their insertion site, and thirty-six percent had flanking repeats greater than 30 bp in length. Most transposed genes (with or without an identifiable donor site) with repeats greater than 30 bp were flanked by direct repeat sequence that was homologous to sequence within the orthologous target site in A. lyrata. One explanation for this pattern might be that the target site was duplicated following the insertion in A. thaliana; such long target-site duplications have been observed in H. pylori genomes after gene transposition [12], and in humans due to retroviral insertion [13]. However, there are simpler explanations, as follows.
In the majority of cases, the repeat sequence flanking the transposed gene consisted of non-transposon host sequence, particularly sequence that corresponds to genic sequence, suggesting that the insertion site had once been part of a tandem repeat, or that the transposed gene had originally resided in a tandem repeat; indeed, the parental gene for three of the nine donor-site transposed genes was part of a tandem duplication (Table 4). Previous work in our lab has demonstrated that mobile genes—transposon and nontransposon alike—in Arabidopsis are often found within tandem repeats from unrelated families [2]; these were called “interruptor” genes. In addition, genes that tend to have duplicated in tandem—such as F-box genes, DEFENSINS, and NB-LRR-type disease-resistance genes—also tend to be those that transpose [2], [14]. Findings by Zhou and coworkers suggested that, in Drosophila, tandem repeats may precede DNA transposition [15]. Other researchers in Drosophila had found the presence of chimeric genes within tandem arrays, and postulated that the mechanism may be via large-loop mismatch repair, where a portion of the DNA sequence between duplicated genes is excised [16]. Further, Cohen and coworkers have demonstrated that tandem duplications are associated with the formation of circles in plants [17] and Drosophila [18], where it is thought that a gene that is part of a tandem duplication can excise by intrachromosomal recombination via the repetitive sequence surrounding it, then potentially integrate into a new region containing sequence homologous to the flanking repeats. In fact, in three of our transposed genes (AT2G34290, AT5G15220, and AT2G26010), a portion of the flanking sequence was found in both the A. lyrata orthologous target sequence and the parental sequence, suggesting the possibility that this same repeat sequence facilitated intrachromosomal recombination and excision from the parental site, then insertion into the homologous target site. Alternatively, ectopic recombination between unlinked sites (say, between one tandem array of a given gene type and another tandem array of the same or similar gene type) may play a role in gene movement; indeed, one hypothesis for the formation of chimeric disease-resistance genes in plants is via ectopic recombination [19].
In over half the cases where a transposed gene had an identifiable parent, no evidence of repeats were found flanking the transposed gene. There are several explanations for this. For instance, the parameters of our experimental design (15/15 exact match) would rule out repeats that were smaller than 15 base-pairs, or repeats that had degenerated beyond what our algorithm could detect. Indeed, in some cases, when we more closely examined the sequence flanking some of the transposed genes which, according to our parameters, were deemed devoid of repeats, we did observe the traces of direct repeats whose sequence similarity would not have been detected by our criteria (data not shown). This is also observed at the parent sites for our transposed genes. In other cases, the donor site contained only one copy of the flanking repeat found at the transposed site. Again, degeneration of the second repeat might have occurred, or the repeat might have been removed entirely via a deletion event [9], [10]. Alternately, recombination with the homologous sequence at the new site might have created chimeric flanking sequences (as illustrated in Figure 2 and Figure 3B). For instance, evidence of flanking direct repeats that are chimeric between the target sequence and the donor repeat sequence is observed in AT2G26010, and AT5G01080, as previously discussed. Particularly after sequence divergence, the original flanking sequence in some instances may not manifest as identifiable direct repeats surrounding the transposed sequence.
Tandemly repeated genes and genes flanked by tandem repeats are a special case because they are regions that are particularly prone to intrachromosomal recombination and the associated circular fragment. As diagrammed in Figure 3, if a sequence in a circular fragment is homologous to a sequence elsewhere in the genome, insertions resulting in flanking repeats are possible. However, reinsertion is likely to be an exceptional event. In most cases, we would expect that the excised circle would simply be lost and thus, these recombination events would result in a net reduction of genomic DNA content. We suggest that this process is the way plants [10] and Drosophila [20] have countered transposon buildup. The gene loss mechanism in plants and other clades able to carry high mutational load is likely to involve the generation of circular fragments of chromosome, and, thus, the potential for transposition.
We calculate the rate of gene transposition as being once every 22,000 years, and hypothesize that the rate of transposition is steady and not punctuated, as the degree of sequence similarity in noncoding sequence between donor and transposed sites varies from 100% identity over all noncoding sequence to 70% identity/50 bp at the very most (Table S2), suggesting that transposition has occurred at different points throughout evolutionary time. This is also consistent with our observation that gene transposition following the divergence of A. lyrata from A. thaliana has occurred at roughly the same rate as it had occurred since following the divergence of A. thaliana from papaya.
Repeats, from a few base-pairs in length to hundreds of base-pairs, are ubiquitous throughout the genome. Via ectopic recombination, these repeats permit the deletion or insertion of a fragment of DNA as small as a few base-pairs, or as large as an entire gene. It seems likely that the genome's potential to re-shape itself in novel ways is built in, not via any special mechanism or adaptation, but as a passive by-product (a spandrel) of the genome's architecture. Obviously the presence of thousands of transposable elements within any genome is a source of recombination; but tandem duplications clearly can play a role as well, particularly at the genic level. Indeed, the fact that many of the repeats surrounding transposed genes in Arabidopsis were associated with what appear to be duplicated gene fragments, and the fact that transposed genes tend to not only form tandem repeats themselves, but insert into tandem repeats, suggests that tandem duplications may particularly facilitate gene excision as well as provide targets for gene insertion.
Automation: Genomes of A. thaliana (TAIR masked repeats 50x, v7) and A. lyrata (JGI unmasked, v1, sequenced by the Weigel lab http://www.phytozome.net/alyrata) are co-annotated to each other, so that features (files of the annotations used are available here http://syntelog.com/t/gray_paper_methods/) that are annotated in one but not the other become new features in their respective genomes (this co-annotation step is performed by software available from our source-code repository: http://bpbio.googlecode.com/svn/trunk/co-anno). New fasta files are created from the original genomic features and the new ones discovered via co-annotation. The fasta file from A. thaliana is blasted against the fasta file from A. lyrata using blastn/e-value of 0.1/word size 7. The blast is then “dissolved” by combining small hits to the same gene pairs into single hits, then keeping only those merged hits with a sum length greater than or equal to 96.
The search for transposed genes is performed using the aforementioned dissolved pairs, such that for each gene in a pair, the algorithm is extended out three genes in either direction, creating a list of (gene +3+3)*2 = 14 genes; next, for that group of genes, the flanking gene method is used to find transposed genes in both query and subject by converting gene names to integer positions to get a list of query-subject pairs; for instance: [(1, 123), (2, 125)]. From this, simple addition and matching is used to find any genes that are “missing,” e.g. the gene at position 124 above is unaccounted for, and flanked by consecutive genes 1, 2 from the ortholog. The integer positions are then converted back to gene names, and the transposed gene and the orthologs of its flankers are reported.
Each putative transposed gene is then checked to see whether it lies within a region in a manually generated list of orthologous regions, since a genes in non-orthologous regions cannot be verified as transposed they are discarded. Each putative transposed gene is then blasted against the non-coding sequence in the orthologous flanking region to determine whether it actually appears in the ancestral position even though it is not annotated as such. No BLAST to the transposed gene over 15 bp can be observed between the flankers in A. lyrata, otherwise it might just be gene loss in A. lyrata (hits_within_0<15). Each putatively transposed gene must be flanked by two genes that are not identical to each other in A. thaliana such that A. lyrata flanker_1≠ flanker_2. If it fails this (e.g. A. lyrata flanker_1 = flanker_2) then it is designated an “interrupter” or “I” gene. Any pseudogene, or any gene that has been identified as repetitive or transposable element sequence of any kind by the Arabidopsis genome database TAIR (http://www.arabidopsis.org/), is removed from the list.
Proofing: Once the list of putative transposed genes is at hand, each gene is visualized in our gene visualization platform GEvo (http://synteny.cnr.berkeley.edu/CoGe/GEvo.pl) via a link provided in the output containing the CoGe identifier of the putative transposed gene as well as the identifier for one of the A. lyrata flankers.
Using blastn parameters of word-size 7 and filtered to exclude simple sequences, each transposed gene was blasted to C. papaya (University of Hawaii v0.4, masked repeats 50x) and V. vitifera (French National Sequence Center v1, masked repeats 50x) to verify the presence of the gene within the genome, masking all non-cd sequence. We then masked the transposed gene itself and blasted 15 Kb on either side of the gene to C. papaya and V. vitifera to find the syntenous regions.
If the gene itself was not found in either papaya or grape, it was labeled “No gene hit” and was discarded from our list. If there was a gene hit to papaya and grape, but the surrounding region of A. thaliana does not hit in the same region in papaya and grape, the gene was labeled “No hits with gene and buffer in same region” and was considered transposed. If there was a hit to the same region of papaya and grape in both the buffer and the gene itself, it was considered to be in the ancestral position and was discarded as not being a true transposition. Each gene had to have a hit in both papaya and grape, and each gene had to be in a nonancestral position in both papaya and grape, otherwise it was discarded from our list. This procedure left us with the 226 confirmed transposed genes that were the basis of this research (Table S1).
The genomic sequence (including non-coding sequence) of all genes considered true transpositions in our list were then blasted to the A. thaliana genome to find the best hit outside of itself. We then examined the best hit to see whether it had sequence similarity with the transposed gene higher than 75% identity across at least 50 base pairs of noncoding sequence. If the best hit fit these criteria, it was deemed to be a putative parental site from which the transposed gene had duplicated (Tables S2, S3).
We used our genome visualization platform GEvo to visually compare the 5′ region ∼500 bp upstream of our target gene to the sequence ∼500 bp downstream of its 3′ region. We limited our search to BLAST hits that were greater than or equal to the e-value of a 15/15 bp exact match, and excluded simple sequences (Tables S2, S3). The 102 non-transposed genes that were selected for the control were genes chosen from specific gene families that had been confirmed by both [2] and this particular study as being underrepresented for gene transposition. Since this portion of the study was performed manually we restricted our analysis to only 102 genes, similar in number to the total number of transposed genes examined for parental sites and flanking repeats (126).
We wrote a program using blastn, word-size 7, with BLAST hits that were greater than or equal to the e-value of a 15/15 exact match, that would look for repeats between 30–400 bp shared between the 2 kb region up and downstream of coding sequence of each gene (Table S4).
|
10.1371/journal.pntd.0000649 | Using Recombinant Proteins from Lutzomyia longipalpis Saliva to Estimate Human Vector Exposure in Visceral Leishmaniasis Endemic Areas | Leishmania is transmitted by female sand flies and deposited together with saliva, which contains a vast repertoire of pharmacologically active molecules that contribute to the establishment of the infection. The exposure to vector saliva induces an immune response against its components that can be used as a marker of exposure to the vector. Performing large-scale serological studies to detect vector exposure has been limited by the difficulty in obtaining sand fly saliva. Here, we validate the use of two sand fly salivary recombinant proteins as markers for vector exposure.
ELISA was used to screen human sera, collected in an area endemic for visceral leishmaniasis, against the salivary gland sonicate (SGS) or two recombinant proteins (rLJM11 and rLJM17) from Lutzomyia longipalpis saliva. Antibody levels before and after SGS seroconversion (n = 26) were compared using the Wilcoxon signed rank paired test. Human sera from an area endemic for VL which recognize Lu. longipalpis saliva in ELISA also recognize a combination of rLJM17 and rLJM11. We then extended the analysis to include 40 sera from individuals who were seropositive and 40 seronegative to Lu. longipalpis SGS. Each recombinant protein was able to detect anti-saliva seroconversion, whereas the two proteins combined increased the detection significantly. Additionally, we evaluated the specificity of the anti-Lu. longipalpis response by testing 40 sera positive to Lutzomyia intermedia SGS, and very limited (2/40) cross-reactivity was observed. Receiver-operator characteristics (ROC) curve analysis was used to identify the effectiveness of these proteins for the prediction of anti-SGS positivity. These ROC curves evidenced the superior performance of rLJM17+rLJM11. Predicted threshold levels were confirmed for rLJM17+rLJM11 using a large panel of 1,077 serum samples.
Our results show the possibility of substituting Lu. longipalpis SGS for two recombinant proteins, LJM17 and LJM11, in order to probe for vector exposure in individuals residing in endemic areas.
| During the blood meal, female sand flies (insects that transmit the parasite Leishmania) inject saliva containing a large variety of molecules with different pharmacological activities that facilitate the acquisition of blood. These molecules can induce the production of anti-saliva antibodies, which can then be used as markers for insect (vector) biting or exposure. Epidemiological studies using sand fly salivary gland sonicate as antigens are hampered by the difficulty of obtaining large amounts of salivary glands. In the present study, we have investigated the use of two salivary recombinant proteins from the sand fly Lutzomyia longipalpis, considered the main vector of visceral leishmaniasis, as an alternative method for screening of exposure to the sand fly. We primarily tested the suitability of using the recombinant proteins to estimate positive anti-saliva ELISA test in small sets of serum samples. Further, we validated the assay in a large sample of 1,077 individuals from an epidemiological survey in a second area endemic for visceral leishmaniasis. Our findings indicate that these proteins represent a promising epidemiological tool that can aid in implementing control measures against leishmaniasis.
| The Leishmaniasis is a widely distributed disease, caused by Leishmania protozoans and transmitted by sand fly vectors. Infected sand flies inject parasites when attempting to take a blood meal. In this process, vector saliva is inoculated together with Leishmania into the host skin. This saliva is composed of molecules that modulate the host's hemostatic, inflammatory and immune responses [1]. Some of these molecules are immunogenic and stimulate strong immune responses in animals including humans [2],[3]. Importantly, the humoral response against sand fly saliva has been proposed as a potential epidemiological marker of vector exposure in endemic areas of Leishmaniasis [4],[5].
Sand fly populations tend to be clustered [5] leading to unequal exposure of human populations. Screening of human antibodies to sand fly saliva could be a useful indicator of the spatial distribution of sand flies in a particular region. Pinpointing areas of high exposure to sand fly bites may be helpful in directing control measures against Leishmaniasis.
Large-scale serological studies to detect vector exposure have been limited by the difficulty in obtaining large amounts of saliva. Additionally, the use of salivary gland sonicate inherits the limitation of potentially considerable variability in stocks of sand fly saliva due to differences in the feeding source and time of collection after feeding [6]. Salivary protein content varies along the feeding cycle and is influenced by the source of feeding used by sand flies [6].
Another limitation of using SGS is a potential lack of specificity of the salivary proteins due to immunogenicity of proteins present in different species. The use recombinant proteins may reduce such a problem by using proteins which exhibit predominant species-specificity.
Two recombinant molecules, rLJM17 and rLJM11, from Lutzomyia longipalpis saliva, were recognized by sera of men, dogs and foxes from endemic areas for visceral Leishmaniasis (Teixeira et al., unpublished data), represent good candidates for large scale testing of human exposure to Lu. longipalpis bites. In this study, we tested a large cohort for exposure to Lu. longipalpis, and validated the results obtained using the recombinant proteins with total sand fly saliva.
Informed consents were obtained from all participants and all clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki. The project was approved by the institutional review board from Centro de Pesquisas Gonçalo Moniz-FIOCRUZ/BA.
Lutzomyia longipalpis, Cavunge strain, were reared at Centro de Pesquisas Gonçalo Moniz-FIOCRUZ, as described elsewhere [7]. Salivary glands were stored in groups of 20 pairs in 20 µl NaCl (150 mM) Hepes buffer (10 mM, pH 7.4), at −70°C. Immediately before use, salivary glands were disrupted by ultrasonication. Tubes were centrifuged at 10,000×g for 2 min and the resultant supernatant (Salivary Gland Sonicate - SGS) was used for the studies.
This study was divided in three phases (Figure 1). Different sets of serum samples were randomly selected from three independent epidemiological surveys previously performed in two endemic areas for Leishmaniasis (one for VL and the other for cutaneous Leishmaniasis). The selection criteria of the subjects enrolled in each survey are published elsewhere [5],[8],[9]. In the first phase, 26 serum samples were obtained from an epidemiological survey of VL in children less than 7 years old living in a region of São Luis, Maranhão State, in northeastern Brazil, where VL is endemic and Lu. longipalpis is prevalent [5]. These samples were selected based on presenting seroconversion against the Lu. longipalpis SGS after a follow up period of six months. The cut off value of the anti-SGS ELISA was established as the mean plus three standard deviations (SD) of the mean optical density (OD) of serum samples of 26 individuals from an urban non-endemic area for both human Leishmaniasis and Lu. longipalpis. Serum samples with OD above this cut-off (0.073) were considered SGS-positive. The same methodology was applied to assess the cut-off values for the recombinant proteins. The objective was to primary set the cut-off values and to verify the concordance of seroconversion against the SGS and the recombinant proteins.
In the second part of the study, we attempted to check if the recombinant proteins were useful to discriminate anti-SGS positivity. To do this, we randomly selected another 80 individuals from the same endemic area, 40 being positive and 40 being negative for anti-SGS, and performed serology against the recombinant proteins. Receiver-Operator Characteristic (ROC) curves were built for each protein separately and for the combination of both. New cut-offs combining highest sensitivity and specificity and the highest likelihood ratio for this discrimination were determined based on the ROC curves. The ROC curves lead us to identify the effectiveness of these proteins for the identification of anti-SGS positivity.
In addition, to evaluate specificity regarding reactivity to other sand flies, we used 40 serum samples obtained from an epidemiological survey conducted in a region endemic for American cutaneous Leishmaniasis (Canoa, a rural village, located near Santo Amaro, Bahia, Brazil). In this region, Lu. intermedia represents the major sand fly species, with L. braziliensis being the main Leishmania species in the area. Both Lu. longipalpis and Lu. intermedia normally live in different ecosystems and only rarely individuals are exposed to both of them. Details of the area, patients and anti-Leishmania delayed type hypersensitivity skin test are described elsewhere [8]. We used data from 40 individuals exposed to Lu. intermedia in a previous investigation [9] and addressed the cross-reactivity to the whole SGS or the recombinant salivary proteins from Lu. longipalpis.
The third part of the study was the validation of the serology for detection of antibodies against the Lu. longipalpis salivary recombinant proteins as a marker of vector exposure. We used a larger panel consisting of 1,077 sera from another population survey done through home visits. Therefore, serum samples were obtained from children residing in two endemic areas for visceral Leishmaniasis (Vila Nova and Bom Viver), in Raposa county, Maranhão State, Brazil. Vila Nova and Bom Viver have an approximate population of 2,600 and 4,307 inhabitants, respectively. Within this population, a total of 1,297 children under 10 years old were identified and, of these, 1,077 children were enrolled in the study (220 individuals withdrawn consent). The flow chart of the third phase of the study is illustrated in Figure S1.
Antibodies to sand fly saliva from endemic area humans, dogs, or fox sera recognize mostly salivary proteins in the range of 15 to 65 kDa. Based on this information we selected nine transcripts coding for salivary proteins from Lu. longipalpis falling in this molecular range (LJM17 [AF132518], LJM111 [DQ192488], LJM11 [AY445935], LJL143 [AY445936], LJL13 [AF420274], LJL23 [AF131933], LJM04 [AAD32197.1], LJL138 [AY455916], and LJL11 [AF132510]) [10]. From the range of recombinat salivary proteins tested rLJM17 and rLJM11 were the best candidates recognized by sera from all three hosts (Teixeira et al. unpublished data). Recombinant proteins were produced by transfecting 293-F cells (Invitrogen) with plasmids (VR2001-TOPO) coding for these different salivary proteins following the manufacturer's recommendations (Teixeira et al. unpublished data). The concentrated supernatant was added to a HiTrap chelating HP column (GE Healthcare) that was then connected to a Summit station HPLC system (Dionex, Sunnyvale, CA) consisting of a P680 HPLC pump and a PDA-100 detector.
ELISA was performed as described before [4]. Briefly, ELISA plates were coated with Lu. longipalpis SGS, equivalent to 5 pairs of salivary glands/mL (approximately 5 ug protein/mL), or with 1 ug of each recombinant protein/mL (when used independently or in combination) in carbonate buffer (NaHCO3 0.45 M, Na2CO3 0.02 M, pH 9.6) overnight at 4°C. After three washes with PBS- 0.05% Tween, the plates were blocked for 1 hour at 37°C with PBS-0.1% Tween plus 0.05% BSA. Sera were diluted 1∶50 with PBS-0.05% Tween and incubated overnight at 4°c. After further washings, the wells were incubated with alkaline-phosphatase-conjugated anti-human IgG (Sigma, Sr. Louis, MO) at a 1∶1,000 dilution for 45 minutes at 37°c. Following another washing cycle, the color was developed for 30 minutes with a chromogenic solution of p-nitrophenylphosphate in sodium carbonate buffer pH 9.6 with lmg/mL of MgCI2.
The concentrations of saliva or recombinant proteins used were determined in a dose- response experiment to assess the optimum signal without loosing specificity. In all experiments, values obtained were subtracted from those obtained the background (i.e. OD values observed in well with only buffer and without SGS or recombinant antigens). The serological experiments were repeated twice with similar results. The laboratory personnel who performed the assays using the recombinant proteins were blinded about the results of ELISA assays for anti-SGS.
The statistical analyzes were performed using the GraphPad prism software 5.0 (GraphPad Prism Inc., San Diego, CA). Data regarding antibody levels before and after SGS seroconversion were compared using the Wilcoxon signed rank paired test. Kruskal Wallis with Dunn's multiple comparisons test was performed to estimate differences of OD values between three or more groups. ROC curves were used to establish the cut-off values based on the identification of the serology value, which presented the highest sensitivity and specificity in the prediction of anti-SGS positivity. Correlations between the antibody titers against SGS and those against rLJM17 and rLJM11 recombinant proteins were checked using the non-parametric Spearman test. For the validation of the serology in the third phase of the study, the calculation of sensitivity, specificity and predictive values were done through contingence tables. In all instances, differences presenting p<0.05 were considered statistically significant.
In the first part of this study, we tested whether rLJM17 and rLJM11 are associated with Lu. longipalpis exposure we measured the reactivity of total SGS, rLJM17, rLJM11 using serum samples from 26 children that seroconverted to SGS-positive in a period of six months. Figure 2A shows anti-SGS antibody levels at time 0 and 6 months, demonstrating a significant increase in the optical density of the samples. Using the same serum samples, assays were then performed with rLJM17, rLJM11 or a combination of both proteins as antigens. Both rLJM17 and rLJM11 were able to reflect the SGS-seroconversion in a variable number of samples (Figure 2B–C).
Combining both proteins considerably increased the detection of anti-SGS seroconversion, as only four samples were negative within those positive for anti-SGS (Figure 2D). In addition, in agreement with these findings, the OD values fold increases were lower for both anti-rLJM11 and anti-rLJM17 compared with anti-SGS (Figure 2E). The serology using the combination of the recombinant proteins displayed a higher fold increase, similar to the pattern observed for anti-SGS (Figure 2E). Western blot analysis performed for a small number of sera showed that some samples that recognize LJM17 do not recognize LJM11 and vice versa (data not shown; Teixeira et al. unpublished data), reinforcing the use of combined antigens to enhance sensitivity.
We further evaluated the effectiveness of the recombinant proteins in predicting anti-SGS seroconversion using a larger sample of individuals (n = 80), in which 40 were negative and 40 were positive for anti-SGS. The combination of both rLJM17 and rLJM11 as antigens incremented effectiveness by 8% compared to rLJM17 tested alone and by 17% compared to rLJM11, estimated by the area under the curves (Figure 3A). Thus, serology using these two combined salivary antigens is suitable to discriminate individuals exposed to Lu. longipalpis saliva (AUC: 0.89; p<0.0001; cut-off 0.054; likelihood ratio: 8.34) compared with the use of LJM17 (AUC: 0.81; p<0.0001; cut-off: 0.022; likelihood ratio: 5.69) or LJM11 (AUC: 0.72; p = 0.035; cut-off: 0.063; likelihood ratio: 2.16) separately (Figure 3B).
Before SGS or the salivary recombinant proteins could be validated as markers of exposure to Lu. longipalpis, it was necessary to assess the specificity by evaluating reactivity towards individuals exposed to other sand flies. Hence, we tested serum samples from an endemic area for cutaneous Leishmaniasis (Canoa, Bahia, Brazil), in which the major species of sand flies is the Lutzomyia intermedia. Both Lu. longipalpis and Lu. intermedia normally live in different ecosystems and only rarely individuals are exposed to both of them. We used data from 40 individuals exposed to Lu. intermedia in a previous investigation [9] and addressed the cross-reactivity to the whole SGS or rLJM17 and rLJM11 from Lu. longipalpis. Furthermore, 40 individuals who lived in an endemic area for Lu. longipalpis were tested for anti-SGS from Lu. intermedia serology (Figure 4). Almost all individuals from the Lu. longipalpis endemic area presented positive serology for this vector, but none of them were positive for anti-SGS for Lu. intermedia (Figure 4). Thirty-eight out of 40 individuals from Lu. intermedia endemic area displayed positive serology for the saliva of this vector, six also recognized Lu. longipalpis SGS (Figure 4), one recognized rLJM11, two recognized rLJM17 and the same two recognized the combination of the two recombinant proteins. Thus, to the end of the second phase of the study we found that both SGS and the recombinant salivary proteins present very low cross-reactivity against a different and wide distributed sand fly. In addition, we established the combination of the recombinant proteins as a potential good predictor of exposure to Lu. longipalpis, since ROC curve analysis showed a sensitivity of 91% and a specificity of 76% with the cut-off value of 0.054 OD, with a likelihood ratio of 8.34.
The final step was to validate the use of these salivary antigens as a reliable marker of exposure to Lu. longipalpis. To do so, we tested a panel of 1,077 samples of unknown anti-SGS status from children from another visceral Leishmaniasis endemic area. Sera positive against rLJM17+rLJM11 displayed a positive correlation with anti-SGS IgG levels (Spearman r = 0.379, p<0.0001; Figure 5A). Additionally, when considering only individuals who seroconverted to SGS (n = 200), this correlation became stronger (Spearman r = 0.491, p<0.0001; Figure 5B). The overall performance of the serology using the combined recombinant proteins was satisfactory, with a sensitivity of 77% (95% CI: 70.5–82.6), a specificity of 88% (95% CI: 85.7–90.1), a positive predictive value of 60% (95% CI: 53.2–65.5), a negative predictive value of 94.4% (95% CI: 92.6–95.9), and a likelihood ratio of 6.43 (Figure 6). We then stratified SGS positive cases in quartiles according to optical density values in order to verify if the efficiency of the serology would increment in those individuals with higher antibody titers against SGS (Figure 6A). Concordant and discordant results from the combined rLJM17 and rLJM 11 serology were calculated for each quartile (Figure 6B). The assay using combined salivary antigens presented a general trend with increased effectiveness of prediction in individuals with higher anti-SGS antibody titers (Figure 6B). Thus, the use of the recombinant salivary proteins was effective in the estimation of exposure to the Lu. longipalpis saliva (Figure 6C).
Age could be an important factor influencing exposure in the endemic areas and thus we tested whether the concentrations of anti-SGS antibodies were influenced by age. In the age range included in the study, there was no difference in the age distribution of positive individuals within stratified quartiles, (Kruskal Wallis test p = 0.065), indicating that the antibody titers may likely represent exposure to the sand fly.
Antibodies against the salivary gland components of blood sucking insects [11],[12], can be used as epidemiological markers of vector exposure [13], as has been shown for Leishmaniasis [4],[5]. Large epidemiological investigations using salivary gland antigens are hampered by the limitation of obtaining large amounts of highly reproducible salivary glands sonicate. Herein we report on the detection of sera reactive to whole SGS using two recombinant proteins from Lu. longipalpis saliva, rLJM17 and rLJM11 and show a positive correlation between the results obtained using SGS and those obtained using rLJM17+rLJM11 as antigens.
Besides the possibility of being produced in large amounts, recombinant salivary proteins bring another advantage to serological tests as they can produce in a highly reproducible fashion. It is known that sand fly saliva protein profile, as well as its relative content, varies at different stages after a meal [6],[14],[15], and this cannot be totally controlled even in standardized colonies.
The combined use of different recombinant proteins is justified since not all SGS-positive sera recognize the same protein bands [5]. In the present study, some serum samples recognized either rLJM17 or rLJM11 when tested by ELISA and this was further confirmed by western blot (data not shown). Such differential recognition may explain the better performance of the test when samples highly reactive to SGS were employed. Other immunogenic salivary proteins are likely candidates to be tested in conjunction with rLJM17 and rLJM11, which may increase the test sensitivity.
Importantly, recombinant molecules selected for use in serology should not cross-react with salivary proteins from other non-vector sand fly species, which may lead to false positive results. In order to test for specificity, we have evaluated sera from one area where Lu. longipalpis is highly predominant to one area where this species is not found. In a large survey in the whole São Luis island, including the three municipalities which comprise areas 1 and 3 of the present report, with the capture of 22,581 specimens Lu. longipalpis (66.4%) of the captured specimens. It was followed by Lutzomyia whitmani (24%) and Lutzomyia evandroi (5.9%), with the remaining 29 species represented 3.7% of the total sample [16]. On the other hand, in the Canoa village (area 2 of the present study) a phlebotomine survey performed at the time of sera collection evidenced a marked predominance of Lu. intermedia, representing 94% of the captured specimens, with a small number of Lutzomyia migonei and Lutzomyia (Nyssomyia) sp. [17]. Comparative sequence analysis of LJM17 from Lu. longipalpis and Lu. intermedia LJM17-homologue showed some areas of high aminoacid conservancy (Teixeira et al. unpublished data). However, cross reactivity with Lu. longipalpis SGS was not observed in animals experimentally exposed to Lu. intermedia SGS [9]. Likely, the tertiary conformation of the LJM17 protein from Lu. longipalpis, is distinct from that of Lu.intermedia accounting for the specificity of the assay.
In conclusion, we have shown here that ELISA employing two recombinant proteins derived from Lu. longipalpis saliva is a powerful tool for detecting specific exposure to vector sand flies in populations. These proteins represent a promising epidemiological tool that can aid in implementing control measures against Leishmaniasis.
|
10.1371/journal.pmed.1002575 | A peer-support lifestyle intervention for preventing type 2 diabetes in India: A cluster-randomized controlled trial of the Kerala Diabetes Prevention Program | The major efficacy trials on diabetes prevention have used resource-intensive approaches to identify high-risk individuals and deliver lifestyle interventions. Such strategies are not feasible for wider implementation in low- and middle-income countries (LMICs). We aimed to evaluate the effectiveness of a peer-support lifestyle intervention in preventing type 2 diabetes among high-risk individuals identified on the basis of a simple diabetes risk score.
The Kerala Diabetes Prevention Program was a cluster-randomized controlled trial conducted in 60 polling areas (clusters) of Neyyattinkara taluk (subdistrict) in Trivandrum district, Kerala state, India. Participants (age 30–60 years) were those with an Indian Diabetes Risk Score (IDRS) ≥60 and were free of diabetes on an oral glucose tolerance test (OGTT). A total of 1,007 participants (47.2% female) were enrolled (507 in the control group and 500 in the intervention group). Participants from intervention clusters participated in a 12-month community-based peer-support program comprising 15 group sessions (12 of which were led by trained lay peer leaders) and a range of community activities to support lifestyle change. Participants from control clusters received an education booklet with lifestyle change advice. The primary outcome was the incidence of diabetes at 24 months, diagnosed by an annual OGTT. Secondary outcomes were behavioral, clinical, and biochemical characteristics and health-related quality of life (HRQoL). A total of 964 (95.7%) participants were followed up at 24 months. Baseline characteristics of clusters and participants were similar between the study groups. After a median follow-up of 24 months, diabetes developed in 17.1% (79/463) of control participants and 14.9% (68/456) of intervention participants (relative risk [RR] 0.88, 95% CI 0.66–1.16, p = 0.36). At 24 months, compared with the control group, intervention participants had a greater reduction in IDRS score (mean difference: −1.50 points, p = 0.022) and alcohol use (RR 0.77, p = 0.018) and a greater increase in fruit and vegetable intake (≥5 servings/day) (RR 1.83, p = 0.008) and physical functioning score of the HRQoL scale (mean difference: 3.9 score, p = 0.016). The cost of delivering the peer-support intervention was US$22.5 per participant. There were no adverse events related to the intervention. We did not adjust for multiple comparisons, which may have increased the overall type I error rate.
A low-cost community-based peer-support lifestyle intervention resulted in a nonsignificant reduction in diabetes incidence in this high-risk population at 24 months. However, there were significant improvements in some cardiovascular risk factors and physical functioning score of the HRQoL scale.
Australia and New Zealand Clinical Trials Registry ACTRN12611000262909.
| In low- and middle-income countries (LMICs), there is an urgent need to develop low-cost strategies for identifying high-risk individuals and delivering lifestyle interventions to prevent diabetes.
We conducted a cluster-randomized controlled trial in a community-setting in India to test whether a peer-support lifestyle intervention could reduce diabetes risk at 24 months.
We identified high-risk individuals on the basis of a simple diabetes risk score.
There was a nonsignificant reduction in diabetes risk in this high-risk population at 24 months. However, there were significant improvements in some risk factors for cardiovascular disease and a measure of quality of life.
The intervention cost was low at US$22.5 per participant over 12 months.
Risk scores for better identifying people at highest risk for diabetes are needed, particularly in resource-constrained settings.
It is important to identify ways to improve program adherence and engagement, possibly by using more flexible modes of program delivery, e.g., at worksites and by text messaging.
| Type 2 diabetes is a major public health problem worldwide [1]. Globally, an estimated 425 million people have diabetes, and the majority of those (79%) are living in low- and middle-income countries (LMICs) such as India [1]. A large proportion of people with diabetes are undiagnosed, and many present with complications at the time of diagnosis [1]. Diabetes imposes a large economic burden on individuals, their families, and national health systems [1]. Therefore, there is an urgent need to develop and implement effective and cost-effective measures to prevent diabetes.
The major efficacy trials have shown that lifestyle interventions targeting physical activity, dietary changes, and weight loss are effective [2–5] and cost-effective [6,7] in preventing type 2 diabetes among people with impaired glucose tolerance (IGT). However, while this is encouraging, the real challenge is to deliver such interventions under ‘real-world’ conditions [8]. The efficacy trials required the expensive oral glucose tolerance test (OGTT) to identify high-risk individuals and involved specialized multidisciplinary teams (e.g., physicians, nurses, dieticians, exercise physiologists) to deliver interventions. These are important factors limiting the translation of findings from the efficacy trials to real-world settings, particularly in LMICs, thereby requiring alternative strategies for identifying high-risk individuals and delivering interventions [9].
Mass screening with an OGTT to identify high-risk individuals is highly challenging in LMICs because of the cost and the limited availability of trained clinical staff and accredited laboratories [10]. Diabetes risk scores are low-cost, noninvasive, and easy-to-administer screening tools, which could reduce the number of OGTTs when used in a stepwise screening approach [11]. International guidelines and expert groups recommend using diabetes risk scores as the first screening step to identify people who may be at high risk, with blood tests undertaken to confirm high-risk status (i.e., prediabetes). These high-risk individuals can then be referred to a lifestyle intervention program [12,13]. However, even these approaches require blood testing on up to 50% of adults, posing difficulties in low-resource settings.
Lifestyle interventions evaluated in the major efficacy trials have involved resource-intensive, individualized counselling delivered on a one-to-one basis or in groups by highly trained health professionals [2–5]. In the Chinese Da Qing IGT and Diabetes Study [3], physicians delivered one individual counselling session and group counselling sessions every week for one month, followed by monthly for three months and every three months for the next 5.8 years. In the United States Diabetes Prevention Program (US DPP) [2], participants in the lifestyle intervention group received 16 individual counselling sessions from case managers within the first 24 weeks following randomization, and then, face-to-face contacts (individual or group) were made every two months for another 2.5 years. Over three years, the intervention costs were US$2,780 per participant [14]. In the Finnish Diabetes Prevention Study [5], participants received one-to-one individualized counselling sessions from nutritionists. Seven sessions were delivered in the first year and one session every three months thereafter until the end of the study at six years. Exercise physiologists guided participants to increase their physical activity through individualized resistance training sessions. In the Indian Diabetes Prevention Programme (IDPP) [4], although the intervention was less labor- and resource-intensive (US$225 per participant over three years) [6] than other efficacy trials, it was delivered on a one-to-one basis by physicians, dieticians, and social workers. These intervention strategies are not feasible for wider implementation in real-world settings in LMICs, where the burden of diabetes is substantial [1] and where the availability of highly trained health professionals for program delivery is very limited [15]. Peer support is an alternative strategy to encourage people to make and sustain healthy lifestyle changes [16]. Peer support refers to the provision of practical, social, and emotional ongoing support from nonprofessionals for complex health behaviors [17]. A recent systematic review has shown that peer support is effective in bringing behavior change in prevention and management of various health conditions, including HIV/AIDS, maternal and child health, and mental health, as well as diabetes, cardiovascular disease, and other chronic conditions [17]. Peer-support interventions are low-cost, culturally appropriate, and potentially scalable [17].
The Kerala Diabetes Prevention Program (K-DPP) was a cluster randomized controlled trial (RCT) of a peer-support lifestyle intervention implemented in a community setting in India [18]. In this paper, we aimed to examine whether the intervention could reduce diabetes incidence at 24 months among high-risk individuals identified on the basis of a diabetes risk score.
The study was approved by the Health Ministry Screening Committee of the Government of India; ethics committees of the Sree Chitra Tirunal Institute for Medical Sciences and Technology (SCT/IEC-333/May 2011), Trivandrum, India; Monash University (CF11/0457-2011000194); and The University of Melbourne (1441736) in Australia. Written informed consent was obtained from all study participants.
This study is reported in accordance with the Consolidated Standards of Reporting Trials guidelines for cluster RCTs (S1 Checklist) [19]. The details of the K-DPP study design have been described in detail elsewhere [18], and the protocol is available at https://link.springer.com/article/10.1186/1471-2458-13-1035. Briefly, K-DPP was a cluster RCT conducted in 60 polling areas (clusters) of Neyyattinkara taluk (subdistrict) in Trivandrum district, Kerala state. Polling areas are well-defined and identifiable locations demarcated with landmarks such as hills, rivers, roads, streets, etc. by the Election Commission of India [20]. A cluster design was chosen for the study, as the risk of contamination would otherwise be high among individuals from the same community.
At the time of study enrolment, Neyyattinkara taluk had 603 polling areas across four legislative assembly constituencies (LACs). Although there were maps available to locate polling areas in each of the four LACs, there was no single map connecting the four LACs, which could show the contiguous polling areas across the borders of LACs. To reduce the risk of selecting contiguous polling areas, we removed 244 polling areas that were located along the borders of the four LACs and selected at random 60 of the remaining 359 polling areas. The 60 polling areas were then randomly assigned in a 1:1 ratio to the control group (received an education booklet on general lifestyle advice) or the intervention group (peer-support lifestyle intervention) by an independent statistician using a computer-generated randomization sequence. After randomization, in both the intervention and control groups, there were two contiguous polling areas. Therefore, one polling area from each pair was replaced with a nearby polling area, which were at least two kilometers apart. Participants were masked to group assignment until the completion of the baseline assessment. Field staff members administering questionnaires (no group-specific questions were included in the questionnaires) and undertaking measurements at baseline and follow-ups, along with laboratory technicians and investigators, were masked. Field staff members administering the process evaluation questionnaire to intervention participants were not masked.
From the electoral roll of each of the 60 polling areas, 80 individuals (age 30 to 60 years) were selected randomly and were approached through home visits. Eligibility criteria included no history of diabetes or other chronic illness that might affect their participation in the trial, being literate in the local language (Malayalam), not being pregnant, and not taking medications known to affect glucose tolerance (glucocorticoids, antiretroviral drugs and antipsychotics). Those satisfying the eligibility criteria were screened using the Indian Diabetes Risk Score (IDRS), which comprises four simple parameters: age, family history of diabetes, physical activity (regular exercise or strenuous work), and waist circumference [21]. At the time of study enrolment, the IDRS was the only risk score from India that had been previously evaluated in a cohort study, with a score of ≥60 being a strong predictor of incident diabetes in Asian Indians [22]. Therefore, we chose the IDRS to screen and recruit our trial participants. Those with an IDRS score ≥60 were invited to attend a community-based clinic to undergo a 75-gm OGTT. Clinics were conducted in local neighborhoods in community buildings (e.g., schools, library halls, church halls). The OGTT was performed according to the World Health Organization (WHO) guidelines [23]. A venous blood sample was taken after an overnight fast for at least eight hours, and a second blood sample was collected two hours after oral ingestion of 75-g glucose dissolved in 250–300 ml of water. Those with fasting plasma glucose (FPG) ≥7.0 mmol/l or 2-hr plasma glucose (2-hr PG) ≥11.1 mmol/l or both were diagnosed to have diabetes based on the American Diabetes Association (ADA) criteria [12]. They were then referred to a healthcare facility and were excluded from the study. The remaining individuals were enrolled in the trial irrespective of their baseline glucose tolerance. If the participant had not fasted for the recommended time, they were asked to attend another clinic in a nearby neighborhood on a different day.
Detailed information on the development and cultural adaptation of the intervention program have been reported elsewhere [24,25]. Briefly, the main theory underpinning the intervention program was the Health Action Process Approach model [26] with more emphasis given to collectivistic rather than individualistic strategies during the intervention design phase. The intervention program was adapted from the Finnish Good Ageing in Lahti Region (GOAL) program [27] and the Australian Greater Green Triangle (GGT) Diabetes Prevention Project [28] through situational analysis, needs assessment, and cultural translation [24,25]. This adaptation process was guided by the Intervention Mapping Approach [29]. The program utilized the core functions of peer support identified in the US Peers for Progress Program [16] and incorporated behavior change strategies that were identified from the needs assessment study [25]. The intervention model and program were tested and further refined following piloting with two groups in 2012–2013 [18].
The 12-month peer-support program consisted of 15 group sessions: an introductory session delivered by the K-DPP team; two education sessions conducted by local experts; and 12 sessions delivered by trained lay peer leaders. An introductory session was planned to introduce the group participants to the program and its mentoring style. The needs assessment study showed that at least one peer group session per month would be optimal and feasible to deliver the intervention [25], and thus 12 peer group sessions were planned. The needs assessment study also emphasized the importance of including sessions on diabetes prevention and management by local experts, as the knowledge on these among people with prediabetes in Kerala was low [25]. Furthermore, during the pilot phase [18], peer leaders were selected from within the groups, and their level of diabetes-specific knowledge was also limited. Therefore, we decided that information would be delivered by experts through two education sessions, and the peer leaders’ role would be to help participants to translate the information into their daily lives. To minimize resources, the education sessions were delivered to participants from two to three neighborhoods within close proximity. Participants were encouraged to bring family members along to further extend the reach of the education.
Participants were assessed at baseline, 12 months, and 24 months. During each assessment, field staff members administered standardized and validated questionnaires to collect measures of sociodemographic characteristics, lifestyle behaviors, medical history, and health-related quality of life (HRQoL). Self-reported levels of physical activity were measured using the Global Physical Activity Questionnaire [31]. Intake of fruits and vegetables were assessed using a food frequency questionnaire [32]. HRQoL was assessed using the 36-item Short-Form (SF-36) health survey [33]. The SF-36 is divided into eight scales (physical functioning, role limitation—physical, role limitation—emotional, bodily pain, general health, mental health, social functioning, and vitality) and two domains (physical component summary and mental component summary). Scores for each of the scales and domains range from 0 to 100, with higher scores indicating better quality of life. The SF-36 data were converted into a six-dimensional health state called the Short Form 6 Dimension (SF-6D), whose score ranges between 0.29 (worse health) and 1.00 (full health). Following the 12-month intervention period, a process evaluation questionnaire was administered to intervention participants by field staff members who were different from those administering main questionnaires. During each assessment, anthropometry (height, weight, fat percent, muscle mass, waist circumference, and hip circumference) and blood pressure were measured, and blood samples were taken for the OGTT, HbA1c, and lipids according to standard protocols [34]. Individuals diagnosed with diabetes on the OGTT at 12-month follow-up were referred to healthcare facilities for treatment and care. However, they were still followed-up, although an OGTT was not performed at 24 months; instead, FPG alone was measured. Blood samples were centrifuged within 30 minutes of collection and transported in boxes with dry ice to a nationally accredited laboratory. Plasma glucose was measured using the hexokinase method on a COBAS 6000 analyzer, with kits supplied from Roche Diagnostics, Switzerland. HbA1c was measured using the high-performance liquid chromatography method on a D-10 BIORAD analyzer and lipids by enzymatic methods on a COBAS 6000 analyzer, using kits supplied by Roche Diagnostics, Switzerland. Low-density lipoprotein (LDL) cholesterol was estimated using the Friedewald equation [35] for participants with triglycerides ≤4.52 mmol/l, and for the rest, values obtained from the direct method were used in the analysis.
The primary outcome was the incidence of diabetes at 24 months, diagnosed by an annual OGTT, according to the ADA criteria (FPG ≥7.0 mmol/l and/or 2-hr PG ≥11.1 mmol/l) [12]. Participants who were diagnosed with diabetes by a physician and taking antidiabetic medications (‘clinical diagnosis’) subsequent to entry in the trial were also included in the primary outcome. Secondary outcomes included weight, waist circumference, waist-to-hip ratio, fat percent, muscle mass, systolic and diastolic blood pressure, FPG, 2-hr PG, HbA1c, lipid profile, IDRS score, ≥5 servings of fruit and vegetables intake per day, physical activity, tobacco use, alcohol use, and HRQoL.
Assuming an annual incidence of diabetes of 18.3% in the control group [4], an intracluster correlation coefficient (ICC) of 0.02 for plasma glucose [36], an average of 17 participants per group, at 5% significance with 80% power, allowing a loss to follow-up of 10%, the numbers of participants and polling areas per study group required were 510 and 30 respectively, to detect a relative risk reduction (RRR) of 30% at 24 months [18]. Funding dictated that the primary outcome be measured at 24 months’ follow-up. Since ICC values for diabetes incidence were not available from published studies, ICC for plasma glucose from a previous study [36] was used. However, a positive ICC for diabetes incidence was not observed in the trial. Hence, the 32% inflation of sample size for a design effect was redundant.
Baseline characteristics of clusters and participants are summarized using mean and standard deviation (SD) or median and interquartile range (IQR) for continuous variables and frequency and percentage for categorical variables. The analyses observed intention-to-treat, i.e., participants and clusters were analyzed according to the group to which they were allocated. There were a few changes to the analysis plan specified in the study protocol [18]. For the primary outcome analysis, instead of logistic regression and Cox-proportional hazards regression, to estimate the relative risk (RR) (and 95% confidence interval [CI] and P value) at 24 months, we used log binomial models estimated by generalized estimating equations (GEE) with an exchangeable working correlation structure and robust standard errors to account for clustering by polling areas. This approach gave improved interpretability of the intervention effect (as increased cumulative incidence (RR) rather than increased odds) and accorded with the use of RR in the protocol’s sample size. As diabetes was only observed systematically at 12- and 24-month time points a discrete time proportional hazards model was considered appropriate in place of the Cox model but this model provides little extra information above and beyond the log-binomial model for diabetes incidence at 12 and 24 months. We also conducted post hoc subgroup analyses by baseline glucose tolerance: normal glucose tolerance (NGT), isolated impaired fasting glucose (IFG), and IGT defined by the ADA criteria [12] and the WHO criteria [23]. To examine the heterogeneity of intervention effect by subgroup, an interaction term between the intervention assignment and subgroup was included in the GEE models, and its significance was tested using the Wald test. The subgroup analyses were done because the current literature on diabetes prevention programs has been largely limited to people with IGT [37], yet the target population in the real world is much broader.
For the analysis of continuous secondary outcomes, mixed-effects linear regression models were used and included outcomes at baseline, 12 months and 24 months, and included all participants with outcome data available at one or more of these timepoints. Skewed variables were log-transformed before analysis. Study group (intervention vs. control), timepoint (follow-up vs. baseline) and a study group-by-timepoint interaction were specified as fixed effects. Random effects were specified for polling areas, to account for the clustered study design, and for participants, to account for correlation between the repeated measurements on the same individual. The P value of the study-group-by-timepoint interaction was used to test the difference in change between study groups. For categorical secondary outcomes, the log binomial model was used.
We assessed the sensitivity of the primary outcome analysis to missing outcome data using multilevel multiple imputation (MMI), accounting for clustering [38]. We performed 10 imputations using GEE to fit log binomial imputation models for missing outcomes and with study group and the following baseline covariates included as auxiliary variables: age, sex, education, occupation, monthly household expenditure, current tobacco use, current alcohol use, fruit and vegetable intake (in servings/day), leisure time physical activity, family history of diabetes, body mass index, waist-to-hip ratio, fat percent, muscle mass, systolic blood pressure, diastolic blood pressure, FPG, 2-hr PG, HbA1c, LDL cholesterol, and triglycerides. The log RR (and its standard error) was computed on each multiply imputed dataset, and the results were combined to obtain the multiple imputation estimate using Rubin’s rule [39]. MMI was performed using the R Jomo package [40].
The costs associated with delivering the peer-support intervention over 12 months were estimated across five major categories (training sessions for peer leaders and LRPs, group sessions, resource materials, administrative costs, and community activities). The K-DPP personnel (intervention manager and intervention assistant) were interviewed to estimate the amount of time they spent on various intervention activities. Personnel costs were calculated based on the actual salary (or remuneration) paid to the intervention manager, intervention assistant, local experts, and LRPs. Nonpersonnel costs (travel, food and logistics, rent for venues, phone calls, designing and printing charges for resource materials, and administrative costs) were estimated based on the actual expenditure. The cost figures were obtained from the finance registers. The cost estimates in Indian Rupees (INR) were converted to US dollars using an exchange rate of INR58.6 = US$1 for the year 2013 [41].
A two-sided P value <0.05 was considered statistically significant for all analyses. Analyses were performed using Stata version 14.2 (StataCorp LP, College Station, Texas, USA), R 3.4.3, or Microsoft Excel 2016 (Microsoft corporation, Redmond, Washington, USA).
Participants from the 60 polling areas were recruited between January 20, 2013, and October 27, 2013. Fig 1 shows the trial profile. A total of 3,689 individuals were contacted through home visits, of whom 137 (3.7%) did not satisfy the age criteria and were therefore excluded. Of the remaining 3,552 individuals, 131 (3.7%) declined participation, and 3421 were assessed for eligibility, of whom 835 (24.4%) were not eligible. Of 2,586 eligible individuals screened with the IDRS, 1,529 (59.1%) had a score ≥60, of whom 1,209 (79.1%) attended community-based clinics and underwent an OGTT. After excluding 202 (16.7%) individuals with diabetes, 1,007 (507 in the control group and 500 in the intervention group) were enrolled in the trial. Baseline characteristics of clusters and participants were similar between the study groups (Table 1). Participants’ mean age was 46.0 years, and the majority were male (52.8%), educated up to secondary school (75.6%), and employed (72.3%). According to the ADA criteria [12], 11.5% had IGT, 57.5% had isolated IFG, and 31.0% had NGT. The corresponding figures for the WHO criteria [23] were 11.5%, 22.5%, and 66.0%, respectively. The prevalence of several cardiovascular risk factors was high at baseline, as reported previously [42]. All clusters and 95.7% (964/1007) of participants were followed-up at 24 months (95.1% in the control group; 96.4% in the intervention group).
Of the 15 total program sessions, participants attended a median of 9 (IQR 3 to 13) sessions; 10.8% attended all 15 sessions, 62.4% attended seven or more sessions, and 89.2% attended at least one of these sessions. Twenty-nine out of 30 groups delivered all of their 12 peer group sessions, according to the intervention protocol. Nearly two-thirds (61.8%, 309/500) of intervention participants reported that they had regular contact with peer leaders outside the formal group sessions during the 12-month program, with a mean number of contacts being 11.3 (SD 8.1). More than half (57.2%, 286/500) reported participation in community activities, including yoga sessions, kitchen gardening, and walking groups. Among those who did not attend formal program sessions (n = 54), one-third (33.3%) still reported that they had regular contact with their peer leaders outside the group sessions, and 16.7% also reported participating in community activities. Overall, only 27 participants (5.4%) did not have any exposure to the program.
After a median follow-up of 24 months, overall, diabetes developed in 147 participants (144 were diagnosed on the OGTT and 3 were clinically diagnosed): 17.1% (79/463) of participants in the control group and 14.9% (68/456) of participants in the intervention group. The RR was 0.88 (95% CI 0.66–1.16), p = 0.36. The RR did not change appreciably after excluding those with baseline HbA1c ≥6.5% (n = 52) (ADA [12] and International Expert Committee [43] cutoff value for diabetes) (RR 0.93, 95% CI 0.66–1.31, p = 0.66) and was similar to the results obtained using MMI (RR 0.86, 95% CI 0.61–1.13), p = 0.29.
The RR in the IGT subgroup (ADA and WHO criteria: 0.66, 95% CI 0.45–0.98, p = 0.038) was lower than that in the isolated IFG (ADA criteria: 0.95, 95% CI 0.68–1.33, p = 0.77; WHO criteria: 0.98, 95% CI 0.68–1.42, p = 0.92) or NGT (ADA criteria: 1.23, 95% CI 0.36–4.26; WHO criteria: 1.15, 95% CI 0.66–2.01, p = 0.63) subgroups. However, there was no evidence in favor of an interaction between baseline glucose tolerance and study group on diabetes incidence (ADA criteria: p = 0.24; WHO criteria: p = 0.11).
Table 2 shows the changes in clinical and biochemical characteristics by study group at 24 months. The IDRS score reduced in both study groups, but the reduction was greater in the intervention group by 1.50 points (p = 0.022). Table 3 shows the changes in behavioral characteristics by study group at 24 months. Compared with the control participants, intervention participants were more likely to consume ≥5 servings of fruit and vegetables per day by 83% (p = 0.008). Intervention participants were 23% less likely to consume alcohol compared with the control participants (p = 0.018), and the amount of alcohol consumed was lower among intervention participants (p = 0.030). Table 4 shows the changes in HRQoL variables by study group at 24 months. Compared with the control group, the intervention participants had a greater increase in physical functioning score of the HRQoL scale by 3.9 points (p = 0.016). The 12-month changes in secondary outcomes by study group are given in S2, S3 and S4 Tables.
Table 5 shows the costs associated with delivering the peer-support intervention over 12 months. The total intervention costs amounted to US$11,225 (US$22.5 per participant). The group sessions were the largest cost contributor (52.8% of total costs), followed by designing and printing charges for resource materials (21.7%), administrative costs (13.8%), and training of peer leaders and LRPs (11.7%). Personnel costs accounted for 26.7% of the total costs. Community activities incurred no program costs.
We recorded no adverse events related to the intervention.
To our knowledge, K-DPP is the first RCT from a LMIC to evaluate the effectiveness of a peer-support lifestyle intervention delivered mainly by lay people in a community setting. This study showed that the intervention resulted in a non-significant reduction in diabetes incidence at 24 months in a high-risk population identified on the basis of a diabetes risk score. However, there were significant improvements in some cardiovascular risk factors, including IDRS score, fruit and vegetable intake, and alcohol use, and physical functioning score of the HRQoL scale.
The trial was powered for a 30% RRR for diabetes incidence at 24 months and observed a 12% RRR (nonsignificant), which was lower than that reported in other effectiveness trials. A meta-analysis by Ashra and colleagues assessing the effectiveness of 13 pragmatic lifestyle interventions implemented in routine clinical practice showed that the pooled estimate of RRR was 26% [44]. In the Study on Lifestyle intervention and Impaired glucose tolerance Maastricht (SLIM) study, the RRR was 58% at three years [45]. In the Joetsu Diabetes Prevention Trial, the RRR varied from 27% (nonhospitalization with diabetes education and support) to 42% (short-term hospitalization with diabetes education and support) at three years [46]. In the Spanish Diabetes in Europe—Prevention using Lifestyle, Physical Activity and Nutritional Intervention (DE-PLAN) project, which was implemented in primary care settings, the RRR was 36% at four years [47]. In a mobile phone effectiveness study conducted in India, the RRR was 36% at two years [48]. In the Diabetes Community Lifestyle Improvement (D-CLIP) translational trial conducted in India, the RRR was 32% at three years [49]. The lower effect in our study could be attributed to the following reasons. In previous studies, most (if not all) participants had IGT, while in K-DPP, participants were identified on the basis of a risk score, and the majority had isolated IFG or NGT, albeit with a high burden of cardiovascular risk factors [42]. So far, from the limited recent literature available [49–51], there is no proven intervention to reduce diabetes incidence among those with isolated IFG. Furthermore, 24 months’ follow-up may not have been long enough to allow for an intervention effect to be observable, and thus a longer-term follow-up has been planned.
In the control group, there was a decline in fruit and vegetable intake and the reported level of physical activity at 24 months. These are consistent with findings from other recent longitudinal studies conducted in Kerala, showing that the proportion of people meeting the recommended intake of fruits and vegetables and level of physical activity is continuing to decrease over time in the absence of any intervention [52,53]. There was a greater increase in physical functioning score of the HRQoL scale in the intervention group at 24 months. Previous studies have shown that improvement in HRQoL is likely to be mediated by improved physical activity and weight loss [54].
In our study, the cost of delivering the peer-support lifestyle intervention over 12 months was US$22.5 per participant, a large percentage (52.8%) of which was accounted for by the group sessions, the main mode of formal program delivery. This is less than one-third of the intervention costs incurred in IDPP (US$75 per participant per year) [6]. The lower cost could be mainly attributed to the fact that, in IDPP, health professionals (physicians, dieticians, and social workers) were involved in delivering the intervention, while in K-DPP, the intervention was delivered mainly by lay peer leaders. In IDPP, the personnel cost was US$36 per participant, whereas in K-DPP, this cost was only US$6. However, the effect size in IDPP (28.5%) was higher than that in K-DPP (12%). It is possible that resource-intensive lifestyle interventions are more effective than low-resource interventions in reducing diabetes progression, at least in the IGT population (S5 Table and S1 Fig). If the K-DPP intervention were implemented as a real-world program, the unit costs for each individual would be much lower. This is because the one-off costs (e.g., training of peer leaders and LRPs and printing charges of resource materials) would be distributed over a much larger number of individuals, and the relative travel and administrative costs would also be lower. In K-DPP, travel and administrative costs accounted for one-third (33.3%) of the total costs. This is because the K-DPP personnel and local experts had to travel to the field, spending around three hours for every return trip. However, these costs would be relatively lower in a program setting and if the program were delivered at scale. Moreover, the very important K-DPP community activities incurred no program costs, as they were led by peer leaders with assistance from LRPs.
The K-DPP trial has a number of strengths. The study was conducted in the Indian state of Kerala, which has a high prevalence of diabetes (approximately 20%) and several other cardiovascular risk factors [52,55,56]. The state is in the most advanced stage of epidemiological transition compared to other Indian states [57], and it is also the harbinger of the future for the rest of India in relation to the burden of chronic diseases [55,56]. Therefore, Kerala provides an ideal setting for the implementation and evaluation of a diabetes prevention program in a community setting in India. As far as we are aware, K-DPP is the first diabetes prevention trial to deliver a peer-support lifestyle intervention program mainly by trained lay people in a low- and middle-income setting. In the D-CLIP trial from Chennai, India, although peer support was provided by community volunteers, it involved a team of health coaches and fitness trainers in the delivery of intervention. Also, metformin (500 mg twice a day) was added to those at highest risk of developing diabetes [49]. Although metformin was found to be equally effective as lifestyle intervention in previous studies [2,4] and is cheap, the current evidence base supports its use only in combination with lifestyle interventions [58]. Other strengths of our trial include a very high follow-up rate at 24 months (97.5%), use of a rigorous study design, and a much better representation of women (nearly half the participants were women) compared to previous diabetes prevention trials in India [4,48,49,59]. However, there are also some study limitations. In subgroup analyses, balance of potentially confounding characteristics between the subgroups compared is not guaranteed, and the power may have been insufficient for such analyses. Data on behavioral risk factors (tobacco use, alcohol use, physical activity, and fruit and vegetable intake) were collected using questionnaires that were not validated by objective measures and are likely to be subject to response bias. It is possible that social desirability and acquiescence biases associated with the intervention may have resulted in the small differences observed in some of the behavioral outcomes at 24 months. Furthermore, we did not adjust for multiple comparison, and given the likelihood of type 1 errors [60], changes in secondary outcomes observed should be interpreted cautiously.
In efficacy trials of behavioral or social interventions, recruitment of highly selected individuals, resource-intensive interventions, and close monitoring to ensure compliance will almost always overestimate the outcomes that will actually be achievable under ‘real-world’ conditions [61]. However, given that the efficacy of lifestyle interventions to prevent diabetes among high-risk individuals, particularly among those with IGT, has been quite well established, it is now important to determine their effectiveness in real-world settings, in which the target population is likely to be much broader if program participants are not recruited on the basis of clinical testing [62]. Our study findings have some important implications for policy and future research with regards to diabetes prevention in India (and perhaps also in other LMICs). First, using a risk score rather than the OGTT to identify high-risk individuals was part of our strategy to develop a low-cost diabetes prevention program. While the IDRS with a score of ≥60 identified individuals with a high burden of cardiovascular risk factors [42], the majority had isolated IFG or NGT and not IGT. The results of our subgroup analyses suggest a trend towards greater reduction in diabetes incidence among those with IGT compared to those with isolated IFG or NGT. As mentioned previously, so far, lifestyle interventions have not been shown to be effective in reducing diabetes risk among those with isolated IFG [49–51]. Further research is required to determine the optimal cutoff for the IDRS to identify those at highest risk of developing diabetes. Alternatively, risk scores that are better at picking up people with IGT could be developed. Second, given the high burden of cardiovascular risk factors in the trial population [42] and improvements observed in some of these at 24 months, it is important to evaluate the potential longer-term benefits of the intervention on both diabetes incidence and cardiovascular risk. Finally, although the K-DPP intervention was low-cost and delivered mainly by lay people in community neighborhoods with support from local self-government bodies, it is important to do more research on how to increase program adherence and engagement, possibly by using more flexible modes of program delivery, e.g., at worksites and by text messaging. This research should also consider how continued program implementation beyond the current 12-month program for group participants can be supported by developing partnerships with other kinds of community organizations or partnerships that could deliver the intervention at scale in Kerala and elsewhere in India in the future.
In this low- and middle-income setting, a low-cost peer-support lifestyle intervention resulted in a nonsignificant reduction in diabetes incidence at 24 months in a high-risk population identified on the basis of a risk score. However, there were significant improvements in some cardiovascular risk factors and physical functioning score of the HRQoL scale.
|
10.1371/journal.pcbi.1001107 | Human Leg Model Predicts Ankle Muscle-Tendon Morphology, State, Roles and Energetics in Walking | A common feature in biological neuromuscular systems is the redundancy in joint actuation. Understanding how these redundancies are resolved in typical joint movements has been a long-standing problem in biomechanics, neuroscience and prosthetics. Many empirical studies have uncovered neural, mechanical and energetic aspects of how humans resolve these degrees of freedom to actuate leg joints for common tasks like walking. However, a unifying theoretical framework that explains the many independent empirical observations and predicts individual muscle and tendon contributions to joint actuation is yet to be established. Here we develop a computational framework to address how the ankle joint actuation problem is resolved by the neuromuscular system in walking. Our framework is founded upon the proposal that a consideration of both neural control and leg muscle-tendon morphology is critical to obtain predictive, mechanistic insight into individual muscle and tendon contributions to joint actuation. We examine kinetic, kinematic and electromyographic data from healthy walking subjects to find that human leg muscle-tendon morphology and neural activations enable a metabolically optimal realization of biological ankle mechanics in walking. This optimal realization (a) corresponds to independent empirical observations of operation and performance of the soleus and gastrocnemius muscles, (b) gives rise to an efficient load-sharing amongst ankle muscle-tendon units and (c) causes soleus and gastrocnemius muscle fibers to take on distinct mechanical roles of force generation and power production at the end of stance phase in walking. The framework outlined here suggests that the dynamical interplay between leg structure and neural control may be key to the high walking economy of humans, and has implications as a means to obtain insight into empirically inaccessible features of individual muscle and tendons in biomechanical tasks.
| Biological neuromuscular systems are generally able to perform a specified movement task in several ways – as they have significantly more degrees of freedom than mechanical constraints. Understanding how humans resolve these redundancies to drive individual muscles and tendons in typical joint movements is of interest in the fields of biomechanics, neuroscience and prosthetics. Many experimental studies have uncovered neural, mechanical and energetic features of individual muscle and tendon function in common tasks like walking and running. However, a unifying theoretical framework that explains the many independent empirical observations is yet to be established. In this work, we show that leg muscle-tendon morphology and neural co-ordination, together, enable efficient ankle movements in walking. This finding provides quantitative insight into the operation and performance of posterior-leg muscles and tendons in walking, and motivates the idea that different muscle-tendon units take on different mechanical roles to best actuate the ankle in gait. Results reported have implications both for better understanding neuromuscular co-ordination in gait, and for the design of lower limb prosthetic and orthotic technologies.
| A common feature in biological neuromuscular systems is the redundancy in joint actuation. Redundancies in actuating a joint with a prescribed force and motion can be classified at three levels. Joints can be actuated by multiple muscle-tendon units (MTUs) working simultaneously. At any instant, energy for MTU work could come from the series elastic tendon or from the active muscle. Each muscle has many sensors and can be controlled by multiple neural pathways acting together. Understanding how these redundancies are resolved in typical joint movements has been a long-standing problem in biomechanics, neuroscience and prosthetics [1], [2].
There is a large literature (reviewed in [3]) on objectives that might drive the way humans resolve neuromechanical redundancies. Several objectives ranging from metabolic cost, efficiency, and mechanical economy to fatigue and active muscle volume have been proposed as driving factors. Direct measurements in humans have revealed some details pertaining to the ‘inner workings’ of individual muscles and tendons resulting from the resolution of neuromechanical redundancies in natural execution of common tasks. Electromyography (EMG) has long quantified neurally stimulated electrical activity (activation) in individual muscles, and indicated which MTUs contribute to joint dynamics during the course of a movement [4]. Recently, ultrasonography has resolved ankle plantar flexor and knee extensor MTU strain into muscle strain and tendon strain during walking, running, and jumping [5]–[7]. Novel approaches using powered exoskeletons to replace leg muscle work have helped estimate the metabolic efficiency of ankle joint actuation in walking [8], [9]. Together, the above studies have uncovered critical neural, mechanical and energetic aspects of individual muscle and tendon contributions to joint actuation.
However, the abundance of research on the (a) driving objectives underlying and (b) empirical observations on redundancy resolution is not accompanied by a unifying theoretical framework that relates the two. There is a need to explain the breakdown of joint actuation, possibly driven by one or more of the above driving objectives, into observed individual element contributions.
Previous studies [10], [11] have proposed that the optimality of neural control for prescribed objectives can resolve individual muscle-tendon contributions to joint actuation in walking. These studies model leg MTUs with morphological parameters based on literature estimates, assert a control objective such as tracking biological joint mechanics or minimizing metabolic cost of transport, and obtain optimal muscle activation profiles for the specified objective. While the importance of neural control in determining the operation of individual muscles and tendons is undisputed, such approaches neglect the fact that many sets of activation patterns can correspond to similar values for the driving objective - making it difficult to uniquely resolve individual muscle activation profiles from an overall mechanical or energetic prescription. Further, several control objectives may be operating in tandem to generate neural activations given the highly non-linear, multi-input multi-output nature of the system - making it difficult to obtain optimal neural activations using a top-down approach. These observations reduce the utility of such approaches for explaining empirical results and making testable predictions on the workings of individual muscles and tendons within the system.
An alternative proposal for resolving individual muscle-tendon contributions to joint actuation in walking is found in optimal design. A starting point for such an approach lies in a recent study by Lichtwark & Wilson [12]. They propose that optimal muscle-tendon design for efficient actuation of an isolated MTU can explain empirically observed muscle and tendon strain profiles within the MTU. The empirically realistic nature of this proposal may directly stem from the well-documented fact that compliant tendons enable muscles to produce force economically [13], [14]. However, this proposal does not scale to explain the breakdown of joint actuation amongst individual elements, as the forces produced by individual MTUs are not known a priori for a given joint actuation.
Thus, existing optimal control and optimal design approaches are limited, albeit in different ways, by the very joint actuation redundancies they seek to address. Extra sources of information are needed to address this problem. EMG data contains information about muscle activity, and could potentially be used as a source of biologically realistic neural control commands to muscles. This promises to circumvent the above-mentioned difficulties in obtaining optimal muscle activations. Further, having muscle activation profiles could also enable a more systematic study of the effects of MTU structure (design) on the breakdown of joint actuation amongst individual elements. In other words, estimating muscle activations from the data allows a consideration of both neural control and muscle-tendon design, in tandem, on the operation of individual muscles and tendons.
Motivated by the above ideas, we have developed a theoretical framework to (a) address how the load of actuating a joint is shared amongst the many MTUs, (b) elucidate features of leg design and neuromuscular control enabling the breakdown and (c) clarify functional advantages arising from the load sharing. As a case study, we examine ankle joint actuation in human walking. We model the three primary leg MTUs contributing to ankle action in walking (Figure 1). Each MTU is characterized by (a) Hill-type muscle dynamics [15], (b) a common non-linear tendon model [16] and (c) a bilinear excitation-activation relation [3] - all of which are assumed to be internally consistent. These relations are parameterized with a minimal set of twelve muscle-tendon morphological features (representing leg MTU design). We conduct a computational exploration of the muscle-tendon design space for correspondence to well-known biological objectives. Specifically, for each set of system parameters, we actuate the model with joint state and muscle activations from healthy human gait data (Methods) to characterize the resulting joint torque and metabolic consumption. An overview of the modeling scheme is presented in Figure 1.
Our results are organized into five sections. First we present our estimates of muscle activations from EMGs recorded during human walking. In the second section, we characterize the leg parameter space by ability to produce human-like ankle torques and economy. We show that there is a unique parameter vector that is able to accomplish both, and that this unique vector corresponds to the maximum metabolic economy. Third, we present the optimal leg parameters, compare them with biological values and discuss their influence on metabolic economy. Fourth, we present model plantar flexor muscle and tendon strain predictions, compare them with two sets of independent empirical recordings and use them to evaluate mechanical power breakdown between muscle and tendon within each MTU. In the fifth section, we present metrics regarding the breakdown of ankle actuation amongst the two different plantar flexors.
Muscle activation is an indicator of a muscle's force-generation capability, indicated by the proportion of troponin bound to calcium [17]–[19]. It is driven by neurally stimulated electrical activity in the muscle. Since EMG data is a qualitative indicator of muscle electrical activity [4], it contains valuable information about individual muscle activity and can be useful in understanding the breakdown of joint actuation. However, quantitative uses of EMG data have been limited by variability in the signal and measurement artifacts. Here we show that considering dominant biophysical characteristics of the muscle activation build-up along with the randomness inherent in the EMG measurement yields repeatable and reasonable activation estimates.
Classic EMG analysis involves rectification and low-pass filtering [20], [21]. But low-pass filters smear out the filtered signal, leading to loss of both phase and amplitude information, particularly turn-on and turn-off of muscle activity [22]. Recently Sanger proposed a probabilistic method to resolve the signal variability and noise floor related problems in analyzing EMG signals [22]. In this paper, the muscle electrical activity driving the EMG signal was modeled as a jump-diffusion process:(1)where is a diffusion process with rate , is a jump process with frequency and represents a uniform distribution indicating that is a uniform random variable when there is a jump. The measured EMG signal was modeled as a random process with an exponential density and rate given by :(2)
Propagating the probability densities in a classic recursive Bayesian manner, to estimate the that best describes the observed EMG signal, Sanger reported excellent temporal resolution of EMG turn-on/turn-off during forced maximal contraction tasks. However, the biophysical relevance to analyzing EMG from dynamic tasks is limited by (a) the sharp, near-instantaneous turn-on and turn-off in the Sanger estimates, and (b) the lack of amplitude-buildup when the muscle is on (Figure 2).
We attribute this to differences between the modeled jump-diffusion process and the true buildup of muscle active state in normal tasks (Supplementary Text S2). The constant frequency and uniform amplitude of the jump process [22] compromises the history dependence of active-state buildup, causing sudden jumps in the estimates when the EMG signal turns on/off. Further, the Sanger model has the same jump rate for source and sink or for activation and deactivation. This neglects the differences in activation and deactivation time constants that are critical to muscle activation build-up [19]. Thus the Bayesian approach proposed in [22] appears to estimate the times when muscle electrical activity turns on/off, and not the muscle active state because activation dynamics (relating electrical activity to cross bridge formation) are not explicitly included.
One way to account for the activation dynamics would be to incorporate them directly into the jump-diffusion model and numerically evaluate a solution. We chose a simpler approximation, and applied the activation dynamics on the muscle electrical activity from Sanger's model to estimate muscle active state . Activation dynamics was specified by the classic bilinear form [3]:(3)
This differential equation models the history dependence in build-up of muscle activation, and captures differences between activation () and deactivation () time constants with the ratio . Notes on the biophysical relevance of our estimation procedure are available in Supplementary Text S2.
The muscle activation profiles estimated using our two-step procedure are shown in Figure 2. The intermediate Bayesian estimate has a step-like shape as it primarily captures the turn on and turn off of the muscle electrical activity measured by EMG. The estimated activations have profiles that are qualitatively expected from known temporal features of ankle muscle force build-up [4]. Further, the synergistic soleus and gastrocnemius muscles have similar profiles. Random step to step variations in EMG signals do not drastically change the estimated activation profiles. A repeatable ensemble average was obtained in as few as eight trials in cases of minimal motion artifact. The ensemble average estimates (Supplementary Text S2) show little variability in turn-on/turn-off times, and show greater variability in amplitude features (particularly when activation is high). The method and resulting estimates were found to be quite robust to normal, day-to-day variations in electrode placement for a given subject. We used our estimates of neurally stimulated muscle activations observed in walking to conduct the computational exploration (illustrated in Figure 1) of muscle-tendon morphologies.
Using the muscle activations and joint kinematics estimated from normal walking data, we actuated the leg muscle-tendon model parameterized by a set of morphological features . The parameter vector comprises the tendon reference strain , the tendon shape factor , the muscle maximum isometric force and the tendon slack length for each of the three ankle MTUs. We randomly generated sets of leg muscle-tendon parameter vectors, (from a uniform distribution with bounds stated in Supplementary Text S1), and computed both the model ankle torque profile, and metabolic energy consumed, , for each set:(4)
The resulting errors between model and human ankle torques are plotted against the model metabolic consumption (Figure 3).
Notable features of the plot include (a) the overall L shape, (b) a vertical boundary evidently representing the minimum energy that model muscles have to expend given the inputs, regardless of torque match and (c) an evidently systematic horizontal boundary below the population representing the best match between model and data. Each point along this horizontal boundary corresponds to a different metabolic consumption for the same level of error between model and human dynamics. A published empirical estimate of the range of metabolic consumption for ankle actuation in walking [8] is indicated, and is seen to be well-approximated by points exhibiting near-minimal economies, close to the the vertical boundary.
Remarkably, this overall parameter-space characterization reveals that the empirically observed realization is among the most economical of the many ways to produce human-like torque. Thus the human leg and the nervous system controlling it resolve the load-sharing redundancies in actuating the ankle most economically.
Points that best approximate human-like dynamics and optimal human-like metabolics lie near the bottom horizontal and left vertical boundary respectively. Thus points representing a logical intersection of the model's ability to best produce both human-like dynamics and metabolics lie in a small region at the lower-left corner (indicated by box in Figure 3). Points in this region not only have similar values of the torque and metabolic cost but also have similar values for the morphological parameters defining them. The coefficients of variance amongst parameter values in the corner region, listed in the caption of Figure 3, are low for most of the parameters (details in Supplementary Text S3). Further, all points outside the corner region compromise on either torque match, or economy, or both.
Thus, parameter vectors defining the corner region points can be identified computationally by encoding the simultaneous realization of two objectives (torque match and optimal economy) into a multi-objective problem. Solutions for such problems are generally sets of points that simultaneously realize both objectives as best as possible. These solutions, known as Pareto solutions, typically form a frontier along which the two objectives can be traded off against each other to varying degrees. In the special case that both objectives logically intersect at a mathematically sharp corner, there is a single strong Pareto optimal solution that best fulfils both objectives without any tradeoffs. As demonstrated above, our problem resembles this special case - within systematic limits of experimental precision, data variability and functional relevance. Thus it is possible to interpret our problem within the strong Pareto optimal framework, and simplify standard multi-objective optimization methods (such as Aggregate Objective Functions, Pareto ranking, evolutionary algorithms, or cost-constraint techniques [23]) to solve for the biologically realistic parameter vectors.
Our simplified approach relies on the observation that biologically realistic muscle-tendon morphological parameters (henceforth referred to as ) should (a) produce the normal human walking mechanics, and (b) minimize metabolic cost. To solve for we take a two-step path: (a) restrict the search to parameter vectors that enable the model to produce human-like torque (horizontal boundary), and (b) look, within the restricted space, for parameter vectors that optimize economy. Thus, the problem of finding is akin to a constrained optimization, performed by generating candidate parameter vector populations and iteratively focussing the search on the biologically realistic left corner (Methods).
For each of the five subjects, we used the training gait data to obtain activation and joint state estimates, automated the above exploration to find corner region parameters (listed in Supplementary Text S3) and defined the model with the optimal vector to be the ‘trained model’. We cross-validated the trained model against variations in input data (Supplementary Text S3) and proceeded to characterize the biological relevance of the trained model morphology.
The optimal leg morphological features for each subject were seen to fall within physiological ranges [18]. To gain insight into non-apparent features underlying the solution, we extracted both functional and geometrical features that significantly influence muscle-tendon action and the associated metabolics: (a) tendon stiffnesses and (b) muscle-tendon rest length ratios.
To compute these metrics, the trained model dynamics were solved numerically (Methods) to obtain muscle lengths , muscle velocities , tendon lengths , muscle-tendon unit force profiles and model ankle torques . Tendon stiffness was approximated as the best fit slope of the tendon force-length relation defined by the computed morphologies , and (Methods). Only regions of the tendon – curve where force was over of the peak force were considered to prevent the non-linear toe regions from artificially reducing the stiffness estimates. Geometric metrics were computed using the optimized morphological features. Since the ankle metabolic cost guiding identification of biologically realistic morphological parameters is dominated by stance-phase activity of the powerful soleus and gastrocnemius muscles, we focus on predictions for these two plantar flexor muscles.
Table 1 highlights the MTU structure trends. Notably, the model soleus and gastrocnemius tendon stiffness values (kSOL and kGAS) are quite compliant and lie within literature ranges [12], [24]. While the stiffness trends encapsulate effects of parameters , and , the effect of slack length is captured in a geometric effect described in the last two rows of Table 1. The ratio of muscle rest length to the computed tendon slack length is conserved for both plantar flexor muscles across subjects. This trend is consistent with published human cadaver studies as well [25].
Next, we sought to understand the significance of the optimal morphologies (specifically as embodied in the above in tendon compliance and the conserved / ratio trends) to metabolic economy. For this, we compared a metabolic efficiency metric accounting for the effects of tendon elasticity against the efficiency of muscle positive work alone. Muscle positive work efficiency was computed based on the metabolic cost of muscle mechanical work during active shortening (Equation 12). For comparison, a net joint level mechanical efficiency based on the total metabolic cost of performing mechanical work at the joint (inclusive of muscle work during active shortening, active lengthening and passive tendon contributions) was calculated (Equation 13). In the latter case, the metabolic and mechanical calculations are not restricted to muscle positive work phases, as the MTU dynamics can allow tendons to perform positive mechanical work at the joint even when the muscle cannot.
Table 2 details the resulting muscle positive work efficiency and the overall joint mechanical work efficiency. The average stance phase efficiency of muscles doing positive work (without regard to storage and release of elastic energy) is . This is consistent with empirically measured performance of isolated skeletal muscle doing positive work [8], [16]. Though the plantar flexor muscles themselves perform at ordinary efficiencies, accounting for tendon elastic energy contributions boosts their efficiency in performing joint mechanical work to a high net ankle mechanical efficiency of during stance phase (Table 2). To ensure this is not an over-estimate due to neglect of tendon viscosity, we recalculated with a nominal viscous loss of of the tendon elastic energy [18] - and obtained a joint work efficiency, still times higher than positive muscle work efficiency. The observation that accounting for elastic energy affords a dramatic increase in efficiency of joint work is qualitatively consistent with another recent report [8].
Thus the biologically realistic morphologies correspond to compliant tendons that store and release elastic energy to enhance joint work with little extra metabolic cost to muscles. As the elastic storage and release is timed to allow muscles to work efficiently, there is an optimal tendon slack length that is tuned to muscle optimal length and the input activation profiles. In summary, our exploration of the muscle-tendon morphological space predicts that the optimal muscle-tendon morphologies enable the nervous system to drive ankle muscles in high performance regimes.
We queried the model for further details regarding individual muscle and tendon operation regimes. Plantar flexor muscle and tendon length estimates from the model are shown in Figure 4. Across subjects, both soleus and gastrocnemius muscle strains were noticeably less than tendon strains. Plantar flexor tendons are stretched slowly over most of stance, and released quickly before toe-off just as the muscles shorten rapidly. In accordance with observations in the previous section, we see that the optimal morphologies enable the timely storage and release of tendon elastic energy (stretching and shortening of tendons), giving rise to efficient (near-isometric) muscle operation. The model's plantar flexor muscle strain predictions are qualitatively consistent with trends reported in independent ultrasonography-based in vivo measures [5], [6].
Further, the model captures the diversity represented in the in vivo data from different studies. For the gastrocnemius muscle, model profiles (Figure 4, Panel B) are consistent with ultrasound recordings reported in [5] for some subjects, and with the measures from [6] for other subjects (Figure 4, Panel C). Specifically, there are differences in early stance action that appear to arise largely from differences in early stance ankle angle, and orientation of the foot at the moment of ground impact. There are also differences in the degree of peak shortening towards toe-off. Thus, our results suggest that qualitative trend variations among different in vivo measures [5], [6] may arise from subject-to-subject gait variations and not necessarily due to differences in the ultrasonography techniques.
Beyond these qualitative observations, model soleus muscle peak strains (Figure 4, Panel A) are quantitatively consistent with those published in [5]. But quantitative differences exist between model predictions for the gastrocnemius muscle (Figure 4, Panels B and C) and the two sets of in vivo measurements. Model gastrocnemius peak shortening strains range from , while [5] and [6] report peak shortening strains of and respectively. To understand the reason for these differences in muscle strains, we studied the tendon and MTU strain profiles. Interestingly, model tendon lengths (Figure 4) and MTU lengths (not displayed) agree quantitatively with both sets of in vivo measures. However, this does not translate to quantitative agreement between model and the in vivo muscle strains. Since muscle length is a geometric function of the tendon and MTU lengths, the quantitative differences can be attributed to inconsistencies between the model's geometry and the complex in vivo geometry. Sources for discrepancy include (a) dimensions of the subjects studied, and (b) differences between our lumped element model geometry and the true anatomical geometries, arising possibly from the two-dimensional nature of our analyses (no volume or shape considerations) and from other model simplifications like constant pennation angles. Nevertheless, the overall trends in model muscle and tendon strain predictions are robust to these errors, and empirically realistic.
The value of our modeling effort extends well beyond enabling comparisons between our theory and published empirical measurements. Difficulties in directly measuring individual muscle and tendon forces within a muscle-tendon unit have precluded resolution of how the total MTU power output breaks down between the muscle and the tendon. Our analysis provides estimates of individual muscle and tendon forces, and therefore enables calculation of muscle power and tendon power within each MTU - as displayed in Figure 5. The most striking feature of these plots is that much of the MTU power arises from the tendons not the muscles. In particular, during the late stance positive power generation period, tendons provide over 80% of the MTU power across subjects. This is consistent with the above observations of tendon strains being much larger than muscle strains for both plantar flexors. Overall, the soleus MTU has higher peak MTU powers than the gastrocnemius MTU. This granularity of information motivates a more detailed study of similarities and differences in the operation of the different muscles and tendons.
The synergistic soleus and gastrocnemius muscles are similar in that they shorten significantly right before toe-off, and move with low velocities, as is expected from their compliant tendons. But there are two easily apparent differences in the movement of these two muscles - during mid and late stance respectively.
First, the length estimates of the two plantar flexors are very different in mid-stance (Figure 4). In particular, the soleus lengthens (eccentric operation) during mid-stance while the gastrocnemius appears characteristically isometric ( GC). Thus the soleus absorbs mechanical work in mid-stance, while the gastrocnemius holds the tendon in place at the muscle end and does little mechanical work. This observation is consistent with ultrasound literature reports [5], [7].
Moreover, there are differences in late stance operation of the two muscles, which are apparent from an analysis of muscle velocities (Figure 6, Panel A). During pre-toe-off shortening, the soleus operates at a peak velocity of , while the gastrocnemius operates at a larger peak velocity of (). These peak toe-off velocities fall in well-recognized ranges. Muscle efficiency is known to peak around for a wide range of muscle lengths, while muscle mechanical power peaks around [13], [26]. Within precision of these empirical numbers, our results suggest that stance-end muscle operation may be driven by peak efficiency for the soleus and peak mechanical power for the gastrocnemius.
Motivated by this idea, we compared each muscle's positive mechanical work and metabolic consumption during the positive work phase of late stance. Table 3 reports ratios of positive mechanical work, metabolic energy cost and the resulting efficiencies of the two muscles in late stance. While the relation between soleus and gastrocnemius mechanical work and metabolic cost had varying trends across subjects, soleus is consistently more efficient than the gastrocnemius. In other words, soleus achieves a much bigger bang (mechanical work-wise) for its buck than the gastrocnemius. Further, the fact that the mechanical work ratios are low () for most subjects - despite the fact that soleus is nearly times as large (in cross-section area) as gastrocnemius - suggests that the gastrocnemius may be more powerful than soleus on a per fiber basis (due to the velocity difference noted above). The above results argue that soleus may be an economical force producing muscle, while gastrocnemius fibers may be more powerful and metabolically demanding than soleus fibers. Details of the metabolic and mechanical powers of the two muscles are available in Supplementary Figure S1.
To further elucidate roles of the two plantar flexors, we studied the metabolically optimal breakdown of ankle torque between the two (Figure 6, Panel B). The ratio of peak soleus and gastrocnemius torque contributions to ankle actuation is an average of across subjects. This ratio does not directly follow either the ratios of the optimal values or the metabolic costs of the two muscles. It is likely due to a combination of the , metabolic costs and the muscle activations. Interestingly, the most efficient partitioning of ankle torque amongst the synergistic plantar flexor muscles appears commensurate with the ratios of soleus and gastrocnemius stiffness reported in Table 1. This suggests that the soleus and gastrocnemius tendon extensions may be similar, which is just what we see in Figure 4.
Finally, the muscle operation and load-sharing results arise uniquely from the optimal parameter vectors. A point along the horizontal boundary of Figure 3 - that has a greater metabolic consumption than the biologically realistic corner points - also corresponds to (a) different muscle velocities than the optimal corner points (one-to-one relation between metabolic cost and muscle velocities), (b) different forces generated for the same activations, and (c) different (non-optimal) load-sharing solutions (see Supplementary Text S3).
Our results describe how humans resolve redundancies within and between MTUs involved in ankle joint actuation. We have demonstrated that there is a unique leg morphology which (a) most economically relates activations and angles from gait data with torques therein, (b) produces the above data via plantar flexor muscle motions, tendon motions, and metabolic performance that are consistent with experimental observations and (c) resolves empirically inaccessible features ranging from individual muscle forces and metabolic demand to mechanical power and working efficiencies. This morphology (defined by maximum isometric forces, tendon material properties and slack lengths for the ankle MTUs) is anatomically realistic, and Pareto optimal for the two objectives of torque match and efficiency.
To understand features of the morphology that enable this multi-objective optimality, we make a few observations about the solution. The optimal muscle isometric forces result in the most efficient breakdown of joint torque amongst the different MTUs. The tendon slack lengths balance the capacity for a timely buildup of force in response to the activations against the need to cycle tendon elastic energy for efficient force generation. Finally the optimal tendon material properties make for just the right stiffness values to produce the required joint torque but with just enough compliance to reduce muscle metabolic cost. These features indicate that leg muscles and tendons are designed to enable a metabolically optimal realization of human-like ankle mechanics under neural controls observed in normal walking.
Interestingly, the optimal parameters for any one MTU do not arise independently of those for the other MTUs, as both efficiency and torque match are net objectives for all the MTUs operating together. Rather the optimal structural parameters are a solution for the system as a whole to achieve the two objectives. Therefore, unlike the Lichtwark & Wilson study [12] that predicted the most efficient force generating design for an isolated MTU, our results predict the leg structure that most efficiently breaks down ankle torque amongst the different MTUs, and then the muscles and tendons within each MTU - all in an empirically consistent manner.
Further, the load-division implied by the optimal leg morphology also reveals a role division amongst the different muscles. The gastrocnemius muscle has a very compliant tendon - allowing it to work isometrically like a clutch in mid-stance, store energy in the tendon slowly and release it rapidly in late stance to produce high mechanical power on a per fiber basis - akin to a catapult (as proposed in [5]). The soleus muscle, on the other hand, has a stiffer tendon and larger maximum isometric force - making it very inefficient to generate high power by rapid shortening in late stance (as the metabolic cost increases commensurately with shortening velocities and scales with ). Thus soleus operates at lower muscle velocities in late stance, and can be thought of as an efficient force generator. To our knowledge, this is the first observation of differences in late stance operation of the two plantar flexor muscles in human walking, and remains to be tested in future experiments. Interestingly, previous studies have found differences in energy management by adjacent leg extensor muscles in insect locomotion [27], [28]. Thus differences in morphology of adjacent synergistically controlled MTUs may diversify MTU function across species. For smaller muscles like the tibialis anterior, which contribute little torque or mechanical power, the efficiency objective appears flat across the MTU design space. Nevertheless, these muscles may have important roles to play in fine control or sensing - that could be explored in a future study. Much of the biomechanics literature has focused on single objective problems. It has been acknowledged that multiple objectives could be acting in tandem [3]. Our results motivate the novel idea that one overall objective (of economically producing human-like torque) can give rise to different objectives (power, efficiency, control) for each individual element in the system.
It is possible that the neural controller may be ‘managing’ the different muscles and tendons spanning the ankle by ‘assigning’ different roles to each - based on their morphology - to efficiently accomplish ankle actuation in walking. In other words, the dynamical interplay between neural control (modeled here with estimated activations from human EMG data) and leg structure (modeled here with MTU morphologies) may in itself be optimal for the overall objective of efficiently generating ankle torque. This idea stands in contrast to previous proposals of the optimality of neural control alone [10], [11], [29] or of MTU structure alone [12] for a prescribed control or performance objective. The interaction between neural control and muscle-tendon unit mechanics may be facilitated by any subset of many neural pathways - particularly reflex pathways. However, there are many possible reflexes (muscle force [30], fascicle length and velocity [31], [32]) that can modulate impedance of ankle muscles at any given point in the gait cycle. This neural pathway redundancy has posed a challenge to decipher when and by how much each reflex pathway may contribute to activation of any given muscle. A systematic approach to quantify the role of specific reflexes and resolve this redundancy is desirable.
Our framework has implications as a starting point for such an endeavor. Since every reflex pathway is sensitive to specific changes in muscle state (force, length and velocity), inspecting the dominant trends in our muscle state and activation estimates provides insight on possible pathways contributing to the observed state changes. For instance, a period of muscle stretch and low activation followed by a period of isometric behavior and a coincident rise in activation is likely to correspond to a stretch reflex (gastrocnemius in mid-stance period of walking). A period of similarly shaped force and activation profiles may involve positive force feedback (soleus in late stance of walking). Such observations generate hypotheses on how impedance is modulated within the neuromuscular system. Forward dynamical simulations with perturbation analyses could used to test such hypotheses and quantify contributions of different reflexes to legged dynamics. Insights gained from such efforts are of interest for applications in the control of assistive devices [15]. Also, understanding the reflex responses that (along with tendon and MTU dynamics) modulate leg extensor muscle impedance after heel-strike may add perspective to studies on neuro-motor control during mechanical contact [33].
Finally, the framework described in this study also has implications as an analytical tool to probe empirically inaccessible metrics to understand regulation, roles, operation and performance of individual elements in gait. The first steps would be to extend the theory across muscles and joints for walking. Difficulties in obtaining inputs from gait and EMG data for deeper and more proximal muscles could be overcome via a forward dynamical simulation approach wherein both the timings of muscle activity, along with the muscle-tendon morphological parameters, are evaluated for our two objectives. If feedback control loops linking muscle state to activation are also included, perhaps other objectives of dynamical stability could be considered in tandem - in a similar framework, to quantitatively characterize the interplay between neural control and leg morphology. Accounting for feed-forward contributions to this interplay constitutes an important challenge that needs to be overcome. Another natural extension would be to characterize different tasks with our framework. A question of fundamental interest is to understand whether the same leg morphology is energetically optimal for the neural controls and joint mechanics across tasks. An affirmative answer would suggest that, for any specified task, humans select the joint mechanics that minimizes metabolic cost for the legs they have. A negative outcome would imply that human leg morphology and neuromuscular co-ordination are specifically energetically optimal for self-selected-speed walking.
This study was conducted in strict accordance with the principles expressed in the Declaration of Helsinki. The study was approved by the MIT Committee on the Use of Humans as Experimental Subjects (protocol number 0903003157). All subjects provided written informed consent for the collection of data, subsequent analysis and publication of results.
To investigate the leg dynamics underlying the data, we modeled the major muscle-tendon units contributing to ankle function in normal walking. Anatomically, this corresponds to the big MTUs responsible for ankle joint rotation in the sagittal plane - the soleus and gastrocnemius plantar flexors with the Achilles tendon split amongst them, and the tibialis anterior dorsiflexor (Figure 1). Both the medial and lateral heads of the gastrocnemius muscle were represented as one effective muscle, since they act synergistically in gait. Other muscle-tendon units spanning the ankle joint were not included as their contribution to ankle torques and energetics in normal, level-ground walking is minuscule [4]. The muscle-tendon dynamics actuating the ankle joint are outlined below.
For each MTU, we minimized the number of free model parameters by (a) using literature values where they are known to be reliable ( [30]), (b) fixing values in documented general ranges when dynamics are insensitive to precise values ( [35]), and (c) taking advantage of inter-relations between parameters (example, muscle optimal length and tendon slack length are inter-related by subject dimensions, so was set to a scaled nominal value from [35]). Set values for the above three parameters for each of the three MTUs are provided (with sensitivity notes) in Supplementary Text S1.
The other twelve parameters each correspond to a key morphological feature (slack length, reference strain and force, shape factor) of the three modeled muscle-tendon units. They are known to be difficult to measure in vivo [36], cadaver measurements are rather unreliable [25] and there is no fool-proof procedure for scaling nominal values from literature to subject dimensions [37]. Thus the model is characterized by twelve free parameters, denoted as .
The leg muscle-tendon unit dynamics, characterized by , relates neurally commanded muscle activation () and joint angle to joint torque . Considering a specified , active state profiles for the three muscles and joint angles from the data (or equivalently the muscle-tendon unit lengths ) constrain Equations 5 and 6 for each unit. The relations can be simultaneously solved for and (implicitly ), which in turn can be used to evaluate and for each unit, and thus to calculate the model ankle torque .(9)
Each possible vector specifies a certain (a) model ankle torque, (b) distribution of that ankle torque amongst the MTUs, (c) division of mechanical work and MTU strain amongst the associated muscle and tendon. Since each parameter vector exacts different mechanical work from the model muscles, it also causes them to expend different amounts of metabolic energy to actuate the ankle.
Muscle metabolic consumption is known to result from the heats of activation, maintenance, shortening, resting, and other molecular processes involved in muscle force generation [38]. While many schemes have been proposed to comprehensively account for these different components [10], [12], [39], they depend on setting several parameters correctly. To avoid accuracy and sensitivity issues that accompany multi-parameter metabolic calculations, we used empirically-based heat measures from classically accepted and well-reproduced muscle metabolic studies [13], [26], [38]. The data, reproduced in Supplementary Text S1 (along with sensitivity bands), relates the normalized metabolic power required for isolated muscle activity with the normalized contractile velocity.
At any time , if muscle is activated to level , and is contracting at velocity , then the instantaneous (denormalized) metabolic power consumed by that muscle is:(10)where is implicitly a function of the parameters due to the specified muscle-tendon dynamics, is the function in Supplementary Text S1, and the maximum isometric force when muscle is activated to level during a natural task is . Overall metabolic energy cost for the muscle is the time integral of its instantaneous metabolic power:(11)
To avoid numerical errors in computing cost, metabolic power was only accounted for when , which is a small enough approximation that it does not affect accuracy. Total metabolic cost of actuating the ankle muscles in 1 gait cycle is for the three muscles to .
In summary, given the data-driven inputs and model dynamics, each set of morphological parameters defining the dynamics specifies a model torque profile, and a model metabolic consumption - as indicated in equation 4.
For analysis, two efficiency metrics were calculated for stance phase of the gait cycle:
|
10.1371/journal.pntd.0005393 | Water, Sanitation and Hygiene (WASH) and environmental risk factors for soil-transmitted helminth intensity of infection in Timor-Leste, using real time PCR | No investigations have been undertaken of risk factors for intensity of soil-transmitted helminth (STH) infection in Timor-Leste. This study provides the first analysis of risk factors for intensity of STH infection, as determined by quantitative PCR (qPCR), examining a broad range of water, sanitation and hygiene (WASH) and environmental factors, among communities in Manufahi District, Timor-Leste.
A baseline cross-sectional survey of 18 communities was undertaken as part of a cluster randomised controlled trial, with additional identically-collected data from six other communities. qPCR was used to assess STH infection from stool samples, and questionnaires administered to collect WASH, demographic, and socioeconomic data. Environmental information was obtained from open-access sources and linked to infection outcomes. Mixed-effects multinomial logistic regression was undertaken to assess risk factors for intensity of Necator americanus and Ascaris infection.
2152 participants provided stool and questionnaire information for this analysis. In adjusted models incorporating WASH, demographic and environmental variables, environmental variables were generally associated with infection intensity for both N. americanus and Ascaris spp. Precipitation (in centimetres) was associated with increased risk of moderate-intensity (adjusted relative risk [ARR] 6.1; 95% confidence interval [CI] 1.9–19.3) and heavy-intensity (ARR 6.6; 95% CI 3.1–14.1) N. americanus infection, as was sandy-loam soil around households (moderate-intensity ARR 2.1; 95% CI 1.0–4.3; heavy-intensity ARR 2.7; 95% CI 1.6–4.5; compared to no infection). For Ascaris, alkaline soil around the household was associated with reduced risk of moderate-intensity infection (ARR 0.21; 95% CI 0.09–0.51), and heavy-intensity infection (ARR 0.04; 95% CI 0.01–0.25). Few WASH risk factors were significant.
In this high-prevalence setting, strong risk associations with environmental factors indicate that anthelmintic treatment alone will be insufficient to interrupt STH transmission, as conditions are favourable for ongoing environmental transmission. Integrated STH control strategies should be explored as a priority.
| We present a detailed analysis of WASH, environmental and demographic factors associated with intensity of STH infection in Manufahi District, Timor-Leste, using qPCR. Investigation of risk factors for intensity of STH infection is rarely undertaken, and prior analyses have used microscopic-based eggs per gram of faeces (epg) measures, which are of lower diagnostic accuracy than qPCR. Additionally, few analyses have investigated combined WASH and environmental risk factors in association with STH. This is important due to the extensive potential interrelatedness of environmental, social, behavioural and host factors in any given setting influencing STH survival and transmission. This analysis uses categorical intensity of infection variables for Necator americanus and Ascaris spp., and advanced statistical modelling to adjust for multinomial intensity outcomes, dependency of observations, effects of poverty, and confounding from other measured variables. As such, this analysis provides a comprehensive assessment of risk factors for STH in Manufahi District, Timor-Leste. This is of importance for development of policy and programmatic decisions; risk factors need to be considered not only for their clinical and statistical significance, but more broadly in terms of what may represent modifiable pathways for STH transmission.
| Surprisingly little evidence convincingly demonstrates the benefits of water, sanitation and hygiene (WASH) interventions on reducing soil-transmitted helminth (STH) infections [1,2]. Yet it is widely believed that WASH improvements together with anthelmintics could break STH transmission cycles in settings in which anthelmintics alone are insufficient [3,4]. There has been inadequate epidemiological investigation of the role of improved WASH in reducing the STH burden, but there is a growing need for evidence to enable more effective investment in WASH and integrated strategies for STH control.
Intensity of STH infection is important to assess in epidemiological analyses. STH are highly aggregated in humans, with a small number of people harbouring large numbers of helminths, and the majority harbouring few or none [5]. As with prevalence, intensity of worm burden is marked within various groups of the community such as different age groups and gender [6]. This well-described phenomenon is a key feature of this macroparasite relationship with the human host. For quantitative investigations it is therefore problematic to use solely prevalence of infection as an indicator of STH burden or transmission, because large changes in intensity may only be accompanied by small changes in prevalence [6]. STH do not reproduce within the host; infection intensity depends on the time and extent of exposure [7]. Where STH are endemic, maximum worm intensity usually occurs at ages five to ten for Ascaris lumbricoides and Trichuris trichiura, and in adolescence or early adulthood for hookworm [6]. Whilst the reasons for this are unknown, it may be due to behavioural and social factors, nutritional status, genetic and immunological factors [5,8–11]. There is evidence that some individuals are predisposed to heavy or light STH infections [9,12]. Intensity of T. trichiura infection reacquired by an individual after treatment has been found to be significantly correlated with the intensity of infection prior to treatment [13]. Additionally, intensity of infection with STH has been identified as substantially greater when any of the species occurred in combination with one or more of the others [14], probably also due to exposure, genetic and immunological factors, which could then act in determining risk of associated morbidities. Despite this knowledge, there is much focus on the use of prevalence to measure STH infection endemicity.
The relationships between intensity of STH infection and risk factors have been inadequately explored, yet could provide useful information as to why intensities differ by host age, environment, and helminth species. Because a key feature of the STH life cycle is the soil-dwelling stage, STH survival, development and transmission potential all rely on a complex assortment of environmental, social, behavioural and host factors. Therefore, in addition to investigating associations between WASH and STH, community-based associations must be considered within their environmental context [15,16]. Although more evidence is required, STH associations with WASH have been systematically appraised [2]. Studies have additionally identified temperature, rainfall, soil porosity and pH, vegetation and elevation ranges as influencing N. americanus larval development and STH transmission [16,17]. We have previously separately reported on WASH [18] and environmental [19] risk factors for STH prevalence in Manufahi District, Timor-Leste. Given exposure-related risks, and associations between heavy-intensity infection and morbidity, this analysis was conducted to investigate whether WASH- and environmental-related risk factors in this district may also be associated with infection intensity, using categories derived from quantitative PCR (qPCR), a highly sensitive and specific diagnostic technique [20]. By combining data on both WASH and environmental risk factors this analysis provides a more complete picture of risks and thereby augments the current knowledge of risk factors for STH in Timor-Leste. Knowledge of WASH risk factors will be used to inform control strategies in this country. Whilst many environmental risk factors may not be modifiable, the inclusion of these factors will enable targeting of control strategies to areas of greatest need. This is one of very few extensive investigations of combined WASH and environmental risk factors for STH undertaken. It is additionally the first epidemiological analysis of risk factors undertaken using categorised intensity of STH infection from qPCR.
From 24 communities, 2827 eligible people provided baseline survey data, of whom 2152 participants (1038 males, 1114 females) completed both an individual questionnaire and provided a stool sample and were included in this analysis (Table 1, [18]). Using our infection intensity cut-points, more than half (52%) of participants had heavy-intensity N, americanus infection; 10% had heavy-intensity Ascaris infection (Table 1). There was very low prevalence of water or sanitation infrastructure, and most households owned few assets. Most heavy-intensity Ascaris infection occurred in children (Fig 1). Heavy-intensity N. americanus infections were more spread across age groups (Fig 2). Heavy-intensity Ascaris infection varied significantly by socioeconomic quintile (P = 0.012); N. americanus infection intensity did not (P = 0.468).
Environmental factors were associated with N. americanus infection (Table 2). Of particular note, precipitation, measured in centimetres, was significantly associated with a six-fold increased risk of moderate-intensity (adjusted risk ratio [ARR] 6.1, 95% confidence interval (CI) 1.9, 19.3), and seven-fold increased risk of heavy-intensity infection (ARR 6.6, 95%CI 3.1, 14.1), compared to no infection. Sandy-loam soil around the house was associated with more than two-fold higher risk of moderate-intensity (ARR 2.1, 95%CI 1.0, 4.3), and heavy-intensity infection (ARR 2.7, 95%CI 1.6, 4.5), respectively, compared to other soil types. Increasing elevation above sea-level was associated with slightly reduced risk of heavy-intensity infection (ARR 0.90, 95%CI 0.83, 0.97), but was not associated with moderate-intensity infection (ARR 0.94, 95%CI 0.83, 1.1). Increasing normalised difference vegetation index (NDVI) was associated with increased risk for heavy-intensity infection (ARR 1.1, 95%CI 1.0, 1.1). Soil acidity was not included in N. americanus regression models (P>0.2 on univariable analysis).
Co-infection with Ancylostoma spp. was associated with four-fold higher risk of heavy-intensity N. americanus infection (ARR 4.1, 95%CI 2.1, 8.0). G. duodenalis was marginally non-significant for heavy-intensity infection (ARR 0.71, 95%CI 0.49, 1.0). Due to the sex by age interaction term results are reported separately for females and males within age groups. Relative to no N. americanus infection, a significant gradient of increased risk of heavy N. americanus infection intensity with increasing age group was evident for females (ARRs increasing from 3.2 to 9.6; see Table 2), however this was less evident for moderate-intensity infection (with the exception of being aged 65 years or older having four-fold increased risk of infection; ARR 4.4, 95%CI 1.6, 11.9). For males, relative to no infection, being aged 18 to 64 years was significantly associated with more than three-fold increased risk of any intensity infection (moderate-intensity ARR 3.3, 95%CI 1.3, 8.7; heavy-intensity ARR 3.6, 95%CI 1.8, 7.3). Sex in participants aged one to five years (i.e. reference group) was not associated with intensity of infection. A gradient of generally increasing risk of moderate- and heavy-intensity infection was also evident with worsening socioeconomic quintile (being significant across most subgroups for heavy-intensity), with people in the poorest quintile having more than twice the risk of infection for both intensity levels (moderate-intensity ARR 2.0, 95%CI 1.1, 3.7; heavy-intensity ARR 2.2, 95%CI 1.3, 3.6).
Few associations were found between WASH variables and STH outcomes in adjusted analyses. Of note is that a shared piped water supply was associated with strongly reduced risk of heavy-intensity infection compared to an unprotected stream (ARR 0.32, 95%CI 0.12, 0.84), and use of surface water was associated with twice the risk of moderate-intensity infection compared to an unprotected stream (ARR 1.9, 95%CI 1.1, 3.2). Boiling household water was associated with half the risk of moderate-intensity N. americanus infection compared to not boiling water (ARR 0.52, 95%CI 0.34, 0.80). Having one preschool-aged child in the household was protective against heavy-intensity N. americanus infection (ARR 0.57, 95%CI 0.40, 0.82). For moderate-intensity infection having one preschool-aged child in the house was not significant (ARR 0.81, 95%CI 0.52, 1.3), but having more than one was associated with reduced risk (ARR 0.57, 95%CI 0.34, 0.94). People reporting three or more bowel motions during the previous 24 hours (indicating diarrhoea) was associated with reduced risk of heavy-intensity infection compared to people who reported less than three bowel motions (ARR 0.40, 95%CI 0.17, 0.96). People who reported having access to anthelmintic drugs and people who reported actually taking deworming treatment within the previous 12 months, was not associated with risk of infection in adjusted models, despite these factors being highly significant in univariable analysis for heavy-intensity infection. Methods of post-defecation anal cleansing, and shoe wearing, all of which were highly significant in univariable analyses for heavy-intensity infection, did not emerge as risk factors in adjusted analyses.
Factors significantly associated with Ascaris infection were age, and environmental variables, particularly alkaline soil and elevation above sea level (Table 3). Alkaline soil was significantly associated with highly reduced risks of moderate-intensity (ARR 0.21, 95%CI 0.09, 0.51), and heavy-intensity Ascaris infection (ARR 0.04, 95%CI 0.01, 0.25, note low numbers) compared to acidic soils. Neutral pH soil showed no association with risk of infection. Increasing elevation was associated with Ascaris infection, with observations of a mild gradient of increasing risk with increasing infection intensity (moderate-intensity ARR 1.3, 95%CI 1.2, 1.4; heavy-intensity ARR 1.4, 95%CI 1.2, 1.7). Increasing NDVI was also associated with mildly increased risk of heavy-intensity infection (ARR 1.2, 95%CI 1.1, 1.4). No WASH variables emerged as risk factors for Ascaris infection. Increasing age was associated with reducing risk of both moderate and severe infection intensity on a gradient that was significant for many age groups (particularly for heavy-intensity infections). Sex and socioeconomic status were not risk factors for Ascaris infection intensity.
This analysis presented the first investigation of combined WASH, environmental and demographic factors for intensity of STH infection in Timor-Leste. Using PCR-derived intensity of infection categorisation, similar infection intensity profiles to previous epg-based profiles [25] were found for each of Ascaris and N. americanus, with the most intense Ascaris infections in children, declining intensity and prevalence in adulthood, and prevalence and intensity of N. americanus being high in both childhood and adulthood. For N. americanus, heavy-intensity infections occurred in older age groups, although at low proportions. Whilst current risk factor models were not separately analysed by age groups, these results are in agreement with previous findings of different age-specific risk factors for different STH species in the study area [18]. This highlights the potential role of exposure-related risk factors, although other factors, such as acquisition of some level of immunity, may play a role [25].
It has previously been hypothesised that sex and age associations with STH are strongly related to exposure-associated behaviours [26]. Females showed a highly significant, increasing gradient for risk of heavy N. americanus infection with increasing age. Although less significant, a gradient was also evident for moderate-intensity N. americanus infection. Whilst overall male sex in those aged one to five was not a significant risk factor for N. americanus infection intensity compared to females of this age group, there was again an observation of greater heavy-intensity infection in males aged 18 to 64 relative to males aged one to five. These observations suggest that there are additional age- and sex-related factors occurring. This may include age as an expected indicator of time-accumulation given STH do not multiply in the host [14] and the longevity of N. americanus [27]. Alternatively, there could be exposure behaviours in older adults (compared to children) that are important to identify as they may be amenable to modification. There could also be differences in host immunity, particularly at different ages. Increased animal and soil contact through agricultural activities represents a direct potential transmission pathway (particularly in males) that requires further exploration. Further investigation into the female-age group association with heavy-intensity infection need to be undertaken; this could reflect particular household-related practices undertaken by women but not men. Further activities, such as constructing daily activity diaries, would be valuable to enable further insights in this setting. Alternatively, the findings of different sex and age patterns may be indicative of other factors such as host genetics [26].
Mixed-effects multinomial models were used to investigate the statistical relationship between intensity of STH infection, and WASH and environmental risk factors, whilst accounting for heterogeneity within village and household random effects. The lack of autocorrelation identified in semivariograms after accounting for large-scale environmental trends indicates that environmental variables explained the majority of spatial correlation in the data [19]. In our adjusted models incorporating WASH, demographic and environmental variables, environmental variables were generally associated with the greatest ARRs for infection intensity for both N. americanus and Ascaris spp. Precipitation was associated with increased risk of N. americanus infection of any intensity, but not Ascaris intensity. It is important to note that the precipitation variable included in these analyses was derived from 50-year averaged data from the driest month [19]; it is not reflecting seasonality, which could have had an impact on N. americanus survival rates in the soil [17]. Seasonal fluctuations could affect transmission potential but are not considered likely to have a strong influence on infection patterns given the longevity of N. americanus in the human host [17].
High rainfall contributes to suitably moist conditions for eggs and larvae to survive in soil, including the propensity for N. americanus larvae to remain near the soil surface and thus be available for human infection [17], but for Ascaris, excess rainfall may have negative impacts, possibly because the eggs sink lower in the soil as rainfall drains away. Our analysis showed strong associations between sandy-loam soil and highly increased risks of N. americanus infection, yet conversely, no significance in adjusted models for Ascaris spp. Observational associations between hookworms and sandy soil have been reported since the early 1900s (reviewed in [17]). Significance of soil type and rainfall likely reflect an important difference in life cycles and transmission potential between these two STH. N. americanus survive in the external environment as motile ensheathed larvae, but Ascaris spp. are present as (non-motile) eggs. The interrelated features of large-particle “sandy” soil, which tends to be less dense, aids both larval motility and water draining during/after rainfall, being therefore more amenable to N. americanus larval survival [17,26] and subsequent transmission potential. Ascaris eggs, on the other hand, are more susceptible to extremes, being able to dessicate in dry soil and to retard development in extremely wet soils [28]; this supports the lack of association between Ascaris spp., sandy soil and precipitation in this analysis. These factors, plus the shorter developmental time to infectivity in the soil of N. americanus compared to Ascaris [16], may contribute to the considerably greater prevalence of N. americanus.
We have previously reported on a protective association observed between alkaline soil type and Ascaris infection [19]; in this current analysis there was some evidence of a gradient of increasing infection intensity, although numbers were low and this finding therefore requires verification. Other studies use soil acidity data in spatial analyses [29,30], with one study reporting associations between acidic soil and increased infection risk ([30]; although this study used categories of pH that were all considered acidic compared to our definitions which defined neutral soil type as pH 6.6 to 7.3 [31]). Generally, soil acidity information is still rarely collected, yet this is an important potential determinant that could vary with precipitation, and other ecological or land use factors. Further analysis of pH ranges in epidemiological studies will contribute to knowledge of the optimal conditions for survivability of these helminths.
Differences in motility and survivability also potentially explain the direct association between increased elevation and Ascaris intensity of infection, with downhill runoff and draining after rainfall potentially facilitating survivability (and hence transmission) of those Ascaris eggs that remain in soil at higher elevations (i.e. those that do not get washed away); it would be plausible that Ascaris eggs that are washed downhill may be washed into rivers and streams, or lie within saturated environments that are less conducive to development. For N. americanus this was an inverse relationship for heavy-intensity infection; the protective association seen from elevation may reflect lower temperatures at higher elevations (as temperature was not included in multivariable analyses due to its high correlation with elevation). Negative correlations between hookworms and elevation, and, less consistently, positive correlations between A. lumbricoides and elevation, have previously been reported (reviewed in [17]).
Given high STH prevalence, poverty and poor existing WASH infrastructure [18], and the large quantity of risk factors investigated, few WASH risk factors have emerged in these analyses. Homogeneously poor access to improved WASH resources in study communities would limit our ability to find major associations [18], and is the most likely explanation for this. No significant WASH risk factors for Ascaris intensity of infection were found in adjusted models. For N. americanus, the protective association with boiling water against moderate-intensity infection is slightly surprising. There is evidence that N. americanus larvae can survive and remain infective for several days in water (decreasing with duration of water exposure) [32]. Whilst there is negligible published evidence for N. americanus infection via ingestion, this finding points to faecal contamination of drinking water sources as a possible exposure pathway. Water supply effects were also not significantly associated with intensity of N. americanus infection in the expected direction, with different levels of risk between surface water and an unprotected spring; both of which are unimproved water sources [33]. This could possibly be due to location of communities downhill from springs (thus positioned for gravity-fed flow), whilst communities may be generally uphill from surface water. There may additionally be a greater tendency for people to remove footwear when going to surface water, compared to (potentially smaller) springs. Alternatively, this may reflect heavy N. americanus contamination in the vicinity of particular water sources in the study area. Self-reporting error or a misunderstanding of water source definitions used in our study are also possible explanations. The protective effect of shared piped water, but not other ‘improved’ sources such as tubewells, is of interest and may reflect a heightened level of hygiene awareness in situations where multiple households use the same source, or alternatively, high correlation between some other variable and this one (although confounding and collinearity were investigated). The general lack of WASH associations, particularly with levels of sanitation, is similar to results that we have reported previously for prevalence [18]; however it was not previously clear whether this was because prevalence models were age-stratified, which could have affected power to detect effects. Lack of WASH risk factors therefore most likely reflects homogeneously poor access to WASH infrastructure, with flow-on impacts on amenable hygiene behaviours, in these communities, or a true lack of association with STH in this district. Alternatively, with a multinomial outcome, analyses could have adequate power to detect only moderate-large associations (see limitations).
Prevalence of N. americanus has previously been reported to be significantly associated with low socioeconomic status in this study area [18]. As has previously been identified socioeconomic status in this community reflects relative poverty that was still measurable within a general setting of poverty [18] and it is interesting that, for N. americanus, slightly higher estimates of association were seen for socioeconomic strata in heavy-intensity relative to moderate-intensity infection. This highlights an advantage of investigating socioeconomic status in defined districts on high-resolution (i.e. village-level) scale as opposed to national scale; it has been previously reported that between- and within-village heterogeneity may limit the usefulness of socioeconomic proxies in aggregated large-scale analyses [28]. The greater level of detail from this multinomial model provides additional insight into the N. americanus-poverty relationship. An interesting protective effect was the presence in the household of preschool-aged children; possibly this reflects adoption of hygienic behaviours when there are young children to protect them from disease exposures. The finding of reduced risk of heavy infection in people who reported three or more bowel motions is not surprising given that diarrhoea causes dilutive effects on quantities of helminths [34]. Our finding that recent deworming was not significantly associated with infection in adjusted models may be due to self-report error, with possible confusion about medications received.
The risks associated with environmental variables have important implications for STH control. The high rainfall, mountainous, tropical environment combined with high levels of poverty, poor WASH infrastructure and behaviour, and the longevity of STH eggs and larvae survival in soil [16], provides a fertile environment for STH transmission in this district. This is a challenge for helminth control because environmental variables themselves are not modifiable. Despite this, awareness of high-risk factors can influence other activities, primarily hygiene- and sanitation-associated behaviours to manage environmental risks. This provides a strong justification for investment in WASH activities irrespective of their individual statistical significance in risk factor analyses, as this is an exposure-reduction pathway that can potentially be manipulated. As current evidence for hygiene behaviours on STH control is sparse (reviewed in [1,2]), further research on the hygiene behaviours that could have greatest impact in this scenario needs to be undertaken as a priority.
This analysis is an important contribution to an ongoing RCT that will assess the benefits of augmented albendazole with WASH for STH control in Manufahi District, Timor-Leste [21]. As well as a detailed understanding of baseline WASH infrastructure and behaviours upon which to benchmark trial-related improvements in WASH, the knowledge of environmental factors is an essential prerequisite for effective targeting of interventions.
This is an observational analysis and, as such, cause and effect cannot be determined. As has been noted previously [18] much of the WASH data collected involved self-report of infrastructure and behaviours. Presence, type and cleanliness of household and village latrines were verified by interviewer observation. Self-reporting is a frequently-encountered drawback of measuring WASH characteristics. Further, extensive heterogeneity in assessing WASH behaviours on STH outcomes makes assessment of WASH characteristics challenging [15,35]. An important research priority is to develop specific WASH measurement guidelines for STH control. Power calculations indicated power to detect low associations for N. americanus and moderate associations for Ascaris infection intensity in multinomial models.
There are particular strengths to this study. This is one of very few epidemiological investigations of risk factors for STH infection intensity; this is particularly important to assess for environmental factors, given the links to STH transmission dynamics and correlations with morbidity. In this paper a community-based risk analysis is presented that combines high-resolution environmental, WASH and demographic variables in adjusted models. Advanced statistical techniques have been used to adjust for multinomial intensity outcomes, dependency of observations, effects of poverty, and confounding from other measured variables. As with all analyses, there is the possibility of residual confounding from unmeasured factors. However this provides the most comprehensive assessment of STH risk factors that we have identified in any setting.
A further strength is the use of PCR; a highly sensitive and specific technique [20] that is increasingly used for STH diagnosis. PCR-derived intensity of infection categorisation is a recent development, and requires further validation in different epidemiological settings [23]. Notwithstanding the need for further refinement of cut-points, different risk factors for moderate and heavy-intensity STH infections were found in this study area, with some evidence of a scale of increasing risk for factors such as soil type. This contributes useful, and highly relevant, information on risk factors within these communities. Use of infection intensity to determine risk factor associations requires more investigation. In particular, use of prevalence alone could mask significant intensity-related associations. This may mean that key evidence for WASH benefits may be overlooked in epidemiological studies that use prevalence of infection as the outcome. The possibility that WASH significance may be underreported in this way has been inadequately explored.
With intensity of STH infection as the outcome, a comprehensive risk analysis of environmental, WASH and demographic variables is presented for communities in Manufahi District, Timor-Leste. Strong risk associations with environmental variables were identified. However, generally few associations with WASH risk factors were evident. This raises the importance of accurate measurement of WASH, and the need for clear guidelines on measuring WASH epidemiological research. This result also has important implications for STH control activities. Even in the absence of WASH significance, WASH infrastructure and behavioural-related activities are the only identified mechanism that could reduce or prevent transmission in an environment of high STH transmission potential. In this setting, anthelmintic treatment alone will not interrupt STH transmission; this provides a strong justification for application of integrated STH control strategies in this district.
This analysis used baseline data from 18 communities in a cluster randomised controlled trial (RCT), supplemented with data from an additional six communities, in Manufahi District, Timor-Leste (Australian and New Zealand Clinical Trials Registry ACTRN12614000680662) [21]. STH have recently been reported as endemic in this community, with prevalence of N. americanus of 60% and Ascaris spp. of 24%, as detected by qPCR [18].
The University of Queensland Human Research Ethics Committee; the Australian National University Human Ethics Committee; the Timorese Ministry of Health Research and Ethics Committee; and the University of Melbourne Human Research Ethics Committee approved the study protocol. Participant informed consent processes included explaining the study purpose and methods, and obtaining signed consent from all adults and parents or guardians of children under 18 years [21]. Children aged less than 12 months were excluded [21].
The RCT commenced in May 2012. Detail on the RCT design is provided in the trial protocol [21]. A baseline survey of 18 communities involved in the RCT, and six additional communities, was conducted between May 2012 and October 2013. All communities surveyed were rural, and agrarian occupations predominated. Manufahi District has terrain varying from flat coastal plains to relatively mountainous inland areas (with elevation exceeding 1100 metres in some communities). It is a tropical region, with very high average rainfall of 190cm [19] and a wet season extending for close to ten months of the year. The average annual temperature is 24.5°C [19].
A single stool sample per participant was collected and fixed in 5% potassium dichromate. Multiplex qPCR was used to analyse stool samples for the presence and intensity of STH infection. Details on the qPCR diagnostic method are provided elsewhere [20].
Village, household and individual level questionnaires encompassing a broad range of potential WASH and socioeconomic risk factors were administered by trained field workers [18,21]. Interviewer observation of household and village latrines, their type and cleanliness was undertaken; all other questions were self-reported. Data were collated and entered into a Microsoft Access database and extracted to STATA 13.0 (Stata Corporation, College Station, Texas) for error checking.
Individual-level data were linked to questionnaire and parasitological outcomes and household GPS coordinates [18,21]. Principal component analysis was used to create a wealth index, based on ownership of household assets (animals, transport and appliances), house floor type, reported income, and presence of electricity [18,22]. Using eigenvalues above 1, four principal components were retained and used to produce a final wealth score which was categorised into quintiles of relative socioeconomic status [18].
Outcome variables were intensity of N. americanus and Ascaris infection, which were analysed separately. Intensity of infection was derived from qPCR DNA cycle threshold (Ct) values, and categorised into two groups: (i) heavy-intensity, and (ii) moderate- to light-intensity infection (hereafter called “moderate intensity”) using algorithms generated from seeding experiments to correlate Ct-values to eggs per gram of faeces (epg) equivalents. Full detail of this method is provided elsewhere [20,23]. Exposure variables were WASH variables from study questionnaires, grouped into domains of related variables (e.g. household sanitation; household water supply; household hygiene; household socioeconomic status), and environmental variables that were sourced separately. Environmental variables were selected for analysis based on reported prior relationships with STH development [17], and availability via open-access sources. Temperature, precipitation, elevation, soil texture, soil pH, landcover and vegetation data were selected for analysis (Table 4) and processed using the geographical information system ArcMap 10.3 (ESRI, Redlands, CA) [19]. Very few environmental analyses incorporate information on soil texture and soil pH; it has been possible to incorporate these variables due to soil surveys conducted in the study region between 1960 and 1965 [24]; soil type was not considered to have changed dramatically since that time. A range of environmental variables related to the above factors was produced according to long-term average data, seasonal periods, and spatial resolution [19], with household as the data point, and a 1 km buffer applied (whereby the median raster value within a 1 km radius of the household was used [19]). Quality checks and exploratory analyses were undertaken to determine the most suitable version of each variable for analysis. Separate assessment of spatial autocorrelation was undertaken using semivariograms of residuals from multivariable models of selected environmental variables, with household and village random effects [19]; no additional autocorrelation was identified [19]. The analysis of environmental covariates in this study was limited to risk factor investigation. Predictive risk maps for STH infection in Manufahi District are published separately [19].
Variables were investigated for multicollinearity according to likely relationships determined from literature, using tetrachoric analysis and the STATA “collin” user written package, according to the type of variable. Temperature and elevation were collinear; each variable was analysed in separate univariable models and subsequent variable selection was based on lower Akaike’s Information Criterion (AIC), indicating better predictive performance of the model. Chi-squared tests were conducted to compare intensity of infection by age, sex and socioeconomic quintile. Using categorised intensity of infection as the outcome, univariable and multivariable mixed effects multinomial regression was undertaken, with household and village random effects to account for dependence of observations. Regression analyses were undertaken for N. americanus and Ascaris spp. separately.
Regression models were not age-stratified due to insufficient numbers for some combinations of outcome and explanatory variables. Univariable regression was undertaken for each risk factor, with inclusion of variables in multivariable regression if they had P<0.2 on the Wald test in univariable analyses. All multivariable models included age group, sex, and socioeconomic quintile as covariates. Forward stepwise variable addition was used with variables retained if P<0.1 within, then across, domains of variables, until the most parsimonious adjusted model for each outcome was achieved. A categorised age variable, and a sex*age interaction term, were investigated, as the association between sex and the outcome was anticipated to vary by age group. Interactions were investigated by developing models without, then with, the interaction term and comparing these using the likelihood ratio test, with P<0.1 being the inclusion criterion for the interaction. Applying this criterion, the interaction term was retained in the final N. americanus model, but not the Ascaris model. A 5% significance level was used, however this analysis reports results of up to 10% significance, which is important for epidemiological interpretation. Analyses were conducted using generalised structural equation models in STATA 14.1 (Stata Corporation, College Station, Texas). Due to uncertainty regarding the linearity of the association of continuous environmental variables and the infection outcomes, quadratic terns were also investigated in all models; however as none of these quadratic terms were significant in the adjusted models, these results are not presented. Post-analysis power calculations indicated 80% power, with a 5% significance level, to detect relative risks of 1.2 to 1.8 for N. americanus infection intensity (depending on level of intensity), and, reflecting lower prevalence overall, relative risks of 2.7 to 3.9 for Ascaris infection intensity.
|
10.1371/journal.ppat.1006600 | TRIM32-TAX1BP1-dependent selective autophagic degradation of TRIF negatively regulates TLR3/4-mediated innate immune responses | Toll-like receptor (TLR)-mediated signaling are critical for host defense against pathogen invasion. However, excessive responses would cause harmful damages to the host. Here we show that deficiency of the E3 ubiquitin ligase TRIM32 increases poly(I:C)- and LPS-induced transcription of downstream genes such as type I interferons (IFNs) and proinflammatory cytokines in both primary mouse immune cells and in mice. Trim32-/- mice produced higher levels of serum inflammatory cytokines and were more sensitive to loss of body weight and inflammatory death upon Salmonella typhimurium infection. TRIM32 interacts with and mediates the degradation of TRIF, a critical adaptor protein for TLR3/4, in an E3 activity-independent manner. TRIM32-mediated as well as poly(I:C)- and LPS-induced degradation of TRIF is inhibited by deficiency of TAX1BP1, a receptor for selective autophagy. Furthermore, TRIM32 links TRIF and TAX1BP1 through distinct domains. These findings suggest that TRIM32 negatively regulates TLR3/4-mediated immune responses by targeting TRIF to TAX1BP1-mediated selective autophagic degradation.
| TLR3/4-mediated signaling needs to be effectively terminated to avoid excessive immune responses and harmful damages to the host. In this study, we provide genetic evidence to show that the E3 ubiquitin ligase TRIM32 negatively regulates TLR3/4-mediated innate immune and inflammatory responses. Trim32-/- mice are more sensitive to the inflammatory death upon Salmonella typhimurium infection. We found that TRIM32-TAX1BP1-dependent selective autophagic degradation of the adaptor protein TRIF effectively turned off TLR3/4-mediated innate immune and inflammatory responses. Our findings reveal a novel mechanism for terminating innate immune and inflammatory responses mediated by TLR3/4.
| The innate immune system is the first line of host defense against pathogen invasion. After detection of structurally conserved components of the invading pathogens by so-called pathogen recognition receptors (PRRs), the host cells initiate a series of signaling cascades which ultimately induce the transcription of downstream antiviral genes, such as type I interferons (IFNs) and inflammatory cytokines, to induce innate immune and inflammatory responses as well as facilitate adaptive immunity [1,2,3,4]. However, excessive immune and inflammatory responses cause tissue damages and serious diseases such as septic shock [5].
Toll-like receptors (TLRs) are evolutionarily conserved PRRs that play critical roles in host defense against various pathogens. TLRs contain an extracellular domain, a transmembrane domain, and a conserved cytoplasmic toll/IL-1 receptor (TIR) domain. Upon ligand stimulation, the TIR domains of TLRs mediate their homo- or hetero-dimerization [6], and act as platforms to recruit downstream TIR domain-containing adaptor proteins and other signaling molecules, leading to the activation of transcription factors such as IRF3 and NF-κB. These transcription factors collaborate to induce the transcription of a series of downstream antiviral genes [7]. Most TLRs except TLR3 and TLR4 signal through the TIR-containing adaptor MyD88. TLR3, which recognizes viral dsRNA and plays important roles in innate antiviral responses, signals through the TIR-containing adaptor TRIF but not MyD88 [8]. TLR4, which recognizes LPS of bacteria and is essential for innate and inflammatory responses to infected bacteria, signals through MyD88 to activate NF-κB and TRIF to activate NF-κB and IRF3 [9]. Double knockout of TRIF and MyD88 results in completely abolishment of LPS-induced activation of NF-κB, whereas TRIF-deficiency results in abolishment of LPS-induced activation of IRF3 [9].
Protein degradation is one of the main strategies which have been employed by host cells to inactivate proteins in biological processes. Autophagy is an essential homeostatic process by which damaged organelles, protein aggregates, and invading cytoplasmic microbes are sequestered in double-membraned autophagosomes and delivered to the lysosome for degradation [10]. There are growing evidences that autophagy can be highly selective [11]. Selective autophagy depends on the cargo receptors, including p62, TAX1BP1, NDP52 and so on, which are able to bind to special cargoes and dock onto the forming phagophores [11]. Certain selective autophagy receptors have been reported to be involved in regulation of immune responses. For examples, the cytosolic DNA viral sensor cGAS and intracellular Salmonella typhimurium can be degraded via p62- and TAX1BP1-dependent selective autophagy respectively [12,13]. Whether selective autophagy is involved in the regulation of other immune processes are unknown.
The tripartite motif-containing proteins (TRIMs) of E3 ubiquitin ligase families have been demonstrated to play critical regulatory roles in regulation of immune responses [14,15]. TRIM32 has been reported to mediate K63-linked polyubiquitination of MITA/STING and regulates innate immune responses to RNA and DNA viruses in human cell lines [7]. In this study, we generated TRIM32-deficient cells and Trim32 knockout mice, and found that TRIM32 negatively regulated TLR3/4-mediated innate immune and inflammatory responses. Biochemical and cellular analysis revealed that TRIM32 mediated selective autophagic degradation of TRIF through TAX1BP1. Our findings suggest that TRIM32-TAX1BP1-dependent selective autophagic degradation of TRIF is an important negative regulatory mechanism of TLR3/4-mediated innate immune and inflammatory responses.
Previously, it has been demonstrated that TRIM32 mediates K63-linked polyubiquitination of MITA/STING and regulates virus-triggered induction of downstream antiviral genes in human cell lines [7]. To investigate the functions of TRIM32, we utilized TRIM32 gene knockout mice (S1A & S1B Fig). We found that TRIM32-deficiency had no marked effects on the mRNA levels of downstream antiviral genes Ifnb1 and Isg56 as well as inflammatory cytokine genes Tnfa and Il6 induced by Sendi virus (SeV) or herpes simplex virus 1 (HSV-1) in mouse embryonic fibroblasts (MEFs), bone marrow-derived macrophages (BMDMs) and dendritic cells (BMDCs) (S1C Fig), suggesting that TRIM32 does not regulate virus-triggered signaling in primary mouse cells. However, we found that TRIM32-deficiency potentiated poly(I:C) (a synthetic dsRNA ligand for TLR3)- and LPS (a ligand for TLR4)- but not PGN (a ligand for TLR2)- or R848 (a ligand for TLR7)-induced transcription of downstream genes Ifnb1, Isg56, Tnfa and Il6 in BMDMs, BMDCs and mouse lung fibroblasts (MLFs) (Fig 1A). Consistently, poly(I:C)- and LPS-induced phosphorylation of IRF3 and IκBα (hallmarks for IRF3 and NF-κB activation respectively) was dramatically increased in Trim32-/- MLFs in comparison to their wild-type counterparts (Fig 1B). LPS has been reported to induce both MyD88- and TRIF-dependent signaling, which usually results in serious harmful inflammation in vivo. Monophosphoryl lipid A (MPLA), a derivate of LPS, mainly induces TRIF- but not MyD88-dependent signaling, leading to some protective immune responses in vivo instead [16,17,18]. Interestingly, TRIM32-deficiency also increased MPLA-induced transcriptions of these genes in BMDMs (Fig 1C). These results suggest that TRIM32 negatively regulates TLR3/4- but not TLR2/7-mediated signaling in primary mouse cells. Consistently, overexpression of TRIM32 markedly inhibited poly(I:C)- and LPS-induced activation of the IFN-β promoter, ISRE and NF-κB in human HEK293-TLR3 and HEK293-TLR4 cells respectively in reporter assays (Fig 1D). In similar experiments, TRIM32 did not inhibit IFNγ-induced activation of the IRF1 promoter (Fig 1E).
To investigate the role of TRIM32 in TLR3/4-mediated innate immune and inflammatory responses in vivo, age- and sex-matched Trim32+/+ and Trim32−/− mice were intraperitoneally injected with poly(I:C) plus D-galactosamine or LPS. D-galactosamine is an agent usually used to enlarge inflammatory damage of liver, since poly(I:C) alone is insufficient to cause inflammatory death of mice. As shown in Fig 2A, poly(I:C)- and LPS-induced production of IFN-β, TNFα, and IL-6 was significantly increased in the sera of Trim32−/− compared to Trim32+/+ mice. Consistently, more serious inflammation was observed in the lungs of Trim32−/− mice injected with poly(I:C) plus D-galactosamine or LPS (Fig 2B). Trim32−/− mice showed an early death onset and a significantly higher percentage of lethality within 40 hours in comparison with their wild-type counterparts after injection of poly(I:C) plus D-galactosamine (Fig 2C) or LPS (Fig 2D). It has been reported that poly(I:C) and LPS are able to induce TRIF-dependent cell death which might contribute to poly(I:C)- and LPS-induced death of mice [19,20]. Therefore, we also explored whether TRIM32 is involved in TLR3/4-mediated and TRIF-dependent cell necrosis. The results showed that TRIM32-deficiency had no marked effects on poly(I:C)- and LPS-induced cell death in cell viability assays (S2 Fig). We have also explored the role of TRIM32 in MPLA-induced TRIF-dependent protective immune response in mice. Unlike LPS, MPLA does not cause much inflammatory response, and MPLA alone is insufficient to cause inflammatory death of mice. Therefore, we used D-galactosamine to enlarge the inflammatory response induced by MPLA. Interestingly, though MPLA-induced increased transcriptions of type I IFNs and inflammatory genes in Trim32-/- cells (Fig 1C), MPLA plus D-galactosamine-induced serum inflammatory cytokine levels and inflammatory death were markedly decreased in Trim32-/- mice (Fig 2E & 2F), suggesting that TRIM32 plays an important role in MPLA-induced TRIF-dependent protective immune response in vivo.
We have also explored the role of TRIM32 in immune and inflammatory responses to Salmonella typhimurium infection. As shown in Fig 3A, Trim32-/- mice carried less Salmonella typhimurium in their livers and spleens compared with that of their wild-type littermates at 8 days post oral administration of Salmonella typhimurium, suggesting that Trim32-/- mice exhibited more efficient clearance of invaded Salmonella typhimurium than the wild-type mice. Consistently, a larger number of viable immune cells existed in the spleens of Trim32-/- mice (Fig 3B). Trim32-/- mice produced much higher levels of inflammatory cytokines including TNF-α and IL-6 (Fig 3C) and showed much more serious inflammatory damage of their small intestinal villus (Fig 3D) after oral adiminstration of Salmonella typhimurium, which led to a higher sensitivity to Salmonella typhimurium-induced loss of body weight and inflammatory death of Trim32-/- mice (Fig 3E & 3F). These results suggest that TRIM32 negatively regulates TLR3/4-mediated innate immune and inflammatory responses in vivo.
We next investigated the molecular mechanisms of TRIM32 in the regulation of TLR3/4-mediated signaling. Reporter assays showed that TRIM32 inhibited TRIF-, but not TBK1- and IRF3-mediated activation of ISRE (Fig 4A). Furthermore, both overexpression and endogenous coimmunoprecipitation experiments indicated that TRIM32 interacted with TRIF (Fig 4B and 4C). In addition, we routinely found that TRIM32 dramatically destabilized TRIF but not TRAF3, TBK1 or IRF3 in our co-transfection experiments (Fig 4D). Endogenous experiments indicated that TRIM32-deficiency markedly attenuated poly(I:C)-induced degradation of TRIF in MLFs (Fig 4E). These results suggest that TRIM32 mediates the down-regulation of TRIF, which is an adaptor protein specifically utilized by TLR3/4 but not other TLRs.
Since TRIM32 is an E3 ubiquitin ligase, we examined whether TRIM32 destabilizes TRIF via the ubiquitin-proteasomal pathway. Unexpectedly, the E3 enzyme-inactive mutants of TRIM32, TRIM32(C40S) and TRIM32(ΔRING), destabilized TRIF as efficient as the wild-type TRIM32 (Fig 4F). Furthermore, TRIM32 failed to catalyze polyubiquitination of TRIF (Fig 4G). Consistently, reconstitution of either wild-type TRIM32 or TRIM32(ΔRING) in Trim32-/- cells could inhibit poly(I:C)- and LPS-induced transcription of Ifnb1, Tnfa and Il6 genes to similar levels (Fig 4H). These data suggest that TRIM32 mediates the down-regulation of TRIF independent of its E3 ligase activity.
Protein degradation is one of the main strategies involved in inactivating proteins in biological processes. Two major systems exist for protein degradation, including the ubiquitin-proteasome and autophagy-lysosome pathways. We found that TRIM32-mediated degradation of TRIF could be inhibited by the lysosomal inhibitor NH4Cl and the autophagic inhibitor 3MA but not the proteasomal inhibitor MG132 (Fig 5A), suggesting that TRIM32 probably mediates degradation of TRIF via an autophagic pathway. Confocal microscopy experiments showed that poly(I:C) stimulation caused aggregation of GFP-LC3 (a marker of autophagy) in HEK293-TLR3 cells (Fig 5B). Poly(I:C) stimulation also caused colocalization of TRIF with GFP-LC3 in HEK293-TLR3 cells (Fig 5C). To further confirm that autophagic degradation pathway is involved in TRIF degradation upon stimulation, we used MG132 and 3MA to pre-treat Trim32+/+ and Trim32-/- BMDMs for 2 hours before poly(I:C) stimulation. The results showed that 3MA pre-treatment had no marked effects on TRIF level in both un-stimulated Trim32+/+ and Trim32-/- cells, but attenuated poly(I:C)-induced degradation of TRIF in Trim32+/+ but not Trim32-/- cells (Fig 5D). MG132 pre-treatment markedly increased TRIF level in un-stimulated cells and also attenuated poly(I:C)-induced degradation of TRIF in both Trim32+/+ and Trim32-/- cells (Fig 5D). TRIF has been reported to be degraded by TRIM38 via the ubiquitin-proteasome dependent pathway, and TRIM38-deficiency increases TRIF level in un-stimulated cells and attenuates poly(I:C)-induced degradation of TRIF [21]. Consistently, knockdown of TRIM38 increased TRIF level in un-stimulated Trim32-/- cells, and also attenuated poly(I:C)-induced degradation of TRIF to a larger extent in these cells (Fig 5E). Taken together, these results suggest that TRIM32 is involved in the autophagic degradation of TRIF induced by poly(I:C) stimulation. Additional experiments showed that knockdown of LC3B, which is an important marker of autophagy and is required for fusion to the lysosomes, markedly inhibited TRIM32-mediated degradation of TRIF (Fig 5F), whereas overexpression of TRIM32 dramatically enhanced the interaction of TRIF with LC3B-II (Fig 5G), which is a basic membrane component of autophagesomes and derived from LC3B-I during autophagy [22]. In similar experiments, overexpression of TRIM32 did not cause the conversion of LC3B-I to LC3B-II (Fig 5H). Moreover, deficiency of ATG7, which is an essential E1 ligase for LC3B-II formation, inhibited poly(I:C)- and LPS-induced degradation of TRIF (Fig 5I). These results suggest that TRIM32 mediates degradation of TRIF through the autophagy-lysosome pathway. Consistently, pre-treatment of cells with balifomycin, an inhibitor for fusion of autophagosomes and lysosomes, markedly attenuated poly(I:C)-induced down-regulation of TRIF (Fig 5J). Endogenous TRIM32 constitutively associated with TRIF in un-stimulated cells, and their association slowly decreased following poly(I:C) stimulation (Fig 5J), suggesting that TRIM32 disassociates from TRIF-containing autophagosomes at the later stage of stimulation.
The autophagic pathways can be distinguished as the canonical or the selective autophagic pathway. The selective autophagy receptors deliver cargoes to the autophagosomes for selective degradation [23]. It has been shown that ULK1/2, FIP200 and ATG13 are critical for initiation of the classical autophagic pathway [23]. We found that deficiency of ULK1/2, FIP200 or ATG13 had no marked effects on the degradation of TRIF induced by poly(I:C) or LPS treatment (Fig 6A–6C), suggesting that poly(I:C)- and LPS-induced degradation of TRIF is not via the canonical autophagic pathway. It has been shown that NDP52 serves as a selective receptor for TRIF and TRAF6 for their selective autophagic degradation [24]. However, in our experiments, we observed that NDP52 failed to interact with and promote degradation of TRIF (Fig 6D & 6E). In similar experiments, NDP52 promoted the degradation of TRAF6 (Fig 6D). Instead, we found that TRIF interacted with other selective receptors including TAX1BP1, OPTN and p62 (Fig 6E), but only overexpression of TAX1BP1, but not OPTN or p62 down-regulated the level of TRIF (Fig 6F). Consistently, knockdown of TAX1BP1 inhibited TRIM32-mediated degradation of TRIF (Fig 6G). In addition, knockdown of TAX1BP1 but not NDP52 attenuated poly(I:C)-induced degradation of TRIF (Fig 6H). Furthermore, knockdown of TAX1BP1 markedly impaired endogenous association of TRIF with LC3 induced by poly(I:C) stimulation (Fig 6I). These results suggest that TAX1BP1 but not NDP52 mediates the selective autophagic degradation of TRIF. Consistent with the biochemical results, qPCR experiments showed that knockdown of TAX1BP1 potentiated poly(I:C)-induced transcription of Ifnb1, Cxcl10 and Il6 genes (Fig 6J).
To explore the mechanism of TRIM32- and TAX1BP1-mediated autophagic degradation of TRIF, we tested a straightforward hypothesis that TRIM32 acts as a bridge protein for TRIF-TAX1BP1 interaction. Confocal microscopy indicated that overexperssion of TRIM32 promoted colocolization of TRIF and TAX1BP1 in certain aggregates that were positive for the autophagosome marker GFP-LC3 (Fig 7A) or lysosome marker GFP-LAMP1 (Fig 7B). Endogenous coimmunoprecipitation experiments indicated that TRIM32-deficiency abolished poly(I:C)-induced association of TRIF with TAX1BP1 as well as attenuated poly(I:C)-induced degradation of TRIF (Fig 7C). These results suggest that TRIM32 acts as a bridge protein for TRIF-TAX1BP1 interaction following poly(I:C) stimulation.
Domain mapping experiments indicated that the NHL (aa360-655) and BBOX (aa66-139) domains of TRIM32 are required for its interaction with TAX1BP1 and TRIF respectively (Fig 7D). Similar experiments indicated that TRIM32 interacted with the middle TIR domain (aa386-475) of TRIF (Fig 7E), whereas TAX1BP1 interacted most strongely with the C-terminal domain (aa476-732) of TRIF (Fig 7E). These results suggest that TRIM32 links TAX1BP1 and TRIF through distinct domains (Fig 7F).
Additionally, we have also explored whether TRIM32 and TAX1BP1 are recruited to lipid rafts of membrane where TLR3/4 recruits TRIF for signaling. Cellular fractionation experiments indicated that membrane-associated TRIF was increased at 0.5 hour after poly(I:C) treatment and then decreased at 1 hour probably because of the degradation of TRIF (S3 Fig). TRIM32 constitutively existed in both cytosol and membrane franction, and poly(I:C) treatment had no marked effects on its distribution. Interestingly, TAX1BP1 only existed in the cytosol either before or after poly(I:C) treatment. Furthermore, TRIF-deficiency had no marked effects on the subcellular location of TRIM32 and TAX1BP1 either before or after poly(I:C) treatment (S3 Fig). These results suggest that TAX1BP1 is not recruited to lipid rafts of membrane, and TRIM32-TAX1BP1-TRIF association occurs in the cytosol after TRIF is dis-associated from the TLR3/4 receptor complexes on the membrane.
In this study, we investigated the role of TRIM32 in TLR3/4-mediated signaling in mouse primary immune cells and in vivo by genetic and biochemical studies. TRIM32-deficiency potentiated poly(I:C)- and LPS- but not R848- or PGN-induced transcription of downstream genes Ifnb1, Isg56, Tnfa and Il6 in BMDMs, BMDCs and MLFs. TRIM32-deficiency also elevated the serum cytokine levels induced by poly(I:C) and LPS, and renders the mice more susceptible to death triggered by administration of poly(I:C) and LPS or Salmonella typhimurium infection. These findings suggest that TRIM32 negatively regulates TLR3/4-mediated innate immune and inflammatory responses.
It has been shown that TRIM32 is an E3 ubiquitin ligase which regulates both DNA- and RNA viruses-triggered induction of type I IFNs in several human cell lines [7]. The current study indicates that TRIM32 is not required for induction of downstream antiviral genes induced by both DNA and RNA viruses in primary mouse cells or in mice. It is possible that TRIM32 functions in different cellular processes between human and mouse cells. TRIM proteins belong to the largest E3 ubiquitin ligase family in mammals, and it has been previously shown that some TRIM family members have distinct functions between human and mouse [25]. In contrast with the observations that the E3 ligase activity of TRIM32 is required for its roles in virus-triggered signaling in human cell lines [7], several results from the current study suggest that the E3 ligase activity of murine TRIM32 is not required for its negative regulatory roles in TLR3/4-mediated signaling. Firstly, the E3 enzyme-inactive mutants of TRIM32 destabilized TRIF as efficiently as the wild-type protein. Second, reconstitution of both the wild-type and E3 enzyme-inactive TRIM32 into Trim32-/- cells inhibited the transcription of downstream genes induced by poly(I:C) and LPS. TRIF is a critical adaptor protein for TLR3/4-mediated innate immune and inflammatory responses. Poly(I:C) or LPS stimulation causes a rapid and dramatic degradation of TRIF to avoid sustained activation of TRIF and expression of type I IFNs and inflammatory cytokines. Previous studies demonstrate that the E3 ubiquitin ligases WWP2 and TRIM38 target TRIF for degradation and inhibit TLR3/4-mediated innate immune responses [21,26]. Both WWP2 and TRIM38 catalyze K48-linked polyubiquitination of TRIF and promote TRIF degradation via the well-established ubiquitin-proteasome system. Instead, TRIM32 promotes TRIF degradation via the autophagic pathway, since the autophagy inhibitor 3MA and lysosome inhibitor NH4Cl but not the ubiquitin-proteasome inhibitor MG132 impaired TRIM32-mediated degradation of TRIF. In addition, TRIM32 promoted the interaction of TRIF with LC3B-II, which is the critical component for autophagosome formation. Interestingly, TRIM32 promotes TRIF degradation via the selective instead of the classical autophagic pathway, since deficiency of components of the selective but not classical autophagic pathway inhibited TRIM32-mediated, as well as poly(I:C)- and LPS-induced degradation of TRIF. Our results also indicated that TRIM38 but not TRIM32 down-regulated TRIF level in un-stimulated cells, whereas TRIM32 contributed to ligand-induced degradation of TRIF. Therefore, TRIM38 and TRIM32 regulate TRIF-mediated signaling through distinct mechanisms.
Several experiments suggest that the selective autophagic receptor TAX1BP1 but not NDP52 is involved in TRIM32-mediated autophagic degradation of TRIF. TAX1BP1 but not NDP52 interacted with TRIF. Overexpression of TAX1BP1 but not NDP52 promoted degradation of TRIF, whereas knockdown of TAX1BP1 but not NDP52 impaired TRIM32-mediated as well as poly(I:C)-induced degradation of TRIF. Furthermore, knockdown of TAX1BP1 markedly impaired poly(I:C)-induced endogenous association of TRIF with LC3.
Our experiments suggest that TRIM32 acts as a link for TRIF and TAX1BP1. Confocal microscopy showed that TRIM32 promoted colocalization of TRIF and TAX1BP1 in certain aggregates which are positive for the autophagosome marker GFP-LC3 or the lysosome marker GFP-LAMP1, while TRIM32-deficiency abolished endogenous association of TRIF with TAX1BP1 induced by poly(I:C). Domain mapping experiments indicated that the BBOX and NHL domains of TRIM32 were required for its interaction with the TIR domain of TRIF and TAX1BP1 respectively, whereas TAX1BP1 interacted with the C-terminal domain of TRIF. These results suggest that TRIM32 links TRIF to the TAX1BP1 autophagosomes through distinct domains.
Based on our data, we propose a working model on the regulatory role of TRIM32 in TLR3/4-mediated innate immune responses (Fig 7G). Ligand binding to TLR3/4 leads to the recruitment of the critical adaptor protein TRIF. TRIF in turn recruits downstream components, leading to activation of several transcription factors and ultimate induction of downstream innate immune and inflammatory genes. Upon activation, TRIF is recruited by TRIM32 to TAX1BP1-containing and LC3-associated autophagosomes for degradation, contributing to termination of TLR3/4-mediated innate immune and inflammatory responses. Our findings suggest that selective autophagic degradation is an important regulatory mechanism for timely termination of innate immune and inflammatory responses mediated by TLR3/4.
All animal experiments were performed in accordance with the Wuhan University animal care and use committee guidelines.
Mouse monoclonal antibodies against Flag (Sigma), HA (Origene), β-actin (Sigma), p-IκBα (CST), p-IRF3 (CST) and p-TBK1 (Abcam); poly(I:C) (Invivogen), LPS (Sigma), R848 (Invivogen), PGN (Invivogen), human IFN-γ (PeproTech), Bafilomycin (Sigma), Monophosphoryl lipid A (Sigma), Z-KAD-FMK (MCE) were purchased from the indicated companies. Luminescent cell viability assay kit (G7570) was purchased from Promega. Mouse antisera to TRIM32 and TRIF were raised against recombinant human TRIM32 and murine TRIF(1–475) respectively. Rabbit antisera to TAX1BP1 and NDP52 were raised against recombinant murine TAX1BP1 and NDP52(1–160) respectively.
Mammalian expression plasmids for Flag- or HA-tagged murine TRIM32 and its mutants, TRIF and its mutants, TRAF6, TRAF3, TBK1, IRF3, TAX1BP1 and NDP52 were constructed by standard molecular biology techniques.
Trim32 gene knockout mice with a CL7/B6 background were provided by Dr. Hong-Liang Li [27]. Genotyping by PCR was performed using the following two pairs of primers: WT-1: GGAGAGACACTATTTCCTAAGTCA;WT-2: GTTCAGGTGAGAAGCTGCTGCA; MT: GGGACAGGATAAGTATGACATCA.
Amplification of the wild-type allele with primers WT-1 and WT-2 results in a 250-bp fragment, whereas amplification of the disrupted allele with primers WT-1 and MT results in a 300-bp fragment.
BMDMs and BMDCs were generated as described [28]. The bone marrow cells (1×107) were cultured in RPMI medium 1640 containing 10% FBS and 10 ng/mL recombinant murine M-CSF (Peprotech) or GM-CSF-containing conditional medium in a 100-mm dish for 5 or 9 days for generation of BMDMs or BMDCs respectively.
Primary lung fibroblasts were generated as described [29]. Primary lung fibroblasts were isolated from approximately 4- to 6-week-old mice. Lungs were minced and digested in calcium and magnesium free HBSS containing 10 μg/ml type II collagenase (Worthington) and 20 μg/ml DNase I (Sigma-Aldrich) for 3 hours at 37°C with shaking. Cell suspensions were filtered through progressively smaller cell strainers (100 and 40 μm) and then centrifuged at 1500 rpm for 4 min. The cells were then plated in culture medium (1:1 [v/v] DMEM/Ham’s F-12 containing 10% FBS, 15 mM HEPES, 2 mM L-glutamine, 50 U/ml penicillin, and 50 μg/ml streptomycin). After 1 hour, adherent fibroblasts were rinsed with HBSS and cultured in media.
Age- and sex-matched Trim32+/+ and Trim32-/- mice were injected intraperitoneally with poly(I:C) (5 μg/g body weight) plus D-galactosamine (1 mg/g body weight) or with LPS (10 μg/g body weight). The survival of the injected mice was monitored every 2 hours.
Age- and sex-matched Trim32+/+ and Trim32-/- mice were orally administrated with Salmonella typhimurium (1×107 pfu per mouse). The body weight and survival of the infected mice were monitored every day.
Blood from mice injected with poly(I:C) plus D-galactosamine or LPS was collected at the indicated times and the serum concentration of TNFα (Biolegend), IL-6 (Biolegend), and IFN-β (PBL) were measured by ELISA kits from the indicated manufactures.
Transfection and reporter assays were performed as previous described [30,31,32,33]. HEK293 cells were seeded on 24-well plates and transfected on the following day by standard calcium phosphate precipitation. Where necessary, empty control plasmid was added to ensure that each transfection receives the same amount of total DNA. To normalize for transfection efficiency, pRL-TK (Renilla luciferase) reporter plasmid (0.01 μg) was added to each transfection. Luciferase assays were performed using a dual-specific luciferase assay kit (Promega, Madison, WI). Firefly luciferase activities were normalized based on Renilla luciferase activities.
Coimmunoprecipitation, immunoblotting and ubiquitination assays were performed as previous described [34,35,36]. For ubiquitination assays, the immunoprecipitates were re-extracted in lysis buffer containing 1% SDS and denatured by heating for 5 min. The supernatants were diluted with regular lysis buffer until the concentration of SDS was decreased to 0.1%, followed by re-immunoprecipitation with the indicated antibodies. The immunoprecipitates were analyzed by immunoblotting with the ubiquitin antibody.
qPCR assays were performed as previously describe [37,38,39,40]. Total RNA from mouse or human cells was isolated using the Trizol reagent (Invitrogen). After reverse-transcription with oligo(dT) primer using a RevertAidTM First Strand cDNA Synthesis Kit (Fermentas), aliquots of products were subjected to qPCR analysis to measure mRNA levels of the tested genes. Gapdh was used as a reference gene. Gene-specific primer sequences were previously described [33,41].
Lungs or intestines from mice were fixed in formalin and embedded into paraffin blocks. The paraffin blocks were sectioned (5 μm) for H&E staining. The immunohistochemistry analysis was performed on the 5-μm sections. The sections were placed on polylysinecoated slides, deparaffinized in xylene, rehydrated through graded ethanol, quenched for endogenous peroxidase activity in 3% hydrogen peroxide, and processed for antigen retrieval by microwave heating for 7 min in 10 mM citrate buffer (pH 6.0). Sections were counterstained with hematoxylin (Zymed Laboratories) for 5 min and coverslipped. Pictures were acquired using a HistoFAXS system.
TRIF+/+ and TRIF-/- HEK293-TLR3 cells (5×107) were treated with poly(I:C) for the indicated times, and then cells were harvested and lysed by douncing for 20 times in 2 ml homogenization buffer (10 mM Tris-HCl [pH 7.4], 2 mM MgCl2, 10 mM KCl, and 250 mM sucrose). The homogenate was centrifuged at 500 g for 10 min for removal of the crude nuclei. The supernatant (S5) was centrifuged at 100, 000 g for 2 hours for cytosol (S100K) and membrane (P100K) generation.
Differences between averages were analyzed by Student’s t-test. P value of less than 0.05 was considered significant.
|
10.1371/journal.pcbi.1004206 | Three-Dimensional Gradients of Cytokine Signaling between T Cells | Immune responses are regulated by diffusible mediators, the cytokines, which act at sub-nanomolar concentrations. The spatial range of cytokine communication is a crucial, yet poorly understood, functional property. Both containment of cytokine action in narrow junctions between immune cells (immunological synapses) and global signaling throughout entire lymph nodes have been proposed, but the conditions under which they might occur are not clear. Here we analyze spatially three-dimensional reaction-diffusion models for the dynamics of cytokine signaling at two successive scales: in immunological synapses and in dense multicellular environments. For realistic parameter values, we observe local spatial gradients, with the cytokine concentration around secreting cells decaying sharply across only a few cell diameters. Focusing on the well-characterized T-cell cytokine interleukin-2, we show how cytokine secretion and competitive uptake determine this signaling range. Uptake is shaped locally by the geometry of the immunological synapse. However, even for narrow synapses, which favor intrasynaptic cytokine consumption, escape fluxes into the extrasynaptic space are expected to be substantial (≥20% of secretion). Hence paracrine signaling will generally extend beyond the synapse but can be limited to cellular microenvironments through uptake by target cells or strong competitors, such as regulatory T cells. By contrast, long-range cytokine signaling requires a high density of cytokine producers or weak consumption (e.g., by sparsely distributed target cells). Thus in a physiological setting, cytokine gradients between cells, and not bulk-phase concentrations, are crucial for cell-to-cell communication, emphasizing the need for spatially resolved data on cytokine signaling.
| The adaptive immune system fights pathogens through the activation of immune cell clones that specifically recognize a particular pathogen. Tight contacts, so-called immunological synapses, of immune cells with cells that present ‘digested’ pathogen molecules are pivotal for ensuring specificity. The discovery that immune responses are regulated by small diffusible proteins – the cytokines – has been surprising because cytokine diffusion to ‘bystander’ cells might compromise specificity. It has therefore been argued that cytokines are trapped in immunological synapses, whereas other authors have found that cytokines act on a larger scale through entire lymph nodes. Measurements of cytokine concentrations with fine spatial resolution have not been achieved. Here, we study the spatio-temporal dynamics of cytokines through mathematical analysis and three-dimensional numerical simulation and identify key parameters that control signaling range. We predict that even tight immunological synapses leak a substantial portion of the secreted cytokines. Nevertheless, rapid cellular uptake will render cytokine signals short-range and thus incidental activation of bystander cells can be limited. Long-range signals will only occur with multiple secreting cells or/and slow consumption by sparse target cells. Thus our study identifies key determinants of the spatial range of cytokine communication in realistic multicellular geometries.
| Cell-to-cell communication is a defining property of multicellular organisms. In particular, the release, sensing and uptake of cytokines, small signaling proteins, by cells is essential for the regulation of the mammalian immune system [1]. Prominent quantitative characteristics of cytokine signaling are high receptor specificity (with Kd ≈ 10−10 nM) and low free cytokine concentrations in the picomolar range [2,3]. The physiological cytokine milieu regulates critical processes like the type and strength of the immune response. Quantitative understanding of such cytokine-driven cellular decisions is beginning to emerge [4–8], yet the underlying spatio-temporal cytokine dynamics remain poorly understood. Cytokines act in a heterogeneous environment, typically with high cell-densities. It is not known how they diffuse under such conditions and, in turn, regulate immune responses. Specifically, how far cytokines can signal away from the producing cell is not clear. Perona-Wright et al. [9] have found that interleukin(IL)-4 is seen by most T cells in the lymph node upon parasite infection, including non-specific ‘bystander’ cells. In this case, many T cells throughout the lymph node could be IL-4 producers. By contrast, several observations suggest more localized cytokine communication [4,10–13].
Given the low measured cytokine concentrations, which are often below 10 pM, the question arises whether and how effective paracrine signals are possible at all, in a situation where only a certain fraction (~25%) of the cells secrete cytokine molecules. Of note, 1 pM is about 1 molecule in 1700 μm3, compared to ~500 μm3 volume of a typical lymphocyte. Higher, systemically elevated cytokine levels arise only in certain immunopathologies, so-called ‘cytokine storms’, where they cause severe damage [14]. However, it has been demonstrated that cytokine concentrations are not always well mixed, and locally higher cytokine concentrations can occur also in ex vivo T cell cultures [12]. Therefore, we asked how and under which conditions such cytokine gradients arise, and if they are able to explain effective paracrine signals.
One possibility to enrich cytokine concentrations would be localized signaling to specific target cells by an immunological synapse [15–18]. Immunological synapses are formed between immune cells by surface proteins after antigen recognition [15,16,19]. They have been observed between various cell types of the immune system, including immunological synapses between T cells and antigen presenting cells (APC, e.g. B cells [10] and dendritic cells [20]), and immunological synapses between T cells and T cells [12]. Many cytokines are secreted preferentially into the immunological synapse [10,21–23], and a range of high-affinity cytokine receptors have been found to be specifically located in the immunological synapse, too [11]. Therefore, it is likely that the synapse has an important function for cytokine signaling, beyond its role for T cell receptor signaling on which theoretical studies have focused [24,25]. Cytokine signaling through immunological synapses might also explain the pleiotropic effects observed for most cytokines, as it would provide specificity of cytokine signaling by restriction of their action. Nevertheless, it is unlikely that paracrine cytokine signals are only possible between cells that are directly connected by an immunological synapse. For instance, Sanderson et al. [26] found that interferon-γ can be seen by bystander cells other than the target cells to which the synapses are formed. To understand which parameters govern autocrine versus paracrine cytokine signaling, we analyzed in this study reaction-diffusion models of cytokine signaling at two scales: through the immunological synapse between two cells and in three-dimensional arrays of many (>100) cells.
For this purpose, we chose the cytokine interleukin(IL)-2 as a model system, a cytokine showing polarized secretion and corresponding receptor expression [10,11,22]. IL-2 was first identified as a T cell growth factor [27], but, paradoxically, is a critical mediator of immune tolerance [28–31]. It is secreted by T helper (Th) cells early after antigenic stimulation and taken up by high-affinity IL-2 receptor (IL-2R) on Th cells and regulatory T (Treg) cells [28–31]. Treg cells mediate immune tolerance and are critical for the prevention of autoimmune reactions [32,33]. IL-2 secretion is digital, i.e. upon receiving an antigen stimulus, only about one quarter of a Th cell population releases IL-2 molecules [34–36]. It is an open question if IL-2 and other cytokine signals act in an autocrine or paracrine manner [31,37]. In response to IL-2 uptake, Th cells and Treg cells upregulate CD25, the α-subunit of the IL-2R. CD25 is often used as an activation marker of T cells, because it precedes proliferation of Th cells and subsequent recruitment of effector immune cells [30,38]. Although IL-2 secreting Th cells upregulate CD25, Long and Adler [37] reported that they lack phosphorylated STAT5, a key intermediate in the IL-2R signal transduction cascade. In the same experiment, other Th cells not secreting IL-2 also upregulate CD25 in response to IL-2, and in addition show fully functional signal transduction [5,37]. These data suggest that the dominant mode of IL-2 signaling is paracrine, in contrast to the presumed function of the immunological synapse in containing secreted cytokines [16,17]. However, unlike T cells, most APC do not express functional IL-2 receptor (IL-2R) [39]. Thus, both the study by Sanderson et al. [26] and the properties of IL-2 signaling suggest a role of the immunological synapse for cytokine signals that goes beyond signal amplification between the two cells associated by a synapse.
In this study, we addressed the question of how and under which conditions paracrine cytokine signals occur despite the measured low bulk concentrations in the picomolar range, and we aimed to define the parameters that control the range of cytokine signaling. To this end, we considered the two key spatial scales, the sub-μm scale of the immunological synapse and the supra-μm scale of cell-to-cell communication. We investigated reaction-diffusion models on these two scales by analytical techniques and advanced finite-element computations in three spatial dimensions [40–44]. To be specific, we utilized a simple, experiment-based mathematical model for IL-2 signaling and gained more general insight through systematic variation of parameters. Our results show that paracrine cytokine signaling is possible in the presence of local concentration gradients combined with nonlinear signal amplification. The spatial range of cytokine signaling can be tuned from purely autocrine via intrasynaptic and short-range paracrine to long-range paracrine. For a wide array of parameters, we found that cytokine gradients in dense multicellular environments range over one to few cell diameters. These computational findings can inform novel experiments probing the spatio-temporal dynamics of cytokine signaling [45].
The binding of cytokines to their high-affinity receptors is followed by receptor internalization and intracellular cytokine degradation, so that cytokine molecules are removed from the medium (Fig 1A). Thus, regulating the strength of cytokine signaling by cytokine receptor expression might also affect the extracellular cytokine concentration and hence, indirectly, signaling. To gain quantitative insight, we first studied a simple reaction-diffusion model, where a cytokine-secreting cell is surrounded by cells that can take up the cytokine. To allow for an analytical solution, we assume the surrounding cells to be placed on a spherical shell with the secreting cell in the center (Fig 1B, see Materials and Methods). For convenience, parameter values are summarized in Table 1. If the target cells are located far away (i.e., their density is low), the cytokine concentration experienced by the target cells is nearly independent of the level of receptor expression (Fig 1C) because the dilution of the cytokine occurs primarily by diffusion in the three-dimensional tissue. On the other and, if the density of target cells is so high that they immediately surround the cytokine secreting cell, the cytokine concentration is practically homogeneous in the small intervening space, as the timescale of diffusion over such a short distance is fast compared to the timescale of cytokine uptake (Fig 1D and S1 Text). As a consequence, the cytokine concentration experienced by proximal target cells is set by the balance of secretion rate by the cytokine-producing cell and uptake rate. The autocrine and paracrine uptake rates Jauto and Jpara depend on the level of cytokine receptor expression on the target cells (Fig 1E), and are practically independent of the cell-to-cell distance even at high cell density (Fig 1F; the low cell-density scenario is independent of the cell-to-cell distance by construction). Interestingly, cytokine concentration (Fig 1C) and uptake rates (Fig 1D) are sensitive to receptor expression on proximal targets cells in the physiologic range of 100 to several 1000 receptor molecules per cell [5]. Thus, this simple model indicates that with a high density of target cells, cytokine receptor expression controls the amount of paracrine cytokine signal.
The model of the previous section assumed homogeneous secretion of the cytokine over the cell surface. However, T cells release IL-2 and other cytokines in a polarized fashion into the immunological synapse [10,21–23]. Therefore, we analyzed a model of cytokine secretion and uptake in the immunological synapse, represented by a small cylindrical region between a Th cell and an opposed APC or second T cell (Fig 2A, see Materials and Methods), extending previous work [46]. The distance between Th cell and opposed cell, in the following referred to as synaptic distance, is in the range of 10 to 40 nm [19,47]. This close contact between Th cell and opposed cell causes a cytokine concentration profile which is almost homogeneous between the two cells in the center of the synapse, and sharply falls off towards the outer boundaries through which cytokine molecules are lost practically irreversibly (Fig 2B, top). In the case of low receptor expression (Fig 2B, top left), the cytokine concentration reaches values in the nM range. Thus, the synaptic cytokine secretion results in locally much higher concentrations than homogeneous secretion (see Fig 1C), in line with experimental data [12]. For comparison, consider cytokine secretion into a cylindrical region with length 2 μm, a typical value for nearby cells but much larger than the immunological synapse (Fig 2B, bottom). In this case, the cytokine concentration falls to less than 20 pM at the surface of the opposed cell. Hence, the very small synaptic distance in a fully formed immunological synapse is crucial for the establishment of high local cytokine concentrations.
The immunological synapse causes two conceptually different types of paracrine signals: Cytokine molecules may bind to cytokine receptors at the opposed cell (Jsynapse) or escape into the extracellular space (Jescape), potentially reaching other nearby cells. Cytokine molecules may also induce autocrine signals by binding to receptors at the secretory cell (Jauto). Fig 2C shows the fractions of Jauto, Jsynapse and Jescape, choosing IL-2 receptor densities that are characteristic of naïve (IL-2Rlow) or preactivated (IL-2Rhigh) IL-2 secreting T cells, and for opposed cells with different IL-2R expression. IL-2Rhigh cells recapture most of the secreted IL-2 molecules, irrespective of the type of opposed cell and the synaptic distance.
Naïve, IL-2Rlow cells show a strong dependence on the synaptic distance (Fig 2C, left). If the synaptic space is sufficiently narrow, Jescape is small; the escape flux could be even further reduced by adhesion molecules sealing off much of the synapse from the extracellular space. If the opposed cell is a second T cell, then the secretory cell and the opposed cell compete for the cytokine molecules. Treg cells outcompete Th cells due to their large receptor number. On the other hand, APCs do not express IL-2R and the IL-2 signals would be purely autocrine. For synapses with somewhat larger synaptic distances, a considerable amount of cytokine molecules can escape (enhanced Jescape) and provide paracrine signals to surrounding cells. Interestingly, the ratio of Jauto and Jescape is most sensitive to the synaptic distance in the physiologic range between 10 nm in the close-contact zone and 40 nm in the outer region of the immunological synapse [19,47]. In all cases, the fraction Jescape of cytokine molecules that escape for paracrine signaling is considerably smaller than in the case of homogeneously distributed cytokine secretion and receptor expression (Figs 2C and 1D). However, even if the cytokine secreting cell is pre-activated and hence autocrine IL-2 uptake is high (IL-2Rhigh), a sizeable fraction of cytokine molecules still diffuses out of the synapse (Jescape ~20%).
In summary, our model reveals two main implications of an immunological synapse for cytokine signaling: A tight synapse causes highly localized cytokine distributions, and it enhances the probability of autocrine recapture. Both properties result from the high aspect ratio of the radius of the cell contact area and the synaptic distance (r and z in Fig 2A). The two properties have opposing effects on the strength of paracrine cytokine signals: While localized cytokine concentrations increase the likelihood of a local paracrine signal, the reduction in effective cytokine secretion reduces the potential of paracrine signals.
The analytically tractable models gave insight into the qualitative properties of paracrine cytokine signaling, and they made quantitative predictions on the consequences of the various time and length scales in the system. For example, the high aspect ratio of the immunological synapse evokes highly localized cytokine concentrations in the vicinity of cytokine secreting cells resembling secretion from a point source (see Fig 2B), and the high diffusion constant in relation to the receptor dynamics makes the system largely independent of the cell-to-cell distance (Fig 1E and 1F). However, the simple models studied above cannot answer the question if effective paracrine signals are possible despite the low bulk cytokine concentrations. To illustrate this problem, consider a classical formula from Berg and Purcell for the timescale of ligand diffusion towards a receptor [48] (Materials and Methods). Measured cytokine concentrations in serum or in supernatants of ex-vivo T cell cultures are typically in the picomolar range [2,3]. Assuming a spatially uniform cytokine concentration of 10 pM and a receptor number of 100 per cell, as is typical for the high-affinity IL-2R on naïve T cells, that calculation reveals that on average, every 7 min a receptor becomes bound by a cytokine molecule. Under these conditions it would take hours to induce a reliable signal, indicating that bulk cytokine concentrations might just be capable of, or even be too low for, stimulating signal transduction. However, it has been reported that IL-2 is subject to appreciable spatial gradients, with much higher concentrations at the surfaces of T cells [12,49].
To investigate the origins and consequences of spatially inhomogeneous dynamics of cytokine signaling, we performed extensive three-dimensional simulations of a T cell population (Fig 3A and 3B). As before, we focus on the cytokine IL-2, for which many parameters, including secretion and receptor expression rates, have been estimated from experiments [5,35,50], and experimentally tested models for the IL-2R dynamics are available [4,5,7]. To account for polarized secretion at the immunological synapse, we do not explicitly model synapse formation but consider the effect of discrete IL-2 sources from which IL-2 escapes into the extra-synaptic space (with rate qeff, corresponding to Jescape in the simplified model of the previous section). The position of the IL-2 source of a producing cell is a randomly chosen point at the cell surface. IL-2 secretion is all-or-nothing [34,36]: only about one quarter of antigen-stimulated T cells release IL-2 molecules, and among these cells, the IL-2 secretion rate is in the range of 10 molecules per second [35]. In accordance with experimental data, already activated IL-2 secreting cells have high IL-2R expression which, for simplicity, we take as constant [37]. Non-secreting cells are assumed to upregulate IL-2R expression in response to IL-2 homogeneously at their cell surface.
Consistent with experimental data [39], APC themselves do not express IL-2R but constitute simply ‘excluded volumes’ with respect to the IL-2 dynamics. To focus on the role of IL-2 uptake by T cells, we do not consider the APC explicitly, but only its consequences for polarized secretion and uptake (see above) (Table 1). Despite this simplification, our simulations consider realistic extracellular volumes (as determined by the cell distances) between the cells as a basis for determining the extracellular concentrations of the secreted cytokines.
Based on these assumptions, we simulated the IL-2 dynamics for a large number of T cells (216 cells in a volume of ~1 nl, Fig 3). Stimulating IL-2 secretion in a fraction of T cells (Fig 3A and 3B), the IL-2 concentration increases rapidly and nearly homogeneously for several hours after stimulation (Fig 3C and 3D and S1 Fig). Then, in response to the high IL-2 concentration resulting from paracrine signaling, IL-2R expression is upregulated in non-secreting cells (Fig 3E) and causes fast IL-2 uptake from the medium. As a result, concentration gradients occur: In large parts of the simulated region, the IL-2 concentration reaches a steady-state at around 10 pM while locally it is more than twice as large (Fig 3C, red regions at 30 h). This inhomogeneity in IL-2 concentration corresponds to receptor upregulation (activation) of non-secreting Th cells: IL-2Rhigh cells are found near the regions with high IL-2 concentration. Analysis of the time course (Fig 3E) shows that all cells upregulate IL-2R levels in response to the increased IL-2 concentration in the first hours after antigenic stimulation and IL-2 secretion. However, as the high-affinity IL-2R is being upregulated, IL-2 becomes increasingly depleted in the medium. As a result, only a fraction of the cells receive a sufficient IL-2 stimulus to sustain high IL-2R expression (IL-2Rhigh cells in Fig 3E), whereas the remaining cells downregulate IL-2R expression (IL-2Rlow cells).
Interestingly, the time courses of IL-2 concentrations at the surfaces of the cells show only small differences between IL-2Rhigh and IL-2Rlow cells (Fig 3F): In the beginning, IL-2 equally rises near IL-2Rhigh and IL-2Rlow cells (see Fig 3D), but as IL-2 depletion sets in, the cells that eventually become IL-2Rlow cells receive slightly less IL-2. Later, at steady-state, the IL-2 concentration is somewhat higher in the microenvironment of IL-2Rlow cells, because they do not consume as many IL-2 molecules. This form of local bistability, which occurs in the expression of IL-2R on Th cells, was observed already in Ref. [4]: Based on a quasi-stationary state assumption, Busse et al. showed that in the model without Treg cells, the IL-2R expression rate responds to the increase of the secretion rate in a digital way and the cells are activated only after a certain threshold is exceeded. A small bistable region around the threshold is observed. These findings were supported by experimental data from primary T cells cultured ex vivo [4]. Thus our present model with the immunological synapse and 3D diffusion matches the bistable system behavior seen in the simpler analytical model.
Taken together, our simulations indicate that the amount of IL-2 escaping from the immunological synapse is sufficient to sustain paracrine signaling in at least a fraction of surrounding cells. However, competition for the cytokine can cause heterogeneity in the response of a cell population and result in bulk IL-2 levels that are much lower than local concentration peaks and in agreement with concentration levels measured by ELISA (see Discussion).
Regulatory T cells constitutively express high levels of high-affinity IL-2R but do not secrete IL-2 [29,30]. To study the effect of Treg cells on the IL-2 dynamics after activation of conventional Th cells, we simulated a T-cell population consisting of antigen-stimulated IL-2 secreting and non-secreting Th cells as well as Treg cells (Fig 4A and 4B). Compared to the situation in the absence of Treg cells (cf. Fig 3), the IL-2 concentration attains a spatially inhomogeneous steady state more rapidly, with the overall IL-2 concentration being lower (Fig 4C and 4D and S2A Fig). Importantly, the non-secreting Th cells do not permanently upregulate IL-2R in the presence of Treg cells because the Treg cells suppress the paracrine IL-2 signal. The comparison with the simulations without Treg cells (Fig 3) imply that Th cells require for sustained IL-2 signaling both a transient strong and a stable weak IL-2 stimulus. The finding that Th cells can sustain IL-2 signaling at low cytokine concentration, but only after initial stimulation with high cytokine concentration, is a spatio-temporal phenomenon similar to hysteresis: Active cells express more cytokine receptors, which bind more cytokine molecules even at lower concentration and thus stabilize the active state once it is achieved. Treg cells can suppress prolonged IL-2 signaling in Th cells by inhibiting the strong initial IL-2 signal and the resulting upregulation of the high-affinity IL-2R.
Having established that an effective paracrine IL-2 signal is possible in our model, and that it can be suppressed by Treg cells, we analyzed to which extent key parameters shape the spatio-temporal dynamics: IL-2 secretion rate, cell-to-cell distance, and fraction of IL-2 secreting cells. Without Treg cells, the number of activated Th cells increases linearly with the effective IL-2 secretion rate qeff until, eventually, all cells in the simulated region become active (Fig 4E, left panel). By contrast, the presence of Treg cells creates a threshold at an effective secretion rate of qeff ~ 20000 molecules/h (about 5 molecules/s), below which there is no paracrine IL-2 signaling between Th cells. The same pattern is observed if we vary the fraction of cytokine secreting cells instead of the effective secretion rate (Fig 4E, middle panel), which reflects digital IL-2 secretion [34,36]. Hence the presence of Treg cells changes the paracrine IL-2 dynamics from a gradual to an all-or-none response: Either the paracrine signal is completely suppressed by competitive uptake, or suppression is overrun and all cells are activated. Note that qeff measures only the IL-2 molecules that escape from the immunological synapse; assuming a tight synapse, this would only be 20% of the total secretion (see Fig 2). However, measured IL-2 secretion rates are ~10 molecules/s [35,50], which is likely to be too small to titrate out the Treg cells in a physiological setting where IL-2 is secreted into the synapse.
Within the range from 2 to 20 μm [12], the cell-to-cell distance (measured between cell surfaces of neighbored cells) does not influence the amount of Th cells that become activated by the paracrine IL-2 stimulus (Fig 4E, right panel). This is because, as anticipated by the analytically treatable model (see Fig 1F), cytokine molecules can reach nearby cells rapidly by diffusion compared to the slower time scales of changes in IL-2R expression and IL-2 internalization. Thus, the exact cell-to-cell distance is unimportant in the physiological range.
Our simulations yielded global elevations in IL-2 concentration only transiently before the target cells expressed high levels of IL-2R; beyond this point, only short-range IL-2 gradients were observed, with local concentrations governing IL-2 signaling. Generally, we expect that the balance between cytokine secretion, dilution through diffusion in the three-dimensional extracellular space and cellular consumption will determine the signaling range. To understand the interplay of these three factors, we performed large-scale simulations of an area containing ~2000 cells, with a single IL-2-secreting Th cell surrounded by non-secreting Th cells which all are potential responders to the IL-2 (Fig 5A). Although we use the specific parameters for IL-2 here, this model is of more general interest and applies to other situations with few signaling cells and many responder cells (e.g., IL-4 secreting Th cells in a B cell population [9]), or can be thought of as representing a cluster of several cytokine secreting cells in a population with a small density of cytokine secreting cells elsewhere.
We found that for the secretion rates estimated for IL-2 [35,50], high IL-2 concentrations are restricted to the microenvironment of the cytokine secreting cell (Fig 5B). Remarkably, although secretion is assumed to be polarized through the synapse, the cytokine concentration is higher along the entire surface of the secreting cell, including the pole opposite to the synapse, than at nearby cells. This is due to the absence of IL-2R on the surface of secreting cells (except for the synaptic space). For larger secretion rates (of the order to 106 molecules/h or 280 molecules/s), the IL-2 signal reaches hundreds of cells. However, with the experimental estimate for the IL-2 secretion rate (10 molecules/s [35,50]), of the order of a 100 secreting cells would be needed to realize such a high rate (assuming an effective secretion rate of 10-20% of the total rate, see Figure 2). Therefore, IL-2 from an individual producer will act locally whereas only large clusters of activated cells could cause long-range signals. The occurrence of two distinct spatial signaling regimes as a function of secretion rate is expected because the cellular uptake rate can be saturated by high cytokine concentrations (akin to an enzymatic Michaelis-Menten rate law where the cytokine receptors function as the enzyme). Below saturation the cytokine signal remains local. Interestingly, the spatial range scales linearly with the logarithm of the secretion rate (Fig 5C). Hence the signaling range exhibits a fold-change response to the effective secretion rate (see Discussion).
To further analyze the properties of cytokine diffusion, we computed the traveling distance of cytokine molecules, i.e. the distance from the cytokine secreting cell at which a ligand is taken up by a receptor. For this purpose, we simulated a pulsed, homogeneous stimulation (see Methods) in a region covering ~5000 cells. We found that despite the apparent short-range induction of effective paracrine signaling, the traveling distance has a broad distribution peaking around four cells away from the cytokine secreting cell in each direction (Fig 5D). Thus, in our reaction-diffusion system, the chemical reactions on the cell surface dominate the diffusion and determine the IL-2 gradient formation. We further compared the distribution of traveling distances with earlier analytical expressions obtained from a reaction-diffusion model of morphogen gradient formation [51]. Despite some differences in the model architecture (see Methods), and despite numerical limitations in simulating an ‘infinite domain’ as assumed by the analytical methods of Ref. [51], our simulations are in good agreement with those analytical results (S3 Fig).
Taken together, our simulations of long-range cytokine diffusion and uptake show that long-range paracrine signals are possible in principle, but require exceptional circumstances (extremely high rates of cytokine production or large clusters of cytokine producing cells) that might not readily occur in vivo.
The spatial regulation of cytokine signaling in the immune system has spurred much interest, particularly in relation to the specificity of cytokine action [4,9–12,22,23,26]. Experimentally, however, cytokine signaling has not been probed directly at fine spatial resolution, although recent advances in synthetic biology could provide new tools in the near future [45]. Here, we used a computational approach to study cytokine signaling in realistic three-dimensional geometries. To this end, we considered two distinct spatial scales. First, we analyzed polarized signaling across narrow junctions – immunological synapses – between immune cells (nm scale). We find that synapses enhance autocrine signaling and signaling towards the cell connected by the synapse, but, importantly, cannot prevent substantial cytokine escape for paracrine communication by mere geometry. Second, we employed advanced simulation tools for partial differential equations to dissect the dynamics of this ‘spill-over’ paracrine signal in dense ensembles of hundreds of communicating cells (μm scale). Using experimentally established parameters for the T-cell cytokine IL-2, we find that cytokine signals emanating from producing cells are short-range (one to few cell-to-cell distances) because of uptake by target cells or competitors. Long-range communication requires coherent secretion by tens to hundreds of producers or/and sparse uptake. Thus we predict that gradients at the cellular length scale are a key property of cell-to-cell communication by cytokines.
We note that the spatial range of diffusible signals is also of relevance for morphogen action [52,53]. In contrast to immune cell signaling with a typical time scale of many hours during which diffusive gradients reach steady state, the transient behavior on shorter time scales is of particular interest for morphogen gradients [51,54].
Cytokine concentrations as measured by ELISA studies in cell supernatants are typically very low, in the picomolar range [3,5,49]. As low cytokine concentrations would imply long signaling times (see Eq 1 below), we hypothesized that paracrine cytokine signals rely on much higher cytokine concentrations in the microenvironment of target cells, which have indeed been detected by live cell imaging [12]. However, it is generally believed that cytokine signaling occurs in the regime of fast diffusion, which is reflected by our parameter values—the typical spatial range of fast diffusion, D/kd (see e.g. [55]), spans 40 cells away from the cytokine secreting cell for our values (see Table 1). Therefore, we analyzed the spatiotemporal dynamics of cytokine signals by more detailed mathematical modeling and simulations. We found that spatial gradients do occur due to nonlinear receptor dynamics and polarized IL-2 secretion at the immunological synapse, despite fast diffusion. This was also quantified, e.g. in terms of the traveling distance of cytokine molecules (Fig 5D). We showed by extensive simulations in three spatial dimensions that such cytokine gradients can mediate paracrine signals targeting cells other than those connected by the immunological synapse, as previously suggested for interferon-γ [26]. Moreover, we analyzed the parameters that control paracrine signaling on the different spatial scales.
It is a long-standing question if cytokine signals are predominantly autocrine or paracrine. IL-2 has initially been thought of as a prototypical autocrine signal facilitating self-activation of Th cells [27,30,56]. More recently, paracrine IL-2 signaling towards Treg cells was identified as essential to prevent autoimmune diseases [29,31], possibly due to competition with autocrine self-activation [4,5,28,57]. Recent experimental observations suggest that also paracrine IL-2 signals towards other Th cells are important for regulation of immune responses, while true autocrine IL-2 signals are suppressed by the intracellular signal transduction pathway [5,37]. A plausible explanation would be that IL-2 secreting cells are constitutively activated, i.e. prone to proliferation and differentiation, due to a strong signal from the T cell receptor, and do not rely on signals via the IL-2R. Th cells not secreting IL-2 may have received a weaker T cell receptor signal, and are only fully activated if they receive additional stimulation from the IL-2R. In this theoretical study, we cannot address the question to what extent such a mechanism is responsible for the activation of T cell populations in vivo. However, our simulations show that the need for paracrine cytokine signals provides several checkpoints for the induction of immune responses downstream of the T cell receptor.
We identified three major control points which are likely important for the fine-tuned regulation of paracrine cytokine signals. First, cytokine receptors have high affinity and are internalized after binding of cytokine molecules. That allows for control of paracrine cytokine signals by expression of cytokine receptors (Fig 1). Second, the effective rate of cytokine secretion, i.e. the paracrine cytokine signal escaping the immunological synapse, sensitively depends on the configuration of the immunological synapse, in terms of the exact synaptic distance (Fig 2). Therefore, we propose that regulation of cytokine signals is an important function of the immunological synapse (see also Refs. [15–18]), along with regulation of the strength of T cell receptor activation [47] and the exchange of microvesicles between T cell and APC [58]. Note that using the synaptic distance is an idealization; in reality the influence of the immunological synapse on cytokine diffusion is more complex, due to its structure consisting of several layers with different types of surface proteins [16,24]. Third, in our simulations, Treg cells efficiently suppress paracrine IL-2 signals, because they express high basal levels of IL-2R, preventing the strong transient cytokine signal. In line with earlier work from us and others [4,5,29,57], this suggests that suppression of IL-2 signals is an important mechanism contributing to immune tolerance mediated by Treg cells. Of note, Treg cells most likely interfere with T cell activation in several other ways, e.g. by release of anti-inflammatory cytokines like IL-10, by forming immunological synapses with T cells, and by other mechanisms yet to be discovered [12,33]. Interestingly, a fourth system property one might expect to have a large influence on the dynamics of the system, the cell density or cell-to-cell distance, is unimportant for the results of our simulations (Fig 4E). This property results from the timescale separation between cytokine diffusion and cytokine uptake (see Fig 1E and S1 Text), and explains recent experimental data [7].
In our model simulations, paracrine cytokine signals are not only characterized by stable cytokine gradients, but also by a rapid and transient cytokine boost occurring in the first hours after stimulation. Such a transient cytokine signal has been observed by single-cell IL-2 capture assays [49,59], and recently also by ELISA in cell supernatants [7], although with conflicting time-scales: IL-2 capture assays evoked a peak in the number of IL-2 secreting cells at 1–6 hr after antigen stimulation [49,59], while Tkach et al. report a peak in the IL-2 concentration measured in vitro after ~50 hr [7]. Our simulations point to an IL-2 peak in the first 10 hr after stimulation, and thus support the earlier suggestion [49] that ELISA studies have limitations in reflecting the time-course of in vivo cytokine signals, although the study of Tkach et al. provides valuable quantitative insight into the dose-response characteristics of IL-2 signals. A reason might be that in culture, cells form thin layers on the bottom of the well, and therefore cytokine molecules are detected by ELISA in the supernatant after a certain delay.
The large-scale simulations resembling a cluster of highly active T cells in the center of a lymphoid organ (Fig 5) reveals a logarithmic, or fold-change response of the spatial signal range with respect to the effective secretion rate. That means, the cell population recognizes relative rather than absolute increases in the stimulus strength (here, the amount of secreted cytokine molecules per time). Fold-changes in sensory biological systems are a classical phenomenon referred to as Weber’s law, and were recently observed in various intracellular signal transduction pathways [60–63]. As a consequence of the fold-change response, sensory systems can act over a broad range of stimulus intensities, from nearly detectable to very intense stimulations. Our computer simulations suggest a similar mechanism for paracrine cytokine signals: Moderate effective secretion by a small fraction of cells allows for short-range signals inside an immunological synapse, larger effective secretion rates may evoke paracrine signals that reach bystander cells in close vicinity but not connected by a synapse, and very high secretion rates or large clusters of secreting cells may evoke an organ-wide cytokine signal or ‘cytokine storm’ [14].
Adaptive immune responses must be rapid and effective in the case of strong infection, but also carefully controlled to avoid autoimmune diseases. In our simulations, the spatial distribution of cytokine secretion and uptake within a population of immune cells had a huge impact on the cellular response, generating multiple layers of plasticity that can be exploited for appropriate regulation of immune responses.
For the simulations of the three-dimensional in silico T cell population (Figs 3–5), a problem specific software was developed in the Heidelberg Numerical Methods Group, based on the open source C++ library deal.II [41]. The system was discretized in time by the damped Crank-Nicolson method. The intercellular area was discretized with an unstructured adaptive mesh, which describes each cell with at least 342 degrees of freedom (64 in the long range simulations in Fig 5) by a Galerkin approach using continuous finite elements (Q1). The discretized system was solved efficiently by controlling the error with adaptive space and time grids by means of the Dual Weighted Residual (DWR) method[42,44]. To allow for larger time steps, the equations were solved in a fully coupled fashion and not with the commonly applied iterative segregating approach. We linearized the nonlinear equations with Newton's method and applied Krylov-Space methods (GMRES) with a geometric multilevel preconditioner [40,43] to solve the resulting linear equations. In the simulations, the secreting Th cells and the Treg cells and the synapse on the cell surface of secreting cells were positioned randomly. We checked the influence of this cell positioning on the simulations with different randomly chosen positions and found that the variations between simulations were negligible.
Our discretized high-resolution numerical data were visualized in cooperation with the Visualization and Numerical Geometry Group from the Interdisciplinary Center of Scientific Computing (IWR) in Heidelberg. For the graphical representation of the three-dimesional scalar data, here the IL-2 distribution in space, two methods were applied, the visualization of isosurfaces using topological methods [64,65] and volume rendering [64]. With the first method specific isosurfaces are visualized by varying the transparency for different isovalues to get an impression of the 3D data set (Figs 5A, S1 and S2). To choose these specific isosurfaces with important features, topological information (Morse complex, persistent homology classes and Betti numbers) is computed. The rendering was performed by using the Visualization Toolkit VTK (http://www.vtk.org) which allows rotation in real time. The second method, volume rendering, produces the image directly from the data without an intermediate geometrical representation. A play with transparency of the whole data set makes the inner structures visible (Figs 3D and 4D). With flexible mapping of the data on colors and opacity, different structures can be visualized efficiently and a realistic representation is obtained (Figs 3C and 3D and 4C and 4D). Difficulties in the data-representation were the wide range of the values over several orders of magnitude and the porous domain (extracellular domain).
The simulations for pulsed stimulation (Fig 5D and S3 Fig) were realized by homogeneous secretion by the cell in the center of the region for a very short time (7 sec) with a qeff such that a concentration corresponding to a single cytokine molecule is released. The simulation is then run until the concentration reaches zero in the whole area. The fraction of the released IL-2 concentration bound by a certain responder cell is equivalent to the probability that the ‘secreted molecule’ was bound. This probability was calculated for the successive layers of responder cells surrounding the secretory cell, in order to obtain the distribution of the traveling distance.
Analytical calculations were supported by Wolfram’s Mathematica. Matlab from Mathworks was used to generate plots and to calculate the special functions applied in Fig 2.
A classical formula derived by Berg and Purcell approximates the characteristic time τ of a ligand diffusing towards a receptor [48]:
τ=14πDρc+14DdRRc=6.9min
(1)
Here, we suppose a cytokine concentration of c = 10 pM, a receptor diameter of dR = 0.1nm, a receptor number of R = 100 per cell, and diffusion constant D and cell radius ρ as in Table 1. Note that in Eq 1 and in the following, cytokine concentrations (nM) are implicitly converted to molecules/μm3 by Avogadro’s constant NA, wherever necessary, as follows: nM = 10-9mol/l = 10-9NA molecules/(1015μm3) = 6/10 molecules/μm3. Note that the time to diffuse towards a T cell (first term in Eq 1) is less than a second, but the mean time to reach a receptor at the cell surface (second term in Eq 1) is in the order of minutes due to the small number of receptors on naïve T cells.
One cytokine secreting cell is either surrounded by a layer of responder cells (‘high cell-density’, see Fig 1B) or placed in a cell-free medium (‘low cell-density’). The cytokine secreting cell has R cytokine receptors, and responder cells have Rresp cytokine receptors, both binding cytokine molecules in their immediate vicinity with rate kon. We assume homogeneous cytokine secretion and uptake, so that the system has radial symmetry. As diffusion is fast (D = 10μm2/s, see Table 1), it reaches a steady state after about L2/D = 0.5 s, where L is the cell-to-cell distance in the case of high cell-density. Thus, it is sufficient to consider the diffusion equation in steady state in the extracellular domain with flux boundary condition at the cell surface:
DΔc(r)=0, r∈[ρ,∞]−4πρ2D∂c∂r|r=ρ=q−konc(ρ)R
(2)
c(r) is the cytokine concentration at distance r from the center of the cell, Δ is the Laplace operator in spherical coordinates, ρ is the cell radius, and q is the cytokine secretion rate. Note that cytokine concentrations are implicitly converted from unit nM to unit molecules/μm3, as above. The boundary condition on the outer boundary is either (low cell-density limit)
c(r→∞)=0
(3)
or (high cell-density limit)
−4π(L+ρ)2D∂c∂r|r=L+ρ=konc(L+ρ)NRresp,
(4)
where N is the number of IL-2 consuming responder cells. In both cases, the problem can be solved analytically for the cytokine concentration c(r) and eventually for the uptake rates Jauto = konc(ρ)R, Jpara = q − Jauto (see S1 Text).
We consider stationary cytokine diffusion in a cylindrical region between a cytokine secreting Th cell and a responder cell, both potentially expressing cytokine receptors (see Fig 2A). This leads to the following boundary conditions at the cytokine secreting cell (z = 0) and the responder cell (z = l):
DΔc(r,z)=0, r∈[0,a], z∈[0,l]−πa2D∂c∂z|z=0=q−konRc|z=0−πa2D∂c∂z|z=l=konRrespc|z=l
(5)
The synaptic distance is l = 20 nm, and the radius of the contact area is a = 2 μm (see Table 1), corresponding to the region where localized IL-2R expression is reported [11]. At the outer boundary of the synapse, we assume c(a,z) = 0, which means that cytokine molecules which escape the cylindrical region do not return to it. The cytokine concentration, and the uptake rates Jauto, Jescape and Jsynapse resulting from this model, can be calculated analytically using Bessel functions (see S1 Text).
We performed simulations in three spatial dimensions (see section ‘software’ above) of our earlier model [4], with some modifications: We consider polarized IL-2 secretion and autocrine uptake at the immunological synapse, by assuming an effective secretion rate at one grid point at the surface of IL-2 secreting cells. Moreover, due to recent experimental observations [5,35], we discard the previously assumed positive feedback from IL-2 uptake to IL-2 secretion, and we set the IL-2 secretion rate to 10 molecules/s and the fraction of IL-2 secreting cells to about 25% (see Table 1). In brief, the model [4] considers interactions of three kinds of cells: Secretory Th cells, responder Th cells and Treg cells. All three cell types express IL-2R molecules on the cell surface. Responder Th cells and Treg cells express IL-2R homogeneously at the cell surface, Treg cells at higher levels than responder Th cells. IL-2 signaling leads to the expression of the α subunit of the IL-2 receptor that is required for high-affinity IL-2 binding in both responder Th cells and Treg cells. Hence both cell types enhance their rate of IL-2R expression (v) upon IL-2 uptake, which we model, following Busse et al. [4], by a Hill equation with a moderate Hill coefficient of 3:
v(t)=v0+v1C(t)3K3+C(t)3
(6)
Here, v0 and v1 are the basal and the IL-2 induced rates of IL-2R expression, K is the half-saturation constant, and C(t) is the number of IL-2/IL-2R complexes, which is a dynamic variable of the model (Table 1). For details and the full model see S1 Text.
|
10.1371/journal.pbio.0050233 | Emergence of Large-Scale Cell Morphology and Movement from Local Actin Filament Growth Dynamics | Variations in cell migration and morphology are consequences of changes in underlying cytoskeletal organization and dynamics. We investigated how these large-scale cellular events emerge as direct consequences of small-scale cytoskeletal molecular activities. Because the properties of the actin cytoskeleton can be modulated by actin-remodeling proteins, we quantitatively examined how one such family of proteins, enabled/vasodilator-stimulated phosphoprotein (Ena/VASP), affects the migration and morphology of epithelial fish keratocytes. Keratocytes generally migrate persistently while exhibiting a characteristic smooth-edged “canoe” shape, but may also exhibit less regular morphologies and less persistent movement. When we observed that the smooth-edged canoe keratocyte morphology correlated with enrichment of Ena/VASP at the leading edge, we mislocalized and overexpressed Ena/VASP proteins and found that this led to changes in the morphology and movement persistence of cells within a population. Thus, local changes in actin filament dynamics due to Ena/VASP activity directly caused changes in cell morphology, which is coupled to the motile behavior of keratocytes. We also characterized the range of natural cell-to-cell variation within a population by using measurable morphological and behavioral features—cell shape, leading-edge shape, filamentous actin (F-actin) distribution, cell speed, and directional persistence—that we have found to correlate with each other to describe a spectrum of coordinated phenotypes based on Ena/VASP enrichment at the leading edge. This spectrum stretched from smooth-edged, canoe-shaped keratocytes—which had VASP highly enriched at their leading edges and migrated fast with straight trajectories—to more irregular, rounder cells migrating slower with less directional persistence and low levels of VASP at their leading edges. We developed a mathematical model that accounts for these coordinated cell-shape and behavior phenotypes as large-scale consequences of kinetic contributions of VASP to actin filament growth and protection from capping at the leading edge. This work shows that the local effects of actin-remodeling proteins on cytoskeletal dynamics and organization can manifest as global modifications of the shape and behavior of migrating cells and that mathematical modeling can elucidate these large-scale cell behaviors from knowledge of detailed multiscale protein interactions.
| The shape of animal cells is largely determined by the organization of their internal structural elements, including the filamentous structures of their cytoskeleton. Motile cells that crawl across solid substrates must assemble their cytoskeletal actin filaments in a spatially organized way, such that net filament growth and cell protrusion occur at the front of the cell. Actin filament dynamics, in turn, influence the overall shape of the cell by pushing on the plasma membrane. In this work, we have explored the ways that variations in small-scale actin filament growth dynamics are coupled to large-scale changes in cell shape and behavior. By manipulating the availability of a family of actin-binding proteins (Ena/VASP) that regulate actin filament growth, we can alter the overall cell shape and motile behavior of epithelial fish keratocytes—unusually fast-moving and regularly shaped cells. We have also found that unperturbed keratocytes in a population exhibit a continuum of shape and behavioral variations that can be correlated with differences in Ena/VASP levels. We have developed a mathematical model that allows us to explain our observations of intrinsic cell-to-cell shape variation, motile behavior, and cell responses to molecular perturbations as a function of actin filament growth dynamics in motile cells.
| The spatiotemporal coordination of the assembly, disassembly, and organization of the actin cytoskeleton is essential for efficient cell migration. The underlying mechanisms by which the actin cytoskeleton is organized and remodeled into specific architectures, which are then conveyed over large scales into observable cell morphologies, remain unclear. However, careful observation of large-scale morphology and behavior can shed light on these mechanisms. The heterogeneity of wild-type populations [1] can be used as a “natural experiment” in which potentially meaningful correlations between observations at molecular and global scales are determined. Because of the complex relationships between the underlying molecular interactions and observable parameters, physical and mathematical modeling is often necessary to interpret such quantitative data in terms of fundamental molecular mechanisms [2]. To achieve a mechanistic understanding of how the global shape and migratory behavior of cells are generated, we used a combination of quantitative analysis of natural cell-to-cell variation and mathematical modeling to help us grasp how the large-scale organization and function of the actin meshwork emerges and propagates from the dynamics of its molecular components.
The actin cytoskeleton can be remodeled by many different families of proteins, including the enabled/vasodilator-stimulated phosphoprotein (Ena/VASP) family, which affects dynamic processes such as growth, capping, and bundling of actin filaments [3], thereby regulating the local spatial organization of the actin cytoskeleton in cells [4–7]. Members of this family—represented by VASP, Mena (mammalian Ena), and EVL (Ena/VASP-like protein) in mammals—are largely functionally interchangeable [8] and have been recognized as important regulators of the actin cytoskeleton during cell migration and axon growth, as well as during filopodia formation, platelet aggregation, and phagocytosis [4,5,7,9–13].
Ena/VASP proteins have been of special interest in the field of cell migration, because they have been found to be both positive and negative regulators of cell speed in diverse motile cell types ranging from the actin-based movement of the intracellular pathogen Listeria monocytogenes to overall amoeboid migration of eukaryotic cells. The central proline-rich region of the Listeria surface protein ActA binds Ena/VASP proteins [14,15], which in turn recruit profilin–actin complexes [9,16] that are necessary for efficient actin monomer addition to growing filaments supporting bacterial propulsion. This mechanism accounts for the dramatic decrease in speed observed in Listeria when Ena/VASP proteins are depleted [8,17] and the speed increase observed when VASP is added to a reconstituted motility system [18]. Analogously, suppression of Ena/VASP protein function has been shown to decrease the speed of migrating neutrophils [10] and chemotaxis efficiency by Dictyostelium discoideum[11]. Conversely, Ena/VASP protein depletion resulted in faster moving fibroblasts due to the reorganization of the actin network, which became highly branched with short actin filaments, leading to more persistent lamellipodial protrusion [5,6]. A functional mechanism for these proteins has emerged, suggesting that Ena/VASP proteins remodel actin networks by enhancing the formation of long actin filaments, competing with capping protein, and potentially decreasing the branching activity of the actin-related protein 2/3 (Arp2/3) complex [4,6,7,19,20]. However, additional studies found no evidence for the latter two activities of VASP [21,22], and its exact molecular functions remain controversial.
Cell morphology represents the global manifestation of the cell's structural organization of the cytoskeleton and thus reflects the specific migratory behavior of different cell types. For example, epithelial fish keratocytes, which are among the fastest locomoting cells, exhibit flat lamellipodia as they glide along two-dimensional surfaces, whereas neutrophils have thicker, more amorphous pseudopodia that allow them to crawl through three-dimensional tissues with speeds comparable to that of keratocytes [23]. Keratocytes have been described as a “fan-” or “canoe-” shaped, exhibiting minor variations in shape and direction during migration [24–26]. With their simple stereotyped shape, keratocytes have been regarded as a good model system to study shape in migrating cells [26–29]. However, not all migrating keratocytes in culture are perfectly stereotyped; a certain fraction naturally exhibits more irregular morphologies [30–32] that have not been studied as well. Following our initial observation that these keratocyte morphologies were correlated with the presence or absence of VASP at the leading edge of the lamellipodium, we investigated how Ena/VASP activity influenced cell morphology as well as motile behavior. We hypothesized that the specific actin filament dynamics produced by actin remodeling proteins, such as Ena/VASP, organize the actin network and contribute to global cell morphology and migration. Quantitative analytical approaches were necessary to discern relationships between numerous perceptible morphological phenotypes and also to detect subtle changes caused by molecular manipulations.
To confirm our initial observation, we measured cell shape, leading-edge shape, filamentous actin (F-actin) distribution, cell speed, directional persistence, and VASP enrichment at the leading edge in a population of keratocytes. Systematic quantitative analysis revealed that these parameters correlated with VASP enrichment at the leading edge, spanning a clear continuum of coordinated phenotypes. Moreover, we have developed a mathematical model that explains the properties of this continuum—in particular, the quantitative correlations observed between the observable, large-scale parameters—in terms of small-scale molecular interactions between VASP and the growing actin architecture. Specifically, our model suggests that the role of Ena/VASP in protecting growing filaments allows for larger-scale cohesion in the actin meshwork, promoting smooth canoe shapes and faster migration. By experimentally manipulating Ena/VASP availability at the leading edge and thus local actin filament growth kinetics due to Ena/VASP activity, we were able to alter the prevailing morphology and trajectory of keratocytes within a population in a way that was accurately predicted by our model. Together, our results suggest that Ena/VASP proteins play a major role in cell morphology and motility by modulating the organization and thus promoting the large-scale coherence of the actin network. Our general approach of using detailed mathematical modeling to connect quantitative measurements of large-scale cell morphological and behavioral features to specific protein biochemical activities should be broadly applicable to many cytoskeleton-associated proteins involved in cell migration.
Populations of primary migrating epithelial fish keratocytes are heterogeneous in cellular morphologies, sizes, and motile behaviors. Most descriptions of keratocytes focus on a subpopulation of cells with stereotyped canoe-like shapes [24–26] and smooth lamellipodial leading edges; however, many have more irregular shapes and rough leading edges [31,32] (Figure 1A and 1B). We initially examined cichlid keratocytes with these extreme morphologies and focused on their leading-edge morphology, which we classified by eye as smooth or rough.
Differences in morphology became more evident when we observed by immunofluorescence that VASP was localized as a uniform thin line at the leading edge of keratocytes with smooth leading edges and did not appear at the edge of cells with rough margins (Figure 1A and 1B). VASP was especially evident at focal adhesions at the rear sides of the cell body of rough polarized keratocytes (Figure 1B) and in keratocytes found in epithelial sheets (unpublished data). When we examined enhanced green fluorescent protein (EGFP)-VASP expression in live migrating keratocytes, we observed a similar localization, with VASP more strongly localized at smooth leading edges (Figure S1). Similar results were observed when the localization of EVL, a different member of the Ena/VASP family, was examined (unpublished data); however, we decided to focus on VASP because its function has been more thoroughly characterized. When individual migrating keratocytes expressing EGFP-VASP spontaneously switched from rough to smooth morphologies, an increase in VASP fluorescence at the leading edge and a decrease at focal adhesions was observed when keratocytes achieved the smooth morphology (Figure S1). Morphology switching was generally an uncommon event on the time scales over which time-lapse sequences were collected (tens of minutes), suggesting that the correlation between VASP localization and cell morphology is stable over the time scale of actin filament turnover in these cells (<30 s) [33]. Our observations, which suggested a relationship between Ena/VASP localization at the leading edge and large-scale cell morphology, prompted us to investigate whether VASP redistribution caused these morphological changes.
To test whether Ena/VASP proteins directly modulated leading-edge shape, we manipulated their availability at the leading edge of keratocytes. To decrease Ena/VASP availability, we used a construct (FP4-mito) derived from the Listeria ActA protein, which localizes to mitochondria when expressed in eukaryotic cells [15,34] and has four proline-rich repeats (P4) that efficiently bind Ena/VASP proteins [14,15,35]. FP4-mito was previously shown to function as an Ena/VASP dominant-negative construct by sequestering and mislocalizing Ena/VASP proteins at the surface of mitochondria thus preventing their function at their normal sites of activity, such as the leading edge, tips of filopodia, and cell–cell contacts in tissue culture cells as well as in developing embryos [4,5,36–38]. As a control, we used a similar construct (AP4-mito) that has been previously used as negative control [4,5,36–38] because it contains point mutations that dramatically reduce binding to Ena/VASP proteins [39] while retaining the ability to localize to mitochondria [5]. When we expressed EGFP tagged FP4-mito in keratocytes, VASP (Figure 2A and 2B) and EVL (unpublished data) were efficiently mislocalized to mitochondria, and a higher percentage of migrating keratocytes, which were observed with time-lapse video-microscopy, exhibited the rough morphology (70%) compared with controls (Figure 2C). Conversely, when EGFP-VASP was overexpressed, a significantly lower percentage of keratocytes (43%) exhibited the rough phenotype (compared to cells expressing EGFP-FP4-mito, p = 0.03, Figure 2C). These results suggest that VASP enrichment at the leading edge can tilt the balance of morphology toward the smooth phenotype.
We also used these time-lapse sequences to examine differences in motile behavior between cells with smooth and rough morphologies. When we measured migration speed, we found that smooth cells were significantly faster than rough cells (Figure 2D, p < 0.01) suggesting that lamellipodial morphology, which can be influenced by VASP availability at the leading edge, is tightly coupled to the migration speed of these cells. Because fish keratocytes have been observed to generally migrate with persistent straight trajectories over long distances in vitro [40], we examined whether directional persistence was related to morphology. We found that smooth cells expressing control constructs (EGFP and EGFP-AP4-mito) had significantly straighter trajectories compared with those of rough cells (p < 0.001, Figure 2E).
Since Ena/VASP availability influenced the fraction of smooth, straight-moving keratocytes within a population, we next examined whether manipulating VASP availability at the leading edge would alter cell trajectories. When Ena/VASP proteins were mislocalized (EGFP-FP4-mito), smooth cells moved in more curved trajectories that were similar to those of rough cells and significantly different from smooth cells expressing control constructs (p < 0.001, Figure 2E). In contrast, when EGFP-VASP was overexpressed, rough keratocytes, which had curved trajectories in controls, maintained straighter trajectories similar to those from smooth cells expressing control constructs and EGFP-VASP. These results suggest that directional persistence was more sensitive to VASP availability at the leading edge than was leading-edge morphology. Taken as a whole, our results show that VASP localization at the leading edge correlates with smooth, fast, and straight-moving keratocytes, and that manipulating Ena/VASP availability alters the morphology and trajectory curvature of keratocytes within a population.
Thus far, we had observed that VASP localization was related to broad classes of keratocyte leading-edge morphologies and that we could manipulate morphology by mislocalizing or overexpressing VASP. We wondered whether morphological variation among wild-type keratocytes might be related to VASP levels at the leading edge, and therefore we performed a detailed, quantitative characterization of a keratocyte population. Instead of using a binary and subjective classification of smooth versus rough, we characterized the natural morphological heterogeneity of keratocytes along several measurable and objective phenotypic continua.
To measure cell morphology rigorously, we determined mathematically the major modes of shape variation by applying the principal components analysis (PCA) to a population of keratocyte shapes represented as aligned, polygonal contours [41]. We found that three primary modes of shape variability accounted for over 95% of all morphological variation: one mode corresponding approximately to cell size, one corresponding to aspect ratio (i.e., whether cells were shaped more like a wide canoe or a rounded “D”), and one corresponding to the position of the cell body along the front–rear direction (see Materials and Methods, Figure S2) [41]. Since we wanted to test whether cell morphology was related to VASP levels, we quantified VASP enrichment at the leading edge of cells by dividing the highest mean VASP fluorescence intensity across the leading edge (“peak”) by the lowest VASP mean intensity (“base”) found interior to the leading edge. This measure of “VASP peak-to-base ratio” is illustrated in Figure 3A and 3B. We found that the population of keratocytes examined (n = 43) displayed a wide and apparently continuous range of VASP peak-to-base ratios (Figure 3C). When we compared cell morphology to VASP enrichment at the leading edge, we found that only the shape mode that correlated with VASP levels described the canoe-to-rounder-D–shape transition. Keratocytes with VASP enriched at the leading edge (high VASP peak-to-base ratios) had a tendency to resemble canoe shapes, whereas cells with low VASP at the edge were more likely to have rounder D shapes (p = 0.0002, n = 43, Figure 3D).
To evaluate the shapes of leading edges quantitatively instead of qualitatively classifying them as smooth or rough, we measured the degree of roughness of the leading edges by calculating the sum of the local curvature at each of 90 points along front of the cell contours (see Materials and Methods). Since curvature at a point is defined as the reciprocal of the radius of the osculating circle, sharply bending curves that are present in rough leading edges osculate small circles and thus have large local curvatures. Our results confirmed our qualitative observation (from Figure 1) that strong VASP localization at the leading edge correlated with smooth edges (p = 0.0003, n = 43, Figure 3E). Additionally, canoe-shaped keratocytes had decreased local curvature and thus smooth leading edges (p = 0.0003, n = 43, Figure 3F). In summary, enrichment of VASP at the leading edge correlated with canoe shape and smooth leading edges, strongly suggesting a morphological continuum related to VASP activity at the lamellipodial edge.
To examine the behavior of live keratocytes with smooth or rough leading edges, we followed their contours, which were generated from each frame of time-lapse sequences of keratocytes overexpressing EGFP-VASP. The shape of the leading edge in rough cells varied widely, whereas smooth cells maintained a constant shape with minor fluctuations (Figure 4 Video S1). In the particular example shown (Figure 4), we also observed that the smooth keratocyte migrated at approximately twice the speed of the rough one.
Our results indicated that the five parameters considered thus far—VASP peak-to-base ratio, cell shape, local leading edge curvature, speed, and directional persistence—all correlated with each other, creating a continuum of keratocyte phenotypic morphologies. One extreme of this continuum contained fast, straight-moving cells with VASP enriched at the leading edge, canoe-like shapes, and smooth leading edges (Figure 2E 2F, and S3). We refer to cells in this end of the continuum as “coherent” to convey their stable morphology and directed movement. The opposite extreme in the continuum of keratocyte morphologies encompassed slow, meandering cells with low VASP at the leading edge, rounder D shape, and rough leading edges, which we denote as “decoherent.”
Because previous studies have indicated that keratocyte leading-edge shape may be related to actin filament (F-actin) density [28], we compared the distribution of F-actin to keratocyte morphology and VASP levels at the leading edge. Keratocytes with high VASP and a coherent morphology had F-actin distributions along the leading edge that peaked in the middle at the front of the cell (Figure 5A), whereas cells with low VASP and a decoherent morphology had uniform F-actin distributions (Figure 5B). We also found that the F-actin density along the leading edge of coherent cells was increased compared with decoherent cells (Figure 5C). To compare VASP enrichment to the enhancement of F-actin in the middle of the leading edge of different cells, we calculated a ratio (referred to as “F-actin peak ratio”) of the mean F-actin intensity values from the middle half of the leading edge (1/4 to 3/4 position along the edge) to the mean of the F-actin values from the rest of the leading edge (positions 0 to 1/4 and 3/4 to 1 along the edge) of each cell (Figure 5C). We found a significant correlation between F-actin enhancement in the middle of the leading edge (F-actin peak ratios) and VASP enrichment (VASP peak-to-base ratios), suggesting that VASP accumulation at the leading edge is associated with the peaked or graded accumulation of F-actin in coherent cells (p < 0.0001, n = 43, Figure 5D).
When we examined the relationship of the Arp2/3 complex to F-actin and cell morphology, we found that Arp3 distribution, as measured by immunofluorescence, corresponded to that of F-actin in both coherent and decoherent cells, which had peaked and flat distributions, respectively (Figure 5E and 5F). When we compared the spatial distribution of the ratio of Arp3 to F-actin to infer the degree of filament branching, no consistent differences in Arp3–to–F-actin ratios were observed between different keratocyte morphologies (unpublished data), suggesting that VASP activity does not significantly affect branching in keratocyte lamellipodia, consistent with previous findings using purified protein systems [21,22], but in contrast to other studies employing cells or purified proteins [4,6,19].
To unify our observations into a functional context, we developed a mathematical model that accounted for self-organization of keratocyte leading edge and VASP-mediated F-actin growth dynamics. This model allowed us to make predictions about keratocyte shape and was based on the following assumptions about actin dynamics and protrusion at the leading edge:
(1) The F-actin network is organized in a dendritic array such that actin filaments are oriented at ±35° relative to the locally normal direction of protrusion [42]. Filaments are distributed over a wide range of angles, but this distribution is doubly-enhanced and peaked at ±35° due to optimal growth conditions for both mother and daughter filaments, the angle between which is 70°. Since mother and daughter filaments are oriented at the same angle with respect to the leading edge [42], we lump all filaments growing to the left and to the right into two groups, and do not explicitly keep track of individual angles.
(2) Growing barbed ends at the leading edge elongate with a rate limited by membrane resistance and local concentration of actin monomers (G-actin) [43].
(3) Arp2/3-mediated filament branching takes place with equal rate per each existent leading-edge filament [28] (Text S1). This per filament rate is equal to the total number of filaments nucleated over the whole leading edge per second divided by the total number of the uncapped leading-edge filaments. The molecular pathway determining this rate is unknown; a plausible mechanism could be based on rapidly diffusive molecules, the total number of which is conserved, controlling the total number of branching events per cell. Assuming that the branching takes place only along the leading edge, each filament has equal probability to become a mother filament. Then, as the total number of growing filament ends increases, the branching rate per filament inversely decreases. A filament at +35° branches off filaments oriented at −35°, and vice versa [42]. We define the leading-edge filament as the filament whose growing barbed end is in physical contact with the membrane.
(4) VASP associates with/dissociates from barbed ends with constant rates and remains associated with elongating barbed ends until it dissociates [6,20,44].
(5) VASP protects barbed ends from capping; unprotected barbed ends are capped at a constant rate [6,20].
(6) The barbed ends of elongating actin filaments undergo lateral flow along the leading edge with a rate proportional to local protrusion [28,45].
(7) The shape of the leading edge is determined by the graded radial extension model [29], according to which the local slope of the leading edge is determined by the ratio of the local normal protrusion rate to that in middle front of the cell.
(8) The length of the leading edge is a constant parameter. At the sides of the leading edge, boundary densities of the uncapped (VASP-free and VASP-associated) barbed ends are constant parameters in the model. These parameters are crucial for the model predictions (discussed below).
These assumptions, which are expressed mathematically in Text S1, lead to equations governing VASP activity, F-actin density, protrusion rate, and leading-edge stability and shape. The analytical and numerical solutions qualitatively explain our experimental observations as follows. In coherent cells, which have high VASP activity at the leading edge and low effective capping rate, the average density of actin filament barbed ends at the leading edge is increased, as well as the proportional VASP density associated with these ends (see Figure 5A–5D; F-actin density, measured along the curve very close to the leading edge in this figure, is proportional to the number of actin filaments per micrometer intersecting with the curve parallel and just behind the leading edge, and therefore is also proportional to the density of barbed ends, assuming that all filaments abutting the leading edge are growing [42]). A simple explanation for this increase in F-actin density in the presence of VASP is that VASP skews the balance between branching and capping. Without VASP, nascent barbed ends emerge at a constant rate, whereas a constant capping rate maintains an average number of growing filaments. VASP protects a fraction of the growing barbed ends from capping, so the effective capping rate per total number of growing ends decreases, increasing the average number of growing filaments (see Text S1 for quantitative details). In addition, when VASP is enriched at the leading edge, actin filaments, which grow for longer time periods before capping, undergo significant lateral flow (illustrated in Figure 6A).
When we investigated the stability of the leading edge of coherent keratocytes mathematically, we found that high VASP activity maintains greater density of barbed ends abutting the membrane at the front, leading to low membrane resistance per filament. This low resistance allows the protrusion rate to become insensitive to F-actin density, and instead limited by G-actin concentration. The even distribution of G-actin along the leading edge, together with the lateral flow of actin filaments, leads to the smooth leading edge of coherent cells (Text S1). In this coherent regime, significant fluctuations of the F-actin density do not cause respective fluctuations of the local protrusion rate, and the leading edge remains smooth. In decoherent cells, which have low VASP activity at the leading edge and a high effective capping rate, elongating filaments are rapidly removed from the leading edge by capping and the density of barbed ends decreases (Figure 5C). Barbed ends, which grow for shorter time periods before capping, undergo slow lateral flow and are not redistributed along the leading edge. In this decoherent regime, fluctuations of F-actin density cause respective fluctuations of the local protrusion rate: high local branching density due to stochastic fluctuations at random locations increases the number of filaments pushing the membrane at the front creating a local protrusive “lobe” (Figure 6B and Text S1). Barbed ends slide faster into and slower out of the lobe, creating a positive feedback between actin filament local focusing and protrusion that causes short-scale instabilities of the leading edge, thus making its shape rough.
When we modeled the F-actin profiles along the leading edge of cells, we found that they depended crucially on the boundary conditions at the sides of the leading edge and on the total branching rate. We assumed that at the sides of the leading edge the cell, where the large adhesions are located, there are specific local conditions generating and maintaining a constant density of uncapped barbed ends. If this fixed boundary density is equal to the average density being maintained along the leading edge by the dynamic balance between branching and capping, then the F-actin density along the leading edge is constant (Text S1). However, if the boundary density is less than this threshold, more nascent filaments branch out closer to the center of the cell. This, in turn, increases the net branching rate at the center, because more nascent filaments branch off the higher number of the existent filaments at the center. The existent growing barbed ends start to effectively compete for resources (because the total number of branching events per second is conserved), and if the F-actin density at the cell sides is kept lower, the center “wins.” This positive feedback leads to the characteristic inverted parabolic profile of the F-actin distribution along the leading edge with maximum at the center (Figure 6C) that matches our observations (see Figure 5A and 5C). The lateral flow is crucial for maintaining the coherent inverted parabolic profile of the F-actin distribution along the leading edge; without it, the barbed ends would cluster irregularly at random locations. The flat F-actin distribution at the leading edge of decoherent cells is due, in part, to the slow and irregular lateral flow along the leading edge.
The characteristic canoe shape of coherent cells is achieved through a graded distribution of extension along the leading edge. Experimentally, we observed that coherent cells with high VASP at the leading edge have increased F-actin density at the leading edge (Figure 5C), which according to our model, leads to increased rates of actin growth and protrusion (Figure E of Text S1). With this high F-actin density peaking in the middle of the leading edge, the rate of protrusion, which is insensitive to the density of barbed ends, decreases very slowly along the leading edge, so the leading edge remains flat and extends far from side to side creating the characteristic wide canoe shape (Figure 6D). At the sides, where the F-actin density decreases significantly, membrane resistance starts to limit protrusion, and the rapidly decreasing protrusion rate leads to high curvature at the sides of the leading edge. In decoherent cells, the overall shape of the leading edge remains parabolic, although with sharper transitions from the center to the curved sides, which are apparent as a rounder D keratocyte leading edge shape (Figure 6D). Because these cells are characterized by lower F-actin densities, which correspond to a qualitatively different region in the density–velocity relation compared with coherent cells (Figure E of Text S1), the protrusion rate in decoherent cells decreases faster from the center to the sides, where protrusion drops to levels that cannot overcome membrane resistance. Consequently, the distance from the center to the sides is less than that in coherent cells, so decoherent cells are narrower from side to side.
Since our model predicted that VASP was responsible for the morphological phenotypes observed, we tested our model by acutely delocalizing Ena/VASP proteins from the leading edge of keratocytes. VASP was delocalized by competition with the pharmacological barbed end capper, cytochalasin D [6,46]. VASP delocalization was often accompanied by a decrease in cell width, suggesting that these two parameters were functionally connected (Figure 7A, B). This result also supported our model, which proposed that low VASP activity at the leading edge resulted in narrow D shaped keratocytes (see Figure 6C, D). In a population of keratocytes, cytochalasin treatment eliminated cells with highest enrichment of VASP at the leading edge (Figure 7C).
Our quantitative comparison of shape showed that cytochalasin treatment eliminated keratocytes with extreme canoe shapes (Figure 7D). Moreover, the observed correlations that established a relationship between cell shape, local leading edge curvature, F-actin distribution, and VASP enrichment at the leading edge of wild-type cells were absent in cells treated with cytochalasin (Figure 7D–7G). Our results show that cytochalasin D, acting as a barbed end capper, antagonized VASP localization at the leading edge and altered the shape of keratocytes and the F-actin network towards the decoherent side of the phenotypic continuum.
During extensive observation of different keratocyte morphologies, we hypothesized that coherent keratocytes with high VASP at the leading edge represented a mature state of cellular organization and migration. We evaluated the contribution of VASP in the generation of smooth lamellipodia in coherent cells by obstructing lamellipodial protrusion and examining its subsequent emergence and recovery. When we placed a barrier in the path of movement of a coherent keratocyte with EGFP-VASP enriched at the leading edge, the front edge of the lamellipodium that reached the barrier became temporarily stalled and the levels of VASP at the leading edge dramatically decreased (Figure 8 and Video S2). When the barrier was removed, the leading edge instantly resumed protrusion and appeared rough with protruding microregions enriched in EGFP-VASP. EGFP-VASP quickly became uniform as the cell continued to regain the original smooth leading edge shape. This rapid redistribution of VASP and thus barbed ends along the leading edge confirms the previously described phenomenon of lateral flow, which is important for the maintenance of coherence, as suggested by our model (Figure 6).
Ena/VASP proteins have not only been implicated in the global determination of migration speeds in different cell types [5,6,10,11,17], but also affect the spatial organization of local actin-related cellular structures, such as lamellipodia that contain a branched dendritic network or filopodia, which possess long actin filaments. The ultrastructure of wild-type lamellipodia and growth cones has revealed long actin filaments, whereas those with depleted Ena/VASP revealed shorter, more branched filaments [4,6]. Lamellipodial structure may also be reorganized to give rise to filopodia by altering actin filament length, through changes in the activities of cappers and antagonizing factors that facilitate filament growth [7,44,47]. Therefore, the balance between the activity of Ena/VASP proteins and capping proteins may determine the type of actin network architecture present in different cell types, which may be observed as changes in cell morphology.
Our initial observations of epithelial fish keratocytes revolved around cell shape and leading-edge morphology. Keratocytes have broad, flat lamellipodia that lack filopodia and have been generally described as having a characteristic fan or canoe shape [24–26] despite the fact that morphological variation is part of the natural heterogeneity of keratocytes obtained from primary cultures [31,32]. Particularly, very little attention has been devoted to less-regular morphologies and to understanding how “coherent,” smooth keratocytes differ from “decoherent,” rough ones. In this study, we have shown molecular differences between these two extreme morphologies and established a strongly correlated suite of morphological phenotypes related to Ena/VASP accumulation at the leading edge. Coherent keratocytes have VASP that is enriched at the leading edge and peaked F-actin distributions along the edge, whereas decoherent cells have sparse VASP and flat F-actin distributions, suggesting that VASP activity at the leading edge modulates the architecture of the actin network, which then becomes evident as the morphological and motile phenotypes observed.
EGFP-VASP delocalization from the leading edge of keratocytes after cytochalasin D treatment showed that Ena/VASP proteins might be binding at or near the barbed end of actin filaments, in agreement with a previous study in fibroblasts, which proposed that this mechanism protects actin filament barbed ends from capping [6]. This proposed anticapping activity of Ena/VASP has been controversial: biochemical studies have demonstrated that Ena/VASP proteins can inhibit actin filament capping by several different barbed binding proteins [20], whereas other in vitro studies showed no evidence of such competition by VASP [21,22,48]. Even though the net result of Ena/VASP activity appears to result in increased actin filament length, the in vivo molecular mechanism of this effect is still unclear. Increased actin filament length by Ena/VASP proteins may stem from direct competition with capping protein for barbed end binding, increased actin filament growth rate, reduced filament branch formation, or a combination of any of these activities [6,19–21]. Our results are more consistent with the hypothesis that a primary function of VASP at the leading edge is to oppose the activity of capping proteins.
A mathematical model helped us understand how the underlying actin network organization and dynamics were influenced by these VASP activities and how that could lead to distinct cellular morphologies. This model pointed to a specific molecular mechanism by which VASP activity increases the length of filaments within the actin network: VASP prevents filaments from being capped, thus allowing them to grow for a longer time. We also experimentally tested the prediction that VASP was needed for the maintenance of the coherent phenotype based on the mechanistic assumption that VASP competes with capping. We treated cells with cytochalasin D to antagonize barbed-end binding by VASP and thus increase filament capping. We observed a drop in VASP density at the leading edge after cytochalasin D treatment and, in agreement with our model, the side-to-side lamellipodial width decreased linearly with a rate of ∼0.1 μm/s, similar to that of the inward actin flow (C.A.W., P.T. Yam, L. Ji, K. Keren, G. Danuser, and J.A.T., unpublished data)]. The decrease in VASP levels at the edge continued for a few tens of seconds during which the keratocyte width shrank by 20%–30%, and then stabilized. Moreover, VASP displacement from the leading edge not only decreased cell width, but also eliminated cells with extreme coherent canoe-shaped morphologies. Together with the altered fraction of keratocyte morphologies observed after VASP mislocalization or overexpression, these data support the idea that VASP activity is important for the maintenance of the coherent morphology.
When keratocyte migration was examined as a function of cell morphology, we found that coherent, smooth cells migrated significantly faster than decoherent, rough cells, which demonstrates that cell morphology is tightly coupled to the speed of migrating keratocytes. These results are consistent with previous descriptions of keratocytes with fast protrusion rates as fan-shaped, whereas cells with slower protrusion rates were described as irregular or fibroblast-like in shape [30]. Our mathematical model suggests that the increased F-actin density at the leading edge, which is observed in coherent cells, creates less resistance per filament as the filament elongates, so the rates of F-actin growth and protrusion accelerate (Figure E of Text S1), leading to the observed faster migration speed. In decoherent cells, which have low F-actin density resulting from low VASP activity at the leading edge, the membrane resistance per filament is large and becomes the limiting factor in the protrusion rate, which becomes very sensitive to the F-actin density and thus cells migrate slower. We observed a strong relationship between keratocyte speed and morphology, which depended on VASP localization at the leading edge. A positive correlation between VASP localization and cell speed or protrusion has also been observed in Dictyostelium [11] and melanoma cells [49], contrary to observations in fibroblasts [5,6]. These conflicting observations suggest that different cell types may distinctly coordinate protrusion with overall cell migration and may have different rate limiting parameters of actin dynamics and cell motility.
When we examined the directional component of velocity in keratocytes, we observed that rough, decoherent, wild-type keratocytes had increased curvature of trajectory compared to smooth, coherent, wild-type keratocytes. Unlike the smooth and regular leading edge of coherent keratocytes, the leading edge of decoherent cells can fluctuate widely during protrusion. In other words, different regions of the leading edge may protrude at different rates in an uncoordinated fashion. This phenomenon may be associated with greater frequency of cell turning, because either the whole left or right half of the lamellipodium would advance faster than the other half, effectively changing the average orientation of the leading edge and consequently changing the direction of migration. Thus, morphological variations manifest themselves during cell migration creating different behavioral patterns. We also found that Ena/VASP protein mislocalization led to increased trajectory curvature. This result is consistent with previous studies showing that intracellular Listeria that were deficient in Ena/VASP recruitment exhibited increased trajectory curvature [50] and VASP-null Dictyostelium displayed decreased cell directionality during chemotaxis [11]. Note, however, that directional control may be mechanistically quite different in these cell types.
Epithelial fish keratocytes can rapidly migrate in a graceful gliding motion, all the while maintaining a relatively uniform and persistent shape. This migratory behavior requires the exquisite coordination of the intricate cellular migration machinery composed of three processes—protrusion, adhesion, and retraction—which are typically dissected separately. This work, in which we focused on the lamellipodial protrusive actin-based machinery resulting in the elongation and capping of actin filaments, is no exception. Future work, armed with broader and more detailed mathematical models, should strive to integrate our increasing understanding of these individual parts of the machinery and to understand how they interact to generate spatiotemporally coordinated cell migration in different cell types. We believe that this work, though limited in scope and susceptible to hidden variables and as-yet unknown molecular players, provides an example of how information from multiple spatial and organizational scales can be successfully brought together to explain part of a complex phenomenon.
Within the reductionist context of this work, quantitative analysis and mathematical modeling were crucial to the understanding of cell shape and motile behavior in terms of the molecular activity of Ena/VASP proteins. In view of the strong correlation between VASP enrichment at the leading edge and the quantitative morphological parameters analyzed in fixed cells, a more quantitative characterization of the morphology (shape, leading-edge curvature) of live migrating cells may be warranted in the future to provide more detailed insights about the dynamics and activity of Ena/VASP. It is important to note that even though our mathematical model was able to recapitulate and provide a self-consistent explanation of our quantitative observations of cell morphology, F-actin organization, and motile behavior, it was only able to do so in a qualitative manner. Ideally, future modeling will be able quantitatively bridge experimental data and theory. Some steps in this direction are discussed in Text S1. Nevertheless, our current model served as an important tool to generate a testable prediction and to interpret the cell morphologies observed.
Overall, cell morphology represents a large-scale manifestation of underlying cytoskeletal organization and dynamics. Regulation and modulation of the actin cytoskeleton are likely to be major biological mechanisms affecting cell migration. Actin-remodeling proteins localize to propulsive structures in morphologically diverse cell types—neutrophils, fibroblasts, neurons, and intracellular bacterial pathogens—where they play crucial roles in the morphogenesis and maintenance of pseudopods, lamellipodia, filopodia, or bacterial comet tails, all of which inherently have different actin network organizations. Ena/VASP proteins, which are capable of enhancing the elongation of actin filaments by competing with capping protein for barbed-end binding, have emerged as important actin-remodeling proteins and strong candidates for the modulation of the underlying actin cytoskeleton that dictates cell morphology and migration.
Keratocytes were cultured from the scales of the Central American cichlid Hypsophrys nicaraguensis as described [51], but the isolated scales were sandwiched between two acid-washed glass 25-mm coverslips and cultured at room temperature in the dark using Leibovitz's L-15 medium (Gibco BRL; http://www.invitrogen.com) supplemented with 14.2 mM HEPES pH 7.4, 10% FBS, and 1% antibiotic-antimycotic (Gibco BRL) before transfection or processing for immunofluorescence the day after isolation.
Keratocytes were transfected using a small-volume electroporator for adherent cells as previously described [52,53]. Coverslips containing keratocytes were placed in fish Hank's Balanced Salt Solution (HBSS) [54] containing 85% NaCl, and 20 μl of plasmid DNA (1 μg/μl) in water were placed directly onto the cells. Keratocytes were immediately electroporated with three pulses of 185 V and allowed to recover for ∼24 h in culture media to allow for expression. Before live cell imaging or immunofluorescence, sheets of keratocytes that had migrated off the scales were washed for ∼5 min in 85% PBS, 2.5 mM EGTA, pH 7.4 to separate individual migrating cells.
Indirect immunofluorescence was performed using rabbit polyclonal anti-murine VASP (2010) and anti-murine EVL (1404) antibodies [4,5]. Keratocytes were fixed in ice cold 2.5% glutaraldehyde, 0.025% Triton X-100 in PBS for 1 min. Autofluorescence was quenched by incubation in 0.1% sodium borohydride in PBS twice for 5 min. Cells were blocked and permeabilized using PBS-BT (3% BSA, 0.1% Triton, 0.02% sodium azide in PBS) before incubation with antibodies diluted in PBS-BT. F-actin was labeled by incubation with fluorescently labeled phalloidin (Invitrogen; http://www.invitrogen.com).
Indirect Arp2/3 immunofluorescence was performed using rabbit polyclonal anti-human Arp3 antibodies as described previously [55,56], except that cells were simultaneously fixed and permeabilized in cytoskeleton buffer containing 0.32 M sucrose (CBS) [57], 4% formaldehyde, 0.1% Triton X-100, and 0.5 μg/ml FITC-phalloidin (Invitrogen) for 15 min.
Images were acquired using an Axioplan microscope (Carl Zeiss Microimaging; http://www.zeiss.com) equipped with a CCD camera (MicroMAX 512BFT; Princeton Instruments; http://www.piacton.com).
FP4-mito, AP4-mito, and mouse VASP in pMSCV [5,17] were subcloned into pEGFP-C1 (Clontech Laboratories; http://www.clontech.com) using standard molecular biology techniques. BglII and HindIII restriction sites were used to subclone FP4-mito, AP4-mito, and murine EVL. HindIII and BspEI were used to subclone VASP.
Because individual keratocytes are heterogeneous in their responses to pharmacological agents, they were treated with 0.5 μM for 5 min; 0.8 μM for 2, 3, and 5 min; or 1.0 μM cytochalasin D (Sigma; http://www.sigmaaldrich.com) for 2 min in culture media.
Time-lapse images were collected at 10-s intervals using a Nikon Diaphot-300 inverted microscope with a CCD camera (MicroMAX 512BFT; Princeton Instruments; http://www.piacton.com). The rear of keratocytes was tracked using the “Track Points” option of MetaMorph software (Molecular Devices; http://www.moleculardevices.com) to measure speed and direction as previously described [31,50,58]. For population speed analysis, tracks were truncated to correspond to the same time (150 s). In this study, a minority of cells imaged using a different objective (n = 11) and persistent circlers (n = 4) were excluded from trajectory analysis. The population used for trajectory analysis included: EGFP smooth, n = 20; EGFP rough, n = 23; AP4-mito smooth, n = 9; AP4-mito rough, n = 16; FP4-mito smooth, n = 11; FP4-mito rough, n = 26; VASP smooth, n = 9; and VASP rough, n = 11. Keratocyte leading-edge morphology was classified as smooth or rough by eye by determining whether each cell was more similar to the smooth or rough reference cells depicted in Figure 1A and 1B. Long trajectories were collected using a low-magnification air objective, which had a resolution suboptimal for detailed cell shape measurements.
To compare immunolocalized Arp3 and F-actin along the leading edge and the enrichment of immunolocalized VASP (VASP peak-to-base ratios) across the leading edge of keratocytes, measurements were obtained using the “linescan” option in MetaMorph and background subtracted. F-actin and VASP distributions along the leading edge were calculated using the cell outline polygons as guides (see “cell shape analysis” section below). For each vertex point along the leading edge of a given cell, intensities were sampled at 20 points (∼2 μm) for F-actin and ten points for VASP (∼1 μm ranging from ∼0.3 μm outside to ∼0.7 μm inside the outlines) spaced one pixel apart along the inward normal and averaged.
Micropipettes were pulled using a P-92 Flaming-Brown micropipette puller (Sutter Instruments; http://www.sutter.com) from 0.5 mm inner diameter (ID)/1.0mm outer diameter (OD) glass capillaries, and positioned with a Narishige MMO-202ND micromanipulator. Keratocytes were transfected overnight using FuGENE6 (Roche Diagnostics; http://www.roche.com) and allowed to recover and express EGFP-VASP for one day. Time-lapse images were acquired using a Zeiss Axiovert 200 inverted microscope with Nomarski differential interference contrast optics and a Cascade II 512B CCD camera (Photometrics; http://www.photomet.com).
Cell morphology was measured by representing cell shapes as polygonal outlines and comparing those outlines with the PCA, as described [41]. Briefly, cell shapes were manually determined by using the “magnetic lasso” tool in Adobe Photoshop to trace the edge of each cell, based on images of fluorescent phalloidin. Each lasso selection was converted into a binary mask and outlines were extracted from those masks to derive a series of (x,y) points corresponding to the cell boundary. Each series was resampled to 150 points, evenly spaced along the cell boundary. Finally, the outlines were mutually aligned to bring the shapes into a common reference frame. The remaining variability in the point positions was then characterized with PCA to derive a small number of highly explanatory modes of shape variation.
This analysis determined that three principal “shape modes,” which are illustrated in Figure S2, are sufficient to explain over 95% of the variability in shape of the 43 untreated cells and 27 cytochalasin D–treated cells. Of these modes, only the second—describing shapes ranging from canoe to D—correlated with VASP distribution. To quantify the shape of an individual cell, we measured its position along this mode in terms of standard deviations from the mean shape.
To calculate the roughness of each cell's leading edge, we used a measure that we refer to as “local leading-edge curvature.” Mathematically, the curvature of a function at a particular point is defined as the reciprocal of the radius of the circle that has the same tangent as the function at that point. A sharply bending curve will share a tangent with a small circle, and thus have a large curvature; in the limit, a straight line is tangent to an infinitely large circle and has zero curvature. The curvature of a parametric plane curve [x(p),y(p)] at a point p can be calculated as (x′·y ′′ – y′·x′′)/(x′2+y′2)3/2, where prime signifies the first derivative at point p and double prime the second derivative. We calculated the curvature at each of 90 points along the leading edge of the keratocyte outlines, using central-difference approximations to the derivatives. To determine the values of “local leading-edge curvature,” we summed the absolute values of the curvatures along the leading edge, and multiplied this by the length of the leading edge to account for the fact that smaller keratocytes will have higher total curvature due to their size alone. (Under this measure, a perfectly smooth semicircle sampled at 90 points would have a value of 90π [≈283]). Overall, rough leading edges have high local leading-edge curvature values and smooth leading edges have low values.
To examine the contours of migrating keratocytes (Figure 4), cell outlines were calculated as described in P.T. Yam, et al. (unpublished data).
The mean speeds per cell for each pair of transfected keratocyte populations (e.g., EGFP versus AP4-mito, EGFP versus FP4-mito, etc.) and for rough and smooth cells (e.g., EGFP rough versus EGFP smooth) were statistically compared using the Mann-Whitney test. Trajectories were evaluated by comparing mean angles between 2 and 45 μm (distance traveled) using the same test. The proportions of smooth and rough cells present in all combinations of populations of transfected keratocytes were compared using the two-sample test for binomial proportions [59]. Linear regression was used to compare the relationship between VASP peak-to-base ratios, cell shape, F-actin distribution (F-actin ratio), and local leading-edge curvature.
Briefly, we modeled the densities of right- (left-) oriented growing barbed ends along the leading edge with functions b+(x,t) (b−(x,t)) for ends not associated with VASP and with functions b̃+(x, t) (b̃−(x, t)) for ends associated with VASP.
According to the model assumptions, the following equations govern these densities:
We considered these equations on the leading edge: −L ≤ x ≤ L. We choose the boundary conditions at x = ±L as follows:
The meaning of these conditions, choice of the model parameters, and methods of solution of equations are thoroughly explained in Text S1.
We described the leading-edge profile with the function y = f(x). The overall steady shape is derived from the Graded Radial Extension model [28,29] according to the formula:
where
is the local protrusion rate, which is a function of the local density of barbed ends.
To investigate the local stability of the leading edge, we solved the system:
where b̄ is the average density of barbed ends and bl is the local density of barbed ends.
|
10.1371/journal.ppat.1006985 | Type I interferon signaling attenuates regulatory T cell function in viral infection and in the tumor microenvironment | Regulatory T cells (Tregs) play a cardinal role in the immune system by suppressing detrimental autoimmune responses, but their role in acute, chronic infectious diseases and tumor microenvironment remains unclear. We recently demonstrated that IFN-α/β receptor (IFNAR) signaling promotes Treg function in autoimmunity. Here we dissected the functional role of IFNAR-signaling in Tregs using Treg-specific IFNAR deficient (IFNARfl/flxFoxp3YFP-Cre) mice in acute LCMV Armstrong, chronic Clone-13 viral infection, and in tumor models. In both viral infection and tumor models, IFNARfl/flxFoxp3YFP-Cre mice Tregs expressed enhanced Treg associated activation antigens. LCMV-specific CD8+ T cells and tumor infiltrating lymphocytes from IFNARfl/flxFoxp3YFP-Cre mice produced less antiviral and antitumor IFN-γ and TNF-α. In chronic viral model, the numbers of antiviral effector and memory CD8+ T cells were decreased in IFNARfl/flxFoxp3YFP-Cre mice and the effector CD4+ and CD8+ T cells exhibited a phenotype compatible with enhanced exhaustion. IFNARfl/flxFoxp3YFP-Cre mice cleared Armstrong infection normally, but had higher viral titers in sera, kidneys and lungs during chronic infection, and higher tumor burden than the WT controls. The enhanced activated phenotype was evident through transcriptome analysis of IFNARfl/flxFoxp3YFP-Cre mice Tregs during infection demonstrated differential expression of a unique gene signature characterized by elevated levels of genes involved in suppression and decreased levels of genes mediating apoptosis. Thus, IFN signaling in Tregs is beneficial to host resulting in a more effective antiviral response and augmented antitumor immunity.
| Type I interferons (IFNs) play a predominant role in the immune response to infectious pathogens. The cellular targets of IFNs have been difficult to dissect because of the ubiquitous expression of the type I interferon receptor (IFNAR). The immune response of mice to lymphocytic choriomeningitis virus (LCMV) is one of the major models for analyzing the action of IFNs. Regulatory T cells (Tregs) have been implicated in the control of LCMV and it has been proposed that IFN may influence their function. The major goal of this study was to define the contribution of IFN signaling on Treg function during different stages LCMV infection. Tregs from mice with selective deletion of IFNAR manifested enhanced suppressive activity during acute/chronic LCMV infection resulting in increased CD8 T cell anergy, defective generation of memory T cells and persistence of virus. Similar effects of IFNAR signaling in Tregs were seen in a tumor model. We identified a unique set of genes in Tregs modulated by IFN signaling that may contribute to the enhanced suppressive function of IFNAR deficient Tregs. IFNs play a beneficial role during acute/chronic viral infections not only by contributing to viral clearance but also by attenuating the function of Tregs.
| Regulatory T cells (Tregs) are a subset of CD4+ T cells, which express the transcription factor Foxp3, and are critical in forestalling both self- and non-self-reactive immune responses [1, 2]. Tregs primarily mediate their suppressive function by targeting conventional effector T cell activation and differentiation, mainly by decreasing the functional activity of antigen presenting cells (APCs) [3]. The critical role of Tregs in autoimmunity is best observed in scurfy mice or patients with IPEX syndrome that are completely deficient in Tregs and succumb to systemic autoimmune disease at a young age. While Tregs must control the activation of T effector cells to prevent autoimmunity, it is also clear that enhanced activation of Tregs may result in the inhibition of host immunity directed against microbes (virus, bacteria, protozoa, fungi and helminth) or tumors leading to poor antimicrobial or antitumor immune response with the persistence of pathogens, and defective tumor immunity [4–6].
Many animal models of bacterial infection are characterized by the expansion of Foxp3+ Tregs including Listeria monocytogenes, Salmonella enterica, and Mycobacterium tuberculosis infections and the suppressive function of Tregs can result in increased bacterial load with systemic tissue invasion [7–9]. Similarly in viral infection, higher frequencies of Tregs are associated with enhanced titers of Hepatitis C virus RNA and Dengue virus [10, 11]. Paradoxically, Tregs have been described to play an early protective role in local infection in animals models of Herpes simplex virus 2 and West Nile virus [12, 13]. During early phases of human immunodeficiency virus infection, Tregs have been postulated to control virus replication in target CD4+ T cells [14]. On the other hand Tregs may play an important beneficial role in preventing exuberant inflammatory responses during infection with parasites such as Pneumocystis carinii [15] and Schistosoma mansoni [16]. Similarly, Tregs protect the host from parasitic infections such as Plasmodium sp., Toxoplasma gondii, as well as infection with the fungus, Candida albicans [17–19]. These complex roles played by Tregs during acute and chronic microbial infections necessitate a delicate balance between the Foxp3+ Tregs and effector T cells to mount effective immune responses against pathogens without the induction of destructive autoimmunity.
The immune response towards viruses and intracellular bacteria are mediated by type I interferons (IFNs) which control the replication of pathogens within host cells. IFNs are members of a multi-gene family of cytokines, which encode IFN-α and IFN-β. Both IFN-α and IFN-β signal through a shared common heterodimeric receptor IFN-α/β receptor (IFNAR) composed of IFNAR1 and IFNAR2 [20]. The interactions of IFNs with the IFNAR mediates activation of Janus family protein kinases to induce the phosphorylation of signal transducer and activator of transcription (STAT). The canonical pathway of Type I IFN signaling is initiated by phosphorylation of STATs (STAT1, STAT2), induction of IFN-regulatory factor-9, resulting in the formation of a tri-molecular complex, IFN-stimulated gene factor-3, which translocates into the nucleus to induce transcription of IFN-stimulated genes through binding of IFN-stimulated response elements [21]. Additionally IFNAR signaling can trigger non-canonical pathways such as activation of γ-activated sequences through homodimerization of STATs (STAT1, STAT3, STAT4, STAT5, STAT6), phosphoinositide-3-kinase/mammalian target of rapamycin pathway, and mitogen-activated protein kinase pathway [22].
IFNs may mediate an array of host protective functions including restricting viral replication [23], activation of NK cell cytotoxicity, maturation of APCs, clonal expansion and survival of antigen-specific CD4 and CD8 T cells during viral infection, promotion of B cell responses, and induction of apoptosis [24–30]. Type I IFNs have proven to be clinically useful in the treatment of chronic viral infections and certain types of leukemias [31]. Detrimental effects of type I IFNs have also been extensively documented during viral infections as well as during bacterial, fungal and parasitic infections [32]. One of the best examples of the complex regulation of antiviral immunity by type I IFNs is lymphocytic choriomeningitis virus infection (LCMV). Blockade of IFN signaling in acute infection with LCMV Armstrong infection results in abrogation of CD8+ T cell responses and defective control of infection [33]. In contrast, blockade of IFN signaling during persistent LCMV Clone (Cl)-13 infection diminished immunosuppressive signals and decreased levels of IL-10 and PD-L1 expressing immunoregulatory DCs. Virus titers in both serum and kidneys were also reduced. The cell type (s) mediating the immunosuppressive effects of IFN have not been defined [33, 34].
Studies on the effects of type I IFNs on Treg function yielded conflicting results [35, 36] and have not used experimental systems to examine the direct effects of IFNs on Treg cell homeostasis and functions. Recently, Srivastava et al (2014) demonstrated that mice infected with LCMV Armstrong manifested a decrease in the absolute numbers of splenic Tregs between days 4 and 7 post infection and that this reduction correlated with the expansion of both CD4+ and CD8+ T effector cells which peak on day 7 post infection. Furthermore, they also demonstrated a selective depletion of wild type (WT) Tregs on day 7 post infection of mixed bone marrow chimeras between WT mice and mice with a global deletion of IFNAR (IFNAR-/-). This latter result is difficult to interpret as our recent studies [37] have shown that IFNAR-/- Tregs in such chimeric (IFNAR-/- x WT) mice are at a competitive disadvantage as are IFNAR-/- Tregs in heterozygous female IFNARfl/fl x Foxp3Cre/WT mice.
In this study, we used IFNARfl/fl x Foxp3YFP-Cre mice to determine the role of IFNAR signaling specifically in Tregs during acute and chronic LCMV infection as well as in models of colon adenocarcinoma and melanoma. We demonstrate that IFNAR signaling in Tregs during the course of both acute and chronic viral infection results in a decrease in their activation status and a decrease in their suppressive function in vivo. The hypersuppressive state of Tregs in the absence of IFNAR signaling results in decreased CD8+ effector T cells, enhanced T effector cell exhaustion, defective generation of antiviral memory CD8+ T cells, and enhanced LCMV persistence. Similarly, in the tumor models, enhanced tumor growth and failure to efficiently generate antitumor T effector cells were observed in the absence of IFNAR signaling in Tregs. The enhanced suppressor function in the absence of IFNAR signaling in Tregs was accompanied by the induction of a gene expression pattern which was similar in the acute and chronic infection models and may be responsible for the heightened suppressor function.
Studies performed by Srivastava et al. (2014), provided preliminary evidence that IFNAR signaling inhibits the function of Tregs during an acute LCMV infection and that the absence of this inhibitory effect resulted in enhanced Treg function and impaired antiviral effector T cell function. Because this study did not definitively prove that the target of the suppressive function of IFNAR signaling was the Foxp3+ Treg, we generated Treg-lineage specific IFNAR deficient mice by crossing IFNARfl/fl mice with Foxp3YFP-Cre mice. We confirmed that the IFNAR was specifically deleted in CD4+Foxp3+ Tregs and not in CD4+Foxp3- T cells, CD8+ T cells or B220+ B lymphocytes (S1A Fig). First, we compared the clearance of LCMV Armstrong from the sera of IFNARfl/fl, IFNARfl/fl x Foxp3YFP-Cre, and IFNAR-/- mice at early time points post-infection. We observed that IFNARfl/fl x Foxp3YFP-Cre mice showed significantly higher viral titers (D3, D7 and D10) than IFNARfl/fl mice, and as expected IFNAR-/- mice had the highest titers among the three groups [33, 38] (Fig 1A). However, IFNARfl/fl x Foxp3YFP-Cre mice cleared LCMV Armstrong on day 14 post-infection.
This result is consistent with the studies of Srivastava et al. (2014) suggesting that the responses of the antiviral effector T cells early during LCMV Armstrong infection are compromised. Indeed, while both the frequencies and absolute numbers of CD8+CD44+ T cells specific for GP33-41 and NP396-404 did not differ between IFNARfl/fl and IFNARfl/fl x Foxp3YFP-Cre mice on day 14 post-infection (Fig 1B and 1C), the production of the effector cytokines IFN-γ and TNF-α was markedly diminished (Fig 1E and 1F). While the frequency and absolute number of CD8+CD44+ T cells specific for GP276-286 were increased, the frequency of IFN-γ producing cells recognizing GP276-286 was still reduced (Fig 1D and 1G). We did observe a modest decrease in the frequency, but not absolute numbers of CD4+Foxp3-CD44+ T cells specific for LCMV GP66-76 (Fig 1H) and this was accompanied by a marked decrease in the production of both IFN-γ and TNF-α by the GP66-76 specific cells (Fig 1I).
Taken together, these studies are consistent with the possibility that the absence of signaling via the IFNAR in Tregs during LCMV Armstrong infection potentiated their suppressive activity resulting in a failure to fully activate LCMV-specific T effector cells [39]. We observed that Treg cell numbers from infected WT and IFNARfl/fl x Foxp3YFP-Cre mice on day 4 and day 5 are similar and that Treg cells from both the WT and IFNARfl/fl x Foxp3YFP-Cre mice decrease similarly on day 7 post-infection (S1B Fig). While Srivastava et al (2014) reported a marked decrease in WT Foxp3+ Tregs on day 7 post infection in mixed bone marrow chimeras, we did not observe a decrease in Tregs on day 5 and the reduction in Treg frequencies and absolute numbers on day 7 was seen in both IFNARfl/fl and IFNARfl/fl x Foxp3YFP-Cre mice (S1B Fig). In contrast, Foxp3+ Tregs frequencies and numbers were increased on day 14 post LCMV Armstrong infection in IFNARfl/fl x Foxp3YFP-Cre mice. Most notably, the percentages and absolute numbers of activated Tregs were increased in IFNARfl/fl x Foxp3YFP-Cre mice compared to IFNARfl/fl mice on day 5, 7 and 14 post Armstrong infection (S1C Fig). In addition to elevated levels of CD44, Tregs in IFNARflf/fl x Foxp3YFP-Cre mice also expressed higher percentages of other activation markers, including Ki-67+, ICOS+ and TIGIT+ (S1D and S1E Fig), consistent with an activated phenotype and greater degree of proliferation at day 5 post Armstrong infection. We did not see any differences in the frequencies of activated CD4+Foxp3- and CD8+ T cells (S1F Fig) in IFNARfl/fl x Foxp3YFP-Cre mice.
The role of Tregs in the maintenance of chronic viral infection and effector T cell exhaustion has been difficult to define as it has been technically challenging to specifically deplete Tregs without the induction of autoimmune disease [40–42]. To evaluate the role of IFNAR signaling in Tregs during persistent chronic viral infection, mice were infected with LCMV Cl-13. We initially examined viral titers by plaque assay in serum at different time points during infection. On days 8, 25, 35 and 43-post infection, IFNARfl/fl x Foxp3YFP-Cre mice had significantly higher viral titers compared to IFNARfl/fl mice, and as expected IFNAR-/- mice had significantly higher titers than IFNARfl/fl mice (Fig 2A). Notably, both the lungs and kidneys of IFNARfl/fl x Foxp3YFP-Cre mice on day 46 post infection had significantly higher viral titers than IFNARfl/fl controls (Fig 2B). These data indicate that IFNAR deficiency specifically in Tregs enhances LCMV persistence.
The persistence of Cl-13 infection among IFNARfl/fl x Foxp3YFP-Cre mice led us to examine the kinetics and activation of Tregs during chronic LCMV infection. The frequencies and absolute numbers of Tregs were higher in IFNARfl/fl x Foxp3YFP-Cre mice on day 25-post infection, but not on day 46-post infection (Fig 2C). However, the activation state of the Tregs as measured by CD44 expression was higher on both days 25 and 46 in IFNARfl/fl x Foxp3YFP-Cre mice compared to IFNARfl/fl mice (Fig 2D). In contrast, no significant differences were observed in the frequencies or absolute numbers of CD4+Foxp3- or CD8+ T cells and their levels of CD44 expression on day 25 post-infection; however, on day 46 post-infection, the percentages of CD4+Foxp3-CD44hi T cells were higher in IFNARfl/fl x Foxp3YFP-Cre mice (S2A–S2D Fig). Furthermore, Cl-13 infected IFNARfl/fl x Foxp3YFP-Cre mice exhibited greater morbidity as manifest by a greater reduction in body weight than IFNARfl/fl mice (S2E Fig). Taken together, these results demonstrate similar to acute infection, failure of signaling via the IFNAR in Tregs results in enhanced Treg activation accompanied by decreased viral clearance.
To determine if the enhanced Treg activation and diminished viral clearance in Cl-13 infected IFNARfl/fl x Foxp3YFP-Cre mice results in decreased antiviral T effector cell responses, we examined the LCMV-specific responses of effector T cells. On day 25-post infection, both the absolute numbers of GP33 and NP396 tetramer positive T cells were similar (Fig 3A and 3B), while the frequencies of IFN-γ and TNF-α producing cells were lower (Fig 3C and 3D). However, on day 46-post infection, the absolute numbers of both antigen-specific CD8+ T cells were significantly decreased (Fig 3A and 3B) and this was accompanied by a marked decrease in the frequencies and absolute numbers of IFN-γ and TNF-α producing cells (Fig 3C and 3D). A moderate increase in the absolute number of CD4+Foxp3-CD44+GP66 Tet+ T cells frequencies was observed in the IFNARfl/fl x Foxp3YFP-Cre mice on day 46-post infection, but not on day-25 post infection (S3A Fig). However, only low levels of IFN-γ and TNF-α were produced and no differences were observed in cytokine production by CD4+Foxp3-CD44+GP66 Tet+ T cells among IFNARfl/fl and IFNARfl/fl x Foxp3YFP-Cre mice on both day 25 and 46-post infection (S3B Fig). Thus, decreased IFNAR signaling in Tregs resulted in reduced number and frequencies of CD8+ virus specific IFN-γ and TNF-α producing cells.
High levels of PD-1 expression are one of the hallmarks of T cell exhaustion. On both days 25- and 46-post infection, the frequency of PD1 expressing CD8+ and CD4+Foxp3- T cells were higher in IFNARfl/fl x Foxp3YFP-Cre mice (Fig 4A and 4B). Two other markers of T cell exhaustion, the transcription factor eomesodermin (EOMES), and CD39 can be co-expressed with PD-1 on exhausted T cells [43, 44]. Higher percentages and absolute numbers of EOMES+PD-1+ cells within the CD8+ and CD4+Foxp3- populations were present in the IFNARfl/fl x Foxp3YFP-Cre mice (Fig 4C and 4D). Correspondingly, PD1+CD39+ frequencies and total numbers were higher in gated CD8+ T cells from IFNARfl/fl x Foxp3YFP-Cre mice (Fig 4E). While the frequencies of gated CD4+Foxp3-PD1+CD39+ T cells did not differ between IFNARfl/fl and IFNARfll/fl x Foxp3YFP-Cre mice, the absolute numbers of CD4+Foxp3-PD1+CD39+ T cells were higher in Treg-specific IFNAR deficient mice (Fig 4F). Similarly, the frequencies of PD-1+ T cells were greater among gated CD8+CD44+GP33 and CD8+CD44+GP276 Tet+ T cells in IFNARfl/fl x Foxp3YFP-Cre mice (Fig 4G and 4H). CD8+CD44+NP396 Tet+ and CD4+Foxp3-CD44+GP66 Tet+ populations from day 46 Cl-13 infected IFNARfl/fl x Foxp3YFP-Cre mice also had higher proportions of PD1 expressing cells (S4A and S4B Fig). In addition, it has been demonstrated that Tregs with higher levels of PD1 expression can mediate enhanced suppression in LCMV infection [45], we also found that Tregs from Cl-13 infected IFNARfl/fl x Foxp3YFP-Cre mice had significantly higher expression of PD1 than controls on days 25 and 35 post infection (S4C Fig). These data demonstrate that the enhanced Treg suppression seen in the absence of IFNAR signaling during Cl-13 infection results in markedly reduced cytokine production by virus-specific CD8+ T cells as well as a phenotype consistent with exhaustion.
Exhausted CD8+ T cells have reduced memory cell potential which is secondary to higher LCMV antigen persistence in infected mice [46]. To determine whether the enhanced T cell exhaustion phenotype observed in IFNARfl/fl x Foxp3YFP-Cre mice is associated with a reduction in the formation of virus-specific memory T cells, we examined the levels of expression of three memory cell makers (CD62L, CD127, CXCR3) on gated CD8+CD44+ GP276 Tet+ T cells (Fig 5A). The frequencies and the absolute numbers of all three memory populations were reduced in IFNARfl/fl x Foxp3YFP-Cre mice on day 46 post infection compared to IFNARfl/fl control mice (Fig 5B–5D). Similar results were seen within the CD8+CD44+GP33 Tet+ population (S5A–S5D Fig). While the number of memory CD8+ T cells is usually inversely correlated with terminally differentiated T cells as measured by KLRG-1 expression [47], the frequencies and absolute numbers of CD8+CD44+GP276/GP33 Tet+ KLRG-1+ T cells were also lower in IFNARfl/fl x Foxp3YFP-Cre mice (Fig 5E and S5E Fig). We further examined the protective capacity of memory CD8+ T cells by re-infecting the day 30 Armstrong infected mice with Cl-13 virus. On day 5 post Cl-13 infection, IFNARfl/fl x Foxp3YFP-Cre mice had reduced frequencies and numbers of CD8+CD44+NP396 Tet+ cells, and in addition GP33- and GP276-stimulated CD8+CD44+ T cells from IFNARfl/fl x Foxp3YFP-Cre infected mice produced significantly less Granzyme B (GrB) positive and GrB/IFN-γ double positive cells compared to CD8+ T cells from control mice (Fig 5F and 5G), however IFN-γ positive cells are tended to be more in infected IFNARfl/fl x Foxp3YFP-Cre mice but they are not significant compared to control mice. Collectively, these results demonstrate that the higher load of virus is associated with a defect in the generation of virus-specific memory T cells in the absence of IFNAR signaling in Tregs.
In order to better understand the molecular basis for the enhanced activation and suppressive function of Tregs from Treg-specific IFNAR-deficient mice during acute and chronic LCMV infection, we performed high-throughput RNA sequencing on sorted CD4+YFP+ Tregs isolated from day 5 Armstrong-infected Foxp3YFP-Cre and IFNARfl/fl x Foxp3YFP-Cre mice. Principal component analysis (PCA) showed distinct clustering of Tregs from Foxp3YFP-Cre mice relative to Tregs from IFNARfl/fl x Foxp3YFP-Cre mice (Fig 6A). A total of 586 genes were significantly differentially expressed (249 genes were down, and 337 genes were up) in IFNARfl/fl x Foxp3YFP-Cre mice (fold change 1.5 and above, adjusted P < 0.05) (Fig 6B). Among the 586 genes, 174 genes were identified in the interferome database [48] (interferome.its.monash.edu.au) as IFN-signaling related (fold change 1.5 and above, adjusted P < 0.05) (S6A Fig), and were excluded from further analysis. We elected to exclude IFNAR regulated genes in order to perform an unbiased downstream analysis, as IFNAR signaling regulates the transcription of up to 2000 genes. The remaining 412 differentially expressed genes were exclusively non-IFN related. Of these, 156 genes were upregulated in infected Foxp3YFP-Cre mice, and 256 genes were upregulated in infected IFNARfl/fl x Foxp3YFP-Cre mice. Gene set enrichment analysis (GSEA) of the 412 non-IFN related genes revealed that the natural Treg vs conventional T cell gene set was enriched to a greater extent in IFNARfl/fl x Foxp3YFP-Cre mice [36 genes out of 42 were in core enrichment, Enrichment score (ES): 0.566, P < 0.01, FDR:0.0] (Fig 6C). We observed that 32 out of 412 non-IFN related genes were differentially expressed (fold change 1.5 and above, adjusted P < 0.05, genes normalized by z-score; 24 genes in IFNARfl/fl x Foxp3YFP-Cre mice and 8 genes in Foxp3YFP-Cre mice were upregulated) and could be classified as Treg-signature genes as previously reported [49, 50] (Fig 6D). Representative upregulated genes in IFNARfl/fl x Foxp3YFP-Cre mice include Areg, Arhpag20, Bub1b, Ccl12, Ccr5, Il1r1, Mki67 (Ki67), Ncf1, Nrp2, Tnfrsf9 (CD137), Tcf19, Uhrf1, and Wnt3; representative downregulated genes in IFNARfl/fl x Foxp3YFP-Cre include Cybb, Dapl1, Fam160a1, Il1r2, and Tnfsf8 (CD153). Some of the above upregulated genes from Tregs include Areg, Ccl12, Mki67, Ncf1, Tnfrsf9, Uhrf1 are well characterized to modulate the enhanced Treg suppressive and proliferative function [49, 50]. Further, we also performed ingenuity pathway analysis (IPA) for non-IFN related genes which resulted in 45-top canonical pathways (adjusted p value < 0.1) (S6B Fig). Specifically, cell cycle: G2/M DNA damage checkpoint regulation pathway (genes involved: Aurka, Ccnb2, Cdk1, and Top2a), cyclins and cell cycle regulation pathway (genes involved: Ccnb2, Ccne2, Cdk1, and E2f1) are enriched positively in IFNARfl/fl x Foxp3YFP-Cre Tregs compared to Tregs from Foxp3YFP-Cre mice, suggesting that cell cycle genes are more functional in Treg-specific IFNAR deficient mice. Furthermore, the c-AMP mediated signaling pathway (genes involved: Akap1, Camk2b, Chrm4, Crem, Fpr1, Prkar1b, and Ptger3) is also enriched positively in IFNAR deficient Tregs. Through IPA, we analyzed non-IFN related genes for top networks based on co-expression, transcription factor binding site predictions and protein-protein interactions. The top two networks identified included cell cycle, DNA replication, recombination, repair, cancer; and cellular movement, hematological system development and function and immune cell trafficking (S6C Fig), differential expression of these associated genes in the pathways are shown (S1 and S2 Tables). Some of the associated genes in the networks include, transcription factors: Depdc1, E2f1, E2f8, Foxm1, Mybl2, and Uhrf1 are downregulated, and pydc4/Ifi16 (interferon gamma inducible protein 16) is upregulated in Tregs from Foxp3YFP-Cre mice; cell cycle kinases: Ccnb2, Aurkb, and Chek1 are also downregulated in Foxp3YFP-Cre mice Tregs; immune cell genes such Tnfrsf9 (CD137), Ccl2, Ccr5, and Il1r1 are downregulated, in contrast C5ar1, CD19, Il1r2, Itga2b (CD41), Ly6c1, and Tlr7 are upregulated in Foxp3YFP-Cre mice Tregs. We also tested whether the reduced cell cycle gene signature in Tregs from Foxp3YFP-Cre mice contributed to a greater degree of apoptosis, but Active caspase-3 staining showed no significant increase in the staining of Tregs from day 5 Armstrong infected Foxp3YFP-Cre mice compared to Tregs from IFNARfl/fl x Foxp3YFP-Cre mice (S6D Fig).
In parallel, we also performed RNA sequencing on Tregs from day 25 post LCMV Cl-13 infection. PCA showed less distinct clustering (Fig 6E), and surprisingly, only 36 genes were significantly differentially expressed (23 genes were downregulated, and 13 genes were upregulated in IFNARfl/fl x Foxp3YFP-Cre mice, fold change 1.5 and above, adjusted P < 0.05) in Tregs from Foxp3YFP-Cre and IFNARfl/fl x Foxp3YFP-Cre mice (Fig 6F). Among those differentially expressed genes, 14 genes were identified in the interferome database [48] as IFN-signaling related (fold change 1.5 and above, adjusted P < 0.05) (S7A Fig). Of the remaining 22 genes, 11 genes were upregulated in their expression in Tregs from Foxp3YFP-Cre mice and IFNARfl/fl x Foxp3YFP-Cre mice, respectively (Fig 6G). Additionally, IPA for non-IFN related genes resulted in 18-top canonical pathways (adjusted p value < 0.1) (S7B Fig), importantly, c-AMP mediated signaling pathway (genes involved: Akap1, and Camk2b) is enriched positively in Cl-13 infected IFNARfl/fl x Foxp3YFP-Cre mice Tregs, this was shown similar pattern in Armstrong infected IFNARfl/fl x Foxp3YFP-Cre mice. One of the top networks for non-IFN related genes include lipid metabolism, molecular transport and small molecule biochemistry (S7C Fig). Few of the associated genes in the network include, kinases: Camk2b, and Hunk and cytoplasmic enzymes: Cpta1, and Scpep1 are down regulated, while proapoptotic factor Erdr1 is upregulated in Foxp3YFP-Cre mice Tregs.
We identified fourteen genes (fold change 1.5 and above, adjusted P < 0.05, normalized by z-score) differentially expressed in Tregs from LCMV Armstrong infected mice which were similarly differentially expressed in Tregs from LCMV Cl-13 infected mice including 7 that were up- and 7 down-regulated (Fig 6G and 6H). We further validated the differential expression of some of these non-IFN related genes during chronic infection which are common to both infection model or unique to chronic infection alone by quantitative real-time PCR (S7D Fig). Several of the upregulated genes include a-kinase anchoring protein 1 (Akap1), calcium/calmodulin-dependent protein kinase II b (Camk2b), Hormonally upregulated Neu-associated kinase (Hunk), Rab4a and Rasgrf2. Akap1 is associated with cAMP signaling, and it can act as a gap junction protein in facilitating the transfer of cAMP from Treg to effector T cells leading to inhibition of T cell receptor (TCR) signaling [51, 52]. Furthermore, CamK2b, Hunk, Rab4a and Rasgrf2 have also been implicated in enhancement of Treg function [53–57]. Taken together, all these upregulated genes participate in heightened Treg suppressive function observed in the Tregs from IFNARfl/fl x Foxp3YFP-Cre mice during both acute and chronic LCMV infections.
Enhanced Treg cell function is very well documented in various tumor models and it has been associated with a poor prognosis [6]. To determine whether the absence of IFNAR signaling was associated with enhanced Treg suppression in a non-infectious setting, we utilized the mouse colon adenocarcinoma MC38 and mouse B16.F10 melanoma models. IFNARfl/f x Foxp3YFP-Cre mice showed higher tumor incidence (MC38: n = 11/11, 100%; B16.F10: n = 5/5, 100%) and significantly increased volume than IFNARfl/fl mice (MC38: n = 8/10 mice, 80%, B16.F10: n = 4/5, 80%) (Fig 7A, left and right panels). Tregs and CD4+Foxp3- cells isolated from tumor infiltrating lymphocytes (TIL) showed comparable frequencies in both strains of mice, however CD8+ T cell frequencies from TIL of IFNARfl/fl x Foxp3YFP-Cre mice were increased. Nevertheless, both CD4+Foxp3- and CD8+ T cells from IFNARfl/fl x Foxp3YFP-Cre mice tended to proliferate less as assayed by Ki-67 expression (S8A–S8C Fig). Importantly, Tregs from IFNARfl/fl x Foxp3YFP-Cre mice TIL expressed significantly higher levels of CD44, enhanced proliferation and expression of PD-1 (Fig 7B). Conversely, both CD4+Foxp3- and CD8+ TIL from IFNARfl/fl x Foxp3YFP-Cre mice expressed lower levels of CD44 and markedly reduced levels of IFN-γ and TNF-α production compared to TIL from IFNARfl/fl mice (Fig 7C: gated on CD4+Foxp3- TILs and Fig 7D: gated on CD8+ TILs). These data strongly suggest that Tregs in TIL from IFNARfl/fl x Foxp3YFP-Cre mice have enhanced suppressor activity in the tumor microenvironment and the phenotype of these Tregs within TIL closely resembles the activated hypersuppressive phenotype observed during acute and chronic LCMV infections.
Tregs mediate a multifaceted role in modulating the immune response to acute and chronic infectious agents. While their beneficial effects in decreasing immune pathology during the resolution phase of many infections is clear, Tregs can also mediate immune suppression resulting in pathogen persistence. During viral infections, rapid activation of the innate immune system generates inflammatory signals that can initially control the infection and ultimately influence the quality and magnitude of the adaptive antiviral effector T cell response. The best characterized innate inflammatory signals are the type I IFNs. During LCMV infection type I IFNs are produced in large quantities immediately following viral infection by plasmacytoid DCs as well as virus infected cells and primarily exert their effect on CD8+ T cells by extending their survival. In this report, we have demonstrated that type I IFNs can also exert beneficial effects by acting on Tregs to down-modulate their suppressive functions both early during the course of acute LCMV Armstrong infection and also later during virus persistence in chronic Cl-13 infection.
Srivastava et al. (2014), have previously examined the effects of type I IFN on Tregs during the course of acute LCMV infection. They concluded that type I IFNs down-modulated Treg function but postulated that the effects of type I IFNs were secondary to a selective decrease in the number of highly suppressive effector Tregs, and secondary to the pro-apoptotic and anti-proliferative actions of type I IFNs early in the course of infection. The major difficulty in the interpretation of this study is that the experimental model they used did not allow them to selectively examine the effects of IFNs on Tregs in the absence of its effects on other cell types. Similarly, one previous study also showed that Tregs were reduced in infected wild type compared to control mice during first week post LCMV-Docile infection, but Tregs expanded to a greater extent from the second weeks onwards. Treg expansion was more pronounced in IL-21R deficient mice, suggesting IL-21 signaling restricts proliferation of Tregs during LCMV infection [58]. The availability of mice with a conditional deletion of the IFNAR in Tregs allowed us to dissect the mechanistic basis of IFNAR signaling in Tregs resulting in their reduced suppressive function and in more efficient antiviral and antitumor immune responses.
Similar to the previous study [39], we found that the generation of antigen-specific CD8+ and CD4+ T cells was comparable in Treg-specific IFNAR-deficient mice and WT controls during acute LCMV infection, but that both the virus-specific CD8+ and CD4+ T cells in IFNARfl/fl x Foxp3YFP-Cre mice produced markedly reduced amounts of IFN-γ and TNF-α accompanied by a slower rate of viral clearance than the controls. Most importantly, we did not observe a decrease in the percentages or absolute numbers of Tregs in the WT control mice that differed from IFNARfl/fl x Foxp3YFP-Cre mice on day 4 or 5 post-infection. We did detect a decrease in Tregs in the IFNARfl/fl X Foxp3YFP-Cre mice on day 7 post-infection, but we observed an identical decrease in Tregs in the control IFNARfl/fl mice. In contrast to the loss of memory/effector Tregs observed in WT mice by Srivastava et al. (2014), we observed an enhanced percentage of memory/effector Treg as defined by CD44 expression on days 5, 7 and 14 post-infection in IFNARfl/fl x Foxp3YFP-Cre mice, as well as higher levels of expression of Ki-67, ICOS and TIGIT. Taken together, our results are most consistent with an enhanced suppressive phenotype of Tregs in the absence of IFNAR signaling in acute virus infection albeit the mice ultimately cleared the infection.
We observed a similar but more profound suppressive phenotype in IFNARfl/fl x Foxp3YFP-Cre mice following Cl-13 infection, as the mice had higher serum titers of virus and also had higher viral titers in lungs and kidneys for as long as 46 days post infection. We observed decreased numbers of antiviral antigen-specific CD8+ T cells accompanied by a profound decrease in effector cytokine production. Increased viral persistence resulted in marked expression of markers associated with T cell exhaustion (PD-1, CD39, EOMES) [43, 44, 46] and decrease in generation of antigen-specific memory CD8+ T cells. The decrease in the formation of memory T cells in IFNARfl/fl x Foxp3YFP-Cre mice in Cl-13 infection should be contrasted to the effects of IL-10 producing Treg cells in augmenting memory T cell formation following Armstrong infection [47]. Slight increases in both the percentages, but not the absolute numbers of Tregs were seen on days 25 and 46 post infection. However, at both times points, we observed a significant increase in the percentages of activated/effector Tregs. To determine if the hyperactivated/hypersuppressive phenotype observed in the absence of IFNAR signaling in Tregs was unique to viral infections, we also examined the responses of IFNARfl/fl x Foxp3YFP-Cre mice to transplantable tumor models. Markedly enhanced growth of the tumor was observed in these mice accompanied by an enhanced percentage of activated PD-1+ tumor infiltrating Tregs. In addition, the activation and cytokine production by both CD4+Foxp3- and CD8+ tumor infiltrating T cells were markedly suppressed. Taken together, these studies demonstrate that IFNAR signaling in Tregs plays a critical role in down-modulating, but certainly not abolishing, their suppressive function and may in viral infections orchestrate the balance between immunopathology and eradication of the virus.
To begin to elucidate the mechanistic basis for the suppressive function of IFNAR-deficient Tregs during both acute and Cl-13 infection, we performed high-throughput RNA sequencing of Foxp3+ Tregs from both controls and IFNARfl/fl x Foxp3YFP-Cre mice on day 5 Armstrong and day 25 Cl-13 infection. A group of Treg-signature genes (Areg, Arhpag20, Bub1b, Ccl12, Ccr5, Il1r1, Mki67 (Ki67) Ncf1, Nrp2, Tnfrsf9, Tcf19, Uhrf1, and Wnt3) were expressed at higher levels in IFNARfl/fl x Foxp3YFP-Cre mice than WT controls on day 5 Armstrong infection. These genes have previously been identified as Treg-Up signature genes [49, 50] and their enhanced expression is consistent with the hyperactivated phenotype of the Tregs at that time point. We did not observe upregulation of this group of genes on day 25 of Cl-13 infection. Interestingly, both in Armstrong and Cl-13 infection, we observed the differential expression of a number of genes in IFNARfl/fl x Foxp3YFP-Cre Tregs which might also play a role in their enhanced suppressive function including Akap1, Camk2b, Rasgrf2 and Hunk. Akap1, which serves as a gap junction protein, is involved in transferring pools of cAMP from Treg to effector T cells resulting in inhibition of TCR signaling [51, 52]. Camk2b participates in the activation of nuclear factor kappa-B, which in turn plays a role in Treg development by stabilizing Foxp3 [53, 54]. Both Camk2b and Hunk have been shown to be have higher levels of expression in human Tregs than conventional effector T cells [55]. Lastly, Rasgrf2 is involved in stimulation of TCR signaling through activation of nuclear factor for activated T cells (NF-AT) [57]. Conversely, we also observed group of genes (Erdr1, Rell1, Tlr7) whose expression is downregulated in Tregs from both IFNARfl/fl x Foxp3YFP-Cre Armstrong and Cl-13 infected mice. These genes have been described as potentially playing a role in apoptosis [59–61] and thus may be involved in the reduced suppressive function of Tregs after IFNAR signaling. We did not observe any differences between IFNARfl/fl x Foxp3YFP-Cre and WT mouse Tregs in expression of Active caspase-3 on day 5 post Armstrong infection again consistent with our data that cell death is not playing a role in reduced Treg suppression in WT mice, although, we cannot exclude the involvement of other cell death pathways. Future studies involving over expression and/or deletion of these genes in Tregs will be needed to specifically implicate one or several of these genes in Treg-mediated suppression during viral infections. We have not yet performed similar gene expression studies in IFNAR sufficient and deficient Tregs derived from the tumor microenvironment and compared such data with Tregs from LCMV-infected mice.
Our study demonstrates that one of the multiple cellular targets of Type I IFNs during viral infection are Treg cells and that the functional result of this interaction is a downregulation of Treg suppressor function. A similar process may take place in the tumor microenvironment and may be responsible for some of the antitumor effects of this group of cytokines [62, 63]. Thus, type I IFNs should be added to the long list of cytokines (IL-1β, IL-4, IL-6, IL-15, IL-21) [64–72] and members of the tumor necrosis factor superfamily (TNFSF) (GITR-L, 4-1BB-L, OX40-L and TNF-α) [73–76] that are purported to decrease Treg suppressive function in autoimmune and infectious disease models. However, the abrogation of suppression is more frequently mediated by the action of the cytokine or TNFSF member on the responder T cells resulting in resistance to suppression [73, 77], whereas we have definitively demonstrated that Tregs are the targets cells in this model. While type I IFNs may attenuate Treg suppression, it remains clear that the major and undefined cellular target for the immunosuppressive effects of type I IFNs in chronic LCMV infection is not the Tregs, as IFNARfl/fl x Foxp3YFP-Cre mice have elevated viral titers, while LCMV Cl-13 infected mice treated with a neutralizing antibody against IFNAR have decreased viral titers [33, 34]. Thus, an approach to target IFNAR in normal hosts for inhibition of Treg suppressive function in chronic infection or in cancer would be difficult.
This study was carried out in strict accordance with the recommendations for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by National Institute of Allergy and Infectious Diseases Animal Care and Use Committee (Protocol No: LI-5E).
Foxp3YFP-Cre (expressing Cre recombinase regulated by Foxp3 promoter) mice were purchased from Jackson Laboratories (Bar Harbor, ME). IFNAR-/- mice were obtained by National Institute of Allergy and Infectious Diseases (NIAID), and maintained in Taconic Farms (Germantown, NY). IFNARfl/fl mice were generously provided by Ulrich Kalinke (Paul-Ehrlich Institut, Langen, Germany), and crossed with Foxp3YFP-Cre mice to generate Treg-lineage specific IFNAR-deficient mice. All strains of mice used in this study were age 8–12 weeks of age), gender matched, and bred in-house.
LCMV Armstrong and Cl-13 viruses (Shevach Laboratory) were propagated in baby hamster kidney-21 fibroblast cells [American Type Culture Collection (ATCC), Manassas, VA]. Viral titers were determined by plaque assay using Vero African-green-monkey kidney cells (ATCC). Viral stocks were frozen at -80 oC until use. Mice were infected with the diluted virus in 1x sterile phosphate buffer saline (PBS) (Armstrong virus, 2x105 plaque forming unit (pfu)/mouse, i.p., or Cl-13 virus, 2x106 pfu/mouse, i.v.). LCMV titers in sera and organs were determined by plaque assay using Vero cells as described [78].
Spleens were harvested from naive (uninfected) and infected mice on indicated days and homogenized the tissues using cell strainer (70 μm, Nest Scientific USA, Rahway, NJ). Red blood cells were lysed using sterile ACK lysing buffer [NH4Cl (0.15 M), KHCO3 (10 mM) and Na2EDTA (0.1 mM), pH 7.3]. Lymphocytes were washed, suspended in sterile complete medium [RPMI medium supplemented with 10% heat-inactivated fetal bovine serum (FBS), L-glutamine (2 mM), sodium pyruvate (1 mM), HEPES (1 mM), non-essential amino acids (0.1 mM), 2-mercaptoethanol (50 μM), and penicillin and streptomycin (100 U/ml)], and total live cells were counted.
Cell surface staining was performed as described [79]. Briefly after harvest, spleen cells (3x106 cells) or tumor infiltrating lymphocytes (TIL) were suspended in sterile complete medium. For surface staining, cells in staining buffer (PBS, 10% heat-inactivated FBS, and 0.05% sodium azide) were incubated for 30 min at 4 oC, then stained with the following surface murine conjugate antibodies: anti-CD4, anti-CD62L, anti-CD39, anti-CD127, anti-IFNAR1, anti-B220 (all are from eBioscience, San Diego, CA); anti-CD8a, anti-CD44, anti-TIGIT (all are from BD Biosciences, San Jose, CA); anti-PD1, anti-KLRG1, anti-CXCR3, and anti-ICOS (all are from BioLegend, San Diego, CA) and live/dead fixable aqua dead cell stain kit (Life Technologies, Carlsbad, CA). For intracellular Foxp3, Ki67, Eomes and YFP detection, fixation and permeabilization were done according to the manufacturer’s guidelines (Foxp3 transcription factor buffer set, eBioscience) and cells stained with anti-Foxp3, anti-Ki67, anti-Eomes (eBioscience), and anti-GFP rabbit polyclonal antibody (Life Technologies, Carlsbad, CA). For intracellular cytokine detection, spleen cells (3x106) in complete medium were stimulated with LCMV peptides (Research Technologies Brach, Protein Chemistry, NIAID): GP33-41 (GP33, 1 μg/ml), GP 276–286 (GP276, 1 μg/ml), NP396-404 (NP396, 1 μg/ml) and GP 61–80 (GP61, 10 μg/ml) along with GolgiStop (2 mM/ml, BD Biosciences) for 5 hrs at 37 oC. TIL were stimulated with cell stimulation cocktail (eBioscience) containing PMA, ionomycin, Brefeldin A, and monensin for 5 hrs at 37 oC. Later cells were washed, fixed and permeabilized and stained with intracellular cytokine antibodies (anti-IFN-γ and anti-GrB, BD Biosciences, and anti-TNF-α, eBioscience) for overnight at 4 oC. For MHC class I tetramer staining, H-2Db GP33, H-2Db GP276 and H-2Db NP396 (NIH tetramer core facility) were used at 1:100 dilutions and staining was done at 4 oC for 1 hr, and for MHC class II tetramers, IAb GP66-77 (GP66) (NIH tetramer core facility) used at 1:75 dilution and staining performed for 90 mins at 37 oC. Cells were washed and acquired by BD LSRII and LSRFortessa (BD Biosciences) flow cytometers with FASCDiva software.
Foxp3YFP-Cre and IFNARfl/fl x Foxp3YFP-Cre mice (four to five mice per group) were infected with Armstrong and Cl-13 virus. On day 5 post Armstrong and day 25 post Cl-13 infection, spleen and lymph nodes were harvested and single cell suspension were prepared. T cells were isolated by labeling single suspension with CD90.2 microbeads (Miltenyi Biotec, San Diego, CA) and purified through LS columns (Miltenyi Biotec). Purified T cells were stained with anti-CD4 (RM 4–5) for 30 minutes on ice and CD4+YFP+ cells (5x105/sample) sorted (purity ~95%) by using FACSAria flow cytometer (BD Biosciences).
Armstrong (day 5) and chronic LCMV infected (day 25) Foxp3YFP-Cre and IFNARfl/flxFoxp3YFP-Cre mice CD4+YFP+ sorted cells were lysed in RLT buffer (Qiagen, Valencia, CA). Total RNA was extracted using Qiagen AllPrep 96 DNA/RNA kit as described by the manufacturer (Qiagen, Valencia, CA), with one exception prior to the extraction, the RLT lysate was homogenized using Qiagen QIAShredder columns (Qiagen) to shear any contaminating gDNA. Samples were then subjected to on-column Dnase I treatment. All steps were performed using PCR amplicon-free laboratory equipment to further minimize background signal during RNA sequencing and library generation. A 150 ng aliquot from each sample was individually adjusted to 50 μl using nuclease-free water. Each sample was processed using Truseq Stranded mRNA Sample Preparation, Rev. E (Illumina Inc., San Diego, CA) using the included barcodes with the following modification: post-amplification libraries were purified with Ampure XP beads twice. The resulting DNA libraries were fragment-sized using a DNA1000 Bioanalyzer Chip (Agilent Technologies, Santa Clara, CA) and quantitated using KAPA Library Quant Kit with universal qPCR Mix (Kapa Biosystems, Wilmington, MA) on a CFX96 Real-Time System (BioRad, Hercules, CA). All eight-ten samples were diluted to a 2 nM working stock and pooled using equal volume amounts. An 11 pM titration point was used to cluster a paired end, RAPID 2-lane flowcell on a Hiseq 2500 DNA sequencer (Illumina). Libraries were run as 2 x 100 bp paired end reads on 2 lanes of an Illumina Hiseq 2500 sequencer, which produced ~28.7 million reads per sample. Reads were trimmed for adapter sequence and filtered for low quality sequence using the FASTX-Toolkit. Remaining reads were mapped to the mouse genome assembly mm10 using Hisat2 [80]. Reads mapping to genes were counted using htseq-count [81]. Differential expression analysis was performed using the Bioconductor package DESeq2 [82]. Further analysis was performed using Partek Genomic Suite (Partek Incorporated) and Ingenuity pathway analysis (IPA) is used for obtaining top canonical pathways, networks based on co-expression, transcription factor binding sites and protein-protein interactions.
GSEA were performed on the set of 412 genes (Armstrong infection) using GSEA v2.2.3 from The Broad Institute [83]. GSEA was run using molecular signature database v.5.2 gene sets, except the C1: positional gene sets, with 1000 permutations and all default parameters except minimum size of 5.
Total RNA samples from Foxp3YFP-Cre and IFNARfl/fl x Foxp3YFP-Cre mice were extracted as described in the NGS gene expression profiling method section. cDNAs were prepared from Superscript IV first-strand cDNA synthesis kit (Thermo Fisher Scientific, Waltham, MA) according to the manufacturer’s instructions. Presynthesized Taqman gene expression assays (Thermo Fisher Scientific) were used to amplify Akap1, Car2, Cpe, Eomes, Erdr1, Gpat2, Rab4a, Rell1, Rasgrf2, Sdc3, Tlr7, and Actb was used as internal control. Real time qPCR was performed using QuantStudio7 Flex Real time PCR system (Thermo Fisher Scientific) using Taqman universal master mix II with UNG (Thermo Fisher Scientific). Target gene expressions were calculated by 2-dct and expressed as relative to Actb.
Murine colon adenocarcinoma cells, MC38 cells (ATCC) and melanoma cells, B16.F10 (ATCC) were grown in complete DMEM medium [DMEM/RPMI supplemented with 10% heat-inactivated FBS, L-glutamine (2 mM), sodium pyruvate (1 mM), HEPES (1 mM), non-essential amino acids (0.1 mM), 2-mercaptoethanol (50 μM), and penicillin and streptomycin (100 U/ml)]. Sex- and age-matched IFNARfl/fl and IFNARfl/flxFoxp3YFP-Cre mice were injected with 2x105 MC38 cells and or 1.25x105 B16.F10 cells diluted in sterile 1xPBS, subcutaneously (right flank region). Tumor growths were measured in regular intervals by digital calipers (Fisher Scientific), and tumor volumes were calculated by the formulas: length x width x depth. On day 18 post tumor implant, tumors were excised in sterile conditions, and TIL were prepared after the mincing the tumor, and digesting with 1x HBSS containing collagenase type IV (0.5mg/ml), Dnase I (0.1 mg/ml) and Hyaluronidase (2.5 units/ml) for 1 hr at 37 oC. Later, digested cells were washed and treated with ACK lysing buffer to lyse RBCs, and then TIL were purified by density gradient centrifugation using buffered percoll (Sigma-Aldrich, 80%/40%).
Flow cytometry data were analyzed using FlowJo software version 10.2 and or 10.3 (FlowJo LLC, Ashland, OR). Graphs were prepared by GraphPad Prism software version 7.0 (GraphPad Software, Inc. La Jolla, CA). Statistical analysis was done through unpaired two-tailed Student’s t-test. All data in the graphs presented as Mean±SEM values, and error bars represent SEM. Data were considered statistically significant when P < 0.05, and represented as * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001.
RNA sequence information reported in this study is deposited in NCBI GEO under the accession number: GSE104517.
|
10.1371/journal.pgen.1001050 | The Extinction Dynamics of Bacterial Pseudogenes | Pseudogenes are usually considered to be completely neutral sequences whose evolution is shaped by random mutations and chance events. It is possible, however, for disrupted genes to generate products that are deleterious due either to the energetic costs of their transcription and translation or to the formation of toxic proteins. We found that after their initial formation, the youngest pseudogenes in Salmonella genomes have a very high likelihood of being removed by deletional processes and are eliminated too rapidly to be governed by a strictly neutral model of stochastic loss. Those few highly degraded pseudogenes that have persisted in Salmonella genomes correspond to genes with low expression levels and low connectivity in gene networks, such that their inactivation and any initial deleterious effects associated with their inactivation are buffered. Although pseudogenes have long been considered the paradigm of neutral evolution, the distribution of pseudogenes among Salmonella strains indicates that removal of many of these apparently functionless regions is attributable to positive selection.
| Pseudogenes have traditionally been viewed as evolving in a strictly neutral manner. In bacteria, however, pseudogenes are deleted rapidly from genomes, suggesting that their presence is somehow deleterious. The distribution of pseudogenes among sequenced strains of Salmonella indicates that removal of many of these apparently functionless regions is attributable to their deleterious effects in cell fitness, suggesting that a sizeable fraction of pseudogenes are under selection.
| One of the most distinctive features of bacterial genomes is their high coding densities, in which genic regions typically constitute more than 80% of the total genome [1]. This is in sharp contrast to many eukaryotes whose genomes contain vast stretches of non-coding DNA and a multitude of transposable and repetitive elements, with protein-coding regions often accounting for only 1% of the genome [2]. The paucity of non-coding regions in bacterial genomes has lead to the idea that pseudogenes would be exceedingly rare [3]; however, recent large-scale analyses have found that virtually all bacterial genomes contain disrupted and eroded genes that have full-length counterparts in other related genomes [4]–[8]. Pseudogenes are particularly prevalent in those bacterial species that have recently become associated with or dependent upon eukaryotic hosts [9]–[10], and in the most extreme cases, pseudogenes can number in the 1,000 s and occupy over half of the genome [11]–[12].
The pseudogenes in bacterial genomes are continually created from ongoing mutational processes and are subject to degradation, and eventual removal, by the further accumulation of mutations. However, the most surprising aspect of bacterial pseudogenes is that their retention time appears to be extremely short. Even in comparison of very closely bacteria, there are very few pseudogenes that are shared among strains typed to the same bacterial species [6], [7], [13]. This observation indicates that bacterial pseudogenes, although often present in high numbers, are deleted at a relatively rapid rate. This feature is again in sharp contrast to eukaryotes, in which pseudogenes often persist over evolutionary timescales and may be shared by distantly related lineages, such as rodents and primates [14]–[17].
Due to the pervasive mutational bias towards deletions that has been observed across bacterial genomes [18]–[19], the rapidly removal of pseudogenes could be caused by the random fixation of background mutations. Because pseudogenes have long been viewed as “a paradigm of neutral evolution” [20], this is the favored hypothesis. Alternatively, pseudogenes could effect a cost and be eliminated from bacterial genomes by an adaptive process. For example, pseudogenes might be detrimental to the organism through energetic costs incurred by the continued transcription and translation of non-functional genes and/or through the production of proteins that are toxic to cells.
In this study, we examine the formation, loss and phylogenetic distribution of disrupted genes in Salmonella. We focus on this bacterial genus because: (1) high-quality genomic sequences of several Salmonella strains have been determined, (2) Salmonella genomes, like those of most other pathogens, possess considerable numbers of pseudogenes [21]–[22], (3) the population structure of Salmonella enterica is essentially clonal [23], allowing the resolution of an unambiguous strain phylogeny, and (4) both experimental [24] and comparative [25]–[27] studies provide evidence of a strong deletional bias in Salmonella, such that genes that are not maintained by selection are rapidly inactivated and eliminated by mutational events. And it is against this background that we test the possibility that the removal of bacterial pseudogenes is adaptive.
Despite the similarity in overall genome sizes, the number of pseudogenes identified in the five Salmonella enterica subsp. enterica genomes can vary by an order of magnitude, ranging from 13 in S. enterica sv. Typhimurium to 147 in S. enterica sv. Gallinarum (Figure 1). The abundance of pseudogenes is not associated with divergence time or phylogenetic affiliations. In fact, the two most closely-related strains, S. enterica sv. Gallinarum and S. enterica sv. Enteritidis, represent nearly the observed extremes of pseudogene abundance (147 versus 21).
Interestingly, the abundance of pseudogenes in the Salmonella genomes is reflected in the genome-wide Ka/Ks ratio, which can serve as a proxy for measuring the level of genetic drift experienced by a lineage. For example, despite their independent origins, the three strains with the highest numbers of pseudogenes also have the highest genome-wide Ka/Ks ratios. This observation is consistent with the expectation that deleterious mutations, such as inactivation of functional genes, are more likely to reach fixation in populations under high levels of genetic drift. In addition to higher rates of pseudogene production, high levels of genetic drift may also result in slower rates of pseudogene removal (assuming that the removal of pseudogenes is favored by positive selection). Consistent with this expectation, highly degraded pseudogenes with multiple inactivating mutations were only found in the three strains with high genome-wide Ka/Ks ratios. But despite the high correlation coefficient between genome-wide Ka/Ks ratios and the abundance of pseudogenes in these genomes (r = 0.74), the correlation does not reach statistical significance (P = 0.15), possibly because of the limited number of genomes examined or the use of a distant outgroup limits the resolution in Ka/Ks ratio estimation (i.e., the difference in high vs. low Ka/Ks groups is underestimated because most substitutions occurred on the branch leading to the outgroup).
Consistent with previous studies in other bacterial genera [6], [7], [13], most pseudogenes in Salmonella genomes are strain-specific. Of the 147 pseudogenes in the S. enterica sv. Gallinarum genome, only five are shared with its closest relative, S. enterica sv. Enteritidis. And of these five, only three share the same inactivating mutations and can be inferred as ancestral. The remaining two shared pseudogenes have different inactivating mutations and were inferred to result from independent events.
To examine mutational processes responsible for creating pseudogenes, we characterized each pseudogene in these Salmonella genomes by the number and type of gene-inactivating mutations. The vast majority of Salmonella pseudogenes (346/378) have only a single inactivating mutation (Figure 2), and among these, short deletions that removed 20% or less of the original open-reading frame predominate (141/346). We observed two cases of complete removal, including a 2,557-bp deletion containing a 1,224-bp gene and a 441-bp deletion encompassing a 189-bp gene. Because we required a pseudogene to be flanked by two conserved genes for its identification, any pseudogene that was removed by a deletion including a neighboring gene would not be recognized by this synteny-based approach. As expected from the mutational bias towards deletions in bacterial genomes, only a small fraction (17%) of the 346 pseudogenes were produced by an insertion; and with the exception of two cases of transposon insertions, all insertions are <10 bp. All remaining cases are due to point mutations: in 131 cases, there is a premature stop codon that reduced the length of the open reading frame by more than 20% and in two cases, a point mutation altered the start codon (one ATG to ATA, and one ATG to ATT).
We identified 32 pseudogenes with more than one inactivating mutation. There is no significant difference in average gene length between pseudogenes containing multiple inactivating mutations and those with a single inactivating mutation. Detailed information of the 378 curated pseudogenes is presented in Table S1.
Because neutral sequences accrue mutations with time, the relative age of a pseudogene is reflected in its number of accumulated mutations. The preponderance of pseudogenes with a single inactivating mutation indicates that most pseudogenes in these genomes are very young (Figure 2). This is also supported by the fact that most pseudogenes are restricted to individual genomes, as expected if they are newly formed. Further attesting to the recency of most pseudogenes is that there is little or no sign of accelerated sequence divergence relative to their functional orthologs (Figure 3).
Given the strong deletional bias in bacterial genomes, it is possible that the lack of old pseudogenes results from the rapid elimination of non-functional sequences by random fixation of mutations alone. If the jettisoning of pseudogenes is largely governed by a strictly neutral process, we expect that the probability of pseudogene removal to be independent to its age. Under this scenario, the age-class distribution is expected to decrease linearly in a log-normal plot (e.g., Figure 2 and Figure 3). However, there is an overabundance of young pseudogenes relative to this expectation no matter which one of the three methods we used to assign age class (phylogenetic distribution, number of inactivating mutations, and level of accelerated sequence divergence). This indicates that a non-neutral force is operating to remove young pseudogenes such that few remain in the genome long enough to accumulate multiple inactivating mutations or to exhibit accelerated sequence divergence rates (as would be expected for non-functional regions that were released from selective constraints). Consistent with this hypothesis, we detected a strong negative correlation between the loss rate and the age of pseudogenes estimated by the number of inactivating mutations (r = −0.99, P = 0.013). The scarcity of old pseudogenes is not likely to be an artifact of the methodologies used to identify pseudogenes (or, as noted above, the methods used to assign pseudogene age): our synteny-based approach is capable of detecting highly degraded pseudogenes harboring more than 10 frameshift mutations [19].
Of the 378 identified pseudogenes among S. enterica genomes, 120 have functional orthologs in Escherichia coli str. K-12 substr. MG1655 for which protein-protein interaction data are available. The average numbers of interacting partners in the protein-protein interaction network (i.e., the connectivity) for pseudogenes having different numbers of inactivating mutations revealed that highly degraded pseudogenes (i.e., those with three inactivating mutations) have, on average, significantly fewer interacting partners that do newly formed pseudogenes (i.e., those with a single inactivating mutation) (Figure 4).
If pseudogenes are completely functionless and their eliminations from bacterial genomes were governed by a strictly neutral process, the time since gene inactivation would not influence the probability of pseudogene removal from a genome. However, examination of pseudogene occurrence across multiple Salmonella genomes revealed deviations from a model of stochastic loss. Several independent lines of evidence, including the phylogenetic distribution of pseudogenes and the pattern of mutation accumulation, each demonstrated that newly formed pseudogenes were purged from bacterial genomes faster than neutral expectation, suggesting that they confer deleterious effects.
A possible explanation for this violation of neutrality is that there is selection for minimizing the size of mutational targets [28]; and since bacteria often have larger effective population sizes than do eukaryotes [29], it might be possible for selection to operate on mutations with extremely small effects (e.g., a 1-kb pseudogene accounts for only about 0.02% of a Salmonella genome). Unfortunately, this hypothesis cannot explain the observed pattern: If selection against inert DNA were the primary factor causing the removal of pseudogenes, we would expect to find fewer, but not necessarily a higher loss rate, for young pseudogenes. The age-distribution pattern predicted by this mutational-target model would, in fact, be indistinguishable from a strictly neutral model.
To account for differences in loss rate among pseudogenes belonging to various age classes requires methods that can accurately determine the relative ages of the pseudogenes present in this bacterial clade. Because most mutations accumulate as a function of time, one method was to use the level of sequence degradation as an indicator of pseudogene age. Because the youngest pseudogenes, i.e., those containing only a single inactivating mutation, have a higher probability of being expressed, their increased loss rate could result from the energetic costs of transcription and translation, which are known to shape the genome organization in prokaryotes [30]–[32] and eukaryotes [33].
Because bacterial cells are haploid, all mutations are effectively dominant since non-functional gene products cannot be masked by the corresponding functional allele as in diploid organisms. In that short indels (i.e., less than 10-bp) that cause frameshift and pre-mature stop codons are two of the most common types of mutations observed among bacterial pseudogenes (see Results and [6]–[7]), it is likely that the products from these altered open reading frames are disruptive to normal operation of cellular networks. This ‘toxic protein’ hypothesis is supported by our inference of protein-protein interactions: We find that those few pseudogenes that have persisted in Salmonella genomes (i.e., those that have accumulated multiple inactivating mutations) correspond to genes with relatively few interacting partners (Figure 4). The low connectivity of these genes can perhaps serve to minimize the deleterious effects of their inactivation. In contrast, for genes with large numbers of interacting partners, alteration of the open reading frames would potentially impact many protein-protein interactions. As such, mutations that remove such pseudogenes would be highly favorable and quickly reach fixation in the population. The negative correlation observed between loss rate and the age of pseudogenes is consistent with this model.
The premise that pseudogenes are removed because their encoded products are either energetically costly or toxic relies on the assumption that, after their initial disruption, these sequences are still being transcribed and translated. The majority of pseudogenes that we analyzed are newly arisen (i.e., have a single inactivating mutation in their coding regions), and since the mutational target of the regulatory portion of a gene is much smaller than the coding region, they are unlikely to harbor mutations that affect their expression. This is confirmed by the fact that most of these pseudogenes have nearly 100% sequence identity to their functional orthologs across the entire upstream intergenic region (i.e., from the end of the anchoring gene to the start codon).
As originally observed in E. coli [34], [35], and recently shown to occur in other bacteria [36], virtually all (even antisense) sequences in bacterial genomes are transcribed. Direct evidence of pseudogene expression is available for several strains of Salmonella. In a global analysis of Typhimurium gene expression using microarrays, Hautefort et al. [37] reported values for the relative expression of about 4,000 genes during host-cell infection. Some pseudogenes were up-regulated (e.g., putA, rffH), and others were down-regulated (e.g., dgoA), more than two-fold under the experimental conditions. An RNA-seq analysis of Typhi found that many pseudogenes were transcribed, albeit at highly reduced levels [38]. In this analysis, nine Typhi pseudogenes– both young and old – were still expressed high levels, but the overall reduction in pseudogene expression was taken to indicate that the majority of pseudogenes were no longer active [38], possibly to ameliorate the deleterious effects that we detected.
To determine if reduced expression fosters the maintenance of pseudogenes in bacterial genomes, we examined the codon adaptation index (CAI, which is an indicator of overall expression levels over evolutionary timescales) of genes in the difference age classes. Paralleling the effect shown in Figure 4, the average CAI is significantly lower in older pseudogenes (average CAI = 0.29 for age class 3 vs. 0.36 for age class 1; p = 0.003, one-tailed unpaired t-test assuming unequal variance). These results are consistent with our expectation that selection acts to remove more highly expressed (and connected) genes once they become pseudogenized.
Mutations in bacterial genomes are known to be highly biased toward deletions [18], [19]. Therefore, it is not surprising to find that accumulation of deletions is the primary force responsible for the erosion of bacterial pseudogenes. However, only a small fraction of pseudogenes detected in Salmonella genomes were found to have lost more than 20% of their original length, despite the high sensitivity of our synteny-based method for pseudogene detection. Given the high incidence of kilobase-sized (and larger) deletions observed during Salmonella experimental evolution [24], the main mechanism for the complete removal of pseudogenes is likely to be large deletions, most of which are large enough to remove neighboring genes and therefore cannot be detected using a local-synteny based approach.
Our systematic characterization of multiple Salmonella genomes indicates that the evolution of bacterial pseudogenes is not strictly neutral such that newly formed pseudogenes have a higher likelihood of being removed. This deviation from the generally accepted view that pseudogenes represent completely neutral regions [20] is likely due to the fact that bacteria have haploid genomes and generally large effective population sizes, therefore increasing the exposure of mutations to selective forces. If pseudogenes are deleterious due either to the energetic costs of transcription and translation or to the dominant-negative effects of anomalous proteins, the high efficacy of selection in bacterial genomes is likely to have a role in their removal. This is consisitent with our finding that those Salmonella genomes with the lowest genome-wide Ka/Ks ratios denoting a relatively high efficacy of selection harbor the lowest numbers of pseudogenes.
Because all bacterial groups, as well as those Archaea examined, display a mutational pattern that is biased towards deletions [18], [19], [33] and their haploid genomes would be more susceptible to dominant-negative effects that pseudogenes might impart, it is likely that the process of adaptive removal of pseudogenes is pervasive among prokaryotes. And given the evidence for selection on intron size in some eukaryotic genomes, presumably due to the energetic cost of transcription [39], these effects need not be restricted to those cellular organisms with haploid genomes, and pseudogene degradation and removal may be found to be operating under similar principles in representatives from all domains of life.
We obtained the complete genome sequences of six Salmonella enterica strains from NCBI GenBank [40], including S. enterica subsp. enterica serovar Enteritidis str. P125109 (NC_011294), S. enterica subsp. enterica serovar Gallinarum str. 287/91 (NC_011274), S. enterica subsp. enterica serovar Choleraesuis str. SC-B67 (NC_006905), S. enterica subsp. enterica serovar Typhimurium str. LT2 (NC_003197), S. enterica subsp. enterica serovar Typhi str. CT18 (NC_003198), and S. enterica subsp. arizonae serovar 62:z4,z23:– (NC_010067) as the outgroup. This set of genome sequences were selected because: (1) the low level of divergence allows for reliable sequence alignment and thus confident inference of gene inactivation events, (2) the phylogenetic relationship among the six strains allows for straightforward assignment of age-class for pseudogenes base on their phylogenetic distribution pattern, and (3) the sequencing was performed by high-coverage whole-genome shotgun sequencing with the Sanger method, which provides high accuracy in homopolymer regions. The last point was of particular importance because our preliminary analysis suggests that sequencing errors in homopolymer regions are a major factor that contributes to erroneous pseudogene annotations in several other S. enterica genome sequences. The difficulties involved in distinguishing true frameshift mutations from sequencing errors prohibit a more comprehensive taxon sampling.
To identify orthologous gene shared among the six S. enterica genomes, we performed all-against-all NCBI-BLASTN [41] searches with an e-value cutoff of 1×10−15 for all annotated protein-coding genes. A set of custom Perl scripts written with Bioperl modules [42] were used for data parsing and processing. The BLASTN results were supplied as the input for OrthoMCL [43] to perform ortholog clustering. The algorithm is largely based on the popular criterion of reciprocal best hits between genomes and has been shown to perform well by a benchmarking study [44].
To infer the phylogenetic relationship among the six S. enterica strains, we aligned the nucleotide sequences of the 2,898 single-copy genes shared by all six strains using MUSCLE [45] with the default parameters. We used TREE-PUZZLE [46] to infer the distance matrix and the phylogenetic tree based on a concatenated alignment with 2,772,598 sites. The changes from default setting in TREE-PUZZLE include: (1) use exact parameter estimates, (2) estimate the nucleotide frequencies and transition/transversion ratio from the data set, (3) use a mixed model with one invariable and eight Gamma rates for rate heterogeneity, and (4) estimate the fraction of invariable sites and the Gamma distribution parameter from the data set.
We calculated the genome-wide Ka/Ks ratio for each of the five ingroup strains to estimate the level of genetic drift experience by the lineage. This ratio is a good approximation for the level of genetic drift because it measures the efficacy of purifying selection in protein-coding region; an elevated level of genetic drift can result in increased incidence of slightly deleterious amino acid replacement, and thus, an increase in genome-wide Ka/Ks ratio. Although positive selection favoring certain amino acid changes can also increase Ka, such scenario is expected to be limited to particular genes and sites rather than driving changes throughout the entire genome [1], [47].
For each of the 2,898 single-copy genes shared by all six strains, we performed multiple sequence alignment at amino acid level using MUSCLE [44] with the default parameters. The resulting protein alignments were converted into codon-based nucleotide alignment using PAL2NAL [48]. To account for possible base composition and codon usage bias in any of the genes examined, we applied the YN00 method [49] implemented in the PAML package [50] to estimate the substitution rates. For each of the five ingroup strains, we calculated Ka and Ks using the outgroup S. enterica subsp. arizonae as the reference. To avoid potential bias in Ka/Ks ratio estimation due to non-sufficient sequence divergence or saturation, we removed genes that have an estimated Ks of less than 0.1 or greater than 1.5 in any of the five pair-wise comparisons. The average Ka/Ks ratio calculated from the 2,290 remaining genes was used to represent the genome-wide estimate for each of the five ingroup strains.
We utilized a synteny-based approach similar to that described previously [19] for pseudogene identification. Although this approach may underestimate the total number of pseudogenes in a genome due to the exclusion of pseudogenes that lack positional homologs in other closely related genomes (which may have originated from horizontal transfer), the stringent requirement allows for confident inference of the gene inactivation events.
To identify pseudogenes with positional homologs, each of the five S. enterica subsp. enterica strains was used as the query against every other genome. The outgroup S. enterica subsp. arizonae was not considered as a query because the ancestral state of any pseudogene identified in this genome cannot be established with our taxon sampling. For each pair of query and subject, we utilized single-copy genes shared by the two genomes as anchors to systematically examine the intergenic regions in the query genome. An intergenic region is flagged as containing a putative pseudogene if an annotated protein-coding gene was found in the syntenic region of other genomes.
For each candidate region, we aligned the query genome to the subject genome using MUSCLE [45] with the default parameters. The two anchoring genes were included to improve the quality of alignment and to allow for examination of the entire intergenic region of the query genome. Possible gene-inactivating mutations, including insertions, deletions, pre-mature stop codons, and/or point mutations in the start codon were inferred based on the annotated gene in the subject genome. The results were manually inspected for the consistency regarding gene synteny and the identified mutations across different reference genomes.
To ensure a high level of confidence when inferring gene-inactivating mutations, we required at least two positional homologs to establish the ancestral state of a pseudogene. During our curation process, the following types of false-positives were removed before the final analysis: (1) the entire open reading frame is intact in the query genome but not annotated as a gene, (2) the putative pseudogene may be explained as an annotation artifact (e.g., the region was annotated as a part of either anchoring genes in the query genome), (3) pre-mature stop codons are the only type of mutations and the protein lengths were reduced by less than 20%, (4) the reference gene is a transposase from a insertion sequence element or of viral-origin (i.e., likely a gene gain in the subject genome instead of a gene loss in the query genome), (5) the phylogenetic distribution of the reference gene suggests that a single gene gain event (e.g., horizontal gene transfer) is the most likely explanation for the absence of corresponding gene in the query genome, and (6) the reference gene is a poorly conserved hypothetical protein and is shorter than 300 bp.
In the rare cases where the identified mutations exhibit inconsistency across different reference genomes or indicate extensive sequence divergence, we extracted the syntenic region from all genomes to perform multiple sequence alignment and deduced the inactivation events based on the most parsimonious scenario. One special case involved a 873-bp inversion within the srfB pseudogene in the Gallinarum genome; we manually corrected the inversion before the multiple sequence alignment to infer other possible inactivation events. Furthermore, the exact boundaries of all indel events that affect the 5′- or 3′-end of a pseudogene were manually verified.
To classify the curated pseudogenes into different age classes, we examined their phylogenetic distribution pattern to characterize the likely time point of gene inactivation events on the phylogeny. Because 302 out of the 378 curated pseudogenes are specific to one genome, we used two additional methods for age class assignments. In the first method, we categorized the pseudogenes based on the number of gene-inactivating mutations that have been accumulated. In the second method, we utilized an ortholog in the outgroup to quantify the level of accelerated sequence divergence in the pseudogene relative to its functional ortholog in another genome. The nucleotide sequence alignments were inferred using MUSCLE [45] with the default parameters and subsequently used as the input for TREE-PUZZLE [46] to calculate distance matrices. Due to the lack of appropriate orthologs in the outgroup, only 227 out of the 378 curated pseudogenes can be classified using this method.
To infer the potential role of a Salmonella pseudogene in cellular fitness, we identified the orthologous gene in Escherichia coli str. K-12 substr. MG1655 (NC_000913), a related enteric strain on which extensive experimental and functional assays have been conducted [51], [52]. For ortholog identification, we used the full-length gene from the closest reference genome as the query to perform NCBI-BLASTP [40] searches. To qualify as an ortholog, we required the BLASTP hit to satisfy all of the following conditions: (1) is the best hit among all of the protein-coding genes in the E. coli MG1655 genome, (2) has an BLASTP e-value of less than 1×10−15, (3) the difference in gene length is no more than 20% of the shorter sequence, (4) the high scoring pairs (HSPs) account for at least 80% of the shorter gene, and (5) the fraction of conserved amino acid sites is at least 60% within HSPs. For pseudogenes that had a corresponding ortholog in the E. coli MG1655 genome, we extracted the protein-protein interaction information from the high quality combined dataset available from an integrated protein interaction database [53], [54] to infer numbers of interacting partners.
|
10.1371/journal.pbio.1001309 | Molecular Requirements for Peroxisomal Targeting of Alanine-Glyoxylate Aminotransferase as an Essential Determinant in Primary Hyperoxaluria Type 1 | Alanine-glyoxylate aminotransferase is a peroxisomal enzyme, of which various missense mutations lead to irreversible kidney damage via primary hyperoxaluria type 1, in part caused by improper peroxisomal targeting. To unravel the molecular mechanism of its recognition by the peroxisomal receptor Pex5p, we have determined the crystal structure of the respective cargo–receptor complex. It shows an extensive protein/protein interface, with contributions from residues of the peroxisomal targeting signal 1 and additional loops of the C-terminal domain of the cargo. Sequence segments that are crucial for receptor recognition and hydrophobic core interactions within alanine-glyoxylate aminotransferase are overlapping, explaining why receptor recognition highly depends on a properly folded protein. We subsequently characterized several enzyme variants in vitro and in vivo and show that even minor protein fold perturbations are sufficient to impair Pex5p receptor recognition. We discuss how the knowledge of the molecular parameters for alanine-glyoxylate aminotransferase required for peroxisomal translocation could become useful for improved hyperoxaluria type 1 treatment.
| Peroxisomes are cell organelles contain proteins involved in various aspects of metabolism. Peroxisome proteins translocate from their site of synthesis in the cytoplasm across the organelle membrane in a fully folded and functional form. One such protein is the enzyme alanine–glyoxylate aminotransferase (AGT). It contains a targeting signal in its C-terminus that is recognized by a receptor protein, Pex5p, in the cytoplasm, which allows its subsequent translocation into the peroxisome. Mutations in AGT cause a disease known as primary hyperoxaluria type 1, in which patients suffer irreversible kidney damage; this disease results, in many cases, from improper targeting of AGT into peroxisomes. To understand better the mechanism of AGT import into peroxisomes and the molecular basis of this disease, we have determined the crystal structure of the complex between AGT and its receptor Pex5p. The structure reveals how overlapping segments of the protein sequence are crucial for both receptor recognition and maintaining the folded structure of the enzyme. Subsequently, we created and studied several mutants of the enzyme, including mutants that are known to cause disease, and found that even minor folding defects in the enzyme prevent its recognition by Pexp5 and its import into peroxisomes. Our data thus provide novel insights into the consequences of mutations in AGT on the catalytic activity of the enzyme, as well as into the mechanisms that cause primary hyperoxaluria type 1.
| Primary hyperoxaluria type 1 (PH1) is an autosomal recessive disorder that generally becomes symptomatic during childhood or adolescence and ultimately leads to renal failure, usually between the ages of 25 and 45 [1]. Although several therapeutic options have been established, the only curative treatment to date is by liver-kidney transplantation [2]. At the molecular level, PH1 is caused by functional deficiencies in the liver-specific, pyridoxal-dependent enzyme alanine-glyoxylate aminotransferase (AGT, EC 2.6.1.44) [3]. AGT catalyzes the transamination of the peroxisomal intermediary metabolite glyoxylate to glycin. Human AGT consists of a 86 kDa homodimer and bears an atypical Lys-Lys-Leu (KKL) peroxisomal targeting signal 1 (PTS1) motif at its C-terminus, which is required for translocation of the enzyme into peroxisomes. The absence of AGT in hepatic peroxisomes, owing to either dysfunction or mistargeting of AGT, causes glyoxylate to escape into the cytosol where it is further metabolized to oxalate and glycolate. The accumulation of oxalate—a compound that cannot be further metabolized in humans—leads to the progressive formation of insoluble calcium oxalate in the kidney and urinary tract, resulting in urolithiasis and/or nephrocalcinosis as the principal clinical manifestations.
To date, around 150 polymorphic variants of the human AGXT gene have been described [4]. These mutations are scattered over virtually the entire encoded AGT sequence and the associated three-dimensional structure of the enzyme (Figure S1). In 2%–20% of human populations in geographically distinct regions, a minor allele haplotype (AGXT-Mi) is found, which encodes an AGT variant with two missense mutations (P11L, I340M). AGT-Mi has around one-third of the catalytic activity of the wild-type enzyme and reduced stability, yet by itself does not lead to a serious clinical phenotype. However, the presence of AGXT-Mi in combination with further mutations causes almost 50% of the reported PH1, demonstrating synergistic disease effects [4]. Only some of the characterized PH1-causing AGXT variants can be directly correlated with AGT enzymatic activity, suggesting that other molecular parameters such as its correct compartmental localization have important implications for AGT function as well. Therefore, it is not surprising that there is no uniform response by PH1 patients to pyridoxine intake, which is thought to stabilize the AGT active site but does not directly affect the localization of the enzyme [5],[6].
On the basis of biochemical and structural data, the molecular mechanism of AGT catalytic activity is well established [7],[8], but the mechanism of peroxisomal AGT targeting is poorly understood. The non-canonical PTS1 Lys-Lys-Leu sequence in human AGT has been described as non-optimal, based on in vitro interaction studies of chimeric proteins formed by fusing the motif with non-human AGTs and other peroxisomal target proteins [7],[9]. Truncation studies of human AGT led to the prediction of an additional binding site within the small C-terminal domain of AGT, proximal to the established PTS1 C-terminus [9]. Another non-overlapping AGT-Pex5p recognition segment was proposed to be located close to the AGT N-terminus [10]. However, in the absence of residue-specific interaction data, it is not known whether additional interactions with the Pex5p receptor are direct or mediated by putative adaptors, or even whether allosteric effects are involved [3],[11].
Moreover, a generalization of the interpretation of available data is virtually impossible, as neither the PTS1 sequence nor a consistent pattern for peroxisomal localization are taxonomically conserved among AGTs from different species [12]. Indeed, depending on the organism, AGTs have been found, partly in parallel, in mitochondria, the cytosol, and peroxisomes [13]. Alternative transcription and translation sites in several AGTs lead to elongated isoforms with an additional N-terminal mitochondrial targeting signal sequence, which overrides the PTS1 required for peroxisomal translocation [3],[14]. Even in the absence of an additional mitochondrial-targeting signal, residual mitochondrial localization has been observed for AGT mutants that tend to aggregate and misfold [4],[14].
The aim of this work has been to unravel the role of non-PTS1 PH1 mutations in AGT mistargeting, to ultimately provide a molecular model for genetically imprinted PH treatment. To identify the complete Pex5p receptor-interaction site, we have first determined the atomic structure of the AGT–Pex5p receptor complex, which forms an elongated Pex5p-(AGT)2-Pex5p assembly. In addition to the established PTS1-binding site, the structure reveals extensive but rather non-specific contributions from sequence segments of the C-terminal AGT domain. To test how perturbations in the AGT structure could result in effects on AGT–Pex5p receptor binding, we mutated several residues of the AGT C-terminal domain near the Pex5p interface and investigated the properties of the resulting mutants by biophysical and functional in vitro and in vivo assays, as well as their ability to bind Pex5p. The interactions observed are highly sensitive to any minor changes in the AGT structure caused by single-residue mutations—including those that have been identified in PH1 patients—demonstrating that non-PTS1 interactions are essential in Pex5p receptor recognition.
To determine the molecular basis of the recognition of AGT by the peroxisomal import receptor Pex5p and its implications in PH1, we purified human AGT and the C-terminal cargo-binding segment of human Pex5p (residues 315–639), referred to as Pex5p(C) [15]. The AGT–Pex5p(C) complex forms with an apparent (1∶1) stoichiometry and has a moderate dissociation constant of 3.5 µM (Table 1). AGT, alone or in complex with Pex5p(C), has a catalytic activity of close to 2,000 µM mg−1 h−1, which is in agreement with previously reported AGT data [16],[17] and suggests that binding to Pex5p does not compromise AGT activity.
We then determined the crystal structure of the AGT–Pex5p(C) complex at 2.4 Å resolution (Figures 1 and S2; Table 2; Text S1). The structure comprises an elongated Pex5p(C)-(AGT)2-Pex5p(C) assembly with overall dimensions of around 140 Å×50 Å×50 Å. The 1∶2∶1 stoichiometry of the complex is in agreement with our isothermal titration microcalorimetry (ITC), gel filtration, and static light scattering data, indicating equal stoichiometric contributions of both protein components (Table S1 and Figure S3).
Each of the two complete AGT polypeptide chains is visible in the final electron density, except for N-terminal residues 1–3 and 1–5, respectively. The overall conformation of the two AGT molecules is identical (Table S2) and the structure shows that they both contain the cofactor pyridoxal-5′-phosphate (PLP) covalently bound to Lys209 (Figure S4). We confirmed the AGT PLP-adduct to be present by spectroscopic analysis of the protein material submitted for crystallization (Figure S4C). The Pex5p(C)-bound AGT dimer superimposes well onto that of the enzyme in the absence of the receptor (PDB entry 1HOC) [8], with a root-mean-squares deviation of 0.41 Å (Table S2). This confirms that AGT dimeric assembly and overall conformation, a prerequisite for AGT catalytic activity [3], is not affected by Pex5p receptor binding.
Well interpretable electron density is visible for most of the two Pex5p(C) receptor molecules (residues 315–639), with the exception of the N-termini (residues 315–323/324), part of the distorted tetratricopeptide repeat (TPR) 4 segment (residues 441–464, 444–460) and the so-called 7C-loop (residues 591–592, 590–596) that connects the 7-fold array of TPR segments with the C-terminal bundle of Pex5p(C) [15]. These regions were either invisible or mobile in previous structures of the same receptor [15],[18], indicating that these sequence segments are generally flexible. Overall, increased flexibility of Pex5p(C), which we attribute to these regions and to the loose arrangement of neighboring TPR domain modules, is reflected in higher root-mean-squares deviations of around 1 Å when Pex5p(C) polypeptide chains of the Pex5p(C)-AGT complex are either superimposed on each other or onto the coordinates of the same receptor from the previously determined Pex5(C)-SCP2 cargo complex (Table S2) [15].
By contrast, there are significant deviations in the overall structure of Pex5p(C) bound to AGT when it is superimposed onto the apo conformation of the same receptor (PDB entry 2C0M). The matching part of the respective structures is limited to residues of the 7-fold TPR array, excluding the C-terminal bundle domain. Hence, the structure of the Pex5p(C)-AGT complex supports the conformational changes of the receptor that have been observed previously on cargo binding [18],[19].
The structure of the AGT–Pex5p complex reveals that the C-terminal AGT domain (residues 283–392) is the exclusive and direct binding module of the Pex5p receptor (Figures 1 and 2). This domain comprises a bundle of the three helices α11 (residues 284–305), α12 (residues 332–343), and α13 (residues 370–387), in which the two longest helices (α11, α13) are in a parallel orientation to each other and the third (α12) crosses helix α13. The three helices are connected by a small two-stranded β-sheet (β8, residues 321–325; β9, residues 358–362) that forms an interface with the N-terminal catalytic AGT transaminase domain. The C-terminal sequence Pro-Lys-Lys-Lys-Leu (residues 388–392), corresponding to the PTS1, immediately follows helix α13.
The overall AGT–Pex5p interface consists of three distinct surface patches (Figures 1 and 2): the first involves the AGT-PTS1 (residues 389–392) that binds, as expected, into the central tunnel-like cavity of the ring-forming array of seven TPR segments of Pex5p(C), generating an interface of 550–600 Å2 (Interfaces Ia and Ib in Figure 1; Figure S5, left panel). The second includes the C-terminal part of the AGT helix α13 that immediately precedes the PTS1 (residues 381–388) and the loop connecting β9-α12 (residues 327–330) that interacts with this part of α13 (Interfaces IIa and IIb in Figure 1; Figure S5, central panel). We refer to this site in AGT as the “extended PTS1” interface, as it is directly upstream of the PTS1. Pex5p interactions from this interface overlap with hydrophobic core contacts by residues from α13 with other parts of the C-terminal AGT domain. Ala383, the most C-terminal AGT residue that is entirely buried within the AGT fold, is preceded by Arg381, which marks the most proximal residue in α13 that contributes to the extended PTS1–Pex5p interface. The third interface is topologically separate from the PTS1 and involves the loop that connects AGT helix α11 and strand β8 (residues 303–307) (Interfaces IIIa and IIIb in Figure 1; Figure S5, right panel). These two additional surface patches, when combined with the PTS1 binding site, increase the overall AGT–Pex5p(C) interface area by almost 2-fold, to around 1,000 Å2 (Table 3). A detailed structural description of all the three interface patches is provided in Text S1.
The three binding sites are topologically preserved in the two AGT–Pex5p(C) modules. However, direct comparison reveals that when using the structure of Pex5p(C) as the basis of superposition, the orientation of the two bound AGT molecules deviate substantially (Figures S5 and S6). If the two protein components are assumed to be rigid bodies, the tilt and twist angles defining their relative orientation [20] change by 27 and 11 degrees, respectively. The difference originates from a limited conformational flexibility with a pivot point at the C-terminus of the AGT helix α13, preceding the PTS1 motif. Owing to the rigidity of the remaining AGT structure, the spatial differences in the superimposed complexes increase to around 20 Å in those parts of each AGT protomer that are most distal to the Pex5p(C) receptor-binding site (Figure S6). Because of these conformational differences, there is little conservation in the specific AGT–Pex5p(C) interactions. With the exception of a few conserved hydrogen bonds formed between three asparagines of Pex5p (Asn415, Asn526, Asn561) and the C-terminal main-chain carboxylate group of Leu392, along with the preceding peptide bond connecting Lys391 and Leu392, the remaining side chains of the AGT PTS1 sequence Lys389-Lys390-Lys391 are either not involved in further specific interactions or, if observed, these interactions are not conserved within the complete Pex5p-(AGT)2-Pex5p complex (Figure 2 and Figure S5). These findings are in agreement with an overall endothermic assembly process under the experimental in vitro conditions, indicating that AGT–Pex5p(C) complex formation is an entropy-driven process (Tables 1 and S1) rather than being dominated by specific enthalpic interactions.
A key finding from our structural data is that binding of the AGT PTS1 motif to the Pex5p receptor is not autonomous from the additional cargo–receptor binding sites, both in terms of sequence connectivity and surface topology. These data could explain why many pathological AGT disease mutations that lead to AGT mistargeting are remote from the Pex5p-binding site. On the basis of our structural data, we argue that even minor folding defects or conformational alterations in AGT could compromise the binding of the AGT composite Pex5p interface, formed by the AGT C-terminal domain and PTS1.
To address this assumption, we mutated several residues in the AGT C-terminal domain close to the Pex5p-binding interface, which we expected to lead to conformational changes in this domain without compromising AGT activity (Figures 2 and S1). The first set of mutations involved two residues from the β9–α12 loop (Ala328, Tyr330) that interact with residues from the C-terminal helix α13 (Leu384 and Lys389). We introduced either more bulky side chains (A328W, Y330W) or removed side chain-specific intramolecular interactions (Y330A). For the second set of AGT variants, we aimed to affect the hydrophobic interactions of the C-terminal helix α13 with other parts of the AGT C-terminal domain. For this purpose, we mutated two residues from this helix (Val376, Leu380) that are completely buried into either an aspartate or proline. Additionally, to provide a structural rationale for established AGT disease mutations, we selected two AGT single residue polymorphisms (G170R, V336D) and the corresponding AGT double mutant G170R/V336D, which have been found in combination with the minor allele haplotype (AGXT-Mi) in PH1 patients. The AGT double mutant G170R/V336D results in a serious pathogenic effect and is non-responsive to pyridoxine treatment [2],[4]. However, the disease-causing mechanism of this AGT polymorphism, like various other mutations, has remained enigmatic. More specifically, the aggravating effect of the V336D mutation from the C-terminal domain in conjunction with the widespread G170R mutation seemed to be inexplicable, as the latter (G170R) is coupled with unwanted mitochondrial import in the AGXT-Mi isoform [21], again by an unknown mechanism of action. A structure of the AGT G170R mutant revealed only minor local conformational changes [22].
First, we attempted to purify all the AGT mutants to test their ability to bind the Pex5p receptor in vitro and to measure their catalytic activities (Table 1). However, the AGT variants with mutations in residues of the C-terminal helix α13 (Val376, Leu380) were insoluble when overexpressed in Escherichia coli, demonstrating that the hydrophobic core interactions of helix α13 are essential for proper folding of the enzyme under the chosen experimental conditions. The same problem of aggregation arose for the pathogenic AGT double mutant G170R/V336D, whereas each of the two single residue variants (G170R, V336D) could be expressed in significant quantities as soluble proteins. Although the aggregated AGT mutants could not be further characterized in vitro, they were used in functional assays to assess their tendency for aggregation in vivo and to investigate the level of peroxisomal targeting from AGT versions with suspected folding defects (see below).
All remaining mutants were purified by affinity chromatography and gel filtration (Figure S3). Proper folding of each protein was confirmed by far-UV circular dichroism spectroscopy (Figure S3). These AGT mutants had catalytic activities similar to the wild-type enzyme irrespective of Pex5p binding with the exception of the G170R mutant, which showed a decrease in activity of around 25%, in qualitative agreement with previous data [17]. Whereas the two pathogenic AGT single-residue mutants (G170R, V336D) did not show a significant change in Pex5p receptor binding, the AGT variants with mutations in the β9–α12 loop showed 2- to 6-fold decreased binding affinities for the Pex5p receptor when compared with the wild-type enzyme (Table 1). The weakest interaction, with a Kd of 19.4±8.3, was found for the Y330A AGT variant, indicating an important contribution of the side chain of Tyr330 to keep the β9-α12 loop in a conformation that is competent for Pex5p receptor binding. As for wild-type AGT, Pex5p binding by all of the AGT mutants is endothermic under the in vitro experimental conditions (Table S1).
To test the functional properties of all selected AGT variants in vivo, we employed a protein import assay in human fibroblasts, using enhanced green fluorescent protein (EGFP)-tagged AGT. When expressing EGFP-AGT without further modification in fibroblasts, we observed that more than 90% of the cells exhibited a punctuated pattern of peroxisomal localization (Figure 3). By contrast, a control version of the enzyme without the PTS1 (ΔPTS1) was evenly distributed in the cytosol without any visible sign of peroxisomal import (Figure S7), confirming that the presence of a PTS1 in AGT is crucial for recognition by the Pex5p receptor.
None of the AGT helix α13 variants showed significant measurable peroxisomal translocation, suggesting that fold defects in AGT lead to an almost complete loss of Pex5p import (Table 1; Figure 3). Aggregation of these AGT versions under in vivo experimental conditions is reflected by the formation of large fluorescent plaques in the cytosol, which are abundant in 28%–100% of transfected cells. This indicates substantial variability depending on the AGT mutant investigated. Whereas AGT(L380P), for instance, aggregates completely (Figure 3B), other AGT mutants (V376D, L380D) reveal a predominantly cytosolic background, suggesting a soluble cellular state with no peroxisomal association (Figure S7). These observations indicate that both misfolding and local conformational changes in the C-terminal domain have a synergistic effect, leading to a loss of AGT targeting to peroxisomes. The data also suggest that indirect effects, arising from altered structural properties of the AGT cargo, rather than direct and specific receptor interactions, are sufficient to abolish proper cargo recognition by the Pex5p receptor for peroxisomal targeting.
A slightly milder effect was observed with the pathogenic double mutant G170R/V336D, with 28% of the protein-forming plaques in the cytosol, and another 28% being properly translocated into peroxisomes (Figure 3). The overall level of non-peroxisomal localization of this AGT mutant is 72%. All remaining AGT variants, including those from the β9–α12 loop and the two pathogenic single-residue mutants (G170R, V336D)< displayed 59%–76% peroxisomal localization, which is in agreement with our in vitro binding data and indicates a weakening but not an abolishment of Pex5p binding. Two AGT mutants from this category (G170R, A328W) showed around 5% aggregation, whereas no significant level of aggregation was measured for the remaining mutants. Taken together, the data show that even minor structural perturbations in AGT have a measurable and significant effect on AGT translocation.
The AGT–Pex5p structure is the second cargo protein–Pex5p receptor complex determined to date, the first being sterol carrier protein 2 (SCP2)–Pex5p [15]. Our data indicate that the dimeric and cofactor-bound arrangement of AGT is preserved and that the enzyme remains functional prior to and upon binding of the Pex5p receptor (Table 1). This observation is in agreement with the unique ability of peroxisomes to import even large and oligomeric cargos as functional protein assemblies [23]–[25]. As our studies have been carried out in the absence of any additional protein components, a potential requirement of adaptor proteins as previously suggested [9],[11] is unlikely. Our data confirm the involvement of segments from the C-terminal AGT domain—previously described as the “PTS1A” binding site [9]—in Pex5p receptor binding, but do not support earlier suggestions that an N-terminal AGT sequence segment contributes to receptor recognition [10].
Comparison of the complexes of Pex5p with SCP2 and AGT allows for the first time the identification of common and diverging principles in target protein recognition (Table 3), beyond the well-established C-terminal PTS1 motif that is shared by most Pex5p cargos [26]. Notably, the measured AGT–Pex5p interaction is about 30-fold weaker than that observed for SCP2. This argues in favor of AGT being highly sensitive to perturbations that affect Pex5p recognition (Table 1; Figure 3) and may mirror the large number of known disease-causing AGT mutations that have been associated with protein mistargeting rather than with catalytic activity effects [4].
The two protein cargo–receptor complex structures reveal that there are almost no specific, conserved side-chain interactions between polar residues from each PTS1 motif with Pex5p, with the notable exception of the very C-terminal leucine residue (Figure 4). This observation is supported by previous findings on AGT that indicate side-chain tolerance at PTS1 position −3 and, albeit more limited, at position −1 [27]. By contrast, our data only partly agree with observations from Pex5p–PTS1 peptide complexes, in which a more extensive hydrogen bond network over several PTS1 residues was observed [25],[28],[29]. Comparison with the available Pex5p–cargo protein complex structures indicates that the adaptability of possible PTS1 conformations to optimize specific interactions with the Pex5p receptor is restricted owing to the additional non-PTS1 protein interfaces that are formed between the C-terminal bundle domain of the receptor and cargo, as previously shown for SCP2 [15],[30] and for AGT in this contribution (Figure 2). Collectively, however, the additional non-PTS1 interactions (marked as IIa,b and IIIa,b in Figure 1) only slightly enlarge the overall Pex5p–AGT interface, in comparison to that observed in the Pex5p-SCP2 complex, in one of the two Pex5p-AGT complexes (Table 3).
The specific Pex5p binding abilities of the PTS1 cargo peptides, corresponding to AGT and SCP2 sequences, are weak, with dissociation constants in the low to sub-µM range [15],[26]. The gain in binding affinity for AGT when the complete protein is used is around 4-fold—3.5 µM instead of 13.5 µM (Table 3). Similarly, a gain in binding affinity of around 6-fold has been found for SCP2–Pex5p assembly when the protein complex is compared with the corresponding PTS1 peptide complex [15]. However, a recent analysis of additional non-PTS1 interactions confirmed that their contribution is only of minor importance, in turn suggesting that SCP2 recognition by the Pex5p receptor is principally driven by autonomous recognition of its PTS1 motif [30]. By contrast, our structural and functional data on AGT–Pex5p show that complex formation is both dependent on the presence of the AGT PTS1 motif and the correct Pex5p binding-competent conformation of the AGT C-terminal domain. Based on these findings, we argue that previously reported problems in establishing in vitro binding with purified protein components and by transfection experiments have failed for several PTS1 protein cargos in vivo owing to contextual defects in protein folding and possibly oligomerization [31],[32]. Moderate binding of the cargo in vivo may facilitate subsequent release of the cargo into the peroxisomal lumen, a process that at present is still less well understood than the mechanism of cargo binding [18],[33],[34].
Further investigation of our structural data of the AGT–Pex5p complex reveals that the sequence segments in AGT that constitute the PTS1 and the hydrophobic core of AGT topologically overlap, whereas in SCP2 the corresponding sequence segments are well separated (Figure 5). Specifically, the PTS1 interactions observed extend to Arg381 (PTS1 position −11, when considering the C-terminal Leu392 as position 0), and the side chains of three residues within the extended PTS1 segment (Ala383, Leu384, Cys387) are also involved in hydrophobic core interactions of the C-terminal AGT domain. The overlapping interactions thus generate a seven-residue segment (381–387) from the C-terminus of helix α13 (Figure 2A) [8] that is involved in both the overall AGT fold and Pex5p receptor recognition. These structural observations indicate that, in contrast to our previous findings on the SCP2–Pex5p complex [15], Pex5p receptor recognition of the PTS1 in AGT is structurally non-autonomous with respect to the remaining fold of the enzyme. Our structural data also explain previous observations on the translocation of AGT molecules that contain mutations in the extended PTS1 motif. These studies showed that diminished binding is caused by folding defects rather than by loss of cargo–receptor interactions that were predicted prior to available structural data [27], indicating that AGT PTS1 binding depends on properly folded AGT and thus is also functionally non-autonomous. This is well illustrated by the strong translocation defects of several extended PTS1 mutations in AGT (L380P; V376P) [27], which are involved in AGT hydrophobic core interactions rather than specific AGT–Pex5p interactions (Figure 5).
In AGT, the additional Pex5p non-PTS1 interactions observed are not as specific as one may expect (Figures 2A and S5) and perhaps explain the moderate overall binding affinity. These findings are further supported by the observation that Pex5p–AGT binding in vitro is an entropy-driven process, suggesting that binding is dominated by order/disorder processes rather than by enthalpy-driven specific interactions.
AGT is an enzyme with a well-established genotype/phenotype database, including about 150 different missense mutations, many of which lead to serious forms of PH. Our structural and functional characterization of the molecular parameters for AGT to be recognized by the Pex5p receptor and its subsequent translocation into peroxisomes offers an opportunity to rationally address functional implications of pathogenic PH-causing missense mutations.
We assume that those PH mutations that lead to irreversible AGT aggregation, irrespective of the presence of the Pex5p receptor, will be difficult or even impossible to treat by chemical intervention as these AGT variants are expected to lose both their enzymatic activity and their ability to be recognized by the peroxisomal Pex5p receptor as a consequence of misfolding. Based on our mapping of known AGT missense mutations on the three-dimensional structure of the AGT–Pex5p complex (Figure S1), we estimate that around half of these lead to fold defects, as they reside in regions that are completely buried within the AGT fold. The fraction of misfolded AGT mutants is probably even higher when associated with the widespread AGXT-Mi gene [3], which leads to additional destabilization of the enzyme. Partial rescue of some of these mutations, by adding chaperones or osmolytes for instance [35], may be possible but remains challenging, as most of these additives tend to be non-specific.
On the basis of our data, we further expect that the loss of function of many of the remaining patient mutations (Figure S1) that result in AGT, which does not aggregate or is only partially prone to aggregation, could be potentially restored by proper chemical intervention. As the topology of the AGT active site is well characterized by PLP binding (Figure S4) and the presence of several highly conserved residues, mutations that directly affect AGT enzymatic activity are predictable and their effect can be verified by AGT activity tests [4],[5]. For mutants of this category, it has been shown that pyridoxine treatment may lead to additional active-site stabilization, resulting in a reduction of clinical symptoms such as calcium oxalate crystallization and an increasing preservation of renal function [36].
However, prior to this work, a rational basis for predicting mutations involved in the loss of peroxisomal targeting has been largely missing. A paradigm pathogenic mutation within this category is the G170R/V336D variant located on the AGXT-Mi allele [36], which creates a serious disease phenotype. For this type of mutation, which predominantly affects peroxisomal targeting, it is desirable to identify compounds that would lead to a gain in AGT binding to the Pex5p receptor, by targeting identified AGT–Pex5p interface areas such as the PTS1 site, the extended PTS1 site, and relevant Pex5p-binding surfaces from the AGT C-terminal domain (Figures 1–2 and S5). The knowledge of designed AGT variants compromised in Pex5p recognition, such as AGT(Y330A), may be useful for targeting the restoration of AGT–Pex5p recognition to wild-type levels. The observed limited flexibility in the non-PTS1 binding areas and the lack of optimized interactions within the PTS1 binding site of AGT (Figures S5 and S6) may provide a knowledge-based system by which Pex5p receptor binding can be maximized by compounds that have the potential to improve protein-protein interactions.
Human AGT (major allele haplotype) and human Pex5p(C) (residues 315–639) were expressed from a modified pET24d vector (G. Stier, EMBL Heidelberg) in Escherichia coli BL21(DE3) RIL. The two genes were amplified by polymerase chain reaction (PCR) using primers containing NcoI and KpnI restriction sites, respectively (Table S4). Following the digestion of the PCR products and the vector, the two constructs were created by ligation (Rapid Ligation Kit, Fermentas). Cultures were grown in Lysogeny Broth medium containing 50 mM Tris pH 7.5 and 1% (w/v) glucose, and induced mid-log phase with 0.5 mM isopropyl-β-D-thiogalactopyranosid overnight at 21°C. Both proteins contained an N-terminal hexahistidine–glutathione S-transferase fusion, which is cleavable with tobacco etch virus (TEV) protease. The cleared lysate was loaded onto a nickel-nitrilotriacetic acid column and the purified proteins were eluted with 50 mM Tris pH 8.0, 150 mM NaCl, 2 mM ß-mercaptoethanol, and 500 mM imidazole. Fusion proteins were cleaved with tobacco etch virus protease overnight at 4°C, along with dialysis into 50 mM Tris pH 8.0, 150 mM NaCl, 2 mM ß-mercaptoethanol, and 20 mM imidazole. The samples were then applied to a nickel-nitrilotriacetic acid column and the flow-through was collected. As a final purification step, gel filtration was performed using a Superdex 75 (16/60) column (GE Healthcare).
In vivo analysis of EGFP-AGT was carried out with the expression vector pEGFP-AGT, which was derived from subcloning a PCR amplification product of AGT into the pEGFP-C1 plasmid (Clontech). Point mutations were introduced into pEGFP-AGT by using the Quickchange XL Site Directed Mutagenesis Kit (Stratagene). All primers are listed in Table S3.
AGT point mutants that were tested in vitro were subcloned into a pET151 D-TOPO vector. Expression and purification of these proteins was performed as described above.
The Pex5p(C)–AGT complex was formed by mixing purified Pex5p(C) and AGT and confirmed by analytical gel filtration and static light scattering, using a MiniDAWN instrument (Wyatt). Specific activity measurements of AGT in the presence and absence of Pex5p were performed as described previously [37],[38], using the following concentrations: 100 mM potassium phosphate pH 8.0, 0.15 mM PLP, 10 mM glyoxylate, and 150 mM alanine. To confirm specific binding of the cofactor to the recombinant enzyme, we recorded absorption spectra between 300 and 600 nm. All measurements were performed on an Infinity 1000 spectrophotometer (Tecan).
Pex5p(C) and AGT were mixed in a 3∶2 molar ratio and concentrated to 5 mg/ml. Crystals were obtained by submitting a mix of 1 µl protein and 1 µl reservoir solution, comprising 0.1 M Bis-Tris (pH 5.3), 0.15 M LiSO4, 17% [w/w] PEG3350, to hanging drop vapor diffusion at 20°C. Streak seeding of a drop with 2.5 mg/ml protein concentration was used to obtain single large crystals.
X-ray data were collected at BM14.1 at ESRF, Grenoble. Data were processed with MOSFLM [39] and scaled with SCALA [40]. Five percent of the reflections were randomly selected for cross-validation. The structure of the Pex5p(C)–AGT complex was solved by molecular replacement using the coordinates of apo-AGT (PDB code: 1H0C) and the Pex5p–SCP2 complex (PDB code: 2C0L) as search models with the program PHASER [41]. REFMAC [42] was used to refine the structure, applying translation/libration/screw parameterization [43]. Manual building and structure analysis were carried out in COOT [44]. The structure quality was assessed with MOLPROBITY [45]. Programs of the CCP4 package [46] were used for structure manipulation, analysis, and validation. The coordinates of the structure have been deposited in the Protein Data Bank (code: 3R9A). Tilt and twist angles were calculated using MOD22 [20].
All proteins were dialyzed against 100 mM HEPES (pH 7.5), 150 mM NaCl, and 2 mM ß-mercaptoethanol. ITC measurements were conducted on a MicroCal VP-ITC using 25–46 µM AGT as a sample and 250–460 µM Pex5p(C) as a titration ligand. Experiments were performed at 25°C. Pex5p(C) was injected in volumes of 10 µl in a total of 27 steps, resulting in a 2-fold excess of AGT at the end of each titration experiment. Ligand heating effects by dilution were subtracted, and data were fitted using MicroCal Origin 5.0.
Circular dichroism experiments were performed on a J-810 spectropolarimeter (Jasco). Proteins were dialyzed into 10 mM potassium phosphate (pH 8.0) and 1 mM dithiothreitol. Far-UV spectra were recorded between 190 and 260 nm, using a 1 mm cuvette and a concentration of 0.15–0.22 mg/ml protein, as determined by specific absorbance at 280 nm. The machine settings were 1 nm bandwidth, 1 s response, 1 nm data pitch, and 100 nm/min scan speed. Secondary structure content was calculated with the Diochroweb server [47], using the analysis program CDSSTR and reference set 4. All circular dichroism data presented are the averages of three separate experiments.
Human fibroblast cells (strain GM5756T) were cultured as described previously [15] and transfected with pEGFP-AGT variants, using FuGENE 6 Transfection Reagent (Roche Diagnostics). At 24 h after transfection, cells were fixed with 3% paraformaldehyde, solubilized with 1% Triton X-100, and subjected to immunofluorescence microscopy. Polyclonal rabbit antibodies against Pex14p were used to label peroxisomes [48]. Secondary antibodies were conjugated with Alexa Fluor 594 (Invitrogen, Germany). All micrographs were recorded on an Axioplan 2 microscope (Zeiss) with a Plan-Apochromat 63×/1.4 oil objective and an Axiocam MR digital camera and were processed with AxioVision 4.6 software (Zeiss). Statistical analysis was carried out from at least three independent transfections of each AGT expression plasmid. Based on the appearance of the AGT fluorescence pattern, around 100 cells of each experiment were visually categorized into three classes: (i) predominant peroxisomal localization, (ii) mostly cytosolic, or (iii) forming aggregates, as indicated by fluorescent plaques over cytosolic background.
|
10.1371/journal.pgen.1002521 | Genetic and Functional Analyses of SHANK2 Mutations Suggest a Multiple Hit Model of Autism Spectrum Disorders | Autism spectrum disorders (ASD) are a heterogeneous group of neurodevelopmental disorders with a complex inheritance pattern. While many rare variants in synaptic proteins have been identified in patients with ASD, little is known about their effects at the synapse and their interactions with other genetic variations. Here, following the discovery of two de novo SHANK2 deletions by the Autism Genome Project, we identified a novel 421 kb de novo SHANK2 deletion in a patient with autism. We then sequenced SHANK2 in 455 patients with ASD and 431 controls and integrated these results with those reported by Berkel et al. 2010 (n = 396 patients and n = 659 controls). We observed a significant enrichment of variants affecting conserved amino acids in 29 of 851 (3.4%) patients and in 16 of 1,090 (1.5%) controls (P = 0.004, OR = 2.37, 95% CI = 1.23–4.70). In neuronal cell cultures, the variants identified in patients were associated with a reduced synaptic density at dendrites compared to the variants only detected in controls (P = 0.0013). Interestingly, the three patients with de novo SHANK2 deletions also carried inherited CNVs at 15q11–q13 previously associated with neuropsychiatric disorders. In two cases, the nicotinic receptor CHRNA7 was duplicated and in one case the synaptic translation repressor CYFIP1 was deleted. These results strengthen the role of synaptic gene dysfunction in ASD but also highlight the presence of putative modifier genes, which is in keeping with the “multiple hit model” for ASD. A better knowledge of these genetic interactions will be necessary to understand the complex inheritance pattern of ASD.
| Autism spectrum disorders (ASD) are a heterogeneous group of neurodevelopmental disorders with a complex inheritance pattern. While mutations in several genes have been identified in patients with ASD, little is known about their effects on neuronal function and their interaction with other genetic variations. Using a combination of genetic and functional approaches, we identified novel SHANK2 mutations including a de novo loss of one copy of the SHANK2 gene in a patient with autism and several mutations observed in patients that reduced neuronal cell contacts in vitro. Further genomic analysis of three patients carrying de novo SHANK2 deletions identified additional rare genomic imbalances previously associated with neuropsychiatric disorders. Taken together, these results strengthen the role of synaptic gene dysfunction in ASD but also highlight the presence of putative modifier genes, which is in keeping with the “multiple hit model” for ASD. A better knowledge of these genetic interactions will be necessary to understand the complex inheritance pattern of ASD.
| Autism spectrum disorders (ASD) are characterized by impairments in reciprocal social communication and stereotyped behaviors [1]. The prevalence of ASD is about 1/100, but closer to 1/300 for typical autism [2]. ASD are more common in males than females, with a 4∶1 ratio. Previously, twin and family studies have conclusively described ASD as the most “genetic” of neuropsychiatric disorders, with concordance rates of 82–92% in monozygotic twins versus 1–10% in dizygotic twins [3], but a recent study finds evidence for a more substantial environmental component [4]. In the absence of Mendelian inheritance patterns, ASD were first considered to be polygenic, i.e., a disorder caused by multiple genetic risk factors, each of weak effect. More recently, an alternative model was proposed that considered ASD as a group of disorders caused by heterogeneous genetic risk factors influencing common neuronal pathways [5], [6]. It was supported by the identification of apparently monogenic forms of ASD, each affecting a limited number of patients (1–2% for the most replicated genes) [7]–[14]. In this model, eventually a single highly penetrant mutation would be sufficient to produce ASD. However, the occurrence of two or more deleterious copy number variants (CNV) or mutations in a subset of patients also suggested that independent loci could act in concert to induce the development of ASD [9], [13]–[16]. In line with these findings, the recent observation that patients with a deletion at 16p12.1 were more likely to carry an additional large CNV agrees with a “two-hit model” for developmental disorders [17].
The genetic causes of ASD are diverse [18], but the main category of genes associated with the disorder is related to the development and function of neuronal circuits [6], [19]. Mutations of genes coding for synaptic cell adhesion molecules and scaffolding proteins, such as neuroligins (NLGN), neurexins (NRXN) and SHANK, have been recurrently reported in patients with ASD [7]–[10], [13], [14], [20]. These proteins play a crucial role in the formation and stabilization of synapses [21], as well as in synaptic homeostasis [22]. SHANK2 and SHANK3 code for scaffolding proteins located in the postsynaptic density (PSD) of glutamatergic synapses. Deletions of ProSAP2/SHANK3 at chromosome 22q13 are one of the major genetic abnormalities in neurodevelopmental disorders [20], and mutations of ProSAP2/SHANK3 have been identified in patients with ASD, intellectual disability (ID) and schizophrenia [7], [23]–[25]. Mutations of ProSAP1/SHANK2 have also recently been reported in both, ASD and ID [9], [26]. The difference in clinical outcome of mutation carriers has been attributed to the presence of still uncharacterized additional genetic, epigenetic and/or environmental factors [27].
In order to better understand the role of the NRXN-NLGN-SHANK pathway in ASD, we first aimed to describe SHANK2 isoform expression in different tissues of healthy individuals. To investigate the role of this pathway in ASD, we screened for SHANK2 CNVs and coding mutations in a large sample of patients with ASD and controls. We provide genetic and functional evidence that SHANK2 is associated with ASD, and that its mutations affect the number of synapses. Additionally, we report the co-occurrence of SHANK2 de novo deletions and inherited CNVs altering neuronal genes, suggesting that epistasis between specific loci in the genome could modulate the risk for ASD.
In order to characterize all isoforms of SHANK2, we scanned genomic databases for specific Expressed Sequence Tags (ESTs) and spliced isoforms. The human SHANK2 gene (NM_012309.3) spans 621.8 kb and contains 25 exons (Figure 1). The longest SHANK2 isoform (SHANK2E, AB208025) contains ankyrin (ANK) repeats at the N-terminus, followed by a Src homology 3 (SH3) domain, a PSD95/DLG/ZO1 (PDZ) domain, a proline-rich region and a sterile alpha motif (SAM) domain at its C-terminus region. All these domains are involved in protein-protein interactions that bridge glutamate receptors, scaffolding proteins and intracellular effectors to the actin cytoskeleton [28], [29]. Two additional isoforms, ProSAP1A (AB208026) and ProSAP1 (AB208027), originating from distinct promoters, were previously detected in the rat [30], [31]. Finally, the shortest isoform (AF141901), also originally described in the rat, results in premature termination of the transcription before the SAM domain due to an alternative 3′ end in exon 22 [32] (Figure 1A). To validate these SHANK2 isoforms in humans, we used specific RT-PCRs and sequencing (Figure 1B). Almost all tissues tested (brain, liver, placenta, kidney, lung, pancreas and lymphoblastoid cell lines) expressed SHANK2 mRNA, except heart and skeletal muscle, for which no expression was detected. We observed inter-individual differences in the relative amount of SHANK2 mRNA that were confirmed by using independent RT-PCRs and primers (not shown). Such differences have been previously reported for other synaptic genes such as NLGN1-4Y, PCHD11X/Y, and SHANK3 [7], [8], [33] and might be the consequence of polymorphisms located in specific regulatory sequences and/or activity dependent expression of this family of post-synaptic proteins [34]. Notably, exons 19, 20 and 23 were found to be expressed only in brain in all individuals tested (Figure 1C). Such brain specific splicing has been already observed for exon 18 in SHANK3 [7], which is similar to exon 19 and 20 in SHANK2. These ‘brain-specific exons’ code for a region in SHANK2/3 located between the PDZ and the proline rich domains. Finally, in contrast to previous results [26], we detected the longest SHANK2E isoform in all independent samples of human brain, with high expression in the cerebellum (Figure 1, Figure S1). This Shank2E isoform was also expressed in the cerebellum and in the liver of rat embryo at E19 (Figure S1).
Berkel et al. 2010 recently identified two independent de novo SHANK2 deletions in two patients, one with ID and another one with ASD [26]. In addition, whole genome analysis performed by the Autism Genome Project (AGP) using Illumina 1M single nucleotide polymorphism (SNP) arrays detected one additional de novo SHANK2 deletion in a patient (6319_3) with ASD [9] (the second patient described by the AGP, 5237_3, is patient SK0217-003 reported in Berkel et al. 2010 [26]). Recently, a 3.4 Mb de novo deletion including SHANK2 was observed in a female patient with speech and developmental delay [35]. To follow up on these results, we genotyped an independent sample of 260 patients with ASD using Illumina 1M Duo SNP arrays (Table S1). In this sample, we detected a 421.2 kb deletion within SHANK2 in patient AU038_3 with autism and moderate ID (see patient section in Materials and Methods, and Table S2). The deletion covered twelve exons (E5–E16) and altered all SHANK2 isoforms (Figure 2A). No other deleterious variants in the remaining copy of SHANK2 were detected by sequencing. The parents did not carry the deletion, indicating a de novo event. The deletion was validated by quantitative PCR analysis using DNA from an independent blood sample from all members of the family and SNP analysis indicated that the deletion originated on the maternal chromosome (Figure S2). SHANK2 deletions were absent in more than 5000 controls [9], [26] and not listed in the Database of Genomic Variants (DGV; http://projects.tcag.ca/variation/).
To probe for additional mutations, we first sequenced all exons of the longest SHANK2E isoform in 230 patients with ASD and 230 controls. We then sequenced an additional sample of 225 patients and 201 controls (Table S1) for the ProSAP1A isoform that corresponds to the major SHANK2 isoform in the brain. Since we screened all SHANK2 isoforms, we used a nomenclature including the SHANK2E isoform that differed from Berkel et al. 2010 [26]. Within the 9 coding exons specific to SHANK2E, we identified R174C (rs7926203) listed in dbSNP in 2 independent patients with ASD and R185Q in one patient with ASD. For this isoform, no variant was identified in the control sample. Within the ProsSAP1A isoform, we identified 24 non-synonymous variations. When these results are integrated with those obtained by Berkel et al. 2010, a total of 40 variants of ProsSAP1A including 3 already reported in dbSNP were identified (Figure 2B, Table 1, Figure S3). Only two variants (Y967C and R569H) with MAF>1% are detected and there is no enrichment of rare variants of SHANK2 (MAF<1%) in patients with ASD compared with controls. Because variants affecting conserved amino acids in the SHANK proteins are most likely to have a functional effect, we tested whether there was an enrichment of these variants in patients compared to controls. The alignment of the SHANK protein sequences and the conservation of the variants are indicated in the Table S5. In both mutation screening studies, the first performed by Berkel et al. 2010 and the second presented here, we observed an enrichment of variants affecting conserved amino acids in patients compared with controls (Figure 2C, Table S5 and Table S7). Overall, 12 of 15 (80%) of the variants identified only in the patient sample affected conserved amino acids compared with only 6 of 17 (35.3%) in controls (Fisher's exact test 1-sided, P = 0.013, OR = 6.83, 95% IC = 1.19–53.40). Because several independent patients carried these variants (Table 1), the enrichment is even more significant when the number of carriers was considered. The variants affecting conserved amino acids were observed in 29 of 851 (3.4%) patients and in 16 of 1090 (1.5%) controls (Fisher's exact test 1-sided, P = 0.004, OR = 2.37, 95% CI = 1.23–4.70). A total of 8 variants were identified in patients and controls. Among these 8 variants, 2 affected conserved amino acids (R818H and S557N). The variant S557N was observed in 9 of 851 (1.06%) independent families with ASD and in 3 of 1090 (0.28%) controls (Fisher's Exact Test one sided, P = 0.029, OR = 3.87; 95% CI = 0.96–22.29). It affects a conserved serine with a high probability of being phosphorylated and located in the SH3 domain of all SHANK proteins. This domain binds to GRIP and b-PIX, two proteins linking SHANK to glutamate AMPA receptors and actin skeleton, respectively [36]. In our initial mutation screen, R818H was observed in 5 of 230 patients with ASD and 0 of 230 controls. In order to determine if R818H was more frequent in the patients with ASD, we screened an additional sample of 3020 individuals with ASD, 1783 controls from European descent, and the Human Genome Diversity Panel (HGDP) control dataset (Table S3 and S4). R818H was virtually absent outside Europe and had the highest allelic frequency (2.37%) in Finland, but overall its frequency was not higher in patients with ASD compared with controls (ASD 32/3250 (1.0%); controls 27/2030 (1.33%); Fisher's exact test 2-sided, P = 0.28) (Table S3).
Finally, and unexpectedly, during this additional mutation screening, we detected a variation (IVS22+1G>T) altering the consensus donor splice site of exon 22 in a Swedish control, SWE_Q56_508 (Figure 3A). This variant was predicted to disrupt all SHANK2 isoforms by deleting the proline rich and the SAM domain, except for the shortest isoform AF141901, where the mutation is located in the open reading frame (ORF) and should lead to a G263V change. This variant was not observed in 1786 patients or 1407 controls, and is not listed in dbSNP. This control female was part of a previous epidemiological study [37] and had been extensively examined for anthropometrics and cardiovascular risk factors such as blood pressure and levels of all major hormones. In addition, she was ascertained for axis I psychiatric disorders and personality traits using the Temperament and Character Inventory (TCI) [38] and the Karolinska Scales of Personality (KSP) [39]. Notably, despite the predicted deleterious effect of the mutation, this subject had no major somatic or psychiatric health problems. Regarding personality traits, none of her scores for TCI items were different from those found in the general population. KSP assessment showed that her scores for neuroticism (51.3), nonconformity/aggressiveness (56.7), and psychoticism (50.5) were not different from the general population (mean ± SD = 50±10). However, she displayed a high score (61.4) for the extraversion factor, and for one of its subscales, monotony avoidance [40].
In order to establish the functional impact of SHANK2 variations, we performed expression studies in primary neuronal cell cultures after over-expression of wild-type vs. mutant ProSAP1A/Shank2A cDNA (Figure 4). All the variants (n = 16) identified by our first screen of 230 patients and 230 controls were tested: 5 were identified only in patients (V717F, A729T, G1170R, D1535N and L1722P), 6 were detected in patients and controls (S557N, R569H, K780Q, R818H, Y967C and P1586L) and 5 were only found in controls (L629P, A822T, V823M, R1290W and Q1308R). All variants were predicted as damaging by Polyphen2 DIV except Q1308R identified only in controls and predicted as benign [41]. In the patient sample, 5/5 variants affected conserved amino acids in the SHANK proteins compared with only 2/6 in the group of variants identified in patients and controls, and 2/5 in the control group. All mutation sites were introduced into the rat ProSAP1A cDNA and confirmed by sequencing. The effect of the Shank2 variants was further investigated in cultured hippocampal neurons. Upon transfection, Western blot analysis revealed that the different GFP-Shank2 fusion proteins were expressed with the expected size (Figure S4). Results from quantification showed that none of the variants affected the cluster formation of Shank2 protein along the dendrites, the number or the general branching pattern of dendrites (Figure S4). In contrast, 8 variants identified in patients or in patients and control group reduced significantly the density of Shank2 positive synapses per 10 µm dendrite length compared with wild-type GFP-Shank2 (Figure 4). None of the variants identified in controls only were shown to have a significant effect. After Bonferonni correction for the 16 tests, 4 variants significantly affected synapse density. Among these variants, A729T, G1170R and L1722P were identified only in patients and S557N was observed more frequently in patients than in controls. As expected, the majority of the variants leading to reduced synaptic density altered conserved amino acids present in SHANK proteins (7/8), and the majority of variants not changing synaptic density affected amino acids present only in SHANK2 (7/8). The 4 strongly associated (after Bonferroni correction) variants affecting synaptic density modified conserved amino acids in other SHANK proteins. Because the significant threshold of 0.05 is arbitrary, we additionally tested for the quantitative effect of the variant on synaptic density as a continuous trait (Figure S4) and found that variants identified in patients were associated with a significant decrease of synapse density in vitro compared with those shared by patients and controls (Student's t test 2-sided P = 0.022) or those only detected in the controls (Student's t test 2-sided P = 0.0013). As expected, variants affecting conserved amino acids were associated with a higher reduction of synapse density in vitro (Student's t test 2-sided P = 0.014).
To test if additional CNVs may modulate the impact of SHANK2 mutations in the development of ASD, we analyzed the CNVs of patient AU038_3 and the two patients (5237_3 and 6319_3) carrying SHANK2 de novo deletions previously identified by Pinto et al. [9] (Figure 5 and Table S6). In addition to our CNV study group of 260 patients with ASD and 290 controls, we used the CNV dataset from the AGP, which includes 996 patients with ASD and 1287 controls genotyped with the Illumina 1M SNP array [9]. Remarkably, all three patients with SHANK2 de novo deletions also carried rare inherited genetic imbalances at chromosome 15q11–q13 (Figure 6), a region associated with Angelman syndrome, Prader-Willi syndrome and other neuropsychiatric disorders, including ASD [42]–[61]. This region is characterized by recurrent deletions/duplications with breakpoints generally located within five segmental duplications named BP1 to BP5, which act as hotspots of non-allelic homologous recombination. In the BP5 region, patients AU038_3 and 5237_3 carried the same 496 kb duplication of the nicotinic receptor CHRNA7 gene (29.8–30.3 Mb, hg 18; maternally inherited in patient AU038_3 and paternally inherited in patient 5237_3). This small CHRNA7 duplication was present in 13 of 1257 patients with ASD (1.03%) compared with 9 of 1577 controls (0.57%) (Fisher's exact test, 2-sided P = 0.19).
These duplications are considered of uncertain clinical significance since they were previously detected at similar frequencies in patients with epilepsy (6 of 647, 0.93%), in controls (19 of 3699, 0.51%) [50], and in subjects referred for chromosomal microarray analysis (55 of 8832, 0.62%) [51]. In contrast, larger 15q13.3 deletions (∼1.5 Mb) between BP4 and BP5, encompassing the CHRNA7 locus have been associated with disorders such as ID, epilepsy, schizophrenia, and ASD [43], [46]–[48], [50], [52]–[54], [57]–[59]. In the BP4 region, the same two patients AU038_3 and 5237_3 also carried two independent deletions of the rhoGAP ARHGAP11B gene. Loss of ARHGAP11B was detected in 8 of 1257 patients with ASD (0.64%) and in 4 of 1577 controls (0.25%) (Fisher's exact test, 2-sided P = 0.15). Patient 5237_3 carried a large deletion (235.2 kb) of the full gene, transmitted by the mother. Patient AU038_3 carried a smaller deletion of 49.8 kb of the first two exons, transmitted by the mother. Both deletions overlap the segmental duplications of BP4 and have been reported to accompany the majority of microduplications involving CHRNA7 [51]. However, in patient 5237_3, the two CNVs are present on distinct parental chromosomes since the CHRNA7 duplication and the ARHGAP11B deletion are paternally and maternally inherited, respectively. Finally, the third patient, 6319_3, carried a paternally-inherited BP1-BP2 deletion of 468 kb, removing NIPA1, NIPA2, CYFIP1, and TUBGCP5. This deletion was observed in 4 of 1257 patients with ASD (0.32%) and in 4 of 1577 controls (0.25%) (Fisher's exact test, 2-sided P = 0.74). The BP1-BP2 deletion is associated with phenotypic variability and has been reported in individuals with neurodevelopmental disorders [20], schizophrenia [53], [60], ASD [44]–[46], [49], and epilepsy [61]. In a recent screen for large CNVs (>400 kb) performed on 15,767 children with ID and various congenital defects, and 8,329 unaffected adult controls [20], deletions affecting CYFIP1, NIPA1, NIPA2 and TUBGCP5 were associated with neurodevelopmental disorder (P = 4.73×10−6), epilepsy (P = 1.48×10−3) and autism (P = 1.99×10−2).
Several additional CNVs also altered compelling candidate genes for susceptibility to ASD. In patient AU038_3 we detected a previously unreported paternally inherited intronic duplication of CAMSAP1L on chromosome 1q32.1, coding for a calmodulin regulated spectrin-associated protein highly expressed in the brain. Patient 5237_3 carried a de novo deletion altering the coding sequence of the tyrosine phosphatase DUSP22 on chromosome 6p25.3 and a maternally inherited intronic duplication of NLGN1 on chromosome 3q26.3 [9]. These CNVs were observed at similar frequencies in patients with ASD compared with controls. DUSP22 deletions were observed in 8 of 1257 patients with ASD (0.64%) and in 14 of 1577 controls (0.89%), while NLGN1 intronic duplications were observed in 60 of 1257 patients with ASD (4.77%) and in 62 of 1577 controls (3.93%). Finally, patient 6319_3 carried an unreported maternally inherited intronic deletion of contactin CNTN4, a gene on chromosome 3p26.3 associated with ASD [62], as well as a paternally inherited deletion within the protocadherin PCDHA1-10 gene cluster on chromosome 5q31.3. Interestingly, this deletion removes the first exon of both PCDH8 and PCDH9 and was significantly less frequent in patients with ASD compared with controls (ASD: 62 of 1257; controls: 132 of 1577; Fisher's exact test, 2-sided P = 0.0003; OR = 0.57; 95% CI = 0.41–0.78).
We also analyzed the genome of the Swedish control SWE_Q56_508 carrying the SHANK2 splice mutation using the Human Omni2.5 BeadChip array from Illumina (Figure 3B). Two close duplications on 2p25.3 were detected, altering four genes, LOC391343, SNTG2, PXDN and MYT1L. The inheritance of these two duplications could not be investigated, because DNA samples from the parents were not available. However, 2 of 1577 controls also carried of the same close duplications, suggesting that these CNVs are located on the same chromosome. Among the affected genes, syntrophin-γ2 (SNTG2) and myelin transcription factor 1-like (MYT1L) are expressed in the brain. Alterations of SNTG2 and MYT1L have been previously reported in patients with ASD [20], [63], [64] and schizophrenia [65], respectively. SNTG2 is a scaffolding protein interacting with the NLGN3/4X proteins [66] and a component of the dystrophin glycoprotein complex [67]. MYT1L is a myelin transcription factor required to convert mouse embryonic and postnatal fibroblasts into functional neurons [68].
The identification of mutations in synaptic proteins such as NRXN1, NLGN3/4X and SHANK2/3 has demonstrated that a synaptic defect might be at the origin of ASD [5], [6]. Here we confirm the presence of SHANK2 de novo deletions in individuals with ASD, with a prevalence of 0.38% (1/260) in our cohort of ASD patients analyzed with the Illumina 1M SNP array. This frequency is similar to the one reported previously by the AGP in a larger sample of 996 patients with ASD (0.2%) [9]. SHANK2 deletions altering exons were not detected in controls, in agreement with previous findings [9], [26]. As reported for SHANK3 [7], no other coding variations were detected in the remaining SHANK2 allele of the deletion carriers, suggesting that, in some individuals, a de novo deletion of a single allele of SHANK2 might be sufficient to increase the risk for ASD. In one case, a patient carried two rare SHANK2 variants predicted as deleterious and inherited from different parents, indicating that they were separate alleles.
For the remaining SHANK2 variants, patients were heterozygous for non-synonymous rare variations inherited from one of their parents (Figure S3). Since parents were apparently asymptomatic, the causative role of these variants in ASD remains difficult to ascertain. However, we observed a significant enrichment of SHANK2 variants affecting conserved amino acids in patients with ASD compared with controls. This was also the case in the previous mutation screening by Berkel et al. 2010 [26]. The majority of the variants affecting conserved residues and identified in the patients were shown to alter the ability of SHANK2 to increase the number of synapses in vitro. Importantly, the assays performed in this study show that the variants can potentially impact on the function of the protein, but they do not confirm that they have deleterious effects on neuronal function in vivo in people that carry them. However, these results are consistent with previous findings showing that inherited variants of SHANK2 and SHANK3 cause synaptic defects in vitro [7], [69], [70]. Recently, Berkel et al. 2011 showed that two inherited (L1008_P1009dup, T1127M) and one de novo (R462X) SHANK2 mutations identified in patients with ASD affect spine volume and reduced Shank2 cluster sizes [70]. This deleterious effect was also observed in vivo since mice expressing rAAV-transduced Shank2-R462X present a specific long-lasting reduction in miniature postsynaptic AMPA receptor currents [70].
In patients, the only feature associated with carriers of SHANK2 mutations compared with other patients was a trend for low IQ (P = 0.025, OR = 3.75, 95% CI = 1.1–20.0) (Table S8). But, as observed for SHANK3 mutations, this correlation could differ from one individual to another (i.e. the patient with a SHANK2 de novo stop mutation reported by Berkel et al. 2010 presented with high-functioning autism [26]).
Our result also showed that potentially deleterious SHANK2 variants were detected in a heterozygous state in parents and in the general population without causing severe phenotypic consequences. Indeed, we showed that almost 5% of the Finnish population is heterozygous for the SHANK2 R818H variation, which modifies a conserved amino acid and is associated with lower synaptic density in vitro. Furthermore, we identified a SHANK2 splice site mutation in a control female without any apparent psychiatric disorders. Similarly, two frame-shift mutations and one splice site mutation of SHANK2 are listed in dbSNP and in the 1000 genomes project [71]. These nonsense variations should be interpreted with caution since none of them has been validated by Sanger sequence technology. Taken together, variants affecting conserved amino acids of SHANK2 might act as susceptibility variants for ASD, but, in some cases, additional genetic, epigenetic or environmental factors seem to be necessary for the emergence of the disorder.
In order to detect risk and protective genetic factors, we analyzed the CNV burden of the individuals carrying deleterious variations of SHANK2. Notably, the three ASD patients with de novo SHANK2 deletions also carried CNVs on chromosome 15q11–q13, a region associated with ASD [43], [47], [48], [50]–[52], [72]. In contrast, the patient reported by Berkel et al. 2010, who did not meet all the diagnostic criteria for ASD, seemed to have no CNV at chromosome 15q [26]. Although the probability to observe the co-occurrence of a de novo SHANK2 deletion and a duplication of CHRNA7 at 15q is very low, two of the three patients carrying a de novo SHANK2 deletion also carried the CHRNA7 duplication. While the numbers are small, this finding could suggest epistasis between these two loci. The role of CHRNA7 in ASD was recently supported by the observation of low levels of CHRNA7 mRNA in the post-mortem brain from patients with ASD [73]. Interestingly, it was also found that, in contrast to the gene copy number, the transcript levels CHRNA7 were reduced in neuronal cells [74] or brain samples with maternal 15q duplication [75]. Finally, functional studies have shown that NLGN and NRXN, which belong to the same synaptic pathway, are key organizers of the clustering of nicotinic receptors at the synapse [76]–[78]. Therefore the co-occurrence of a deletion of SHANK2 and a duplication of the nicotinic receptor CHRNA7 could act together within the same pathway to increase the risk of ASD in patients AU038_3 and 5237_3. In patient 6319_3 carrying the BP1–BP2 deletion, several genes might also play a role in the susceptibility to ASD. Among them, NIPA1 and TUBGCP5 encoding a magnesium transporter and a tubulin gamma associated protein, respectively, are highly expressed in the brain. However, the most compelling candidate in the deleted region is CYFIP1 [45], [53], which codes for a binding partner of FMRP, the protein responsible for fragile X syndrome. Both CYFIP1 and FMRP are involved in the repression of synaptic translation [79], one of the major biological mechanisms associated with ASD [80]. Therefore, the co-occurrence of a loss of one copy of SHANK2 and CYFIP1 might increase the risk of abnormal synaptic function in patient 6319_3.
If some individuals have a higher risk to develop ASD when a deleterious SHANK2 variant is present, others individuals may experience a protective effect by additional genetic factors. For example, control SWE_Q56_508 carried a SHANK2 splice mutation, but clinical examination revealed no major disorders. In addition, this control individual also carried a partial duplication of SNTG2 and MYT1L. Based on a single control subject, it is not possible to formally prove that these additional hits at SNTG2 and/or MYT1L acted as suppressor mutations, counteracting the phenotypic effects of the SHANK2 splice mutation. However, the encoded proteins may interact with the NRXN-NLGN-SHANK pathway. Both SNTG2 and SHANK2 are scaffolding proteins localized in actin rich structures [81]–[83] and bind directly to neuroligins [66]. Furthermore, mutations of NLGN3/4X identified in patients with ASD decrease their protein binding to SNTG2 [66]. In addition, MYT1L is a myelin transcription factor that is sufficient, with only two other transcription factors, ASCL1 and BRN2, to convert mouse embryonic and postnatal fibroblasts into functional neurons in vitro [67]. Therefore, alterations of SNTG2 and/or MYTL1 might modulate synapse physiology and counteract the effect of the SHANK2 splice site mutation. We recently highlighted the key role of synaptic gene dosage in ASD and the possibility that a protein imbalance at the synapse could alter synaptic homeostasis [6]. In the future, animal models should be developed to test whether the effect of a primary mutation in a synaptic protein complex (e.g. Shank2) can be reduced or suppressed by a second mutation (e.g. Sntg2 or Myt1l). A similar suppressor effect has been demonstrated by the decrease of abnormal behavior of the Fmr1 mutant mice carrying a heterozygous mutation of the metabotropic glutamate receptor mGluR5 [84].
In summary, we confirmed that de novo SHANK2 deletions are present in patients with ASD and showed that several SHANK2 variants reduce the number of synapses in vitro. The genomic profile of the patients carrying deleterious de novo SHANK2 deletions also points to a possible genetic epistasis between the NRXN-NLGN-SHANK pathway and 15q11–q13 CNVs. CHRNA7 and CYFIP1 were already proposed as susceptibility genes for neuropsychiatric disorders [43], [45], [49], [51], and our study provides additional support for this association. Therefore, as previously observed for ID [85], our results suggest that the co-occurrence of de novo mutations, together with inherited variations might play a role in the genetic susceptibility to ASD. Finally, our analyses suggest the interesting possibility that deleterious mutations of neuronal genes (e.g. SNTG2 and MYT1L) could potentially counteract the effect of synaptic deleterious mutations (e.g. SHANK2). The identification of risk and protective alleles within the same subject is one of the main challenges for understanding the inheritance of ASD. Initial results from the 1000 genomes project has estimated that, on average, each person carries approximately 250 to 300 loss-of-function variants in annotated genes and 50 to 100 variants previously implicated in inherited disorders [71]. To date, it is not clear how many loci can regulate synaptic homeostasis and how these variants interact with each other to modulate the risk for ASD [6]. A better knowledge of these genetic interactions will be necessary to understand the complex inheritance pattern of ASD.
This study was approved by the local Institutional Review Board (IRB) and written inform consents were obtained from all participants of the study. The local IRB are the “Comité de Protection des Personnes” (Île-de-France Hôpital Pitié-Salpêtrière Paris) for France; the Sahlgrenska Academy Ethics committee, University of Gothenburg for Sweden; the local IRB of the medical faculty of JW Goethe University Frankfurt/Main for Germany; the Committee #3 of the Helsinki University Hospital, Finland; the “Comitato Etico IRCCS Fondazione Stella Maris” at Stella Maris Institute, Calambrone (Pisa), Italy; the “Comitato Etico Azienda Ospedaliera-Universitaria Policlinico-Vittorio Emanuele”, Catania, Italy.
Patients with ASD and analyzed for CNV analysis and/or mutation screening are presented in Table S1. Patients were recruited by the PARIS (Paris Autism Research International Sibpair) study at specialized clinical centers disposed in France, Sweden, Germany, Finland, UK. The Autism Diagnostic Interview-Revised (ADI-R) and Autism Diagnostic Observation Schedule (ADOS) were used for clinical evaluation and diagnosis. In Sweden, in some cases, the Diagnostic Interview for Social and Communication Disorders (DISCO-10) was applied instead of the ADI-R. Patients were included after a clinical and medical check-up with psychiatric and neuropsychological examination, standard karyotyping, fragile-X testing and brain imaging and EEG whenever possible. All patients were from Caucasian ancestry.
The patient AU038_3 with a de novo SHANK2 deletion is an 11.05 year-old boy diagnosed with autism and moderate ID (Table S2). He was the only child of non-consanguineous parents from European descent. His parents had no relevant personal and familial history of psychiatric or medical illness. He was born at 40 weeks of gestation, after normal pregnancy and delivery. Birth weight, length and occipitofrontal head circumference were 2500 g (5th percentile), 48 cm (22nd percentile) and 31 cm (2nd percentile), respectively. Apgar scores were 7 and 10 at 1 and 5 minutes, respectively. In the first year of life, the pediatrician reports did not mention signs of hypotonia. At 2 months, he was operated for an inguinal hernia. Motor acquisition was apparently normal (sitting at 6 months), but with a late acquisition of walking, at 18 months. Speech was severely delayed, without any apparent regressive phase. Only a few words and sentences appeared when he was 4 y and 6.5 y, respectively. His expressive language remained limited to restrictive sentences, mainly dyssyntaxic. A formal diagnostic assessment for autism was performed when he was 11 years old. The scores of the Autism Diagnostic Interview-Revised (ADI-R) domains were: social 24, communication 23, and behavior 6 (cut-offs for autism diagnosis are 10, 8 (verbal autism) and 3, respectively); the age at first symptoms was before 36 months. Cognitive evaluation with the Kaufman Assessment Battery for Children (K-ABC) showed moderate intellectual deficit (composite score 40). He required assistance with basic activities such as eating and dressing. At examination, he had a normal facial appearance, with a prominent chin. General and neurological examinations were normal, except for hypermetropia and astigmatism. High-resolution karyotype, fragile X testing, MLPA analysis of telomeres and microdeletion/microduplication syndromes, and metabolic screening for inherited disorders of metabolism (urine amino acids, mucopolysaccharides and organic acids, uric acid in blood and urine) were all normal. No significant epileptic event was reported on the electroencephalogram.
The two male patients with de novo SHANK2 deletions reported by Pinto et al. 2010 [9] (5237_3 and 6319_3) shared several clinical features with patient AU038_3. Patient 5237_3 is a Canadian subject diagnosed with autism (based on ADI-R and ADOS) associated with below average non verbal IQ (<1st percentile) and language (<1st percentile). He had minor dysmorphic features including 5th finger clinodactyly and several curled toes, and no history of epilepsy. Patient 6319_3 was recruited in the same geographic area as patient AU038_3 (Grenoble, France) and was clinically diagnosed with PDD-NOS. The ADI-R scores were: social 14, communication 8, behaviors 2 (cut-off for autism: 3); with an age at first symptoms <36 months). He had mild ID as evaluated with the WISC-III (full scale IQ 60, performance IQ 60, verbal IQ 67). His language was delayed (first words 24 m, first sentences 48 m), but functional. He had no history of regression or epilepsy. The physical exam was normal, except for large and prominent ears and flat feet; the neurological exam was also normal. Similarly to patient AU038_3, he had hypermetropia.
The control female carrying the splice site mutation (IVS22+1G>T) was part of a cohort of 172 females recruited for a study on obesity, anthropometrics, and cardiovascular risk factors [37]. In addition, these women were assessed for axis I psychiatric disorders and for personality traits using the Temperament and Character Inventory (TCI) [38] and the Karolinska Scales of Personality (KSP) [39]. This subject had no psychiatric disorders and her TCI and KSP scores were similar to those found in the general population.
To define the genomic structure of the human SHANK2 gene, we used the two reference sequence genes from UCSC (NM_012309 and NM_133266), one human mRNA from GenBank (DQ152234) and three Rattus reference sequence genes from UCSC (NM_201350, NM_133441 and NM_133440). SHANK2 is transcribed in four isoforms described in GenBank (AB208025, AB208026, AB208027 and AF141901) and is composed of 25 exons. Transcript analysis of SHANK2 was performed in human brain regions from four independent controls (two females and two males) and in human tissues (heart, brain, placenta, lung, liver, skeletal muscle, kidney, pancreas and B lymphoblastoid cell lines) using the Clontech Multiple Tissue cDNA panel (Clontech). Total RNA was isolated from control human brain tissues by the acid guanidinium thiocyanate phenol chloroform method and reverse transcribed by oligodT priming using SuperScript II Reverse Transcriptase (Invitrogen). The PCR was performed with HotStar Taq polymerase (Qiagen) and the protocol used was 95°C for 15 min, followed by 40 cycles at 95°C for 30 sec, 55 to 58°C for 30 sec, 72°C for 30 sec to 1 min, with a final cycle at 72°C for 10 min. PCR primers were designed to detect the ANK domain, the SH3 domain, the PDZ domain, and the SAM domain in order to distinguish the four SHANK2 isoforms and are indicated in Table S11. All RT-PCR products were directly sequenced. The expression of SHANK2E isoform was also studied by SYBR-Green real-time PCR approach. The fluorescence was read with the Applied Biosystems 7500 Real-Time PCR System. Each assay was conducted in three replicates. GAPDH was used for the ΔCt calculation and total brain was used as the reference for relative quantification calculation (RQ). The relative RQ of transcripts was calculated as 2−ΔΔCT with the magnitude of upper error as 2−(ΔΔCT−SEM)-2−ΔΔCT and the magnitude of lower error as 2−ΔΔCT-2−(ΔΔCT+SEM). The primers specific to SHANK2E isoform are indicated in Table S11. In situ hybridization was performed essentially as described previously [28]. Transcripts encoding the different ProSAP1/Shank2 cDNAs (ProSAP1/Shank2 starting with the PDZ domain, ProSAP1A, starting with the SH3 domain and ProSAP1E/Shank2E, starting with the ankyrin repeats) were detected with isoform specific S35 labeled cDNA antisense oligonucleotides purchased from MWG-Biotech (Ebersberg, Germany) directed against the ATG regions of the different mRNAs. All variants were evaluated for potential pathogenicity using the HumDIV method for rare alleles of PolyPhen2 [41].
DNA was extracted from blood leukocytes or B lymphoblastoid cell lines. The SHANK2 CNV was detected with the Illumina Human 1M-Duo BeadChip, which interrogates 1 million SNPs distributed over the human genome. For the Swedish control SWE_Q56_508 carrying the SHANK2 splice mutation we used the Illumina Human Omni2.5 BeadChip array. The genotyping was performed at the Centre National de Génotypage (CNG) and the Institut Pasteur. Only samples that met stringent quality control (QC) criteria were included: call rate ≥99%; high confidence score log Bayes factor ≥15; standard deviation of the log R ratio (LRR) ≤0.35 and of the B allele frequency (BAF)≤0.13; number of consecutive probes for CNV detection ≥5; CNV size ≥1 kb. When the QC criteria were met, we used two CNV calling algorithms, QuantiSNP [86] and PennCNV [87], and the CNV viewer, SnipPeep (http://snippeep.sourceforge.net/). To obtain high-confidence calls, the CNVs identified by QuantiSNP were validated by visual inspection of the LRR and BAF values. PennCNV was used to confirm inheritance status of the resulting CNV calls. CNVs were validated by quantitative PCR analysis using the Universal Probe Library (UPL) system from Roche. UPL probes were labeled with FAM and the fluorescence was read with the Applied Biosystems 7500 Real-Time PCR System. Each assay was conducted in four replicates for target region probe-set and control region probe-set. Relative levels of region dosage were determined using the comparative CT method assuming that there were two copies of DNA in the control region. The relative copy number for each target region was calculated as 2−ΔΔCT with the magnitude of upper error as 2−(ΔΔCT−SEM)-2−ΔΔCT and the magnitude of lower error as 2−ΔΔCT-2−(ΔΔCT+SEM). UPL probes and primers are indicated in Table S12. For comparisons between patients and controls, statistical significance for each CNV was assessed using a 2-sided Fisher's exact test.
The 24 coding exons of SHANK2 were amplified and sequenced for mutation screening. The PCR was performed on 20–40 ng of genomic DNA template with HotStar Taq polymerase from Qiagen for all exons the protocol used was 95°C for 15 min, followed by 35–40 cycles at 95–97°C for 30 sec, 55–62°C for 30 sec, 72°C for 30 sec to 90 sec, with a final cycle at 72°C for 10 min. Sequence analysis was performed by direct sequencing of the PCR products using a 373A automated DNA sequencer (Applied Biosystems). Genotyping of R185Q, V717F, A729T, R818H, G1170R, D1535N and L1722P was performed by direct sequencing or Taqman SNP Genotyping Assays system from Applied Biosystems designed with Custom TaqMan Assay Design Tool. All primers are indicated in Table S9. Enrichment of SHANK2 variations in the ASD sample compared with controls was assessed using a 1-sided Fisher's exact test (hypothesizing that cases will show an excess of SHANK2 variants compared to controls).
Rat GFP-ProSAP1A (Shank2A) cDNA was mutated according to the human mutations using the site directed mutagenesis kit (Stratagene). The mutagenesis primers were listed in Table S10. We have tested all the variants (n = 16) identified in our first screen of 230 patients with ASD and 230 controls: 5 were detected only in patients (V717F, A729T, G1170R, D1535N and L1722P), 6 were detected in patients and controls (S557N, R569H, K780Q, R818H, Y967C and P1586L) and 5 were only found in controls (L629P, A822T, V823M, R1290W and Q1308R). All mutated amino acids were conserved among human, rat and mouse ProSAP1/Shank2. All cDNAs were sequenced and subsequently tested for expression by Western blot analysis. After expression of the constructs in Cos7 cells, the cell homogenate was separated on a gel, transferred to a nitrocellulose membrane and subsequently protein bands were detected using a rabbit anti-GFP antibody. Thereafter, the cDNAs were transfected into primary hippocampal neurons. Cell culture experiments of rat hippocampal primary neurons (embryonic day 18–21: E18-21) were performed as described previously [88]. In brief, after preparation, hippocampal neurons were seeded on poly-l-lysine (0.1 mg/ml; Sigma-Aldrich, Steinheim, Germany) coated coverslips at a density of 4×104 cells/well (transfection experiments) or 2×104 cells/well (immunological staining). Cells were grown in Neurobasal medium (Invitrogen, Karlsruhe, Germany), complemented with B27 supplement (Invitrogen), 0.5 mM L-glutamine (Invitrogen), and 100 U/ml penicillin/streptomycin (Invitrogen) and maintained at 37°C in 5% CO2. Hippocampal cells were transfected using Lipofectamine 2000, according to the manufacturer's recommendation (Invitrogen). Fluorescence images were obtained using a camera attached to a fluorescence microscope. For immunofluorescence, the primary cultures were fixed with ice cold 4% paraformaldehyde/1.5% sucrose/PBS for 20 min at 4°C and processed for immunohistochemistry. After washing three times with 1× PBS for 5 min at room temperature the cells were permeabilized for 3 min on ice in a buffer containing 0.1% Triton X-100/0.1% Na-Citrate/PBS and washed again three times with 1× PBS. Blocking was performed with 10% fetal calf serum/PBS for 1 h at room temperature followed by incubation with the primary antibody (mouse anti-Bassoon) overnight at room temperature. After a further washing-step the cells were incubated with the secondary antibody coupled to Alexa555 (red) (Molecular Probes, Invitrogen) for 90 min at room temperature, washed first with 1×PBS and then with aqua bidest for 5 min and mounted in Mowiol (with or without DAPI for staining of the nucleus). All animal experiments were performed in compliance with the guidelines for the welfare of experimental animals issued by the Federal Government of Germany, the National Institutes of Health and the Max Planck Society.
In morphological studies, dendrites were considered primary when processes extended directly from the cell body, and secondary when processes branched off primary dendrites. Twenty transfected neurons were chosen randomly for quantification from at least three independent experiments for each construct. Morphometric measurements were performed using Axiovision Zeiss microscope and Axiovision software with a 40× magnification. For the quantification of excitatory synapse number, cells were counterstained with anti-Bassoon antibodies. From randomly chosen transfected neurons, Bassoon-positive spots from primary dendrites were counted and the length of dendrites was measured. The total number of spines was expressed as density per 10 µm length of dendrite. Measured data were exported to Excel software (Microsoft), and the data of each variant were compared by using the Mann-Whitney U test. The comparisons of synaptic density for each phenotypic or conservation categories were performed using the Student's t test.
|
10.1371/journal.ppat.1000187 | Transmission of Vibrio cholerae Is Antagonized by Lytic Phage and Entry into the Aquatic Environment | Cholera outbreaks are proposed to propagate in explosive cycles powered by hyperinfectious Vibrio cholerae and quenched by lytic vibriophage. However, studies to elucidate how these factors affect transmission are lacking because the field experiments are almost intractable. One reason for this is that V. cholerae loses the ability to culture upon transfer to pond water. This phenotype is called the active but non-culturable state (ABNC; an alternative term is viable but non-culturable) because these cells maintain the capacity for metabolic activity. ABNC bacteria may serve as the environmental reservoir for outbreaks but rigorous animal studies to test this hypothesis have not been conducted. In this project, we wanted to determine the relevance of ABNC cells to transmission as well as the impact lytic phage have on V. cholerae as the bacteria enter the ABNC state. Rice-water stool that naturally harbored lytic phage or in vitro derived V. cholerae were incubated in a pond microcosm, and the culturability, infectious dose, and transcriptome were assayed over 24 h. The data show that the major contributors to infection are culturable V. cholerae and not ABNC cells. Phage did not affect colonization immediately after shedding from the patients because the phage titer was too low. However, V. cholerae failed to colonize the small intestine after 24 h of incubation in pond water—the point when the phage and ABNC cell titers were highest. The transcriptional analysis traced the transformation into the non-infectious ABNC state and supports models for the adaptation to nutrient poor aquatic environments. Phage had an undetectable impact on this adaptation. Taken together, the rise of ABNC cells and lytic phage blocked transmission. Thus, there is a fitness advantage if V. cholerae can make a rapid transfer to the next host before these negative selective pressures compound in the aquatic environment.
| The biological factors that control the transmission of water-borne pathogens like Vibrio cholerae during outbreaks are ill defined. In this study, a molecular analysis of the active but non-culturable (ABNC) state of V. cholerae provides insights into the physiology of environmental adaptation. The ABNC state, lytic phage, and hyperinfectivity were concurrently followed as V. cholerae passaged from cholera patients to an aquatic reservoir. The relevance to transmission of each factor was weighed against the others. As the bacteria transitioned from the patient to pond water, there was a rapid decay into the ABNC state and a rise of lytic phage that compounded to block transmission in a mouse model. These two factors give reason for V. cholerae to make a quick transit through the environment and onto the next human host. Thus, in over-crowded locations with failed water infrastructure, the opportunity for fast transmission coupled with the increased infectivity and culturability of recently shed V. cholerae creates a charged setting for explosive cholera outbreaks.
| Diarrheal disease is the second most common cause of death among children under 5 years of age globally – it is the leading cause of morbidity [1],[2]. The Gram-negative bacterium Vibrio cholerae is a facultative pathogen having both human and environmental stages, and is the etiologic agent of the secretory diarrheal disease cholera [3]. Today, the burden of cholera is estimated to reach several million cases a year in both Asia and Africa, with fewer cases in Latin America [4]. Aquatic reservoirs harbor V. cholerae during extended periods between outbreaks [5], but there is little known about how fast V. cholerae moves from one patient to the next during an outbreak. Transmission between patients may be quite rapid. For example, two devastating outbreaks strike Dhaka, Bangladesh annually. The high burden of disease [6], collapsed water infrastructure, poverty, and crowding make Dhaka an ideal setting for the fast transmission of a facultative pathogen such as V. cholerae. At the host population level, first degree relatives in households are more likely to be infected with V. cholerae [7]. At the pathogen level, the di-annual cholera outbreaks may be clonal [8],[9],[10], and there are rapid shifts in drug resistance patterns [11],[12]. Despite these epidemiological observations that support a model for rapid transmission during an outbreak, little is known about the selective forces that drive facultative pathogens – like V. cholerae – out of the environment and into the next host.
Using the infant-mouse model of cholera, we recently demonstrated that genes induced late in the infection provide a fitness advantage for the transition to aquatic environments [13]. In this study, V. cholerae from cholera patients or in vitro culture were transferred to an aquatic environment. We tested three factors as potential selective forces for driving V. cholerae out of the aquatic environment and into the next host. These factors are shared among several facultative pathogens and are as follows: the viable but non-culturable state, hyperinfectivity, and lytic phage. Escherichia coli, Shigella sonnei, Listeria monocytogenes, Campylobacter jejuni, and V. cholerae are examples of facultative pathogens that lose the ability to culture on standard media upon transfer to aquatic environments [14],[15]. This phenotype was traditionally called the viable but non-culturable state (VBNC) because the cells maintain the capacity for metabolic activities such as protein synthesis, respiration, and have intact membranes despite their inability to culture [16]. However, we prefer to use the active but non-culturable (ABNC) term for reasons explained by Kell et al [17]. The critical debate over terminology is if it is possible for bacteria with a known in vitro growth condition to be viable and (but) nonculturable. Since the answer to this question seems unresolved, the ABNC term is a more conservative definition. In the case of V. cholerae, animals become infected when inoculated with high doses of ABNC bacteria (>106 or >1000-fold above the typical ID50 in animal models) suggesting that ABNC bacteria can be rescued for vegetative growth in vivo [14]. The experimental designs in these studies were unfortunately not overly relevant to conditions in the field; the ABNC state was induced by prolonged incubation at 4°C. ABNC V. cholerae have been observed in rural and urban water samples in Bangladesh between and during outbreaks [5]. ABNC V. cholerae are found as single cells or associated in aggregates with phytoplankton and zooplankton [5],[18],[19],[20]. For these reasons, ABNC V. cholerae are proposed to be the environmental reservoir that maintains V. cholerae between outbreaks and seeds new outbreaks. However, the role this reservoir plays during an outbreak is unclear because cholera outbreaks accelerate faster than the stochastic contribution of V. cholerae from an environmental reservoir [21].
The second factor we measured was hyperinfectivity. This phenotype was discovered when V. cholerae from patients were found to be more infectious in the infant-mouse model than in vitro grown V. cholerae [22],[23]. This phenotype has also been documented in Citrobacter rodentium [24], and can be modeled with mouse passaged bacteria [25]. The role hyperinfectivity plays in transmission is largely unknown, but V. cholerae from patients remain hyperinfectious for at least 5 h in pond water [23]. Models suggest that outbreaks start when an index case consumes V. cholerae from an environmental reservoir, but the acceleration of the outbreak is driven by hyperinfectious V. cholerae. Unlike the stochastic contribution of environmental V. cholerae, mathematical models that incorporate hyperinfectivity produce the steep rise in case numbers that are consistent with the actual rise in cases observed in Dhaka, Bangladesh during an outbreak [21].
The third factor we examined was lytic phage; we note here that this report concerns only lytic vibriophage and not cholera toxin phage or other lysogenic phage. Lytic vibriophage in the environment have been studied from almost the time that V. cholerae was first discovered [26], but recent phage epidemiology papers provide new insights into the role phage play in outbreaks. The percentage of patients passing lytic phage rises as a cholera outbreak progresses; at the same time, phage titers in the environment increase [27],[28]. Towards the end of an outbreak, the vast majority of cholera patients (>90%) void lytic vibriophage in addition to V. cholerae. Over a 5-year study of patients at the International Centre for Diarrhoeal Disease, Bangladesh (ICDDR,B), at least half of cholera patients harbored lytic vibriophage [29]. The ubiquity of lytic phage at the end of an outbreak suggests phage may play an important role in stopping an outbreak. This hypothesis is also supported by mathematical models [30], as well as epidemiological data that indicate household contacts of an index case that does not harbor lytic phage are at an increased risk of infection with V. cholerae [29].
In summary, a cholera outbreak is currently modeled as follows: An outbreak begins with the consumption of ABNC V. cholerae from the environment, is accelerated by hyperinfectious bacteria shed from patients, and is terminated by a rise in lytic phage. This model however does not provide a reason (selective pressure) for V. cholerae to leave the aquatic environment and go to the next host. Contrary to the current model regarding the importance of ABNC V. cholerae for transmission, we show that the loss of culturability is a negative selective pressure for transmission, and non-culturable cells are not the major contributors to infection. Instead we show here that culturable V. cholerae recently shed by patients are the major contributors to infection, and upon prolonged incubation in pond water, lytic phage and ABNC cells rise in the aquatic environment to cooperatively block transmission. In addition, transcriptional analysis suggests that bacteria quickly adjust to the stresses of the aquatic environment, and lytic phage have an undetectable influence on this adaptation. Despite this adaptation, rice-water stool V. cholerae rapidly become ABNC. In the absence of high-titer phage, our results support the model that recently shed hyperinfectious V. cholerae drive cholera outbreaks.
The strains used in this study are provided in Table S1. Strains were grown on Luria-Bertani (LB) agar or in LB broth with aeration at 37°C with streptomycin (SM) 100 µg/ml unless otherwise specified. SM sensitive strains were cultured on LB or a Vibrio spp. selective medium, TTGA [31]; the plating efficiency on TTGA and LB was equivalent (data not shown). The in vitro derived V. cholerae were prepared by growth for 4 h at 37°C with gentle rocking in M9 minimal medium (pH 9.0) supplemented with trace metals, vitamins (Gibco MEM Vitamins, Invitrogen), and 0.5% glycerol [32]; this medium is referred to as ‘M9 pH 9’.
Water was collected from a pond in central Dhaka each day of experimentation using a mechanical pump and intake hose system that collected water approximately 0.5 m below the water surface to avoid fluctuations in osmolarity due to rain water stratified at the top layer of the pond. This pond has historically cultured positive for V. cholerae; however in this study, V. cholerae and phage lytic for V. cholerae were below the limit of detection by standard methods [29] on the days of experimentation. Eighty liters of unfiltered pond water were transferred to a barrel lined with a pond-water washed autoclave bag, an aquarium bubbler was placed in the barrel to oxygenate the water as well as to avoid stratification, the barrels were positioned in an open shed shielded from direct sunlight but freely exposed to the outside air: water temp. 26–28°C, dissolved oxygen ≈6%, conductivity 260–300 µS/cm, total dissolved solutes 137.8 mg/l, salinity = 0.1 ppt, and pH 6.6–6.9. One liter of the water was centrifuged at 2,744 g at room temperature (RT) for 5 min, and filter sterilized through a 0.2 µm filter (FS pond water). This FS pond water was used to resuspend in vitro and in vivo derived V. cholerae, as well as for chemical analysis. After each experiment, bleach was added to each barrel to 0.5% and held for 24 hrs to sterilize the contents. Cultures on LB agar were taken to confirm complete sterilization before the water and bag were disposed. Inorganic chemical analysis on stool supernatant and pond water samples was performed by Dr. R. Auxier at the Center for Applied Isotope Studies (U. of Georgia, Athens, GA). Dr. A. Parastoo at the Complex Carbohydrate Research Center (U. of Georgia, Athens, GA) determined the major sugars in the FS pond water samples using mass spectrometry.
Stool samples were collected from adult patients (>15 yrs of age) with acute watery diarrhea and no prior treatment with antibiotics. The samples were examined by darkfield microcopy to confirm the presence of V. cholerae [29], and were included in the study if >95% of the cells were highly motile and vibrioid in shape. All samples were screened and found to be negative for ETEC, the ratio of V. cholerae to non-V. cholerae bacteria was determined, and the presence of lytic phage was assayed as previously described [29].
Stool samples meeting the inclusion criteria were clarified of mucus and debris by centrifugation at 988 g for 3 minutes at RT, and then V. cholerae were pelleted by 15 minutes of centrifugation at 26,892 g. Bacterial pellets were resuspended in an equal volume of FS pond water at a final concentration of approximately 1×108 CFU/ml; alternatively, pellets were resuspended in RNAlater (Ambion, INC), flash frozen, and stored at −80°C for subsequent microarray analysis. Fifty ml aliquots of the resuspension were transferred to dialysis tubes with a 12 KDa cutoff (Fisher Scientific INC), and the tubes were immediately transferred to the pond microcosm described above. The tubes were kept just below the surface of the water, and bacteria and phage did not traverse the dialysis tubing (data not shown). The time from stool collection in the hospital to incubation in pond water was under 1 h. The collection of the rice-water stool from human subjects was reviewed and approved by both the Research Review Committee and Ethical Review Committee at the International Centre for Diarrhoeal Disease Research, Bangladesh, and by the Human Research Committee at the Massachusetts General Hospital.
V. cholerae were isolated by single colony purification from stool on either LB SM or TTGA media [31]. The in vitro derived V. cholerae were prepared as described above in M9 pH 9 at a final concentration of 1×108 cfu/ml (approximately equivalent to the density in stool). After incubation at 37°C with gentle rocking for 4 h, the cells were then pelleted by centrifugation at 26,892 g for 15 min at RT, resuspended in an equal volume of FS pond water, transferred to dialysis tubes, and placed in the pond microcosm in a manner similar to the stool derived V. cholerae described above. Additionally, a portion of the pellets were stored in RNAlater as above for subsequent microarray analysis. At 5 and 24 h, the contents of the dialysis tubes (patient and in vitro derived) were transferred to sterile centrifuge tubes. For microarray analysis, the contents were centrifuged at 26,892 g for 15 min at RT, and the pellets were stored in RNAlater as above. At the 0, 5, and 24 h time points of collection for patient or in vitro derived cells for microarray analysis, paired samples were simultaneously taken for animal experiments described below.
The competitive index of V. cholerae pre-incubated in M9 pH 7, M9 pH 9, or rice-water stool supernatant was determined using 5 to 6-day-old Swiss Webster mice as described previously [22]. In brief, the O1 El Tor Inaba strain N16961 (LacZ−) of V. cholerae was grown overnight on LB agar with SM, and colonies were resuspended in LB broth. The cells were washed and incubated in M9 pH 7, M9 pH 9, or phage negative stool supernatant. During the incubation of the in vitro samples, stool samples from cholera patients were screened for V. cholerae and processed as described above. After the 1 h incubation of the in vitro grown strains, infant mice were inoculated intragastrically with 105 CFU of a 1∶1 mixture of the paired LacZ+ stool V. cholerae and in vitro grown LacZ− wild-type N16961 strain. At 24 h post inoculation, the small intestine was harvested and the homogenized contents were serially diluted and plated on LB SM, X-gal 40 µg/ml agar plates. After overnight incubation at 37°C, blue and white colonies were counted to determine the competitive index.
To study the infectivity of V. cholerae transferred to the pond microcosm, the ID50 was determined for both stool and in vitro derived V. cholerae after 0, 5, and 24 h of incubation in pond water. At 0 h, stool derived V. cholerae were prepared as described above and serially diluted in LB. The in vitro derived V. cholerae were prepared as described above with the 4 h preincubation at 37°C in M9 pH 9, and subsequent serial dilution in LB. Groups of 5–6 day-old Swiss Webster mice were then inoculated intragastrically with doses that ranged from approximately 1 to 105 CFU per mouse. Mice were euthanized at 24 h post inoculation, and the small intestinal homogenates were plated as described above. Values of ≥1,000 CFU/mouse (limit of detection = 100 CFU) were recorded as positive for infection. A dose-response curve was made by plotting the fraction of infected mice against the log10 of the input V. cholerae cell count – either by CFU or direct counts. The ID50 was estimated from this curve by a standard nonlinear regression using the Hill Equation – the Hill slope was fixed at 1.0 when there were <3 data points between the values of 0.1 and 0.9 on the Y axis. The 95% confidence intervals for the ID50 (CI) and coefficient of determination (R2) are provided. The ID50 for the stool and in vitro derived V. cholerae incubated in the pond microcosm was determined at 5 h and 24 h in the same manner.
The in vivo dynamic between V. cholerae and lytic phage was investigated by the co-infection of both the bacteria and lytic phage in ID50 experiments as described above. Lytic phage isolates were obtained in a pair-wise manner from the same patients that the V. cholerae isolates were obtained. Phage were isolated from stool supernatant by a standard plaque assay on a bacterial lawn made of the V. cholerae isolate from the same stool sample [33]. Phage were picked from 3 serial clear lytic plaques. V. cholerae were prepared for the animal studies by overnight growth and a 4 h incubation at 37°C in M9 pH 9 as described above. The V. cholerae isolated from a given patient and the paired lytic phage were combined for 8 min prior to infection at a phage::bacterium multiplicity of infection (MOI) that reflected what was observed in the rice-water stool and pond microcosm in this project: 0.001 to 5 PFU/CFU. The inocula were then serially diluted and groups of at least five infant mice were inoculated with doses that ranged from approximately 1 to 105 CFU per mouse. The ID50 was calculated for each experiment as described above. At a given dose of bacteria and phage, the burden of infection was determined by calculating the median CFU/ml for each group of at least five mice. This was repeated for a total of 3 strains at all doses of bacteria and paired phage. The three medians were plotted individually, and the average of the three medians was also plotted. A Student's t-test was performed between the average for the no-phage control and each phage dose.
We investigated if colonization of infant mice by V. cholerae in the presence of lytic phage was because the bacteria had become resistant to the phage. One way that bacteria can become resistant to phage is by altering the phage receptor which is most commonly LPS for vibriophage [34],[35]. A basic test for putative LPS mutants is agglutination in LB [36]; this test lacks absolute specificity as other phenotypes can also cause agglutination such as expression of the toxin co-regulated pilus (TCP), but TCP is not expressed in LB [37]. We validated the agglutination assay with LPS extraction and gel electrophoresis of several putative LPS mutants identified by agglutination (below). We chose strain EN159 for this study because it is SM resistant. From each animal coinfected with EN159 (all doses) and the paired EN159 phage (all doses), eight isolates were colony purified (3×) and frozen for further evaluation of phage sensitivity. These isolates were grown in LB SM broth overnight and agglutination of the cells was assessed if the media clarified after 20 min of static incubation. The fraction of the 8 isolates from a given mouse that agglutinated was recorded as a fraction of isolates from a given mouse that were putative LPS mutants [36]. All isolates from mice infected with the highest dose of V. cholerae (1.5×105 CFU/mouse) and all phage MOI's (0, 0.005, 0.1, and 2.0) were further tested for phage resistance by the standard plaque assay.
For validation of the agglutination assay, LPS was extracted from a total of five isolates from 5 different mice infected with the highest bacterial dose (1.5×105 CFU/mouse) and at highest MOI (2.0). As a control, LPS was extracted from a total of five isolates from 5 different mice infected with the highest dose of bacteria (1.5×105 CFU/mouse) and no phage. The input strain also served as an additional control. Cell surface polysaccharides from the eleven strains were isolated and analyzed as described recently [38],[39]. Briefly, Proteinase K-digested whole cell extracts were isolated according to Hitchcock and Brown [40] and analyzed by electrophoresis on 16.5% SDS-polyacrylamide gels. The complete synthesized LPS and the lipid A-core oligosaccharide precursor were visualized by silver staining [41].
RNA was prepared from the samples collected at 0, 5, and 24 h of dialysis in pond water (described above). The frozen suspensions of bacteria in RNAlater (Qiagen) were thawed on ice, spun at 15,000 g for 20 min at 4°C, the supernatant was discarded, RNA was extracted from the pellet using the Qiagen (Valencia, CA) RNeasy Mini Kit, and DNA was removed using the Qiagen on-column RNase-Free DNase set. For qRT-PCR validation, complete DNA removal was achieved using the Ambion (Applied Biosystems/Ambion, Austin, TX) DNA-free DNase Treatment kit. Each RNA sample was spiked with an in vitro transcribed Arabidopsis RNA which served as a reference for color balancing during scanning; the control RNA was provided by the Pathogen Functional Genomics Resource Center (PFGRC) at the J. Craig Venter Institute (formerly TIGR). Labeling of cDNA was performed as described previously [42] with the exception that the reverse transcription reaction used Superscript III (Invitrogen, Carlsbad, CA) at a reaction temperature of 52°C for 1 h and 8 µg RNA. The cDNA from each reaction was split and labeled with either Cy3 or Cy5 (dye swapped). Unless indicated otherwise, at least 4 technical microarray replicates (2 dye swaps) were performed per biological replicate. There were two biological replicates for each condition: patient derived with phage, patient derived with no phage, and in vitro derived.
Microarrays were provided by PFGRC and consisted of glass slides with genes spotted in quadruplicate with 70 bp oligonucleotides for each of 3810 V. cholerae ORFs. Hybridizations were performed as described previously [42]. Microarrays were scanned with a Perkin-Elmer Scanner, and the raw data were analyzed using the Perkin-Elmer Scan Array Express, Imigene, and Spotfire software packages. Cy3 and Cy5 data from each slide were split into the relevant biological groupings as single channel data. All items with a raw intensity of less than 50 were assigned a minimum intensity value of 49.9 [43]. The complete data set was log2 transformed and normalized against all other scans by the 75th percentile. The values for a given gene across all scans were then normalized by the z-score for that specific gene. The normalized data was then compared by ANOVA according to the relevant biologic grouping [44],[45],[46]. For ANOVA analysis between 6 groups, a Bonferroni correction was applied to account for bias due to multiple tests by dividing the desired level of significance (α = 0.01) by the total number of comparisons performed (22,860 = 3,810 genes with 6 comparisons) [44],[45],[46]. Therefore, the corrected false-positive rate was α = 4.4×10−7 which was rounded to α = 1.0×10−7; P values that fell below 1.0×10−7 were considered statistically significant. Cluster analysis was performed by Spotfire with the following metrics: clustered by Unweighted Pair-Group Method with Arithmetic mean (UPGMA), correlated by Pearson Product Momentum Correlation, and ordered by Input Rank. As an independent measure of similarity between biological groupings, Principal Component Analysis (PCA) was performed on all samples using Spotfire. After the ANOVA, all replicates were ungrouped, and the cluster analysis and PCA were performed in an unsupervised fashion with respect to the technical replicates and biological groupings. Fold-changes between two biological groupings were calculated using distinction calculations performed by Spotfire, and fold-changes with P values<6.6×10−6 (Bonferroni corrected) were considered significant. Microarray data are available in the supplemental material (Tables S3, S4, S5, S6, S7, S8, S9 and S10).
There was sufficient sample to obtain cDNA template from three phage positive patients (EN159, EN182, EN191), and three phage negative patients (EN124, EN150, EN174). RNA was isolated as described above, and qRT-PCR was performed as previously described [13]. In brief, cDNA was synthesized from 1 µg of RNA using the SuperScript II First Strand Synthesis System for qPCR (Invitrogen Inc.). The qRT-PCR experiments were performed with iQ SYBR Green supermix (Biorad). Each reaction contained 200 nM primers, approximately 10 ng of the template, and the ROX reference dye. All primer pairs (Table S1) amplified the target with efficiencies of 92% or greater (data not shown). The mean cycle threshold for the test transcript was normalized to the reference transcript sanA [47] and argS. The reference argS was chosen because no expression changes were detected in this microarray project as well as all publicly available V. cholerae microarray databases. Values >1 indicate that the transcript is in higher concentration than the reference.
This project focuses on rice-water stool samples collected from three patients (EN159, EN182, and EN191) who harbored lytic vibriophage for V. cholerae, and the respective phage and V. cholerae isolates from these three patients. In addition, rice-water stool was collected from three patients who did not harbor lytic vibriophage (EN124, EN150, and EN174), and V. cholerae was isolated from each of these patients. Therefore, the biological replicates for each arm of the study were three unless stated otherwise; sufficient numbers of infant mice for ID50 testing were available only for the three patient samples that harbored phage. At the time of collection, all patients were severely dehydrated as defined by the World Health Organization [48].
As V. cholerae passes from the patient into pond there is dramatic shift in osmolarity and in the concentrations of inorganic nutrients and carbon sources. Some of these factors are depicted in Fig. 1. NaCl and KCl are major contributors to osmolarity and both have a decline from 2,600 to 22 ppm (120-fold) and 820 to 6 ppm (140-fold) between the rice-water stool and pond supernatant, respectively. The conductivity difference between the rice-water stool (as well as LB broth) and pond water is approximately a 50-fold decline. Phosphate and fixed nitrogen are typically limiting inorganic nutrients in fresh water ponds. Phosphate and fixed nitrogen (NH4+) decline from 160 to 0.1 ppm (1,600-fold) and 52 to 0.5 ppm (104-fold), respectively. V. cholerae was placed in filtered pond water and then dialyzed in 12 KDa tubing with live pond water. Therefore, carbon sources such as large polymers like chitinous exoskeletons would not be present in the dialysis bags. Carbon sources detected were rhamnose (29 Mol.%; 16 nM), fucose (20% Mol.%; 11 nM), glucose (2.7 Mol.%; 1 nM), and unidentified sugars (48.9 Mol.%). This chemistry collectively framed many of the physiological events that occurred as V. cholerae adapted to the aquatic system. This adaptation and pond microcosm system is not necessarily specific to Bangladesh as the chemical composition shown herein is comparable to pond water used in transition studies with pond water obtained in Boston, MA [13].
The culturability of V. cholerae transferred to the pond microcosm was monitored by culture and direct microscopy counts. We define the non-culturable cells as ‘active but non-culturable’ (ABNC) because there were clear transcriptional changes between 5 and 24 h detected by both microarray and qRT-PCR analysis (below). Thus, our measure of ‘active’ was global transcriptional change. Culturability was rapidly lost upon transfer to the pond microcosm at 5 and 24 h with declines of 63% (SD+/−16%) and 98% (SD+/−1.0%), respectively (Fig. 2A). The V. cholerae isolates from the respective patients were grown in vitro (M9 pH 9) and transferred to the pond microcosm; the declines in culturability in the pond microcosm were similar for the in vitro derived samples compared to the patient derived samples (Fig. 2A). Despite the drop in culturable cells, the total cell numbers remained constant by direct counts (Fig. 2B) for all sample types; the cell number was also constant for phage negative patient samples and the paired in vitro grown strains (Fig. 2B). The culture counts are not available for the phage negative patient samples because two isolates were unexpectedly SM sensitive. The plating efficiency of starting cultures neared 100%. For example, the average concentration of V. cholerae from patients (EN159, EN182, EN191) at 0 h by culture counts and direct counts was 1.0×108 CFU/ml (+/−1.1×108 CFU/ml) and 1.65×108 CFU/ml (+/−0.35×108 CFU/ml), respectively.
The PFU titer was monitored at 0, 5 and 24 h in the pond microcosm (Fig. 2C). At 0 h, the average ratio of phage to V. cholerae for all three patient stools was 2.2×10−6 (SD+/−3.5×10−6 ). At 5 h, this ratio increased by 4 orders of magnitude to 1.0×10−2 (SD+/−1.2×10−2) by culture counts, or 3 orders of magnitude to 1.5×10−3 (SD+/−1.3×10−3) by direct counts. At 24 h, this ratio increased an additional 2 orders of magnitude to 4.0×10−1 (SD+/−3.9×10−1) by culture counts, but remained steady at 3.8×10−3 (+/−3.2×10−3) by direct counts. From 5 to 24 h, this ratio changed because the culturable counts decreased 14-fold. These findings are supported by micrographs that illustrate altered morphology of V. cholerae only in the patient derived samples from phage positive patients (Fig. 3A). Lytic and lysogenic vibriophage have been previously characterized from patients [27],[33],[35],[49],[50],[51],[52]; our phage isolates are consistent in terms of the tropism of those lytic phage previously published [27] because our phage had specificity for the Inaba or Ogawa serotype of the O1 El Tor V. cholerae biotype, and the phage were unable to form plaques on O139 V. cholerae (data not shown). These data indicate the phage receptor may be O1 LPS as has been demonstrated previously [34],[35]. Support for this hypothesis is the generation of LPS mutants in the presence of lytic phage (presented below).
The ID50 for V. cholerae freshly shed from the patients (113 CFU; 95% confidence interval [CI] = 65–196 CFU) was lower compared to the in vitro grown reference (596 CFU; 95% CI = 193–1834 CFU; Fig. 4A). Hyperinfectivity was also observed after 5 h of dialysis between the patient (51 CFU; 95% CI = 13–202 CFU) and in vitro culture (680 CFU; 95% CI = 276–1673 CFU; Fig. 4B). These findings are consistent with competition experiments previously published that suggest V. cholerae maintains hyperinfectivity for at least 5 h after exit from the patient [23]. We tested if hyperinfectivity could be induced by the medium alone (stool-supernatant), and we found that hyperinfectivity could not be induced in vitro by incubation in stool supernatant (pH 9) or minimal media (M9 pH 9) (Fig. S1). Unique to the present study was that the single strain infection experiments revealed that the fraction of mice infected with high doses of patient derived V. cholerae was reduced at 5 h and 24 h compared to the in vitro reference (Fig. 4B–E). Indeed, the ID50 was not able to be calculated for the patient derived samples at 24 h because less than 50% of the animals were infected (Fig. 4D–E). The 24 h time point corresponds with the point when the titer of PFU was highest and the titer of culturable cells was lowest (Fig. 2); note again that the no phage control for this experiment are in vitro derived cells. We hypothesized, and show below, that the incomplete colonization observed is due to the presence of lytic phage in the inocula.
We wanted to investigate the relevance of the ABNC state to the transmission of V. cholerae. To do this we tracked the ID50 over time by both culturable counts and direct counts. We focus here on the ID50 data from the in vitro derived V. cholerae because the phage positive patient derived samples failed to fully colonize at 5 and 24 h. In the context of the pond system, the total cell counts remained constant but the proportion of culturable cells decreased over time. We tested three competing hypotheses: (i) If culturable cells are equally infectious as non-culturable cells, then the ID50 by total cell counts will be constant as the percent of culturable cells decreases. (ii) If culturable cells are more infectious than non-culturable cells, then the ID50 by total cell counts will increase as the percent of culturable cells decreases. (iii) If culturable cells are less infectious than non-culturable cells, then the ID50 by total cell counts will decrease as the percent of culturable cells decreases. As mentioned above, the culture cell counts fell from 100% to 27% to 3% at 0, 5 and 24 h during the experiment. The corresponding ID50 by culturable counts remained constant as the culturable counts decreased at 0, 5 and 24 h (Fig. 4A, B, D). However, the ID50 by total cell counts rose from 596 (95% CI = 193–1834) to 1683 (95% CI = 683–4145) to 7383 (95% CI = 3970–13731) as the culturable counts decreased at 0, 5 and 24 h (Inset table in Fig. 4). Therefore, hypothesis (ii) appears to be correct that culturable cells are more infectious than non-culturable cells, and thus, the major contributors to infection are culturable V. cholerae.
Because lytic phage are present in aquatic reservoirs and in at least half of cholera stool samples, we wanted to determine the relevance of lytic phage to the transmission of V. cholerae. We hypothesized that the reduction in colonization at 5 and 24 h for the patient derived samples (Fig. 4) was caused by the bloom of lytic phage because no such reduction was observed for phage minus in vitro derived V. cholerae. We tested this hypothesis by coinfecting infant mice with V. cholerae isolated from the three phage positive patients (EN159, EN182, EN191), and the paired lytic phage isolate from each patient (described above). The inocula were made by mixing bacteria and phage at various MOI's that were relevant to those observed in rice-water stools and after incubation in the pond microcosm (Fig. 2C). A linear dose response (R2 = 0.99; slope = −0.57; 95% CI = −0.83 to −0.32; Fig. 5A) was observed in the infant mice inoculated with a constant bacterial dose (1–2×104 CFU/mouse) and variable phage dose (1.0×10−3−2.0 MOI). In contrast, at a high dose of V. cholerae (1–2×105 CFU/per mouse) there was a significant reduction in colonization at all MOI tested (Fig. 5B).
The coinfection experiments were also performed as ID50 experiments with variable concentrations of V. cholerae and 4 constant doses of phage (MOI = 0, 0.005–0.1, 0.05–0.1, 0.5–2.0). There was no difference in the ID50 between the no phage control and animals coinfected with phage at an MOI of 0.05–0.1 (Fig. 5C) and 0.005–0.01 (data not shown). However, the ID50 in mice co-infected with phage at a MOI of 0.5–2.0 (18 CFU; 95% CI = 9–36) was significantly lower than the no phage control (65 CFU; 95% CI = 37–111 CFU). These experiments support the hypothesis that phage can limit infection at doses of V. cholerae greater than 103 CFU. However at high phage MOI and low doses of in vitro grown V. cholerae, the phage may have an unexpected positive impact on the ID50 (Fig. 5D). EN159 isolates from experiments in which the phage may have had a positive impact on infectivity (MOI of 2.0; 100–200 CFU) were found to be phage sensitive and not LPS mutants (data not shown). In competition in the infant mouse model, these isolates competed 1∶1 with the input strain suggesting there was no gain of function from prior co-culture with the phage (data not shown).
Although the phage reduced the burden of V. cholerae infection, complete clearance of the bacteria was not observed, as had occurred in the pond microcosm (Fig. 4E). To investigate a reason behind this we closely examined isolates from the EN159 coinfection studies because EN159 is SM resistant. 40–70% of isolates from mice coinfected with EN159 V. cholerae at a dose of 1–2×105 CFU/per mouse agglutinated in LB independent of the phage dose – a phenotype consistent with LPS mutants [36]. No isolates from mice coinfected with V. cholerae at a dose of less than or equal to 1–2×103 CFU/per mouse agglutinated. To confirm that agglutination was indicative of LPS mutations in our system we analyzed LPS from several isolates. The LPS from a total of five isolates from five different mice infected with the EN159 V. cholerae and phage (MOI = 1.0) was compared to the LPS from a total of five isolates from different mice infected with EN159 V. cholerae and no phage. All five mouse passaged isolates from coinfection experiments with phage were resistant to the phage and agglutinated, and the five mouse passaged isolates without phage were sensitive to the phage and did not agglutinate. These phage sensitive colonies demonstrated a wild-type LPS with the typical two band pattern consisting of the lipid A-core oligosaccharide precursor as the lower band and the complete LPS with attached O antigen as the upper band (Fig. 5F). In contrast, phage resistant colonies exhibited an O antigen deficient phenotype (Fig. 5E). We were concerned about lysogeny among phage resistant colonies. Phage sensitive V. cholerae were infected with phage in vitro and subsequent treatment of phage resistant isolates with and without mitomycin-C [53] yielded no phage; this experiment was repeated for all strains and phage in this study. These data suggest there was no lysogeny. However, experiments with additional stresses (osmotic shock, UV, etc.) and phage isolates genetically marked with an antibiotic resistance marker would be required to definitively show the absence of lysogeny. Taken together, these data indicate that at a high dose of V. cholerae, spontaneous LPS mutants will dominate during in vivo colonization in the presence of lytic phage.
Because ABNC V. cholerae have low infectivity, yet represent the predominant state of the bacteria after 5 h of incubation in the pond microcosm, we measured possible transcriptional changes during this transition. The goal was to ascertain whether the bacteria were adapting to the nutrient poor conditions in a manner dependent or independent of their source of origin (patient or in vitro). Samples for the microarray fell into six biological groups: patient derived samples (EN159, EN182) incubated in the pond microcosm for 0, 5 and 24 h (designated T0P*, T5P* and T24P*, respectively; Fig. 6A) and the paired in vitro derived isolates incubated in the pond for 0, 5 and 24 h (designated T0I, T5I and T24I, respectively; Fig. 6A). Both patient (EN159, EN182) samples harbored phage, which is indicated with an asterisk. Two additional patient samples that did not harbor phage (EN124, EN150) were included as controls for transcriptional changes induced by phage (designated T0P, T5P and T24P, respectively; Fig. 7A). The patient samples EN174 and EN191 and in vitro samples EN124 and EN150 were excluded because of insufficient material for microarray analysis. The qRT-PCR validation is provided in Table S2. Cluster analyses in Fig. 6 and Fig. 7 isolated key expression patterns; genes within these groupings are described by biological function in Tables S3, S4, S5, S6, S7, S8, S9 and S10. A complete list of all fold changes is available in Table S10.
This project was designed to concurrently test three critical factors for their relevance to the transmission of cholera. The patient to pond microcosm system allowed us to evaluate (i) the infectivity of V. cholerae as the cells enter into (ii) the ABNC state in (iii) the presence or absence of lytic vibriophage. The ID50 data suggest that the major contributors to infection are culturable V. cholerae. Phage did not affect colonization immediately after passage from the patients because the PFU titer was likely too low. However, V. cholerae failed to colonize the animals after 24 h of adaptation to pond water – the point when the PFU titer and ABNC cells were highest. Taken together, these data challenge the concept that the aquatic environment is an amenable refuge for V. cholerae during transit between human hosts.
The entry into the ABNC state has been challenging to standardize experimentally because it is difficult to sufficiently dilute the cells to the point that they do not self fertilize key nutrients, and at the same time maintain a high enough cell density for tractable experimentation. These problems were overcome by using dialysis tubes containing V. cholerae in suspension in a large volume (80 L) of live pond water. In this system, culturability reproducibly fell by approximately 60% and 98% by 5 and 24 h, respectively, independent of the origin of the bacteria. Defining the physiologic state of cells that do not culture has been controversial. Herein we limit our work to two populations: cells that culture and those that do not culture. We do not differentiate within the population of non-culturable cells that may contain a subpopulation of dead cells. That said, the non-culturable cells are likely to be alive for several reasons: propidium iodide staining for intact membranes indicated the majority of cells in all arms of the study at 24 h had intact membranes (data not shown). Secondly, the RNA yield was similar between 0 and 24 h despite the 98% loss in culturable cells. Finally, the microarray data at 24 h showed continued adaptation in the pond. For example, genes involved with adaptation to low phosphate and fixed nitrogen were induced [59],[60]. Tests for metabolic activity were not performed on all samples. Instead, we define ‘active’ in the context of this project as the capacity for transcriptional change despite a lack of culturability.
Having defined the proportion of non-culturable cells at 24 h, we tested the relevance of ABNC cells to infection. In the context of the pond microcosm, the total cell counts remained constant but the proportion of culturable cells decreased over time. We tested several hypotheses to determine the role of non-culturable cells in transmission. One hypothesis stated that if culturable cells are more infectious than non-culturable cells, then the ID50 calculated by total cell counts will increase as the percent of culturable cells decreases. The data revealed that the ID50 by total cell counts rose as the culturable counts decreased at 0, 5 and 24 h. Therefore, these data suggest that culturable V. cholerae were the major contributors to infection. Previous studies have demonstrated infection with ABNC V. cholerae is possible without an in vitro pregrowth in rich media, but the doses used in these experiments were often quite high [14]. We do not propose diminishing the significance of the ABNC state as ABNC bacteria may still play a vital role in maintaining environmental reservoirs of facultative pathogens between outbreaks. However, our results indicate that the relevance of ABNC V. cholerae during an outbreak may be limited. These results are consistent with other systems that draw into question the role of ABNC cells in infection without in vitro pregrowth [61],[62].
In Dhaka, Bangladesh, lytic vibriophage are common in human patients and the environment [27],[28],[29]. The phage fluctuate in number seasonally in delayed concordance with cholera outbreaks [26],[27],[28], and household contacts of index cases that do not have lytic phage are at an increased risk of being infected with V. cholerae [29]. Despite this epidemiology, the dynamic role that phage play in the environment has not been studied. Phage carried over from the rice-water stool samples bloomed in the pond microcosm by 5 h. There was no significant rise in the phage titer between 5 and 24 h. However, the ratio of phage to CFU increased because of the continued decline in culturable counts between 5 and 24 h. Production of lytic phage is dependent on the growth of its host. Since the bacteria had no net increase/decrease in cell number in the pond system in the presence or absence of phage, it is likely that there was not sufficient growth capacity to make more phage. At the highest dose of V. cholerae, there was only partial colonization of mice infected with patient derived V. cholerae at 5 h, and fewer mice infected at 24 h. These data suggest that phage may reduce colonization, and at 24 h, the negative impact of phage on infectivity is exacerbated by the decline of culturable V. cholerae. Coinfection experiments with V. cholerae and lytic phage confirmed that phage have a negative impact on colonization. At low doses of bacteria and high doses of phage, the bacteria became more infectious. The relevance of this phenotype to the natural environment remains to be determined. The ready generation of LPS mutants in the coinfection experiment provides one mechanism by which bacteria may escape phage, but this is detrimental for the bacteria as LPS mutants are attenuated [63],[64]. This attenuation may be one reason why LPS mutants may not accumulate in the environment. Future studies to elucidate the mechanisms by which lytic phage may influence the infectivity of V. cholerae will add an additional dynamic to consider when modeling cholera transmission. At the most basic level, a better quantification of the infectious dose of V. cholerae in the natural setting, and the seasonal titer of phage and V. cholerae in the environment, will be a starting point for these future studies.
The microarray platform was similar to those previously used, but the method of analyzing the data by ANOVA and PCA to compare biological groups is novel for the V. cholerae field. ANOVA and PCA have been used in this manner in other fields when comparing large numbers of varied sample types [65],[66],[67]. qRT-PCR was used to validate the microarray. The results between qRT-PCR and the microarray were concordant; no false positives were observed by microarray when crosschecked with qRT-PCR. However as expected, the qRT-PCR was more sensitive than the microarray, and the opportunity for false negatives in the microarray analysis is therefore increased. The decrease in sensitivity by microarray was most pronounced with the patient derived samples after 24 h in pond water – with and without phage. One explanation for this decrease in sensitivity is that the RNA may be damaged at 24 h. If true, this would provide some insight into the physiological status of the bacteria at 24 h. qRT-PCR was normalized to an internal reference gene (sanA and argS) whereas the microarray was normalized to the global expression level. Therefore, qRT-PCR is less affected by RNA damage in this case because the references will theoretically also be equally affected by damaged RNA. If the RNA from cells incubated for 24 h is indeed damaged, this suggests that ABNC bacteria may be less capable of maintaining ribonucleic acid integrity. Despite these concerns, the congruence between the 5 h and 24 h samples at the global level demonstrates that the 24 h data are still informative albeit with reduced sensitivity at the level of individual genes.
The microarrays reveal immediate and striking transcriptional adjustment to the pond within the first 5 h. As the bacteria enter the ABNC state, they adapt to the nutrient limited nature of pond water by upregulating genes required for low phosphate and fixed nitrogen conditions. This adaptation is similar between patient and in vitro derived samples. In addition, there is a general down regulation of ribosomal proteins that indicates a general decline in the ability to synthesize new proteins. These congruent expression patterns are contrasted by divergent profiles of patient and in vitro derived samples at the global level. At both the 5 and 24 h time points, the patient derived samples do not converge with the in vitro derived samples by cluster analysis or by PCA. This result was not caused by the lytic phage in the patient derived samples because patient derived samples without phage were also divergent from the in vitro derived samples. These findings are supported by in vitro studies with lytic phage and E. coli and P. aeruginosa that detected changes in less than 4% of genes [68],[69].
The lack of convergence between patient and in vitro derived samples has relevance to vaccine development. One vaccine strategy is to vaccinate with V. cholerae in a ABNC state expressing ‘environmental’ surface proteins. The method behind this strategy was to transfer LB grown bacteria to pond water and allow the bacteria to enter the ABNC state and express a new repertoire of environmental antigens. This strategy is still valid, but we caution that in vitro derived cells incubated in pond water may indeed be ABNC, but they may express a different set of antigens than those from patients incubated in pond water. Thus, it may prove necessary to express surface proteins of ABNC bacteria derived from patients by genetic modification using inducible expression systems. Simple incubation of in vitro derived V. cholerae in pond water may not be adequate.
In summary, the dynamic interaction between lytic phage and bacteria in the pond environment suggests that the model of cholera transmission be reconsidered with respect to the urgency for transmission to the next host. Phage did not affect colonization immediately after shedding from the patients because the phage titer was too low. However, V. cholerae failed to colonize the animals after 24 h of incubation in pond water – the point when the phage and ABNC cell titers were highest. At 24 h, the rise of ABNC cells and lytic phage blocked transmission. The dialysis system was open with respect to small molecules but was closed with respect to the phage. The impact of the phage in the natural environment will be a function of the dilution of the bacteria away from the phage in large bodies of water that are free-flowing open systems such as rivers or closed systems such as ponds. Real-time studies of phage and bacterial titers in such bodies of water will provide a critical temporal factor to consider when gauging the negative selective pressure imposed by lytic phage and the ABNC state. Understanding this dynamic may ultimately demonstrate that the unfavorable conditions in the environment provide the critical selective pressure for toxigenic V. cholerae to maintain its facultative pathogen life-history strategy.
|
10.1371/journal.pbio.2004188 | Adolescent development of cortical oscillations: Power, phase, and support of cognitive maturation | During adolescence, the integration of specialized functional brain networks related to cognitive control continues to increase. Slow frequency oscillations (4–10 Hz) have been shown to support cognitive control processes, especially within prefrontal regions. However, it is unclear how neural oscillations contribute to functional brain network development and improvements in cognitive control during adolescence. To bridge this gap, we employed magnetoencephalography (MEG) to explore changes in oscillatory power and phase coupling across cortical networks in a sample of 68 adolescents and young adults. We found a redistribution of power from lower to higher frequencies throughout adolescence, such that delta band (1–3 Hz) power decreased, whereas beta band power (14–16 and 22–26 Hz) increased. Delta band power decreased with age most strongly in association networks within the frontal lobe and operculum. Conversely, beta band power increased throughout development, most strongly in processing networks and the posterior cingulate cortex, a hub of the default mode (DM) network. In terms of phase, theta band (5–9 Hz) phase-locking robustly decreased with development, following an anterior-to-posterior gradient, with the greatest decoupling occurring between association networks. Additionally, decreased slow frequency phase-locking between frontolimbic regions was related to decreased impulsivity with age. Thus, greater decoupling of slow frequency oscillations may afford functional networks greater flexibility during the resting state to instantiate control when required.
| During the transition from adolescence to adulthood, humans have decreases in impulsivity and increases in cognitive control. These behaviors are supported by a distributed set of brain regions, including the prefrontal cortex, that can be studied by with a variety of brain-imaging tools. Magnetoencephalography (MEG) is an approach that allows us to study spontaneous brain activity at the millisecond timescale, providing unique insight into local neural activity (power) and interactions between brain regions (estimated through phase-locking). Neural circuits exhibit oscillatory activity across a broad range of frequencies. Relatively slower-frequency (4–10 Hz) oscillations are thought to support cognitive control. We found that, during the transition from adolescence to adulthood, power was redistributed from slower frequencies to higher frequencies, with the greatest increase in faster frequency power in the posterior cingulate cortex. We also found that the phase-locking of prefrontal cortex theta band (5–9 Hz) oscillations decreases during adolescence. Mediation analysis of self-reported impulsive behavior suggests that band phase-locking contributes to decreases in impulsivity. This activity pattern may be an intrinsic marker for the ability for control-related brain regions to engage downstream processing networks. Our results indicate that spontaneous neural activity continues to be refined systematically during adolescence and contributes to cognitive maturation.
| The transition from adolescence to adulthood is characterized by significant enhancements in brain function, supporting increased cognitive control and normative decreases in impulsivity [1,2]. Developmental task-based functional magnetic resonance imaging (fMRI) studies indicate that core regions supporting cognitive control (e.g., anterior cingulate cortex [ACC] and anterior insula [aIns]) are engaged in adolescence during cognitive tasks, but their blood oxygen level–dependent (BOLD) signal activation [3,4] and connectivity with other brain regions continue to increase into adulthood [5–7]. As such, brain networks supporting cognitive control are present prior to adolescence; however, the successful instantiation of cognitive control continues to improve [8]. Developmental resting-state fMRI (rs-fMRI) studies analyzing whole-brain connectivity patterns parallel this principle, such that the organization of functional brain networks is relatively stable by childhood [7,9,10], while integration (between-network functional connectivity) continues to refine well into late adolescence and early adulthood, supporting improvements in cognitive control [7].
The majority of developmental research on resting-state functional networks has utilized fMRI (see [11] for a review), providing the field a window into the development of resting-state networks at infra-slow frequencies (0.01–0.10 Hz). However, much less is known about the development of these networks at faster frequencies (i.e., 1–10 Hz oscillations) known to support the cognitive constructs that demonstrate a protracted development [12]. Because fMRI is not sensitive to this timescale of oscillation, magnetoencephalography (MEG) serves as a complementary tool to understand resting-state network development by allowing us to explore this relatively faster oscillatory range.
The correlation between electrophysiology and BOLD has been studied in both human and nonhuman primates, with a consistent finding of correlations between modalities in broadband gamma activity (40–100 Hz) within local neuronal pools during tasks [13,14]. Oscillations in this frequency range play a role in enabling local neuronal synchronization, whereas slower frequency (4–14 Hz) oscillations have been shown to support long-distance integration [15,16]. For example, synchronization of slow frequency oscillations within the frontoparietal (FP) network [17] are associated with cognitive control and have been shown to improve behavioral performance on control tasks [18,19]. Additionally, theta band activity (4–10 Hz) intensifies when control demands are increased [20]. Hence, slow frequency oscillations across control regions may contribute to top-down modulation of processing networks [12,21,22]. For example, long-range interactions from frontal to visual association regions during working memory retention and mental imagery evolved most strongly in the theta and alpha frequency range [23,24]. Moreover, evidence suggests that the prefrontal cortex leads the posterior parietal cortex in sustained visual attention tasks in the theta band [25]. Slower frequency oscillations, often in the theta band, have been shown to organize local neural activity in the gamma band, such that neurons tend to have greater firing rates in the trough of an ongoing slow frequency oscillation, providing a temporal template for neuronal communication [22,26]. As such, the phase of slower frequency oscillations, especially within the theta band, may be critical for coordination of neural activity over long distances [22,27].
In addition to task states, the electrophysiological correlates of control networks defined by BOLD fMRI during the resting state are becoming clearer. Resting-state BOLD networks correlate to the alpha and beta band, as measured with MEG [28]. There is additional evidence suggesting that correlations with BOLD may be greater at even slower frequencies, such as delta and theta bands (1–10 Hz) [29]. Recently, Hacker and colleagues characterized the spatial correspondence in humans of resting-state BOLD fMRI and band-limited power using electrocorticographic recordings, discovering frequency-specific oscillations within association networks in the slow frequency range (3–14 Hz) [30]. In sum, association networks map onto slower frequency oscillations (4–14 Hz) that may support coordinating activity of other brain networks.
Electrophysiological (i.e., electroencephalography [EEG]/MEG) studies have begun to offer insight into development changes in cortical oscillations. The majority of research concerning electrophysiological maturation across development has used EEG, finding age-related decreases in total power (total amount of activity across broadband frequencies) [31] and absolute power in each frequency band [31–34]. Additional work has shown that there is a redistribution of relative power (power in a given band in relation to total power across all frequencies) from lower to higher frequency bands [35], with frontal regions reaching adult levels of power after more posterior processing regions [31,32,36]. Similar posterior-to-anterior gradients have been observed using EEG measures of coherence, an index of regional coupling including both phase and amplitude components [37]. Notably, the curvilinear decreases in the delta and theta bands (i.e., 0.5–7 Hz) are highly correlated with gray matter volume decreases during adolescence [38]. Using MEG, increased amplitude correlations have been observed both within and between functional brain networks at rest throughout adolescence [39]. Although these studies have begun highlighting developmental trajectories of neural oscillations, the poor spatial specificity of EEG and lack of brain/behavior relationships utilizing MEG/EEG have limited our understanding of the regional and functional network development of oscillations and their potential contribution to cognitive development.
We sought to bridge this gap in the understanding of adolescent development, linking the age-related changes in brain network oscillations to cognitive development. In a sample of 68 adolescents and young adults (aged 14–31 years), we employed MEG to explore intrinsic properties related to oscillatory developmental within and between cortical networks, with regard to both power and phase. Specifically, within frequency intervals related to interareal neural interactions (1–49 Hz) [40,41], we examined regional and network-level oscillatory power and functional coupling of well-defined brain networks using the phase-locking value (PLV), similar to recent approaches [42]. Unlike correlation or coherence measures, the PLV ignores the amplitude (power) relationship between 2 oscillators. This enhances the ability to analyze phase relationships between brain regions, which is known to support interareal communication between large neuronal pools [26]. Interareal phase relationships in the theta band increase across multiple components of cognitive control [12], including working memory [43], error commission [44], and conflict.
Similar to previous EEG studies, we found a redistribution of regional power from slower delta band oscillations to faster beta band oscillations, with greater decreases in delta band power anteriorly in the cortex and greater increases in beta band power posteriorly. In terms of phase, we demonstrate age-related decreases in phase-locking of slow frequency (5–9 Hz) oscillations during adolescence, which followed a robust anterior-to-posterior gradient, with the greatest age-related changes in midline frontal regions, an area known have protracted cognitive development throughout adolescence [1,3,7]. Using a priori network membership, we show that the greatest developmental slow frequency decoupling occurred in higher-order association networks, relative to processing networks. Finally, we demonstrate that decoupling of slow frequency oscillations between anterior prefrontal regions and the anterior temporal lobe is related to self-reported impulsivity, a developmentally sensitive measure of cognitive control known to decrease robustly throughout adolescence.
In order to probe developmental changes in functional brain regions and networks, we used a previously defined functional parcellation established from rs-fMRI [45] to parcellate the cortical surface into 333 regions of interest (ROIs) in a sample of 68 individuals aged 14 to 31 years. For each ROI at each frequency (1–49 Hz; 1 Hz intervals), we calculated relative power to probe regional age-related changes in regional power and a PLV between each ROI pair to determine the age-related differences in degree of coupling between the phases of the oscillations between regions (see Fig 1 for workflow overview).
First, we averaged the PLV matrices at each frequency across both ROI dimensions for each frequency and subject. This resulted in one global cortical PLV for each frequency, for each subject. There was no significant main effect of age predicting PLV (β = −0.0004, t = −1.255, χ2(1) = 1.576, p = 0.209). However, there was a significant age by frequency interaction predicting PLV (χ2[48] = 125.56, p < 0.001). A significant negative relationship between global PLV and age at each frequency interval between 5 and 9 Hz (all p < 0.05, false discovery rate [FDR] corrected) emerged, suggesting that phase relationships between regions in the 5–9 Hz frequency band become less coupled throughout adolescence (Fig 2A). No other frequency intervals showed a significant age-related change in PLV (all p > 0.05).
Similar to the PLV analysis, for each subject, we computed relative power at each frequency (1–49 Hz in 1 Hz intervals) for each ROI (see Methods for details). Similar to the PLV analysis, we obtained a measure of global power by averaging relative power across each ROI for each frequency. We observed a significant negative relationship between delta band power (1–3 Hz) and age (all p < 0.05, FDR corrected), such that delta band power decreased with age (Fig 2B). Conversely, beta band power (14–16 Hz and 22–26 Hz) significantly increased with age (all p < 0.05, FDR corrected), supporting previous developmental EEG studies noting a shift in power distribution, such that slower wave oscillations tend to shift towards relatively higher frequencies at rest [31,32,36]. There was no evidence for a significant relationship between 5–9 Hz power and age (t = −0.36, p = 0.71). Moreover, we did not observe a significant relationship between PLV and power (t = −0.01, p = 0.99). These results further support the notion that phase and power are largely orthogonal, providing complementary information in regard to the development of neural oscillations.
To determine the anatomical locus of PLV decreases with age in the 5–9 Hz band, we averaged each individual subject’s PLV matrices in the 5–9 Hz frequency interval. Next, we regressed age onto each ROI pair’s PLV, controlling for motion and power (see Methods) and extracted the beta weight for age from each model. This resulted in a pairwise matrix of beta weights (beta matrix), representing the rate of change across development in slow frequency PLV for each ROI pair.
We examined whether age-related changes in PLV demonstrated anatomical gradients across the cortex. To that end, we obtained a summary rate of change for each ROI by summing down the columns of the beta matrix and regressing each ROIs summed beta weight against its y-coordinate (in Montreal Neurological Institute [MNI] coordinate space) in each hemisphere and x-coordinate, separately. Average distance from each ROI to every other ROI and ROI surface area were included as nuisance regressors in all regression models to control for distance-dependent artifacts (i.e., anatomically proximal regions have artificially inflated PLV). Along the anterior-to-posterior axis, we observed a significant negative relationship between the summed beta weights and the y-coordinate (t = −13.19, p < 10−10), indicating a strong anterior-to-posterior gradient of PLV change, such that frontal regions showed greater decreases in theta band PLV (i.e., more decoupling) with age than posterior regions (Fig 3A and 3B). Regions undergoing the greatest decrease in PLV (top 5%) over development are rank ordered in Table 1. In the lateral-to-medial gradient, we observed a significant negative relationship between the summed beta weights and the x-coordinate in the left hemisphere (t = −6.97, p < 10−10) but only a trend in the right hemisphere (t = 2.01 p = 0.05), indicating slow frequency PLV decreased more rapidly with age along the medial wall. In sum, the greatest rate of decrease in slow frequency PLV occurred in midline frontal regions.
In addition to the PLV analysis, we also characterized regional changes in power throughout adolescence. For each region, we summed the beta weights across frequencies demonstrating a significant Power × Age relationship in Fig 2B, for delta and beta bands separately. Similar to slow frequency PLV, delta band power demonstrated a significant anterior-to-posterior gradient (t = −10.33, p < 0.0001), with the largest age-related decreases in delta power occurring in frontal regions, especially in the frontal operculum (Fig 4A). In contrast to delta power, developmental beta band increases in power followed a posterior-to-anterior gradient (t = 15.86, p < 0.0001), such that the greatest developmental increases in beta band power occurred in medial and lateral parietal regions (Fig 4B). Of note, the posterior cingulate cortex, a hub of the default mode (DM) network, demonstrated the greatest age-related increase in beta band power. Power in the 5–9 Hz frequency interval did not demonstrate any significant age-related increases or decreases (t = −0.36, p = 0.71), nor did 5–9 Hz power demonstrate any significant developmental anterior-to-posterior gradients (t = −1.70, p = 0.09).
To assess developmental changes in the anterior-to-posterior gradient of PLV in other frequency bands, for each subject and each ROI, we regressed age onto PLV and extracted the resulting beta weight for age. Beta weight matrices were generated for each frequency interval (see Methods), summed, and regressed against the ROI’s y-coordinate. We then extracted the beta weight from the y-coordinate regressor in each regression model and plotted this as a function of frequency (Fig 3C). Slow frequency age-related decreases in PLV were most prominent at 6 Hz. To quantify these results statistically, we tested for significant differences in the correlation between ROI beta weights and anterior-to-posterior gradients between a given frequency interval (in 5 Hz bins) by comparing the slopes (i.e., beta weights) of the regression models from each frequency interval to the 6–10 Hz interval (see Methods for more details). A significant difference would be reflected in a z-statistic > 1.645, p < 0.05, one-tailed, indicating that the 6–10 band had a significantly greater negative slope between the summed beta weights for PLV × Age and the anatomical y-coordinate of the region. We did not find evidence for a significant difference for the alpha range (intervals from 11–15 Hz; z = 0.13, p > 0.05). However, for frequencies less than 6 Hz and greater than 15 Hz, we did find a significant interaction (all z > 1.645, p < 0.05), indicating that the greatest gradients in PLV occur within the theta and alpha band regime.
To quantify developmental changes in the anterior-to-posterior gradient of power across all frequency bands, for each subject and each ROI, we regressed age onto power and extracted the resulting beta weight for age. As in the PLV analysis, beta weight matrices were generated for each frequency interval (see Methods) and were regressed against the ROI’s y-coordinate. We observed a negative gradient in the delta regime, whereas a positive gradient existed in the beta band (Fig 4C). Thus, age-related decreases in delta band power were most prominent in frontal regions, whereas age-related changes in beta band power were most prominent in posterior regions.
Next, we aimed to determine whether our developmental effect of an anterior-to-posterior gradients of PLV and power differences with development were specific to the resting state versus a task state. To this end, we analyzed data from the maintenance period of a working memory paradigm in a subset of our sample (n = 28; details of MEG task methods and results in Methods). After extracting pairwise PLVs and regional power for each subject and frequency band within the 5–9 Hz band, we averaged across frequency bands, resulting in 1 phase-locking matrix per subject. Paralleling the resting-state analysis, we regressed age on each pairwise PLV across subjects, controlling for subject head motion. We extracted the beta weight from the age regressor, resulting in a beta weight matrix, representing linear effects of age on changes in PLV during working memory maintenance. To test for an anterior-to-posterior effect as was observed during the resting state, we summed down the columns and regressed the ROI’s y-coordinate on this summed linear age effect. We did not observe an anterior-to-posterior gradient during working memory maintenance (t = −0.02, p = 0.98). Moreover, we did observe the anterior-to-posterior gradient in this subset of subjects (t = −9.31, p < 10−10) during rest. These findings suggest that the strong decreases in 5–9 Hz phase-locking in frontal regions likely are specific to the resting state. Similar to PLV, the age-related effects in delta and beta power were specific to the resting state. We calculated power during the maintenance period of the working memory task across the delta (1–3 Hz) and beta band (14–16 Hz and 22–26 Hz). For each frequency interval and each ROI, we regressed age against power and extracted the beta weights from the age regressor. For each frequency interval, we regressed the y-coordinate against the beta weights. We did not observe an anatomical gradient within the delta band or beta band during the maintenance period of the task (all p > 0.05, FDR corrected), suggesting that age-related effects in power are also specific to the resting state. Together, these results indicate that adolescence is characterized by frequency-specific changes in PLV and power that are specific to the resting state.
In addition to specific regional changes in PLV, we aimed to characterize developmental changes in PLV as a function of networks [45]. For each network combination (e.g., DM-DM, DM-FP, etc.), we obtained the mean beta weight of the linear effect of age on PLV for all ROI pairs of the networks being compared. The resulting heat map is shown in Fig 5A. We then performed a one-way ANOVA to quantitatively assess whether some networks experienced a greater rate of change in PLV with age compared to others. Here, we submitted summed beta weights of within-network interactions (e.g., DM to DM) to the ANOVA. As determined by the ANOVA test, there was a significant difference in the summed beta weight for age effects at the network level (F[12,320] = 9.57, p = 10−10). A subsequent post hoc analysis revealed that the negative linear age effect was greatest for the salience (SAL) network compared to any other network (p < 0.05) (Fig 5B). More generally, a t test between the beta weights within association networks and the beta weights within processing networks revealed that PLV within association networks decreased with age significantly more compared to processing networks (t = −6.51, p < 0.001).
To make inferences concerning significant developmental differences in delta band and beta band power at the network level, we performed a one-way ANOVA on the beta weights by grouping the regions according to a priori network affiliation for the delta and beta regime, separately. With respect to the delta band, we found a significant difference in the average beta weight for age effects at the network level (F[12,320] = 22.71, p = 10−36). A subsequent post hoc analysis revealed that age-related decreases in delta power within networks were greatest for the auditory, SAL, cinguloopercular, and FP networks (all post hoc comparisons were corrected for multiple comparisons using the Tukey method). For complete post hoc results, see Table 2.
With respect to beta band power, we also found a significant difference in the average beta weight for age-related differences at the network level (F[12,320] = 12.52, p = 10−20). A subsequent post hoc analysis revealed that age-related increases in beta power were greatest for somatomotor, auditory, and visual networks. For complete post hoc results, see Table 3.
After determining the gradient and locus of decreased phase coupling from adolescence to adulthood, we analyzed specific ROI pairs driving this decrease. Specifically, we aimed to determine the specific pairwise interactions that contributed to the greatest rate of 5–9 Hz oscillatory decoupling. We first identified the top 5% of ROIs that showed the greatest rate of 5–9 Hz decoupling (developmental hubs) from the regional analysis. From those ROIs, we extracted the top 5% of negative beta weights and plotted the connections, with ROIs grouped by networks (Fig 6), as assigned by [45]. All ROIs from the regional analysis were within higher-order association networks, with 8 belonging to the DM network, 3 belonging to the FP network, 1 belonging to the SAL network, 1 belonging to the ventral attention (VA) network, and 3 belonging to an undefined network, though all regions were within anterior portions of the frontal lobe and are considered part of the limbic network in other parcellations (e.g., ref [46]). With the exception of 2 links, all links from these developmental hubs were to regions of other association networks, indicating that pairwise decreases in 5–9 Hz coupling are largely specific to association networks.
We have demonstrated a strong decrease in 5–9 Hz PLV within midline frontal regions. Given the role of anterior prefrontal cortex and anterior temporal lobes in impulse control [47] and the role of theta (4–10 Hz) oscillations in cognitive control [12], we sought to determine whether decreases in frontal slow frequency PLV were related to decreased impulsivity throughout adolescence. The UPPS-P Impulsive Behavior Scale is a validated self-report 59-item measure of impulsivity [48]. Items are endorsed on a 4-point scale from 1 (agree strongly) to 4 (disagree strongly). After appropriate reverse scoring, scores for each item range from 1 (non-impulsive answer) to 4 (high level of self-reported impulsivity). The UPPS-P can provide scores from specific subscales (e.g., Urgency, Lack of Premeditation, Lack of Perseverance, Sensation Seeking). In the current analysis, we utilized a total impulsivity measure (mean across all items) to increase the precision of each subject’s estimate. Within our sample, total impulsivity scores from the UPPS-P scale (M = 2.02, SD = 0.35; Range [1.32, 2.75]) were consistent with normative variability in impulsivity as reported in previous work [49]. Furthermore, the Cronbach α for the total impulsivity measure in our sample was 0.93, indicating excellent internal consistency. Total impulsivity was negatively associated with age (β = −0.333, t = −2.74, p = 0.008), such that impulsivity decreased significantly with development. To obtain a cluster of regions that significantly decreased in PLV as a function of age, we submitted the individual subject matrices to the network-based statistic (NBS) [50]. The NBS is a common tool used in rs-fMRI studies to identify clusters of suprathreshold links displaying a similar effect (e.g., increasing or decreasing PLV with age). It seeks to control family-wise error rate when mass univariate testing occurs, as in the case of running regression analyses on each ROI pair. Briefly, a test statistic is generated for each ROI pair’s PLV as a function of age. A cluster is identified using a breadth first search, followed by permutation testing to significance based on a cluster’s size.
A cluster composed of 49 regions with 122 links survived the permutation test (1,000 resamples; red links in Fig 7A). Similarly, we performed a median split on impulsivity to break the sample into a high impulsivity group and a low impulsivity group. Individual subject matrices were once again submitted to the NBS, controlling for age. A cluster composed of 13 regions with 14 links survived the permutation test (1,000 resamples; orange links in Fig 7A). Three links comprising 5 distinct regions overlapped between the 2 clusters (PLV × Age and PLV × Impulsivity; yellow links in Fig 7A). For statistical confirmation of overlap between PLV and age with PLV and impulsivity, we subsequently submitted to 3 separate mediation analyses. The fist link (L1) was between the left superior frontal gyrus (MNI coordinates: −15.05, 64.73, 13.29) and the right inferior frontal gyrus (MNI coordinates: 25.07, 7.38, −16.41), the second link (L2) was between the left temporal gyrus (MNI coordinates: −50.60, 9.26, −18.71) and right medial frontal gyrus (MNI coordinates: 12.40, 25.55, −16.38), and the third link (L3) was between the left middle temporal gyrus (MNI coordinates: −44.87, 7.38, −34.85) and the right medial frontal gyrus (MNI coordinates: 12.40, 25.55, −16.38). As a separate means of dimensionality reduction more focused on the a priori network organization, as well as the strong 5–9 Hz decoupling within the SAL network, we also tested mean SAL network PLV as a mediator between age and impulsivity. Mean SAL network PLV was not associated with UPPS-P total impulsivity scores while co-varying age (β = −0.183, t = −1.45, p = 0.152).
In addition to PLV, we also tested delta band power and beta band power for meditation in the relationship between age and impulsivity. Neither delta (minimum p = 0.47, FDR corrected) nor beta-power (minimum p = 0.90, FDR corrected) in any node significantly mediated the relationship between age and impulsivity. Together, these results indicate that resting-state slow frequency phase-locking, not power, contributes to age-related decreases in impulsivity.
Mediation analysis on each link separately revealed that partialing out the variance of each of the 3 ROI pairs significantly attenuated the relationship between age and impulsivity (indirect pathway [path ab], L1: β = −0.133 [95% CI −0.244 to −0.017], p = 0.03; L2: β = −0.154 [95% CI to −0.322, −0.023], p = 0.02; L3: β = −0.130 [95% CI, −0.251 to −0.036], p = 0.003). For statistics on individual paths, see Fig 7B. These findings suggest the observed age-related decreases in impulsivity is, in part, accounted for by the decoupling of slow frequency oscillations during the resting state between the anterior prefrontal cortex and the anterior temporal lobe. However, care should be taken when interpreting the mediation effects, as links demonstrating significant mediation did not survive multiple comparisons corrected when all PLV × Age links were tested together. Regardless, overlapping links between brain/behavior and brain/age relationship suggest that slow frequency PLV, in part, contributes declining impulsivity during adolescence.
Interactions between functional brain networks demonstrate a protracted development well into adolescence and early adulthood [6,7,10] and have been shown to support the maturation of cognitive control [7]. However, the development of resting-state network oscillations and their contribution to cognitive development have not been explored. We found a decrease in theta band (5–9 Hz) phase coupling that was strongest in midline frontal regions, especially in association networks. In parallel, many of the strongest pairwise decrease in resting-state theta coupling occurred between regions affiliated with the DM, FP, and SAL networks. Furthermore, decreased slow frequency coupling between anterior frontal and temporal lobe regions was related to decreased impulsivity with development, providing an oscillatory contribution for decreased impulsivity throughout development.
In terms of oscillatory power, we found a redistribution of power from slower delta oscillations to faster beta oscillations. These findings support and extend prior resting-state EEG [31,34], and concurrent EEG-fMRI studies [51] have reported significant developmental decreases in delta power and increases in beta power [35]. Here, we extend these findings through source localization enabling characterizing of these developmental changes in terms or regions and functional networks. Specifically, there were significant age-related decreases in delta power, most strongly in frontal and opercular regions comprising the SAL and cinguloopercular networks. Conversely, there were significant age-related increases in beta power, most prominent in processing networks. The posterior cingulate cortex, a hub of the DM, demonstrated the greatest age-related increase in beta band power. The DM network demonstrates a protracted development in BOLD connectivity [52], supporting increased specialization and integration of this network with other functional networks [53].
A canonical feature of electrophysiological estimates of power and phase during the resting state is the dominance of oscillations in posterior regions of the brain. The negative slope of age-related decreases as a function of the posterior-to-anterior gradient suggests that frontal regions are becoming more decoupled broadly but most prominently, and statistically significantly, for the 5–9 Hz (theta) band. The post hoc analysis in which we tested for significant differences in the correlation between ROI beta weights and anterior-to-posterior gradients between a given frequency interval (in 5 Hz bins) statistically supports the notion that the anterior-to-posterior gradient is most prominent for the 5–15 Hz frequency interval, which includes the theta (5–9 Hz) interval in which we observed a significant negative relationship between PLV and age. Thus, the gradient analyses, in conjunction with Fig 3A, provide evidence that theta band (5–9 Hz) decoupling is most prominent in midline prefrontal regions.
Similar to early electrophysiological work using EEG to study coherence between cortical lobes [54], we found a protracted development of control networks within the 5–9 Hz frequency interval, particularly within the SAL network, comprised of the anterior cingulate and aIns. Both of these regions are anatomical and functional hubs of the cortex [55,56], with anatomical connections to several major brain networks. Generally, theta band oscillations have been shown to organize higher frequency activity, providing a temporal template for neuronal communication [22,26]. Thus, the phase of theta band oscillations may be critical for the coordination of neural activity [22,27]. Supporting this supposition, a large body of evidence suggests oscillations arising from the SAL network entrain disparate control networks when the need for control is realized [12]. Because adolescence is marked by substantial reductions in behavioral variability that is reliant on control networks [4,57,58], we propose that age-related frontal theta decoupling during the resting state may support the enhanced ability for adults to reliably instantiate control and coordinate regulatory control networks. In support of this, BOLD connectivity studies have found increases in the spatial variability of control and attention networks with development but stability of processing networks [53].
A cluster of frontolimbic regions in anterior prefrontal and anterior temporal lobes also displayed slow frequency decoupling with development. Interactions between these frontolimbic regions and the SAL network had the greatest rate of decoupling of any within- or between-network comparison (Fig 5A). Frontolimbic connectivity is often prescribed a role in impulse control, and when structurally lesioned, leads to greater impulsivity [59,60]. Additionally, recent diffusion tensor imaging and fMRI evidence suggests that frontolimbic connectivity decreases both structurally and functionally throughout adolescence [61,62]. Here, we showed evidence that several interactions between frontolimbic regions were related to impulsivity and also demonstrated significant slow frequency decoupling, confirmed by a mediation analysis. Theta band (5–9 Hz) oscillations may be the mechanism by which these regions communicate to execute impulse control given the role of theta oscillations in the instantiation of cognitive control [12]. Lending support to this proposal, theta band activity tends to flow from frontal regions to more posterior regions [63], suggesting a possible causal association.
Phase-locking should be largely unaffected by power within the same frequency band (but see ref [64]). While age-related changes in PLV and power are related to overarching processes of brain maturation through adolescence, they inform different levels of neural processing. While frequency changes reflect local circuit modifications, PLV reflects the possible interareal effects of these circuit modifications, specifically with regard to coupling across brain regions. Distinct circuit and systems-level modifications are evident through adolescence that would have direct effects on both frequency and coupling (see [65] for a review). At the circuit level, power may be directly affected by maturation inhibitory circuitry supported by increases in GABA, particularly parvalbumin interneurons within the prefrontal cortex [66–68], resulting in greater power within the beta/gamma frequency range [69]. In parallel, and likely indirectly related, there are systems-level changes in specialization of existing connections, such as age-related decreases in frontolimbic connectivity [10,61], that would contribute to the decoupling of slow wave oscillations affecting PLV. As such, developmental decreases in phase-locking may reflect stochastic resonance and/or neural flexibility [70]. If the brain were to maintain a rigid configuration of interactions at this timescale during rest, the ability to explore and switch between brain states would be undermined. Indeed, a prominent theory on the nature of resting state proposes that it serves to allow the sampling of multiple network configurations along an anatomical backbone [71,72]. If this is the case, functional brain networks require flexibility in the form of imperfectly coupled oscillators (i.e., variability) to maintain dynamics in networks at this timescale (millisecond). Several studies have found evidence for increased cortical variability throughout development [70,73,74]. Our findings here support these fMRI-based findings in that decreased phase-locking may represent an overall age-related increase in variability [40,75,76], as well as an overall increase in signal complexity.
A potential limitation of the current study is the depth sensitivity of MEG. The signal-to-noise ratio (SNR) falls with increasing distance from the MEG sensors. However, this limitation exists across all subjects, and thus all ages considered in this study. Given this limitation, we were able to demonstrate decreases in theta band phase-locking within medial wall structures that showed specificity to the resting state versus a working memory task-state.
In sum, our results support and extend previous electrophysiological work characterizing the development of oscillatory power, such that power is redistributed from slower frequency oscillations to faster frequency oscillations. Slow frequency delta oscillations decreased most with age in the frontal operculum, whereas faster beta band oscillatory power increased most strongly in processing networks and the posterior cingulate cortex. Additionally, we found evidence that developmental decreases in slow frequency coupling between control networks supports the transition from adolescence to adulthood that may be related to age-related improvements in impulse control. Age-related decreases in coupling of these oscillations during the resting state may be a mechanism of increased neural flexibility that occurs during adolescence [57,73,74]. As such, future studies should probe frontal theta as a mechanism by which control instantiation is refined during adolescence, using tasks that probe cognitive flexibility, such as task switching and rapid instructed task learning paradigms [77].
All subjects gave written informed consent; parent or guardian consent was obtained for all subjects aged 14 to 17 years. The University of Pittsburgh Institutional Review Board (IRB protocol number: PRO10090478) approved the study, adhering to the Declaration of Helsinki. Subjects were compensated monetarily for their participation in the study.
Of the 81 adolescents and adults we recruited for this study, we include data from 68 subjects, ranging in age from 14 to 31 years (M = 22.51, SD = 5.55). Thirteen subjects were dropped due to unavailable ECG and/or electrooculogram (EOG) data. Based on a questionnaire, none of the subjects—nor their first-degree relatives—currently or previously had a psychiatric or neurological disorder.
For each subject, we acquired a structural MRI to coregister MEG data for analyses in source space. Data from the 68 remaining subjects were pooled from 2 separate protocols within the lab and thus had slightly different structural MR sequences, which would not affect subsequent processing steps. For 28 subjects, structural images were acquired using a sagittal magnetization-prepared rapid gradient-echo sequence (repetition time [TR] = 2,100 ms, echo time [TE] = 3.43 ms, flip angle = 8°, inversion time [TI] = 1,050 ms, voxel size = 1 mm isotropic). For the other 40 subjects included in the second protocol, structural images were acquired using a sagittal magnetization-prepared rapid gradient-echo sequence (TR = 2,200 ms, TE = 3.58 ms, flip angle = 9°, TI = 1,000 ms, voxel size = 1 mm isotropic).
Resting-state MEG data (300 seconds) were acquired using an Elekta Neuromag Vectorview MEG system (Elekta Oy) comprising 306 sensors arranged in triplets of 2 orthogonal planar gradiometers and 1 magnetometer, distributed to 102 locations. The MEG scanner was located inside a 3-layer magnetically shielded room within the University of Pittsburgh Medical Center. The data were acquired continuously with a sampling rate of 1,000 Hz. Head position relative to the MEG sensors was measured continuously throughout the recording period to allow off-line head movement correction. Two bipolar electrode pairs were used to record vertical and horizontal EOG signals to monitor eye movement. A potential confound of developmental studies using MEG is that head size is smaller in younger subjects. Given the sensor locations in the MEG helmet are fixed, smaller heads will by definition have lower signal to noise, as they are further from the sensors. On average, head size is fully developed by 10 years of age [78], which is well below the age of our youngest subject (14 years). We regressed age onto intracranial volume (ICV) and did not observe a significant relationship between ICV and age (t = −1.05, p = 0.29). Additionally, we regressed ICV onto global theta band (5–9 Hz) PLV and did not observe a significant relationship between ICV and global theta band PLV (t = −0.02 p = 0.96).
For artifact removal, we first manually visually inspected every channel across the resting state run for noisy or flat channels and squid jumps. MEG data were then preprocessed off-line using the temporal signal space separation (tSSS) method (10 second correlation window, 0.98 correlation limit), which uses spatial and temporal features to separate signals into components generated within the MEG helmet and components from outside the helmet, which must be artifactual [79,80]. This method greatly increases the SNR of the resulting data [81]. tSSS also performs head movement compensation by aligning sensor-level data to a common reference [82]. This head motion correction procedure also provides estimates of head motion relative to sensor coordinates that we subsequently used for head motion estimates for each subject. Lastly, the raw time series data were down-sampled to from 1,000 Hz to 250 Hz.
An independent components analysis (ICA) approach was used to automatically detect and attenuate heartbeat, eye blink, and eye movement artifacts. ICA was performed on each channel using the Infomax algorithm, with the number of components selected to account for 95% of the variance. The Pearson correlation of the components and the ECG or EOG channel is used to identify artifactual sources through an iterative thresholding method (as implemented in minimum-norm estimate [MNE] Python [83]) and subsequently manually inspected. After removal of the artifactual sources, the data were reconstructed from the remaining components.
MEG sensor data were then projected from the sensors on to the cortical surface to estimate source activities, using the MNE procedure. First, the geometry of each participant's cortical surface was reconstructed from the respective structural MRI using FreeSurfer [84,85]. The solution space for the source estimation was then constrained to the gray/white matter boundary by placing 5,124 dipoles per hemisphere. A forward solution for the constructed source space was calculated using a single compartment boundary-element model. The noise covariance matrix was calculated from a 2-minute empty room scan, in which we acquired data with no subject present. The noise covariance matrix and the forward solution were then combined to create a linear inverse operator to project the resting-state MEG sensor data to the cortical surface. We then warped individual subject data from native space to FreeSurfer average space to facilitate between-subject interpretation of specific regions and networks.
We extracted the time series of resting-state MEG data from a recent parcellation of 333 ROIs covering the entire cortical surface [45]. This atlas was chosen because it comprises major cortical functional networks, including control networks, processing networks, and the DM network and covers the entire cortical surface. Developmental changes in these networks have been observed in fMRI studies [6,7] and are thus candidates for electrophysiological developmental changes at the timescales of which MEG is sensitive.
For each pair-wise relation between ROIs for each subject, a PLV was calculated for each frequency of interest (1–49 Hz in 1-Hz intervals). Phase-locking is a measure of the propensity for 2 signals to maintain a constant phase separation with each other (i.e., synchrony). Therefore, the PLV provides a measure of temporal variability between 2 MEG signals [40].
Here, we binned the data into 100 three-second chunks and obtained 1 PLV across the time windows using a multitapers method with digital prolate spheroid sequence (DPSS) windows (3 tapers), as implemented in MNE python (mne.spectral.connectivity). Three seconds is a sufficiently long segment of data to obtain a reliable estimate of oscillations as low as 1 Hz, as a common recommendation for the minimum number of cycles per window to achieve reliable frequency estimates is 3 [86]. To calculate the PLV at each frequency, 2 time series are spectrally decomposed at a given frequency, given by the equation
PLV=1N|∑n=1Nei(θ1(n)−θ2(n))|
where N is the number of sampled time points and θ1 and θ2 are the phase values at time point n. The PLV was calculated for each ROI pair, resulting in 55,278 PLVs for each frequency and for each subject. A single averaged PLV was then computed by averaging all of the PLVs, ranging from 0 to 1, representing a random phase relationship and fixed phase relationship, respectively.
For each ROI, power was calculated using the Welch method (pwelch function in MATLAB) on the 100 three-second chunks of data, with an overlap of 50%. The relative power at each frequency interval in the range of 1–49 Hz (1 Hz bins) was calculated by dividing the power at a given frequency by the total power (summed power) in the 1–49 Hz interval. This value represents the relative magnitude of each frequency in relation to the total signal.
After ROI × ROI PLV individual subject matrices were calculated at each frequency, individual subject matrices were concatenated forming a 333 × 333 × 49 × 68 four-dimensional matrix. First, we asked whether there were developmental changes in PLV across a broadband frequency range (1–49 Hz). To this end, we averaged the four-dimensional matrix along the first 2 dimensions of the upper triangle, resulting in a single PLV value at each frequency for each subject. A linear mixed-effects model with maximum likelihood estimation was used to examine main effects and interactions predicting PLV. Age and frequency were entered as fixed effects, and random intercepts were estimated for each subject. Significance values for fixed effects were obtained through a likelihood ratio test between models with and without the effects in question (chi-squared test). To test individual frequencies for PLV × Age effects, we regressed PLV against age within each frequency bin and corrected for multiple comparisons using FDR [87]. For visualization purposes in Fig 2A, we performed a median split by age.
First, we asked whether global (across all ROIs) relative power at any frequency interval demonstrated a significant age effect. After relative power was determined for each ROI at each frequency interval, we averaged power across all ROIs for each subject. We then performed a linear regression analysis at each frequency interval (1–49 Hz; 1-Hz bins) and corrected for multiple comparisons using an FDR correction [87]. For visualization purposes in Fig 2B, we performed a mediation split by age.
Once we determined the frequency ranges of significant age effects in phase-locking (theta band: 5–9 Hz) and power (delta band: 1–3 Hz; beta band 14–16 and 22–26 Hz), we sought to determine the specific regions in which phase-locking and power were significantly changing with age. For the analysis of power, for each ROI, we ran linear regression models to determine the rate of change in power within each frequency band as a function of age and extracted the beta weight value from the age regressor. This resulted in a beta weight matrix (ROI × Frequency). We then summed across frequencies within the range of significant effects (e.g., 1–3 Hz for delta band power) for each ROI. For the phase-locking analysis, we ran linear regression models to determine the rate of change in PLV within the theta band as a function of age, controlling for potential confounds, including motion, power, and distance (see below). This resulted in a 333 × 333 matrix of beta weights from the age regressor, representing the rate of change in phase-locking for every ROI pair. To obtain a summary statistic for each ROI, we summed down each column of the matrix, resulting in 333 summed beta weights, which we use to characterize the summed rate of change with age for every ROI across the cortical surface. This process was repeated across frequencies of interest (1–49 Hz) by averaging across frequencies in 5 Hz bins (i.e., 1–5 Hz, 6–10 Hz, …, 46–49 Hz).
We were interested in general trends across the cortical surface. To this extent, we calculated the center of mass for every ROI to obtain a center coordinate and to also get a measure of Euclidean distance between each ROI pair. We the regressed the y-coordinate of the ROI onto the summed beta weights for each ROI (for power and phase analyses separately), controlling for average distance between ROIs and ROI surface area. The average distance between ROIs was included as a nuisance regressor to attenuate the effects of volume conduction. For the PLV analysis, this process was also repeated at each frequency interval and across 5 Hz frequencies bins in the range of 1–49 Hz to determine the specificity of the anterior-to-posterior gradient to the theta band. Specifically, we tested for a significant difference between the slope of each regression model (i.e., beta weights) versus the model including the theta band (6–10 Hz for this analysis) using the following formula [88]:
z=β1−β2SEβ12+SEβ22
where z is equal to the test statistic (values > 1.645 correspond to p < 0.05, one-tailed), β1 is equal to the regression coefficient of the y-coordinate in the 6–10 Hz interval, β2 is equal to the regression coefficient of the y-coordinate in the test interval (e.g., 1–5 Hz), SEβ12 is the squared standard error of the β1 coefficient, and SEβ22 is the squared standard error of the β2 coefficient.
Next, we wanted to identify any trends in specific ROI pairs driving regional decreases in phase-locking. First, we sorted ROIs according to the magnitude of the summed beta weights. We then further probed the top 5% of these ROIs (n = 16), which represents the 16 ROIs undergoing the greatest amount of developmental decrease in phase-locking. Of those 16 ROIs, we further thresholded each ROI’s specific interactions with other ROIs to maintain only the top 5% of each ROIs pairwise beta weight (n = 16 pairwise interactions for each of the 16 ROIs), resulting in a total of 256 pairwise beta weights demonstrating the greatest rate of ROI-ROI decrease in phase-locking.
We wanted to ensure any age-related changes we observed in PLV was not due to changes in the total amount of activity (power) in an area within any given frequency band [64]. First, we extracted a power estimate for each ROI. Specifically, we calculated relative power (see “Power calculation”). We then extracted relative power in the 5–9 Hz frequency band within subjects by taking the mean power within this frequency range for each ROI and dividing by broadband total power (1–49 Hz) for each ROI. For each ROI within each subject, this procedure resulted in relative theta band power. We then averaged across subjects to obtain a mean relative theta band power for each ROI. This value was then plotted against each ROIs y-coordinate to determine the anterior-to-posterior gradient in power across the cortex. Because a significant anterior-to-posterior gradient in power was observed (more power in posterior regions), we included as nuisance regressors the power of each ROI, the interaction between each ROI pair, the log-transformed power of each ROI, and the log-transformed interaction term of each ROI pair into the age models for each ROI pair. Additionally, matching the PLV analysis pipeline, we regressed power onto age at every frequency interval ranging from 1–49 Hz in 1 Hz increments.
During MaxFilter preprocessing, continuous head position estimates are calculated, and any large or sudden head movements are recorded. While MaxFilter performs head movement correction by aligning sensor data to a common reference, it does not account for artifacts generated by head movements, and we wanted to ensure any effects were not a result of head motion artifacts. After extracting the estimated movements from the MaxFilter output, we used the translation vector and rotation matrix for the head position relative to the sensor array (obtained from coregistration) to calculate a three-dimensional head movement vector relative to each sensor at each time point. The norm of this movement vector was averaged across sensors to obtain a single measure of head motion. This motion estimate for each subject was included as a nuisance regressor in all regression models involving the analysis of age-related changes in PLV.
Prior to the neuroimaging visit (M = 43.61 days, SD = 43.33 days), a subsample of participants (n = 62) completed the UPPS-P Impulsive Behavior Scale [48,89–92], either in an online screening (n = 28) or a separate behavioral visit (n = 34). In the current analysis, total impulsivity scores were estimated according to procedures outlined by [48]. We then regressed age onto this total impulsivity score and observed a significant negative linear relationship between total impulsivity and age (see Results).
To determine overlap between links demonstrating a significant PLV × Age relationship and a significant PLV × Impulsivity relationship in a nonarbitrary, data-driven manner, individual subject theta band PLV matrices were submitted to the NBS [50], and a t test was run between adolescents and adults to extract a cluster of regions with a significant decrease in theta PLV with age. We then performed the NBS on the relationship between impulsivity and theta PLV, controlling for age. A total of 3 connections overlapped between the 2 models and were subsequently confirmed using mediation analysis.
To examine whether differences in PLV may account for age-related differences in impulsivity, mediation analysis was performed on PLV values within connections that had significant associations with (1) age and (2) impulsivity (while controlling for age), as defined above. Significance values for indirect effects were obtained using 5,000 draws in a bootstrap procedure [93].
To determine whether resting-state delta band or beta band power mediated the relationship between age and impulsivity, similar to the PLV analysis, we tested each ROI across these 2 frequency bands for mediation effects. Significance values for indirect effects were obtained using 5,000 draws in a bootstrap procedure, as was done previously.
The spatial working memory task was modeled on the classic Sternberg working memory paradigm. Cue stimuli were yellow circles appearing in 1 of 8 possible locations. Each trial began with fixation followed by a presentation of 3 frames (300 ms each) showing one cue stimulus at a time in either the same location or 3 different locations. A blank grid was inserted between the frames for 200 ms to decrease chunking and motion perception. A 1,500 ms (50% of trials), 3,000 ms (25% of trials), or 4,500 ms (25% of trials) delay period was used to minimize habituated preparatory responses.
Following the delay period, subjects made a button press to indicate whether a frame showing 4 circles located among 8 possible locations had occurred in any of the previous cue locations (50% of trials) or were all in novel locations (50% of trials). A total of 144 high load trials and 144 low load trials were distributed across 12 runs, with the order randomized within runs. Intertrial fixation intervals ranged between 1,000 and 4,500 ms, with a short break between runs. The task was designed and run using E-Prime (Psychology Software Tools, Inc., Pittsburgh, PA).
MEG data were first manually inspected for flat or noisy channels that can arise due to sensor malfunction, and these channels were removed from further analysis, as excessively noisy or flat channels may adversely impact further preprocessing steps and data analysis. The maximum number of channels excluded within a single participant was 23. As we did with the resting-state data, we attenuated environmental noise using the MaxFilter software to apply tSSS [80]. If at any time during a trial the total displacement of MEG sensors relative to the head was greater than 5 mm, the trial was rejected from all future analyses. Across all participants, only 38 total trials were dropped for head motion, with at most 4 trials dropped for head motion within a single participant.
The remaining preprocessing steps were applied using tools in the MNE Python package [83]. First, the data were band-pass filtered to the frequency range of interest (1–49 Hz) using a 10-second overlap-add FIR filter. Cardiac, eye blinks, and eye movement (saccade) artifacts are not identified by tSSS because they originate from the subject's body, so we used an ICA method to attenuate these artifacts, similar to the resting-state methods. The shapes of the automatically detected artifactual components were checked visually to verify the selection of artifactual components, and the selection of components was then amended in the rare cases that the automatic procedure failed to identify components that showed clear EOG or ECG patterns. Finally, trials were screened for remaining sensor jumps, muscle artifacts, or saccade artifacts by checking for magnetometer amplitudes that exceeded 2.5 × 10−10 T or gradiometer amplitudes that exceeded 4 × 10−10 T/m; no further trials were rejected by these criteria.
During the experiment, trial event onset times were recorded into a digital stimulus channel through the E-Prime software. The event timings and codes from this channel were checked against E-Prime log files to remove spurious events that occurred in some runs due to software timing synchronization glitches. Based on this verified trial event data, trials with incorrect or omitted responses were removed because we are interested only in trials during which working memory was successfully engaged. In addition, a total of 10 trials across all participants were rejected due to mismatches between stimulus channel event codes and timing reported by E-Prime, with at most 4 trials dropped from a single subject for this reason.
After preprocessing, we extracted the first 1,500 ms of the maintenance period from the task and calculated the PLV between each of the 333 ROIs in the 5–9 Hz frequency range, following the resting-state analysis pipeline. For each ROI pair, we then regressed the PLV onto age, controlling for subject head motion. Next, the beta weight from the age regressor was extracted from each model, and beta weight matrices were constructed. As in the resting-state analysis, we summed down the columns of the matrix to get a summed beta weight representing the total linear age effect. We then regressed this value for ROI against the ROI’s anatomical y-coordinate and did not observe any anterior-to-posterior effects (t = −0.02, p = 0.98).
|
10.1371/journal.pcbi.1004119 | Discovering Anti-platelet Drug Combinations with an Integrated Model of Activator-Inhibitor Relationships, Activator-Activator Synergies and Inhibitor-Inhibitor Synergies | Identifying effective therapeutic drug combinations that modulate complex signaling pathways in platelets is central to the advancement of effective anti-thrombotic therapies. However, there is no systems model of the platelet that predicts responses to different inhibitor combinations. We developed an approach which goes beyond current inhibitor-inhibitor combination screening to efficiently consider other signaling aspects that may give insights into the behaviour of the platelet as a system. We investigated combinations of platelet inhibitors and activators. We evaluated three distinct strands of information, namely: activator-inhibitor combination screens (testing a panel of inhibitors against a panel of activators); inhibitor-inhibitor synergy screens; and activator-activator synergy screens. We demonstrated how these analyses may be efficiently performed, both experimentally and computationally, to identify particular combinations of most interest. Robust tests of activator-activator synergy and of inhibitor-inhibitor synergy required combinations to show significant excesses over the double doses of each component. Modeling identified multiple effects of an inhibitor of the P2Y12 ADP receptor, and complementarity between inhibitor-inhibitor synergy effects and activator-inhibitor combination effects. This approach accelerates the mapping of combination effects of compounds to develop combinations that may be therapeutically beneficial. We integrated the three information sources into a unified model that predicted the benefits of a triple drug combination targeting ADP, thromboxane and thrombin signaling.
| Drugs are often used in combinations, but establishing the best combinations is a considerable challenge for basic and clinical research. Anti-platelet therapies reduce thrombosis and heart attacks by lowering the activation of platelet cells. We wanted to find good drug combinations, but a full systems model of the platelet is absent, so we had no good predictions of how particular combinations might behave. Instead, we put together three sources of knowledge. The first concerned what inhibitors act on what activators; the second concerned what pairs of activators synergise together (having a bigger effect than expected); and the third concerned what pairs of inhibitors synergise together. We implemented an efficient experimental approach to collect this information from experiments on platelets. We developed a statistical model that brought these separate results together. This gave us insights into how platelet inhibitors act. For example, an inhibitor of an ADP receptor showed multiple effects. We also worked out from the model what further (triple) combinations of drugs may be most efficient. We predicted, and then tested experimentally, the effects of a triple drug combination. This simultaneously inhibited the platelet’s responses to three stimulants that it encounters during coronary thrombosis, namely ADP, thromboxane and thrombin.
| Cells are subject to diverse stimuli in vivo, and combine these inputs to generate appropriate biological responses. Activators and inhibitors of various targets work together in different configurations to elicit valuable and sometimes unpredictable outcomes, both natural and therapeutically induced. Many therapeutic approaches combine multiple agents acting on different targets, for example in cardiovascular disease[1], cancer[2–4], and infection[5]. Ideally, we would have a full systems model of every clinically important signaling process, helping us to predict and define potent combinations. However, in many systems, such a model is largely absent. Accordingly many workers seek to simply study the combination effects without considering additional information regarding the signaling network. Thus, screens for novel agents can take a systematic approach[6,7], but are limited usually to comparing the inhibitor combinations to the effects of single agents, without considering wider aspects of the signaling system.
However, the discovery of synergistic effects is not trivial. There is a large set of compounds that target distinct proteins, and considering the pairwise or higher order combinations of all of these is a very substantial task. Accordingly, such screens are frequently performed under a very limited set of experimental conditions. However, in many physiological contexts, cells may be subject to diverse challenges, and it would therefore be ideal for a synergistic combination of drugs to be effective under not just one, but under many alternative conditions. To meet this challenge, systems biology approaches seek to develop integrated computational predictive models of an entire signaling process, and ultimately of a cell, tissue or organism. These models are valuable but often challenging, since their construction requires extensive experimental data, and for this reason they are often developed under relatively limited and controlled settings, such as that of a well characterized cell line. Thus, there is still a requirement to develop more efficient screening methods that by-pass the need for a complete model of a given system, but which capture the essential functional components of that system, as might be relevant in a therapeutic or other practical setting. In order to accelerate the discovery of critical combinations of factors, scientists can either take a bottom-up approach, starting with pairwise combinations and making combinations more complex, or a top-down approach starting with a set of factors and winnowing down the system to the essential components, such as was done to successfully choose 4 transcription factors from 24 that govern the generation of pluripotent stem cells.[8]
High intracellular levels of cAMP maintain platelets in a resting state[9], with prostaglandin I2 (PGI2) and nitric oxide (NO), sustaining the production of cAMP via Gs[10] or limiting its degradation through the cGMP-dependent action of phosphodiesterase III[11]. On the other hand, platelet activators inhibit adenyl cyclase and reduce cAMP via GαI, while βγ subunits of Gi type proteins activate PLC and phosphoinositide 3-kinase (PI3K). The coordinated activity of different types of G proteins is required to modulate platelet behaviour. Platelet activation through G proteins involves Gαi Gαq and Gα12/13[12], with the thrombin receptor, PAR1, acting through all three [13–15] and favouring Gαq-mediated calcium mobilization over Gα12/13 signaling when stimulated with thrombin-receptor activating peptide (TRAP) [16]. TxA2 receptors couple to Gαq, Gα12 and Gα13 [14,17,18]. Platelet responses to epinephrine are mediated by the α2A-adrenergic receptors[19], acting in mice through the Gαi family member Gαz[20]. ADP signalling in platelets, important for sustained aggregation[21], is via GPCRs P2Y1 (coupled to Gαq in mice[22]), and P2Y12 (coupled to Gαi2 in mice[20]). The activation of GPVI (the only non-GPCR receptor targeted in our study) by Collagen or CRP leads to Lyn and Fyn phosphorylation of the FcR gamma-chain[23], allowing Syk docking[24] and activation of phospholipase C (PLC)γ2 [25] and Phosphoinositide 3 kinase (PI3K) [26,27].
Our goal was to develop efficient and practical methods to identify combinations of platelet inhibitors that would be robust in inhibiting platelets under multiple conditions, and would provide insights into platelet signaling networks. We sought to expand inhibitor combination screening by the incorporation of additional information that might give some insights into the performance of the platelet as a system.
The first step in developing our method was to investigate which inhibitors act against which activators[28]. Intuitively combinations of inhibitors are likely to be markedly synergistic when they are acting on parallel pathways. However, it has been shown that under certain feedback conditions, strong synergistic effects will be seen between upstream and downstream points that are located serially along a pathway [7]. Thus, we had no strong expectations of which combinations might show the strongest synergy. We noted that the available consensus that defines the relationships among activators and inhibitors of most signaling systems is frequently based on primary observations that are accumulated in the scientific literature in a piece-meal fashion. Since separate studies may often apply either subtly or grossly different experimental conditions, it is not ideal to simply take the accepted consensus of opinion to pair activators and inhibitors together on the basis of their literature defined targets, but it is of interest to re-evaluate these relationships in a systematic way. The second step in identifying useful combinations was to experimentally evaluate synergistic effects[29,30]. Synergy is defined as a functional interaction between two reagents that shows a much greater effect than expected, based on the known effects of the two reagents alone. There are multiple different definitions of what is precisely meant by synergy[31], and these different definitions may be considered to lie on a spectrum of tests, ranging from weak tests that provide only a suggestion of synergy, and strong tests that provide more robust evidence for such synergy. Typically, the more robust tests rely on the analysis of multiple doses of the two compounds alone and in various combinations. Such synergy studies may rely on analysis of synergies among inhibitors[1,6,7]. However, synergy studies are not confined to examine synergy among inhibitors, even when inhibition is the primary therapeutic goal. Investigation of synergies among activators[32] can assist in defining the profile of inhibitory effects of single and combination inhibitors, which reduce not only the main effects of the activators, but also provide information regarding their synergistic effects.
Since activator-inhibitor relationships, activator-activator synergy and inhibitor-inhibitor synergy each provide insights into the complex network of interacting factors that help in choosing inhibitor combinations, we set out to develop a practical framework integrating all three approaches (S1 Fig.). We integrated this information into a predictive model, and evaluated whether predictions of the model could accelerate the discovery of compound combinations effective at targeting platelet inhibition. This approach predicted a triple combination of compounds that was experimentally validated.
Informed consent was obtained from all subjects for the donation of blood samples for the purpose of platelet function analysis, with study approval obtained from the Royal College of Surgeons in Ireland Research Ethics Committee (REC679b).
Experimental methods followed a previous study[33]. Washed platelets were prepared from venous blood of consenting healthy donors drawn via phlebotomy into 15% (v/v) acid-citrate-dextrose (ACD) anticoagulant (38mM citric acid anhydrous, 75 mM sodium citrate, 124mM dextrose). Blood was centrifuged at 150 x g for 10 minutes at room temperature and platelet rich plasma (PRP) was collected and acidified to pH 6.5 with ACD. 1 μM prostaglandin E1 (PGE1) was added prior to centrifuge PRP at 720 x g for 10 minutes. The resulting pellet was resuspended in JNL buffer (6 mM dextrose, 130 mM NaCl, 9 mM NaHCO3, 10 mM sodium citrate, 10 mM Tris base, 3mM KCl, 0.81 mM KH2PO4 and 0.9 mM MgCl26H2O, pH 7.35) adjusting the concentration to 3x105 platelets/ μl. Washed platelets were supplemented with 1.8 mM CaCl2 immediately prior to the experiment.
The ADP release assay used white 96-well plates (white plates with white wells; Sigma-Aldrich, Ireland). Platelets were incubated with inhibitors for 10 minutes at 37°C on orbital slow shake using a Wallac 1420 Multilabel Counter (Perkin Elmer). 10 μl cocktail (K) or activators were then added and allowed to activate platelets for 10 minutes in the same conditions used with the inhibitors. 10 μl of the detection reagent Chrono-lume (Chronolog; Labmedics Limited, UK) were added and sample luminescence detected after an additional 5 seconds with rapid shaking measuring arbitrary absorbance units (AAU).
The compounds used as platelet activators were CRP (Ca, triple-helical Collagen-related peptide from 0.013 to 30 μg/ml; purchased from Dr Richard Farndale, Cambridge, UK), U46619 (Xa, from 0.003 to 6 μM; Santa Cruz Biotechnology, Germany), TRAP (Ta, Thrombin Receptor Activator Peptide sequence SFLLRN from 0.25 to 16 μM; Sigma-Aldrich, Ireland), Epinephrine (Ea, from 0.001 to 30 μM; Chronolog, Labmedics Limited, UK), and ADP (Aa, from 0.137 to 100 μM; Chronolog, Labmedics Limited, UK). Hill coefficients and response to single agents was evaluated in 4 donors. EC50s and EC90s were determined with GraphPad Prism software, which uses the equation Y = Bottom + (Top-Bottom)/(1+10(LogEC50-% inhibition)*HillSlope). The 2xEC50s were obtained by simply doubling the EC50s. In the case of ADP, to avoid doses higher than 20 μM that might interfere with the assay (S4 Fig.), 10 μ M was used instead of the actual EC50 (∼50 μM). The letter used to represent each compound denoted the selected dose for each, the letter followed by “2” to denote a dose that is double the selected dose, and the letter followed by “90” to denote a dose that causes the 90% activation (S1 Table).
A mother solution of the “activator cocktail” (K), which is all the activators at their selected doses (0.025 μ M Epinephrine (Ea), 0.5 μ M U46619 (Xa), 1 μg/ml CRP (Ca), 4 μM TRAP (Ta), and 10 μM ADP (Aa)) was prepared and serial 1:2 dilutions were used to stimulate platelets. Its EC50, was found to be 0.1636 fold the concentration of the mother solution (S2 Fig.), and this dose was used for cocktail activation in tests of inhibitor synergy. The rationale for choosing this dose was that this was the dose that gave a 50% activation of platelets, which should be relatively sensitive to inhibition by inhibitors or inhibitor pairs: if a higher concentration of the cocktail had been used, it is possible that the platelets would be consistently activated in a way that masked many inhibitory effects or inhibitor combination effects. It is slightly less than the five-fold reduction that would be obtained were the doses to be crudely divided by the number of activators. These doses lie below the individual EC20 values for all five activators (S2 Fig.).
To determine inhibitor IC50s, we evaluated ADP release induced by different doses within a range specified in parentheses. Inhibitors used were Wortmannin (Pi, from 0.137 to 100 nM; Sigma-Aldrich, Ireland), SQ29548 (Xi, from 2.195 nM to 1.6 μM; Enzo Life Sciences, UK), BMS200261 (Ti, from 0.000685 to 0.5 nM; Sigma-Aldrich, Ireland), Yohimbine (Ei, from 15.625 nM to 2 μM; Sigma-Aldrich, Ireland), and MRS2395 (Ai, from 0.137 to 100 μM; Sigma-Aldrich, Ireland). All were dissolved in water except MRS2395, which was dissolved in ethanol, where the ethanol proportion was equal to or less than the 0.37% of the total volume. Platelets were pre-incubated with the inhibitors and then stimulated with the activator cocktail. Cocktail-stimulated platelets were almost completely insensitive to Wortmannin inhibition and therefore the IC50 for Wortmannin was determined on platelets stimulated with 1 μg/ml of CRP.
The 10 consenting healthy donors were all Caucasian between 24 and 42 years of age. Each plate harboured four types of treatments (single agents, activator/activator combinations, inhibitor/inhibitor combinations, activator/inhibitor combinations) and two types of controls (resting and cocktail-activated platelets). Two different arrangements of wells were used in order to limit position effects and, since the results for the two plate layouts broadly correlated, a dataset was assembled from 10 consenting healthy volunteers.
To account for donor/plate variation, analysis was of the rank within each donor of the observed ADP level. Statistical analysis was performed using STATA version 12.0 [34] and the fitting of the final models confirmed using R [35]. Visualisations of data for Fig. 1 and for S3 Fig. (below), were constructed using R[35].
The visualizations were performed using either the basic visualization package or the gplots package in R. The clustering (S3 Fig.) was performed using the hclust function of R, which performs hierarchical clustering (each object is assigned to a cluster, and then the two most similar objects/clusters are joined in one cluster; and so on iteratively until one cluster is created). A one-tail Wilcoxon test was used to test the significance of whether activator-activator and inhibitor-inhibitor combinations were superior to either of the double doses of the component reagents. Raw data were converted to logarithms to the base 10 for visualisation. A small number of duplicate treatments within an individual (ADP for group 1 and Epinephrine for group 2) were replaced by their respective means.
Main effect terms were held fixed, while interaction terms were added using a forward stepwise multiple regression approach (adding terms that significantly improved the model, p<0.05). The pair-wise interactions were tested by fitting pair-wise interaction terms, along with main effect terms. We present results for synergies of inhibitors (the two inhibitors together inhibit much more strongly than expected) or activators (the two activators activate much more strongly than expected); other significant synergistic interactions were not seen.
We defined significant interaction as observation that the double doses of the activators on their own BOTH have significantly less activating effects than the combination in single doses (two Wilcoxon one-tailed tests with P<0.05 for each, Fig. 1). This approach may be beneficial when reagents lack clear dose response relationships[31]. It is equivalent to a limiting case of Loewe additivity, effectively sampling a single point on the isobole when activators have similar potency [30,31].
To integrate the three strands of information, we took the significant interactions identified in the double Wilcoxon test for synergy, and the significant activator-inhibitor combination terms identified from the stepwise linear regression modelling. We brought those forward into an integrated model, including the main effects for each activator and inhibitor.
The inhibitor-inhibitor and activator-activator testing component of the statistical study design was based on a sequential test, namely to test inhibition combination first against one double dose (one-tailed test, p < 0.05), and then against the second double dose (second one-tailed test, p<0.05). No algorithms are available to calculate the power of this approach. Nevertheless, the study design may be informed by the assumption, when two inhibitors each confer a roughly equivalent effect, that this test is equivalent to a test of the inhibitor combination versus either double dose. Assuming a log ADP intensity of 5.2 for a double dose of inhibitor, and 4.9 for a dual inhibitor combination (s.d. = 0.2), in order to have 90% power to detect a significant difference (two-tailed, p< 0.05), a sample size of 10 subjects is required.
Input, analysis code and output is given in two alternative statistical analysis environments, R and STATA. The same results are obtained using either. The input is the complete analysis dataset presented in the main paper.
We investigated reagents thought to act primarily on six proteins in pathways of major therapeutic interest in the inhibition of platelet function[33], denoted by single letters as follows: Thromboxane Receptor (X), Thrombin PAR1 Receptor (T), P2Y12 ADP receptor (A), Epinephrine Receptor (E), PI3 Kinase (P), and GPVI Collagen Receptor (C). The suffix “a” was used to indicate a reagent that activated the protein, and “i” for a reagent thought to inhibit it (so that Xa denotes Thromboxane Receptor activator and Xi its inhibitor). There was no inhibitor available for GPVI, and an inhibitor of PI3 kinase was included because of its inhibitory effects on GPVI stimulated activation. Dose response curves for the activators and inhibitors used in the study (S2 Fig.) were used to select doses for use in combination studies (S1 Table). Visualization of the assay results indicated strong donor variability (Fig. 2). Accordingly, subsequent analysis was performed on the rank of the assay result within each donor dataset, thus correcting for donor effects during analysis.
Activator-inhibitor combinations are summarized in Fig. 3A, with more detailed plots in Fig. 4. The expectation was that effects would largely be seen along the diagonal, corresponding to the a priori pairing of activators and inhibitors. In order to make it easier to see to what extent pairings match or depart from that expectation, we adjusted the data for visualisation purposes, where the values represent the mean values in panel A, minus the mean value observed for the single dose activator alone. Two of the combinations strongly match our expectations (Xa/Xi, and Ta/Ti). However, any combinations involving the ADP inhibitor (Ai) showed a marked departure from expectation, since its extent of inhibition of ADP activation (Aa) was markedly less than that of Ca and Xa (Fig. 3A and 3C). In spite of markedly inhibiting Ca and Xa, Ai did not manage at that same dose to prevent some activation by Aa (Fig. 3A) This suggests that it is not acting as a very efficient inhibitor of its intended target, but may be acting via other mechanisms. Overall, epinephrine (Ea) had weak activatory effects and its inhibitor yohimbine[36] (Ei) had weak inhibitory effects, which may explain why the model did not detect synergies involving this activator-inhibitor pair. It is possible that the doses of epinephrine defined in advance were inappropriate for the particular donors in this study. To evaluate the significance of the observed combination effects, we carried out multiple regression modelling. The regression model was fitted by including a parameter for the main effect for each of the activators and inhibitors. Each additional significant activator-inhibitor combination term (given a value of 1 if the experiment included both the activator and inhibitor; zero otherwise) between a particular inhibitor and a particular activator was added as a parameter in a stepwise fashion until no additional significant terms (p<0.05) could be added. An initial model that included only activator and inhibitor effects alone explained 68% of the variance (S2 Table). This rose to 73% when specificity of action was considered, by including four additional significant activator-inhibitor combination terms (S3 Table). We considered whether a Boolean representation of activator-inhibitor relationships (e.g. that inhibitor Ai cancels out entirely the effect of activator Ta) would model the data adequately. However, a Boolean model of the activator-inhibitor relationships explained less of the variance in the data and provided a significantly poorer fit (p<10–5; S4 Table).
Significant inhibition (Fig. 3B and 3D) was observed for two activators by the inhibitors normally associated with their receptors (Ti/Ta and Xi/Xa). While GPVI Collagen receptor activation (Ca) is thought to be strongly mediated by PI3K [33], inhibiting PI3K (with Pi) had similar effects on Ca as it had on Xa and Ta responses, indicating that Pi is not highly specific for GPVI inhibition, and that its target PI3K may be a convergence point for different signalling routes. Most strikingly, the presumed ADP P2Y12 inhibitor Ai (MRS2395) inhibited other activators (Ai/Ca; Ai/Ta, and Ai/Xa) significantly, and more strongly than it inhibited ADP activation. This may be consistent with either a central role for the P2Y12 receptor in mediating signalling via many receptors, or with an alternative target of action of the drug. Regardless of the mechanism of the observed effect, this first strand of evidence highlights the influence of Ai on multiple activators. This suggests that Ai is a promising candidate to include in a set of compounds to inhibit platelets in combination.
Significant synergy was defined here as a much greater effect of a combination of two reagents than the double doses of either reagent (requirement to pass two one-tailed Wilcoxon tests, each with p<0.05). While more conservative than other approaches[37], it avoids statistical difficulties when effect sizes of different reagents are imbalanced, sampled from non-equivalent points on their respective dose response curves, or where reagents do not have standard dose response curves. Activator-activator synergies are summarized in the bottom left triangle of Fig. 3B, and the same observations after adjustment for differences in main effects of activators in the bottom left triangle of Fig. 3D. The detailed results are shown in Fig. 5. Fig. 3D displays the difference of the activation or inhibition from the most effective double dose of either the first or the second agent within the combination. Two significant activator-activator synergies were identified: activators of the ADP and collagen receptors (Aa and Ca) synergised significantly, and activators of the ADP and thromboxane receptors (Aa and Xa) synergised significantly. This second strand of evidence suggests that concurrent inhibition of platelets activation elicited by Aa, Ca and Xa may be useful in lowering the activation of platelets in the presence of multiple activators. Again, it particularly points to an important role for the ADP receptor in activation.
We tested the effects of inhibitors on the activation of platelets by a cocktail of all five activators, since such a cocktail may be physiologically relevant, and may be more sensitive to inhibitor synergies. The cocktail activation of platelets showed a steep dose response consistent with likely cooperative (synergistic) activity (S2 Fig.). We chose a dose of this cocktail that yielded 50% activation (see Methods), intended as a non-saturating combination activator to be used in inhibitor experiments. While it is likely that this cocktail is more dominated by particular activators, it was notable that, while double doses for four of the five inhibitors had difficulty overcoming the activatory effect of this cocktail, eight of the ten inhibitor combinations lowered platelet activation somewhat (Fig. 3D). This indicated that the doses of activators used in the cocktail were showing sensitivity to inhibitor combinations, but much less sensitivity to double doses of single inhibitors. Thus, the dose of cocktail employed in the study appeared to be appropriate for the purpose of detecting synergies among inhibitors, avoiding saturation effects.
As before, synergy was defined for each pair of inhibitors whenever the combination of inhibitors had a significantly greater effect than either of the inhibitors in a double concentration (Wilcoxon p<0.05 for both comparisons). We observed three significant inhibitor-inhibitor synergies, which involved the pairwise combinations of the inhibitors of Thromboxane Receptor, Thrombin Receptor and PI3K (Fig. 3B and 3D; Fig. 6; Xi/Ti, Xi/Pi, Pi/Ti). This third strand of evidence provides a different perspective from the activator-inhibitor and activator-activator combinations, raising the question of how to reconcile these findings into a single model that makes useful predictions.
The goal of anti-platelet therapy is to effectively inhibit platelet activation exposed to multiple challenges. We wished to define what combination of inhibitors would most effectively inhibit platelet activation brought about by several stimuli. In particular, a researcher faced with all the visually displayed information in Fig. 3 would typically find it hard to anticipate what the likely effect of three way combinations might be. Ideally, the different strands of information should be weighted in a sensible way, that is proportional to the degree of evidence supporting each set of data, to predict an outcome of interest to the investigator. To address this, we created an integrated model. The primary data we used in building the model involved pairwise and main effects, but does not provide direct experimental information regarding three-way or higher order synergies. While pairwise synergies are typically the most important [38,39], it is still of interest to investigate further synergy. To combine the three strands of information, we took (i) the linear regression model derived from the activator-inhibitor combination analysis, that already included all main effects and four activator-inhibitor combination effects, and added (ii) the two significant activator-activator synergy and (iii) the three significant inhibitor-inhibitor synergy terms identified above. These parameters were then fitted together in a unified multiple regression model predicting platelet activation. The resulting “integrated model” thus considers simultaneously all the platelet activation data, comprising resting and cocktail activated controls, single doses, and the various combinations of activators and inhibitors (Fig. 7A; S4 Table). As expected, adding the two additional strands of synergy data resulted in a significantly better fit to the data (p<0.0001, S4 Table).
Fig. 7A provides a visual representation of the model that can help advance understanding and interpretation of drug combination effects in platelets. We set out to exploit this integrated model to make predictions of the most effective trios of platelet inhibitors. We considered the scenario where a platelet is challenged by all five activators: collagen, epinephrine and activated thrombin, plus ADP and thromboxane release from adjacent platelets, as may occur during coronary arterial platelet plug formation in the presence of a ruptured atherosclerotic plaque. The integrated model (S4 Table) was applied to predict the ADP release for each of the 32(25) possible three-way combinations of the single dose inhibitors. This enabled us to predict how well each combination could inhibit platelet activation (S5 Table). The most effective predicted combinations all included Ai (the ADP receptor inhibitor). Of these combinations, the most effective trio of inhibitors identified was a combination therapy targeting ADP, thrombin and thromboxane signalling (Ai, Xi and Ti). We experimentally tested whether Ai, Xi and Ti together strongly inhibit the five-activator cocktail. As a comparison, we also considered whether adding a PI3K inhibitor (Pi) to Ai and Xi would be as efficient; this acts as a control combination, since the integrated model predicted that it would not result in such a strong inhibition of platelet activation (S5 Table). Fig. 7B indicates that while the Ai/Xi/Ti combination favoured by the model exhibited a marked inhibition of platelet activation, the less favoured Ai/Xi/Pi combination showed much less inhibition (p = 0.0003). This experimental validation of the model indicates that the integration of these three sources of data into a single model can aid in pinpointing higher order effective drug combinations. The model is also useful when trying to determine how much of the pattern of platelet activation in the system remains unexplained, for example by assessing model fit and exploring donor response variability (See S1 Text).
Our method demonstrates that a systematic approach to considering pairwise reagent interactions can lead to the discovery of particular combinations of importance in modulating biological activity, identifying a triple combination of platelet inhibitors that is particularly effective. It is of interest to also integrate our findings with what is known previously of platelet signaling (Fig. 8), so that we not only identify useful combinations of inhibitors, but also advance understanding of platelet signaling. TXA2R and PAR1 are the only known activators of G12/13 in platelets. PI3K is not a downstream effector of G12/13 and co-activation of both Gi and G12/13 is sufficient to activate platelets[40]. Thus, the synergy of Pi with both Xi and Ti makes sense, as two independent pathways (G12/13 and PI3K transmitted) are being targeted in parallel. This suggests that the engagement of both pathways may be required for full activation. By the same logic, since they share a common effector pathway, it is not surprising that there is no significant synergy between Xa and Ta. However, paradoxically, the inhibitors Xi and Ti synergise strongly. This suggests that activation and inhibition states of these two receptors are not simple on-off switches. In endothelial cells TRAP causes the engagement of Gq prior to the engagement of G12/13 [16]. There may be relatively subtle dose dependent effects, such that the spectrum of G12/13 and Gα inhibition by a single versus a double concentration of Ti is not resulting in a balanced increase in the inhibition of both pathways. Alternatively, the difference between the lack of activator synergy and the presence of inhibitor synergy could reflect the presence of more than two conformational states of a receptor being induced by activators and inhibitors. This would be consistent with a multiple state model for the thromboxane receptor studied in a platelet-like cell system [41] where certain inhibitors, including Xi, act as inverse agonists, actually downregulating constitutive activation of the receptor. One explanation for the multiple inhibitory effects seen with Ai (MRS2395) is that it is a “dirty” compound with multiple targets, that is not as efficiently targeting P2Y12 as might be expected. Dirty compounds in principle may have the potential to exhibit multiple synergisms resulting from their diverse targets, but we noted that Ai did not synergize significantly with any of the other four inhibitors. Finally, in our activator-inhibitor screen we observed that while Pi(Wortmannin) predictably inhibited Ca (CRP-induced) response [26,27], its inhibitory effects were seen across multiple activators, most notably Xa (U46619-induced) response, in spite of the fact that the existing literature suggests that TXA2 mediated signalling might not immediately involve PI3K (Fig. 8). This paradox may potentially be explained by a second wave of signalling and secretion via PI3K following the initial induction of activation [42]. It is also possible that the platelet signalling network is altered in the inhibition experiments by the presence of the three additional activators (Ea, Ca and Aa), thus potentiating the synergy of the two inhibitors. The two most plausible explanations, of alternative receptor states versus alternative network wiring, may not necessarily be mutually exclusive, since alternative receptor states are likely to represent responses to alternative states of the signaling networks either intracellularly or extracellularly.
Linear modeling defined the activator-inhibitor effects, and in general such model parameterisation needs to be approached with some care to ensure that statistically sensible parameters correspond to biologically interpretable ones. The linear statistical modeling was then used to integrate the different effects of activator-inhibitor, activator-activator, and inhibitor-inhibitor effects only after synergistic activator-activator and inhibitor-inhibitor effects were predefined in a manner consistent with Loewe isobole analysis, comparing combinations to double doses of both constituents. This avoids some of the dangers of linear modeling in inferring statistically significant synergies under some model which does not correspond robustly to Loewe additivity. Overall, the combined experimental and modeling approach may miss some important interactions that would be detected if we had performed the analysis across the dose response curves of each reagent combination. Given the complexity of platelet signaling, we think it likely that other synergies will emerge at different doses, and with larger sample sizes, or different stimulatory or inhibitory conditions. Nevertheless, we believe our approach is a relatively efficient way of establishing the most critical features of the signaling system, particularly when ensuring that all assays are carried out on the limited material provided by each donor in the study. Statistically, our approach appears relatively robust but clearly is open to further development, in particular moving away from a two-stage analysis (defining synergy effects separately from activator-inhibitor effects, and then combining these). Future models that estimate the synergism simultaneously with the activator-inhibitor effects may increase the efficiency of such studies, and widen the applicability to a wider set of scenarios, for example testing the effects of genetic activatory and inhibitory factors on a phenotype.
Integrated modelling of activator-activator, inhibitor-inhibitor and activator-inhibitor combinations may accelerate the discovery of compound and drug combinations that will more effectively target disease states, not only in platelet signalling, but in other potential applications, including cancer therapeutics. Many drugs that are highly successful in the clinic may have a broader mechanism of action than initially hypothesised, often contributing to their clinical efficacy. The systematic approach implemented here provides direct observations of activator-inhibitor relationships that ignores pre-conceived notions regarding the specificity or generality of action of drugs. Thus, in our study, we had prior beliefs concerning the specificity of particular agents in preventing the activation of platelets by certain activators. However, the fact that these pre-conceptions were partly disproved under the particular conditions of our study did not prevent the study design and the computational modelling from identifying a useful triple combination. Clinically used anti-thrombotic regimens provide partial support for the proposed combination identified here, routinely combining inhibition of both ADP and thromboxane signalling[43]. Adding a thrombin receptor inhibitor to these two, as suggested by the integrated model and its experimental validation, is also indicated as a useful three-way combination by a separate study which indicated its apparent synergistic advantages[44]. Clearly, this experimental test of our prediction is relatively limited, considering only two three-way combinations for comparison. Applying modeling to define higher order combinations is likely to be of particular value in experiments with larger numbers of agonists and antagonists, where the number of three-way combinations becomes impractical to screen efficiently.
One approach to screening for synergy that has the potential to actually define whether the reagents are acting in serial or in parallel, is to investigate the response profile of synergy derived from investigating the compounds at different concentrations[7]. While our approach cannot resolve whether factors are in serial or in parallel, it does appear to be efficient at identifying interesting combinations. To get a deeper understanding of how the combinations work, they could be studied in combination with analyses of intermediate components in platelet signaling, such as the phosphorylation states of various proteins. Full systems modelling of the dynamics of intermediate signalling factors may more exquisitely and accurately achieve a similar goal to this study, but would need to model the activation states and kinetics of the “hidden” layer of receptors in Fig. 8, However, this requires collecting quantitative information on the states of these receptors in the presence of multiple combinations of activators and inhibitors. In many clinical contexts such data is difficult to collect, and thus a useful systems model is absent, and may be difficult to develop. Accordingly, synergy modelling integrated with activator-inhibitor combination screens provides a key step in moving beyond the capabilities of current synergy screens[32]. When novel therapeutic inhibitors of blood associated targets are likely to be prescribed in combination with existing therapies, and there are manipulable agonists of the multiple pathways targeted, we advocate initial ex vivo studies to define the combinatorial landscape and make predictions to help in the design of in vivo synergy combination trials in human subjects.
|
10.1371/journal.pcbi.1002415 | System-Level Insights into Yeast Metabolism by Thermodynamic Analysis of Elementary Flux Modes | One of the most obvious phenotypes of a cell is its metabolic activity, which is defined by the fluxes in the metabolic network. Although experimental methods to determine intracellular fluxes are well established, only a limited number of fluxes can be resolved. Especially in eukaryotes such as yeast, compartmentalization and the existence of many parallel routes render exact flux analysis impossible using current methods. To gain more insight into the metabolic operation of S. cerevisiae we developed a new computational approach where we characterize the flux solution space by determining elementary flux modes (EFMs) that are subsequently classified as thermodynamically feasible or infeasible on the basis of experimental metabolome data. This allows us to provably rule out the contribution of certain EFMs to the in vivo flux distribution. From the 71 million EFMs in a medium size metabolic network of S. cerevisiae, we classified 54% as thermodynamically feasible. By comparing the thermodynamically feasible and infeasible EFMs, we could identify reaction combinations that span the cytosol and mitochondrion and, as a system, cannot operate under the investigated glucose batch conditions. Besides conclusions on single reactions, we found that thermodynamic constraints prevent the import of redox cofactor equivalents into the mitochondrion due to limits on compartmental cofactor concentrations. Our novel approach of incorporating quantitative metabolite concentrations into the analysis of the space of all stoichiometrically feasible flux distributions allows generating new insights into the system-level operation of the intracellular fluxes without making assumptions on metabolic objectives of the cell.
| Fluxes in metabolic pathways are a highly informative aspect of an organism's phenotype. The experimental determination of such fluxes is well established and has proven very useful. To address some of the limitations of experimental flux analysis, such as when the cell is divided in multiple compartments, stoichiometric modeling provides a valuable addition. The approach that we take is based on stoichiometric modeling where we consider the thermodynamic feasibility of many different possible routes through the metabolic network of Saccharomyces cerevisiae using experimentally determined metabolite concentrations. We show that next to conclusions on single biochemical reactions in the metabolic network, we obtain system-level insights on thermodynamically infeasible flux patterns. We found that the compartmental concentrations of and NADH are the causes for the system-level infeasibilities. With the current advances in quantitative metabolomics and biochemical thermodynamics, we envision that the presented method will help gaining more insight into complex metabolic systems.
| Metabolic fluxes give immediate insights into the metabolism of a cell [1], [2]. Metabolic flux analysis has proven to be useful, for example for the determination of enzyme functions [3], for the identification of regulatory mechanisms in response to environmental perturbations [4], or as a tool in metabolic engineering [5]. The most common method to quantify metabolic fluxes uses labeled substrates, and the measured label distribution in intracellular metabolites is interpreted together with measured uptake and production rates by means of a metabolic network model [6].
Despite successful quantification of fluxes with flux analysis in different conditions, the method has several limitations. For example, it is limited to the main branches in central carbon metabolism and fluxes cannot be resolved per compartment [7], despite compartmentation being a highly relevant aspect of eukaryotes [8]. Moreover, today's flux analysis rests on a number of a priori assumptions, e.g. on reaction reversibilities or on relevant parts of the network [7], [9].
To improve flux quantification, we need additional constraints on the possible flux distributions in a metabolic network. In stoichiometric network analysis, the metabolic network is modeled as a collection of biochemical reactions where all internal metabolite concentrations are assumed constant [10]. Next to the typical constraints, such as uptake and excretion rates, reaction reversibilities and maximum flux capacities, the field recently began to incorporate thermodynamic information, whereby statements on feasibility of reaction fluxes or flux distributions can be made based on calculation of changes in Gibbs energy using metabolite concentrations [11]–[13]. For example, using flux balance analysis (FBA) and related approaches, metabolite concentrations were used as additional constraints to predict fluxes in the non-compartmentalized organism E. coli [14], [15] or in a model of liver metabolism [16].
Here, we develop a novel approach to integrate metabolite data into metabolic network flux analysis, to get additional insight into the compartmentalized flux physiology of Saccharomyces cerevisiae. The method combines network embedded thermodynamic (NET) analysis [12], elementary flux mode (EFM) analysis [17]–[19], and experimentally determined metabolome data. We employ EFM analysis instead of flux balance analysis because the collection of the generated flux modes can yield insight into all feasible flux distributions, as compared to the single thermodynamically feasible flux solution that is obtained with thermodynamically constrained FBA. Additionally, assumptions on a metabolic objective function of the cell are not required.
Our new approach to analyze the compartmentalized central metabolic network of S. cerevisiae using quantitative metabolite data acquired under glucose batch growth conditions allowed us to generate novel insight into the system-level causalities underlying the intercompartmental redox metabolism. Specifically we show that the and NADH concentrations in the cytosol and the mitochondrion do not allow for the ethanol-acetaldehyde redox shuttle to be active under the investigated condition. Further, we identified a number of maximal reaction activities that could be used as constraints for flux analysis or FBA. We envision that our method becomes a useful tool to unravel system-level insights about a complex metabolic system from metabolome data.
Our approach uses elementary flux modes (EFMs) to describe metabolic flux distributions. The concept of EFMs is well known for stoichiometric network analysis, and it provides a way to explore the flux solution space of a metabolic network that is commonly addressed with flux balance analysis (FBA). With all the EFMs of a metabolic network, any stoichiometrically possible flux distribution can be obtained by a non-negative linear combination of the EFMs [20]. Since we want to evaluate only thermodynamically feasible flux distributions, we demonstrate, as a first step towards the development of our approach, a new property of EFMs, which is that every thermodynamically feasible flux distribution is a non-negative linear combination of thermodynamically feasible EFMs.
The mass balanced flux solution space of a stoichiometric metabolic network can be described with a non-negative linear combination of its EFMs:(1)where any flux distribution is a sum of EFMs with coefficients .
As we show in the proof provided in Text S1, a thermodynamically feasible flux distribution only consists of thermodynamically feasible EFMs:(2)where the thermodynamically feasible flux distribution is only composed of EFMs from the feasible set, .
The mathematical proof demonstrates that by eliminating infeasible EFMs we do not loose feasible flux distributions, because any feasible flux distribution can be composed of only feasible EFMs. Specifically, in the hypothetical case that an infeasible EFM is part of a feasible flux distribution it must involve a cancellation or directionality change of a specific reversible reaction. In this case, the flux distribution can be decomposed into one or more feasible EFMs, and a feasible combination of an infeasible EFM with another EFM. The combination of the infeasible EFM and another EFM must then be either a feasible EFM by itself, or it must be possible to achieve the feasible combination by other feasible EFMs. Hence, Eq. (2) allows us to exclude thermodynamically infeasible EFMs from the complete set of EFMs without excluding thermodynamically feasible flux distributions. It is important to note that the resulting flux solution space defined by the feasible EFMs can still contain infeasible flux solutions because it is possible that the combination of multiple feasible EFMs leads to an infeasible flux distribution.
Exploiting that EFMs allow us to exclude thermodynamically infeasible EFMs, we aimed at developing an approach to generate novel insights into the complex flux physiology of the central metabolism of the yeast S. cerevisiae. Therefore, we assembled a 230 reaction stoichiometric network of its central carbon metabolism and amino acid synthesis pathways (cf. Materials and Methods) encompassing the cytosolic and mitochondrial compartments and many parallel pathways. With our approach we aim to obtain additional insight into the metabolic network operation, therefore we build upon current knowledge by defining the reversibilities in our model as they are defined in the original model [21].
First, we needed to calculate all EFMs. As the EFM calculation is computationally demanding, we initially applied steps to constrain the mass-balanced solution space as much as possible upfront, before we started enumerating EFMs (Fig. 1). Thus, in a first step, we performed flux variability analysis (FVA) [22] on the basis of the measured uptake and production rates of external compounds and biomass, to determine the reversibility of each reaction under the investigated physiological conditions, that is, for growth on glucose. From FVA, we obtain a minimum and maximum achievable flux for each reaction. A reaction is reversible if both a negative and positive flux can be achieved, else it is unidirectional. Reactions that have a minimum and maximum flux that are either positive or negative, and cannot be inactive, are reactions that are always active in the respective direction. Using this approach, we could classify 67 initially reversible reactions as unidirectional (Fig. 2, Dataset S1).
Next, we employed measured metabolite concentrations from glucose batch cultures, and NET analysis to identify additional reaction irreversibilities [12], [23]. For the metabolite data, we assembled published and unpublished data from glucose batch experiments, and generated a consensus data set to define lower and upper concentration limits for 55 metabolites (see Materials and Methods and Dataset S2). For NET analysis, we used the reaction activities inferred from FVA. With the consensus metabolite data set, we obtained constraints on the reversibility of three additional reactions (see Fig. 2). Another iteration of FVA and NET analysis with the obtained constraints as input did not yield any further constraints.
For the obtained condition-specific constrained metabolic network, we computed the EFMs and obtained 71.266.960 EFMs. Using NET analysis and the consensus metabolite data set, we classified 38.420.207 (54%) EFMs as feasible and 32.846.753 (46%) EFMs as infeasible. Assuming that the flux solution space in the metabolic network of an organism can be approximated by the number of EFMs, our result shows that roughly at most half of the solution space is thermodynamically feasible.
With the finding that 54% of all the EFMs are thermodynamically feasible we reduced the number of EFMs that can constitute a thermodynamically feasible flux distribution considerably. In a first analysis step towards generating insights into the flux distribution, we searched the feasible EFM set for reactions that only use a subset of the possible reaction directions compared to the complete set of EFMs. For each reaction in each EFM we determined whether a backward, inactive or forward reaction activity was used. Then, the possible reaction activities for each reaction were compared between the complete set of EFMs and the feasible set of EFMs.
Here, we found that the oxaloacetate transport from the mitochondrion to the cytosol is never used in an EFM of the feasible set, meaning that it has to be inactive under the investigated growth condition. Indeed, we find no contradicting evidence for the prediction when comparing with the experimental observation that a knock-out of the corresponding gene OAC1, whose translated protein is responsible for the respective oxaloacetate transport reaction, does not have an effect on growth rate under glucose batch conditions [24], [25]. Further, as we found that all EFMs with acetaldehyde transport out of the mitochondrion are infeasible, we conclude that during growth on glucose, acetaldehyde can only be transported into the mitochondrion. It is important to note that the EFMs with active oxaloacetate transport, or acetaldehyde transport out of the mitochondrion, are not infeasible because of the metabolite concentration constraints on the respective single reaction only, since single reaction infeasibilities are removed in the first NET analysis step before EFM generation (see Fig. 1). Instead, as we will show later, the infeasibility is the result of a system of coupled reaction activities, where all individual reactions need to be thermodynamically feasible simultaneously.
Next, we wanted to test whether the feasible set of EFMs differs from the infeasible set in terms of reaction rates. Such a comparison is possible by normalizing the reaction rates in each EFM to the glucose uptake rate of the EFM (all EFMs have glucose uptake, EFMs without glucose uptake are internal cycles and they were removed because they are physiologically meaningless [26]). Principal component analysis (PCA) of the normalized EFMs shows a clear difference between the feasible and infeasible EFMs in principal component 2 (PC 2 in Fig. 3). The reactions with the highest loadings in this component are alcohol dehydrogenase in both the cytosol and the mitochondrion (ALCD2x, ALCD2m) and acetaldehyde and ethanol transport to the mitochondrion (ACALDtm, ETOHtm), and these reactions are likely involved in causing thermodynamic infeasibilities. Note, that although there is a separation between feasible and infeasible EFMs because there are no feasible EFMs in the center of the graph, by combination of feasible EFMs it could still be possible to obtain a feasible flux distribution that would be projected in this area of the PCA. Therefore, the loadings of PC1 are not considered.
Next, we searched for the highest and lowest rate of each reaction in the complete set of EFMs and in the feasible set of EFMs to define the flux ranges that can be achieved in terms of flux per unit of glucose uptake. Any flux value in this range can in principle be achieved through a combination of EFMs. When comparing the flux ranges that can be realized by the feasible EFMs with the flux ranges of the complete set of EFMs, we find that eight reactions cannot assume the full range for thermodynamic reasons (see Fig. 4), with four of these reactions already having shown high loadings in the second principal component (see Fig. 3). These quantitative flux constraints result from metabolite concentrations and thermodynamics, and can be applied as constraints in flux balance analysis.
In this work, we demonstrated that an in vivo thermodynamically feasible metabolic flux distribution is only composed of thermodynamically feasible elementary flux modes. This EFM property allowed us to integrate EFMs and NET analysis into a novel approach to study the system-level properties of complex metabolic networks on the basis of quantitative metabolome data. As exemplified with a compartmentalized model of central metabolism in S. cerevisiae and cell-averaged metabolome data generated under glucose batch conditions, 46% of the 71.3 million EFMs were found thermodynamically infeasible, leading to direct insights into reaction directionalities, to constraints in several metabolic rates, and to the identification of reaction patterns that must be inactive due to a thermodynamic infeasibility.
This work builds on earlier work that integrated FBA, thermodynamics and quantitative metabolite data [14], [15], and extends it by using EFMs, allowing us to identify the reasons underlying the infeasibilities without making a priori assumptions on the metabolic objectives of the cell, such as maximization of biomass production, as is the case with FBA. Additionally, in our study we considered a compartmented metabolic network of S. cerevisiae to analyze cell-averaged metabolite data. Notably, the results we obtain are directly related to compartmentation, as can be seen from the identified infeasibility patterns that involve both compartments. The identified infeasibility of the ethanol-acetaldehyde redox shuttle has been previously identified using NET analysis [12] and manual consideration of the system. In this work we demonstrate that by using the flux patterns that are obtained from the EFMs we systematically identify such infeasibility patterns. Although the system-level constraints and their underlying causes can be rationalized without using EFMs, we need the generated EFMs to determine the infeasible patterns that are part of a stoichiometrically balanced flux distribution. In addition, because the number of EFMs can be considered approximately proportional to the flux solution space, we find that roughly half of the flux solution space is thermodynamically infeasible due to systems of reaction activities.
With the recently developed new group contribution method to estimate thermodynamic properties on a genome-scale [35], the recently increased availability in thermodynamic properties through experimental methods [36], the advances in quantitative metabolomics [37] and the now available methods to calculate EFMs also for large stoichiometric network [38], [39], we envision that the here presented approach will be helpful to shed light on metabolic flux physiology in more complex metabolic system such as higher cells simultaneously growing on multiple carbon substrates, where the applicability of classical flux analysis methods are still rather limited.
We used experimental data on metabolite concentrations for Saccharomyces cerevisiae obtained from four independent experiments with at least two replicates [40]–[42] with equal growth medium but under different cultivation conditions (bioreactor, shake flask, 96-well). Based on the data from the independent experiments, we constructed a consensus data set, where for each metabolite a minimum and maximum concentration was defined. The minimum and maximum concentrations were determined from all the replicates of measurements for each metabolite. To reduce the effect of outliers on the ranges, when more than 3 replicates were available, we removed the values higher than the third quartile +1.5 IQR (inter quartile range), and values lower than the first quartile −1.5 IQR. Physiological data was obtained for S. cerevisiae on glucose as carbon source from one of the four experiments. In Dataset S2 we describe the details of the experimental conditions of the data sets, the obtained concentration ranges and physiological data.
The stoichiometric metabolic network model describes the core central carbon metabolism of S. cerevisiae in the cytosol and mitochondrion with 230 reactions and 218 metabolites (see Dataset S3), and was developed on the basis of the genome-scale metabolic model iND750 that contains 1149 reactions and 646 metabolites [21]. For our model, we selected the cytosolic and mitochondrial reactions belonging to glycolysis/gluconeogenesis, pentose-phosphate pathway (PPP), TCA cycle, anaplerosis, pyruvate metabolism, and oxidative phosphorylation. The reversibility of each reaction was taken from Duarte et al. [21]. A cytosolic malate synthase was added to complement the glyoxylate shunt in the cytosol [43]. A citrate synthase was added to the cytosol since this is supported by localization studies [44].
To allow the model to synthesize all amino acids that are required for biomass, we added the following pathways: For L-alanine, two biosynthetic routes from pyruvate were included: cytosolic and mitochondrial alanine transaminase reactions, which were assumed to solely produce but not degrade L-alanine [45], [46]. Furthermore, L-glutamate could be produced via three alternative pathways: cytosolic or mitochondrial NADP-dependent glutamate dehydrogenase from alpha-ketoglutarate or mitochondrial NAD-dependent glutamate synthase from alpha-ketoglutarate and glutamine [47]. For glycine synthesis, we implemented three pathways such that it could be synthesized in the mitochondria via (i) alanine-glyoxylate transaminase [48], or in the cytosol by (ii) glycine hydroxymethyltransferase from serine [49], or (iii) from L-threonine via threonine aldolase [50]. As the latter reaction was assumed to be reversible, it could also be used to produce L-threonine, and such it constitutes a second possibility to produce L-threonine next to the linear pathway from L-aspartate. For all other amino acids, the model contains only one linear cytosolic pathway consisting of consecutive enzymatic reaction steps. Here, no alternative paths exist or they are excluded based on biochemical literature as it was done also to construct models for -based flux analysis [45], [46].
The model further includes transport reactions across the mitochondrial membrane for metabolites that participate in reactions in both the cytosol and the mitochondria. Additional transport reactions that were not contained in iND750 (i.e. for L-glutamate, alpha-ketoglutarate, homocitrate, glyoxylate, and 2-oxobutanoate) were added to properly connect additionally included alternative pathways for amino acid synthesis to the metabolic network.
The biomass composition was adopted from iND750 besides that trehalose and glycogen were discarded since carbohydrate storage was not considered in our model. Lumped reactions for synthesis of the remaining biomass constituents, i.e. lipids, nucleotides, and cell wall components from the corresponding precursors were determined based on the biomass composition as provided in iND750. In the model, carbon molecules that can be exchanged with the environment are glucose, glycerol, pyruvate, acetate, ethanol, succinate, and .
The model is not proton balanced. The reason for this is that it is close to impossible to do the proton balancing correctly (e.g., for transport reactions). Thus, we did not want to add any potentially wrong constraints on the model and therefore did not account for proton balancing, with one exception. We only balanced the protons around the respiratory chain by replacing the cytosolic protons produced and consumed in the reactions CYOR_u6m, CYOOm and ATPS3m by a unique species “hcyt”. As a result, ATP generated in ATPS3m can only occur through the respiratory chain.
With NET analysis we can determine the feasibility of a flux distribution based on ranges for the concentrations of the involved metabolites. A flux distribution is thermodynamically infeasible when one or more reaction activities conflict with the calculated Gibbs energy of reaction range(s). Conversely, a flux distribution is feasible when no conflicts are found. It is important to note that a metabolite concentration is constrained in NET analysis by any reaction that has the metabolite as a reactant. Therefore, a flux distribution can be infeasible due to propagated constraints in a pathway. The NET analysis implementation constrains metabolite concentrations of metabolites that occur in multiple compartments as a sum of the compartment specific concentrations, corrected for their volume. Therefore, the compartmental distribution of such metabolites is left free. The compartmental volume fractions of the cytosol and mitochondrion are set to 0.35 and 0.1, respectively. The NET analysis approach is similar to other thermodynamic analysis approaches [14], [15]. A main difference from other approaches is that with NET analysis we aim at checking flux distributions for thermodynamic feasibility, and at estimating ranges of Gibbs reaction energies and metabolite concentrations.
For NET analysis we used the concentration ranges defined from the experimental data. For all other metabolites in the network we assumed a default range with a minimum concentration of 0.0001 mM and a maximum of 120 mM, except for carbon dioxide (“co2tot”), phosphate (“pi”) and diphosphate (“ppi”) that were constrained to a range of 1 mM to 100 mM [51], [52], and oxygen (“o2”) that was constrained to 0.001 mM to 0.1 mM [53]. By using such large metabolite concentration ranges we account for the noise in the metabolite concentration data and uncertainties in Gibbs energies of formation. Typical uncertainties in formation energies are in the order of 0.02–2 kJ/mol [35], which are overshadowed by variations in metabolite concentration data. The compartmental pH values were set to 5, 6.5 and 7 for the external, cytosolic and mitochondrial environment, respectively [54], [55]. The ionic strength was set to 0.15 M for all compartments. For the correct consideration of transport thermodynamics in NET analysis, we defined the specific transported species for each transport reaction where possible, and calculated transport reaction values according to Jol et al. [56].
Computations for FVA, NET analysis and EFM generation were done using MATLAB (The Mathworks). For optimization of FVA problems we used the LINDO API library (LINDO Systems Inc.) and for NET analysis we used anNET [23] in combination with the LINDO global solver. To generate EFMs we used the Java implementation from Terzer and Stelling [19] on a quad-core system (3 GHz) with 128 GB memory. To test the thermodynamic feasibility of each EFM we used anNET, which was modified to run in an automated way on a cluster of computers encompassing on average 60 CPUs (3 GHz). Testing EFMs for thermodynamic feasibility was computationally intensive and took approximately 14 days.
To find the reaction activity patterns that cause infeasibility, we considered each infeasible EFM separately and performed an iterative analysis. In NET analysis of the EFM, we removed consecutively each reaction's activity constraint from the NET analysis optimization and determined the feasibility. If the activity pattern became thermodynamically feasible, the reaction activity that was removed was identified as part of the pattern. Then we continued with removing the next reaction activity, while keeping the activities that were identified as part of the pattern. We continued this process for all the reaction activities. This process led to a set of reaction activities, which is a subset of the activities in the analyzed EFM, of which the removal of one activity leads to a feasible system. The set of reaction activities is only infeasible as a whole system, and no single reaction can be marked infeasible by itself. All possible infeasible reaction patterns may not be found when multiple patterns are present in an EFM, because the order of reaction activity removal determines which pattern is found. The obtained infeasible patterns cover all the infeasible EFMs.
Flux balance analysis with maximization of biomass production, maximization of ATP production and maximization of the ratio of ATP production over the sum of squared fluxes was performed according to Schütz et al. [28] with the physiological data used for FVA as constraints. The computations were done using MATLAB (The Mathworks) using the LINDO API library (LINDO Systems Inc.) for optimization.
|
10.1371/journal.pcbi.1005247 | Evolution at ‘Sutures’ and ‘Centers’: Recombination Can Aid Adaptation of Spatially Structured Populations on Rugged Fitness Landscapes | Epistatic interactions among genes can give rise to rugged fitness landscapes, in which multiple “peaks” of high-fitness allele combinations are separated by “valleys” of low-fitness genotypes. How populations traverse rugged fitness landscapes is a long-standing question in evolutionary biology. Sexual reproduction may affect how a population moves within a rugged fitness landscape. Sex may generate new high-fitness genotypes by recombination, but it may also destroy high-fitness genotypes by shuffling the genes of a fit parent with a genetically distinct mate, creating low-fitness offspring. Either of these opposing aspects of sex require genotypic diversity in the population. Spatially structured populations may harbor more diversity than well-mixed populations, potentially amplifying both positive and negative effects of sex. On the other hand, spatial structure leads to clumping in which mating is more likely to occur between like types, diminishing the effects of recombination. In this study, we use computer simulations to investigate the combined effects of recombination and spatial structure on adaptation in rugged fitness landscapes. We find that spatially restricted mating and offspring dispersal may allow multiple genotypes inhabiting suboptimal peaks to coexist, and recombination at the “sutures” between the clusters of these genotypes can create genetically novel offspring. Sometimes such an offspring genotype inhabits a new peak on the fitness landscape. In such a case, spatially restricted mating allows this fledgling subpopulation to avoid recombination with distinct genotypes, as mates are more likely to be the same genotype. Such population “centers” can allow nascent peaks to establish despite recombination. Spatial structure may therefore allow an evolving population to enjoy the creative side of sexual recombination while avoiding its destructive side.
| For a novel genotype to establish in a population, it must (1) be created, and (2) not be subsequently lost. Recombination is a double-edged sword in this process, potentially fostering creation, but also hastening loss as the novel genotype is being recombined with other genotypes, especially when rare. In this study, we find that spatial structure may affect both the creative and destructive aspects of recombination in rugged fitness landscapes. By slowing the spread of high-fitness genotypes, spatially restricted mating and dispersal may allow diverse subpopulations to arise. Reproduction across the borders of these subpopulations—at “sutures”—may create genetic novelty. Depending on the topography of the fitness landscape, such novelty may be in the domain of attraction of a new, higher peak; the population may “peak-jump” to an area of genotype space unlikely to be explored by mutation alone. Lineages founded by peak-jumping events are particularly prone to early extinction, as recombination with unlike genotypes may disrupt the rare allele combination and thereby produce low-fitness offspring. However, these fledgling peak lineages may be protected from early extinction by mating within small homotypic clusters—in “centers”. Thus, spatial structure may allow a population to create rare genotypes via recombination, and allow those rare genotypes to persist despite recombination.
| Sexual recombination has long been a puzzling evolutionary strategy (see [1,2]). Recombination has the potential to create novel high-fitness genotypes in a population, but also to destroy high-fitness lineages by recombining them with genetically distinct lineages. Whether recombination speeds or slows adaptation depends largely on the relative strengths of its creative and destructive effects.
One of the earliest adaptive explanations for recombination is the Fisher-Muller effect, in which beneficial alleles in different lineages can recombine into a single lineage, speeding adaptation [3,4]. The Fisher-Muller effect exemplifies the creative aspect of sex, and many studies have shown faster adaptation due to Fisher-Muller dynamics [5–8]. However, the Fisher-Muller effect assumes that beneficial alleles remain beneficial when recombined into new genetic backgrounds. This assumption is necessarily broken in multi-peaked fitness landscapes [9], which arise when genetic interactions among loci yield multiple high-fitness allele combinations separated by valleys of low-fitness intermediate genotypes. In such landscapes, the adaptive effects of recombination are more complex.
Studies on two-locus rugged landscapes focus on escape from suboptimal peaks, and have found that modest levels of recombination may speed adaptation slightly, while substantial recombination slows or halts adaptation entirely [10–12]. However, studies on rugged landscapes with more than two loci yield conflicting results, variously reporting recombination as slowing adaptation [13], speeding adaptation [14], or having complex effects dependent on the topography of a fitness landscape, the population inhabiting it, and the time scale considered [15–17]. Studies on empirical fitness landscapes report recombination as speeding adaptation [6,18] or having complex effects dependent on the fitness topography and rate of recombination [15]. The varied results described above may partly depend on the genetic variation that a particular landscape supports. If there are multiple suboptimal peak genotypes, these competing lineages may interact. Depending on the topography of the fitness landscape, recombination between individuals on different suboptimal peaks may create an offspring in the attractive domain of a novel peak, termed “peak-jumping” [15,19]. Thus, in topographies that permit peak-jumping, when subpopulations occupy different suboptimal peaks, recombination may allow peak-jumping to novel, higher peaks [19,20].
What conditions might enable a recombining population to maintain the diversity required for peak-jumping? Restricted mating and dispersal (which we call “local reproduction”) may promote population-wide diversity by slowing the spread of high-fitness genotypes and creating competitive refugia for lower-fitness genotypes [21,22]. However, the same spatial restriction that allows population-wide diversity also impedes recombination between those diverse types, as mating occurs largely within monotypic clusters. Martens and Hallatschek [22] show that recombination between spatially abutting lineages (which we call “sutures”) can be sufficient to speed adaptation due to Fisher-Muller effects in their smooth landscape model. In some rugged landscapes, recombination at sutures may allow peak-jumping. However, lineages founded by peak-jumping events are particularly prone to early extinction as recombination may disrupt the rare allele combinations and consequently prevent establishment—recombination with the majority genotype may pull fledgling peak populations off their precipices and into the valley between [23]. On the other hand, recombination within monotypic clusters (which we call “centers”) may allow high fidelity of rare allele combinations, but also prevent the creation of such rare allele combinations as no effective recombination is occurring. Which effects of sutures and centers dominate, and in what circumstances? In this paper, we examine the combined effects of recombination and local reproduction on adaptation on rugged landscapes.
In our simulation, a population inhabits an n × n square lattice. Each lattice point may be empty or may house one organism. Organisms have a haploid genotype of L loci, where the allele at each locus is either a 0 or a 1. Each genotype has an associated survival probability (SG). Unless otherwise indicated, populations are initialized with individuals of the genotype farthest from the optimal genotype (that is, G0 such that H(G0, Gopt) = L, where H is the Hamming distance operator and Gopt is the optimal genotype), with each lattice point having an SG0 probability of starting occupied. Evolution occurs via discrete update steps described below, and simulations conclude when the optimal genotype reaches a predefined frequency, or when a predefined number of epochs have occurred, where an epoch is defined as n × n updates.
At each update, a point is chosen at random. If this focal point houses an individual of genotype G, the individual dies with probability 1 − SG, and the lattice point becomes empty. If the focal point is already empty, then a birth event can occur. For a birth event, two parents are needed. The first parent is chosen from a pre-defined dispersal neighborhood about the focal point, and second parent is chosen from a pre-defined mating neighborhood about the first parent. For simplicity, we set the sizes of these two neighborhoods equal, and call the radius of this neighborhood the “reproductive distance”. If there are no parents who satisfy the criteria, no birth event occurs. We focus on two extreme cases. In our “local reproduction” condition, a focal point’s neighborhood is defined by the lattice points immediately to the north, east, south and west (the Von Neumann neighborhood); in our “global reproduction” condition, the neighborhood is defined as the entire lattice, minus the focal point.
Once the parents are chosen, an offspring genotype is formed by recombination and mutation. To simulate recombination, one of the two parents is chosen at random to contribute the allele at the first locus, and between-locus crossover occurs with probability r. Thus r = 0 yields no crossing over, while r = 0.5 yields independent assortment of parental alleles. To simulate mutation, each locus of the recombined offspring’s binary genotype changes its allelic state (0→1 or 1→0) with probability μ. Finally, the offspring is born, and inhabits the initially-empty lattice point.
To investigate the interplay of recombination and reproductive distance, we use a 4x2 factorial design: four recombination probabilities and two neighborhood sizes. For each factorial combination, we simulate replicate populations evolving on a multi-peaked rugged landscape. Our default fitness landscape is defined to allow peak-jumping; that is, there exist two suboptimal peaks (0011 and 1100) which can recombine to produce the optimal genotype (1111). We will relax this contrivance later in our results. In this 4x2 experiment, all populations are initialized on a suboptimal peak (0000), and all parameters (lattice size, initial density, mutation rate, etc.) are held constant. We find that the qualitative effect of recombination–whether it speeds or slows the traversal of the rugged fitness landscape–can depend on whether reproduction is localized (Fig 1), and this interaction between recombination and reproductive neighborhood is significant (p<0.001, Manly’s permutation test [24]). When reproduction is global, slight recombination speeds peak establishment while substantial recombination slows peak establishment. However, when reproduction is local, all rates of recombination speeds peak establishment.
To investigate why the effect of recombination may depend on reproductive distance, we focus on two aspects of a genotype's spread through a population: discovery (i.e., first appearance of the genotype in the population) and establishment (i.e., first appearance without subsequent loss). Both the discovery and establishment of the optimal genotype are affected by the interaction between recombination and reproductive distance, and the rate at which simulations discover and establish the optimal peak genotype appears to be biphasic within each parameter set (S1 Fig). The first phase, presumably due to discovery via recombination, shows rapid discoveries and subsequent establishments of the optimal genotype. The second phase, presumably due to discovery via mutation (indeed, when r = 0 this is the only phase), shows slower discoveries and a substantial lag between discovery and establishment. Both phases onset earlier when reproduction is global, yet most global reproduction simulations lag behind their local reproduction counterparts. This is because, when recombination is nonzero, the majority of local reproduction simulations discover and establish in the first phase, presumably by recombination; while the majority of global reproduction simulations discover and potentially establish in the second phase, presumably by mutation. This tortoise-hare pattern is also seen in mean relative fitnesses of populations over time: at shorter observation times, the global reproductive schemes are likely to be higher-fitness; at longer observation times, the local reproductive schemes are likely to be higher fitness (S2 Fig).
Local reproduction seems to allow quicker discovery and quicker subsequent establishment of the optimal peak. To investigate why, we focus on discovery and establishment separately. For a peak genotype to establish in a population, it must (1) be created, and (2) not be subsequently lost.
On rugged fitness landscapes, populations may become trapped on a suboptimal fitness peak. It is also possible for a population to discover multiple distinct suboptimal peaks before any single peak genotype has fixed. Localized reproduction may promote the coexistence of multiple peaks by increasing the time-to-fixation of a newly discovered peak. Thus, localized reproduction may foster the diversity of genotypes required for peak-jumping via recombination (e.g., the creation of peak genotype 1111 due to recombination between suboptimal peak genotypes 0011 and 1100) [19]. However, localized reproduction precludes peak-jumping unless the peak lineages are physically close. Physical proximity could result if two expanding peak lineages eventually abut, allowing meaningful recombination at the suture between the distinct genotypes. Such sutures between subpopulations may allow repeated discovery of genotypes in the domain of attraction of a higher fitness genotype. Indeed, in a representative simulation of intermediate recombination with local reproduction from Fig 1, multiple suboptimal peak genotypes coexist (0011 and 1100), and the globally optimal genotype (1111) is repeatedly created at the sutures between these subpopulations (Fig 2B, S1 Video). In a parallel representative run with global reproduction, no such sutures exist, because an intermediate genotype, once discovered, quickly sweeps to near fixation (Fig 2A, S1 Video).
Does local reproduction encourage sutures between subpopulations? To test this, we simulate a two-locus landscape with two peak genotypes (10 and 01) and two valley genotypes (00 and 11, the latter of which is lethal). The population is initialized on genotype 00, and we track how frequently genotype 11 is created, and how it is created. We find that genotype 11 is created by recombination more frequently in local rather than global reproductive schemes, while it is created by mutation at approximately the same frequency in the two schemes (S3 Fig).
Once a peak genotype is discovered, it may be lost due to subsequent recombination with unlike types, lowering the genotypic fidelity of its lineage [25]. When recombination rates are high, such loss may prevent a genotype from establishing [10,11,26]. However, spatially segregated populations may harbor population “centers”, in which mating pairs are likely to be genetically similar, preserving genotypic fidelity. Such centers may allow rare genotypes to persist in a population despite recombination. To examine the effect of centers on the establishment of a novel peak genotype, we model adaptation on a two-locus landscape in which a population may escape from suboptimal peak genotype 00 by crossing an adaptive valley (genotypes 10 and 01) to optimal peak genotype 11. We find a three-way interaction between recombination, reproductive distance, and centers (p = 0.03, Manly’s permutation test). Frequent recombination slows the establishment of the optimal peak genotype in global but not local reproductive schemes (S4 Fig, top row). However, if ‘centers’ are prohibited—that is, if a rare peak genotype (i.e., a peak genotype comprising less than 1% of the population) happens to select a homotypic neighbor as a mate, the mate is replaced with a random individual in the population—then the local and global reproductive schemes have similar results: when recombination is frequent, valley-crossing is effectively prohibited (S4 Fig, bottom row).
Nonspatial analysis of the two-locus rugged landscape suggests that valley-crossing is effectively prohibited when the recombination rate exceeds the selective advantage of the distant peak, as genetic loss due to recombination outpaces selection [10,11,26]. We too find a threshold above which valley-crossing is effectively prohibited, unless centers are provided by local recombination. The adaptive effects of local inbreeding have been investigated since at least Wright, who focused on the resultant decrease in the effective population size [27,28]. The corresponding increase in drift may allow subpopulations to cross adaptive valleys through sequential fixation [29], which may speed valley crossing for the population as a whole [30]. Here, we focus rather on the local decrease in the effective recombination rate (that is, the actual change in linkage disequilibrium due to recombination [31]) which occurs in ‘centers’, and protects rare allelic combinations regardless of their origin.
While recombination may allow a population to more quickly climb a local peak, it can also trap populations on suboptimal peaks [17]. However, recombination may aid escape from suboptimal peaks if the landscape topography permits peak-jumping [14,19,32]. For peak-jumping to occur, multiple suboptimal peak genotypes must coexist in a population. For peak-jumping to substantially speed adaptation, distant peaks cannot be easily accessible by mutation. Thus, there is a limited range of mutation rates in which peak-jumping speeds adaptation: mutation rates must be high enough to create a diversity of genotypes, but not so high that all genotypes are easily accessible. By slowing the spread of high-fitness genotypes, local reproduction allows greater variation at lower mutation rates, and therefore expands the window in which recombination speeds adaptation (Fig 3). Similarly, larger lattices are more likely to allow variation, as more time is required for a fitter genotype to displace a less-fit genotype. Indeed, the larger the lattice, the more recombination speeds adaptation (S5 Fig).
Local reproduction promotes the coexistence of distinct types in a population, and recombination between distinct types may speed adaptation. Thus, at intermediate levels of recombination (r = 0.1), local reproduction expands the range of mutation rates for which recombination speeds adaptation. This expanded range persists at high levels of recombination (r = 0.5), while the corresponding range for global reproduction disappears entirely. Without the centers provided by local reproduction, high levels of recombination trap populations on suboptimal peaks. The protective effect of centers is robust to occasional global reproduction (S6 Fig).
Sutures should be most effective when recombination between two suboptimal peaks can create offspring in the attraction basin of a third, higher peak, allowing for peak-jumping. Centers should be most effective when novel peaks are discovered via peak-jumping, as recombination between the nascent peak and the majority genotypes can create low-fitness offspring. Thus the ability of sutures and centers to modulate the effects of recombination—to harness the creative aspect and mitigate the destructive aspect—may also be sensitive to the particular topography of a rugged landscape.
The full topographies of some naturally occurring fitness landscapes have been measured for small subsets of their genotype spaces [33]. De Visser et al. [15] generated 5-locus empirical fitness landscapes by introducing deleterious mutations into the asexual fungus A. niger, and measuring the fitness effects of five individual mutations and all combinations thereof. Two complete 5-locus fitness landscapes were generated, with 32 genotypes each (though the landscapes are not completely independent as they share four of their five loci of interest). Both landscapes were found to be rugged, with multiple local maxima and minima. However, only one of the landscapes (which we call PJ+) had suboptimal peaks which could recombine into the attraction basin of the optimal peak; the other landscape (PJ−) did not. De Visser et al. found that recombination generally slows or halts the establishment of the optimal genotype in either landscape, though there was a window of very infrequent recombination that could speed adaptation in PJ+ and very slightly and rarely speed adaptation in PJ−(see [15], supplement B1). We create landscapes parallel to PJ+ and PJ−for our model (e.g., replacing relative fitness with relative survival probabilities), and simulate evolution as before. We find a significant three-way interaction between recombination, reproductive distance, and fitness landscape topology on the waiting time for optimal genotype establishment (p<0.001, Manly’s permutation test). On PJ+, recombination slows or prevents the establishment of the optimal genotype when reproduction is global, but never slows or prevents adaptation when reproduction is local (Fig 4, top panel). On PJ−, whose topography is less conducive to landscape exploration via recombination, we find similar results to PJ+ when reproduction is global, but high recombination (r = 0.5) still slows the generation and establishment of the optimal genotype when reproduction is local (Fig 4, bottom panel).
In our test landscape and in two empirically-derived landscapes, sufficiently high rates of recombination prohibit the establishment of a novel high-fitness peak when reproduction is global, but this destructive side of recombination is alleviated when reproduction is local. Moreover, in landscape topographies that allow peak-jumping (our test landscape and, to a lesser extent, PJ+), recombination can speed the establishment of novel high-fitness peaks. Thus, the landscape topography affects the ability of local reproduction to mediate the effects of recombination: accentuating exploration via “sutures” while mitigating recombinatory destruction of rare genotypes via “centers”. We suggest the greatest effect of sutures occurs when peak-jumping is possible, and the greatest effect of centers occurs when novel peaks are created via peak-jumping. The prevalence of such topographical features and spatial restrictions—and therefore how relevant “sutures” and “centers” are to natural populations—remains an empirical question. It is possible, though, that by creating “sutures”, spatially structured populations may efficiently explore rugged landscapes via recombination, and by creating “centers”, those same populations may permit the establishment of novel peaks despite recombination. Spatially structured populations may therefore harness recombination’s constructive effects while mitigating its destructive effects on adaptation in rugged landscapes.
|
10.1371/journal.ppat.1007420 | Differential induction of interferon stimulated genes between type I and type III interferons is independent of interferon receptor abundance | It is currently believed that type I and III interferons (IFNs) have redundant functions. However, the preferential distribution of type III IFN receptor on epithelial cells suggests functional differences at epithelial surfaces. Here, using human intestinal epithelial cells we could show that although both type I and type III IFNs confer an antiviral state to the cells, they do so with distinct kinetics. Type I IFN signaling is characterized by an acute strong induction of interferon stimulated genes (ISGs) and confers fast antiviral protection. On the contrary, the slow acting type III IFN mediated antiviral protection is characterized by a weaker induction of ISGs in a delayed manner compared to type I IFN. Moreover, while transcript profiling revealed that both IFNs induced a similar set of ISGs, their temporal expression strictly depended on the IFNs, thereby leading to unique antiviral environments. Using a combination of data-driven mathematical modeling and experimental validation, we addressed the molecular reason for this differential kinetic of ISG expression. We could demonstrate that these kinetic differences are intrinsic to each signaling pathway and not due to different expression levels of the corresponding IFN receptors. We report that type III IFN is specifically tailored to act in specific cell types not only due to the restriction of its receptor but also by providing target cells with a distinct antiviral environment compared to type I IFN. We propose that this specific environment is key at surfaces that are often challenged with the extracellular environment.
| The human intestinal tract plays two important roles in the body: first it is responsible for nutrient absorption and second it is the primary barrier which protects the human body from the outside environment. This complex tissue is constantly exposed to commensal bacteria and is often exposed to both bacterial and viral pathogens. To protect itself, the gut produces, among others, secreted agents called interferons which help to fight against pathogen attacks. There are several varieties (type I, II, and III) of interferons and our work aims at understanding how type I and III interferon act to protect human intestinal epithelial cells (hIECs) during viral infection. In this study, we confirmed that both interferons can protect hIECs against viral infection but with different kinetics. We determined that type I confer an antiviral state to hIECs faster than type III interferons. We uncovered that these differences were intrinsic to each pathway and not the result of differential abundance of the respective interferon receptors. The results of this study suggest that type III interferon may provide a different antiviral environment to the epithelium target cells which is likely critical for maintaining gut homeostasis. Our findings will also help us to design therapies to aid in controlling and eliminating viral infections of the gut.
| During viral infection interferons (IFNs) are the predominant cytokines made to combat viral replication and spread. Following binding to specific receptors, IFNs induce a JAK/STAT signaling cascade which results in the production of interferon stimulated genes (ISGs). These ISGs will then establish an antiviral state within the cell and will also alert surrounding cells and immune cells to assist in viral clearance [1]. There are three classes of IFNs. Type I IFNs are produced by all cell types and are recognized by the ubiquitously expressed heterodimeric receptor IFNAR1/IFNAR2. Type II IFNs are only produced by immune cells [2,3]. Type III IFNs are made by all cell types but the IFNLR1 (or IL28Ra) subunit of the heterodimeric receptor IFNLR1/IL10Rβ is restricted to epithelial and barrier surfaces and to a subset of immune cells [4–9]. Despite the fact that type I and type III IFNs are structurally unrelated and engage different receptors, signaling downstream of both receptors exhibits a remarkable overlap and leads to the induction of a similar pool of ISGs. These observations originally led to the hypothesis that type I and III IFNs were functionally redundant.
This model has been challenged more and more in recent studies which highlight that the cell type specific compartmentalization of IFNLR1 provides type III IFNs a unique potential for targeting local infections at mucosal surfaces. For example, in vivo data on enteric virus infection of the murine gastrointestinal tract showed that responsiveness to type III IFN is necessary and sufficient to protect murine intestinal epithelial cells (IECs) against rotavirus and reovirus infection [10–12]. On the contrary, type I IFN was necessary to protect against viral infection of cells in the lamina propria and against systemic spread [10,11]. Likewise, it was demonstrated that fecal shedding of norovirus was increased in IFNLR1-deficient, but not IFNAR1-deficient, mice, showing that type III IFN uniquely controls local norovirus infection in the gut [13,14]. Similarly, in the respiratory tract, type III IFNs are predominately produced upon infection with influenza A virus [15–19]. However, as infection progresses type I IFN comes into play to reinforce viral inhibition by inducing a pro-inflammatory response [20].
Differences in the antiviral activity conferred by both cytokines appear to be not only driven by the spatial restriction of their receptors but also by intrinsic subtle differences in signal transduction. It was demonstrated, in human hepatocytes and lung epithelial cells, that type I IFN confers a more potent antiviral protection compared to type III IFNs [5,21–23]. Additionally, it was shown in human IECs that type III IFN partially requires MAP kinase activation to promote an antiviral state while type I IFN was independent of it [24]. Although it has been reported in many studies that very similar ISGs are induced upon type I or type III IFN stimulation of cells, work mostly performed in hepatocytes revealed that both cytokines induce these ISGs with different kinetics [21,25–27]. Type III IFN mediated signaling was found to be associated with a delayed and reduced induction of ISGs compared to type I IFNs [25,26]. Similar differences in the magnitude and/or kinetics of ISGs induction upon type I versus type III IFN treatment were observed in human primary keratinocytes, airway epithelial cells and in Burkitt's lymphoma derived B (Raji) cells, as well as in murine intestinal and lung epithelial cells and immune cells [20,28–31].
The molecular mechanisms leading to this delayed and reduced induction of ISGs upon type III IFN treatment remains unclear. As these differences in kinetics of ISG expression between both IFNs could not be directly explained by their signaling cascades an alternative explanation was proposed where type III IFN receptor is expressed at lower levels at the cell surface. This lower receptor expression level could provide a biochemical explanation for the observed differences in delayed kinetics and weaker amplitude of ISG expression compared to type I IFN. However, to date, there is no direct experimental evidence for this model. Similarly, whether the observed differences between both IFNs is intrinsic to both specific signal transduction pathways and whether it is restricted to some cell types (e.g. hepatocytes) or represents a global signaling signature in all cells expressing both IFN receptors has not been fully addressed.
In this study, we have investigated how type I and III IFNs establish their antiviral program in human mini-gut organoids and human IEC lines. We found that type I IFN can protect human IECs against viral infection faster than its type III IFN counterpart. Correspondingly, we determined that type I IFN displays both a greater magnitude and faster kinetics of ISG induction compared to the milder, slower type III IFN. By developing mathematical models describing both type I and type III IFN mediated production of ISGs and by using functional receptor overexpression approaches, we demonstrated that the observed lower magnitude of ISG expression for type III IFNs was partially the result of its lower receptor expression level compared to the type I IFN receptor. Inversely, the observed delayed kinetics of type III IFN cannot be explained by receptor expression level indicating that this property is specific to type III IFN and inherent to its signaling pathway. Our results highlight important differences existing between both type I and type III IFN-mediated antiviral activity and ISG expression which are not only the result of receptor compartmentalization but also through intrinsic fundamental differences in each IFN-mediated signaling pathway.
We have previously reported that both type I and III IFNs mediate antiviral protection in human IECs [24]. To address whether type I and type III IFN have a different profile of antiviral activity in primary non-transformed human IECs, as reported in human lung cells [22], we compared the antiviral potency of both IFNs in human mini-gut organoids. Colon organoids were pre-treated with increasing concentrations of either type I or III IFNs for 2.5 hours and subsequently infected with vesicular stomatitis virus expressing luciferase (VSV-Luc). Viral infection was assayed by bioluminescence and results showed that both IFNs induced an antiviral state in a dose-dependent manner. We observed that type I IFN was slightly more potent in protecting against viral infection at higher concentration compare to type III IFNs. Type I IFN could almost fully inhibit viral infection while type III IFN was only able to reduce infection to around 80% (Fig 1A). Interestingly, the concentration of type I IFN necessary to provide 90% of relative antiviral protection (EC90) was significantly lower than the one for type III IFN (Fig 1B).
To determine whether type III IFN requires a prolonged treatment to achieve similar antiviral protection as observed with type I IFN, we performed a time course experiment in which human colon organoids were pre-treated for different times with either IFN prior infection with VSV-Luc (Fig 1C). We found that approximately 2 hours pre-treatment with type I IFN was sufficient to reduce VSV infection by 90% (10% remaining infection), while type III IFN required around 5 hours to achieve a 90% reduction of infectivity (Fig 1C and 1D). Interestingly, 24 hours of pretreatment was necessary for type III IFN to almost completely prevent VSV infection (Fig 1C). These results strongly suggest that both type I and type III IFN could have similar potency but that type III IFN requires more time to establish an antiviral state.
We next addressed how long after infection IFN treatment is still able to promote antiviral protection. Colon organoids were infected with VSV-Luc and treated at different times post-infection with either type I or III IFNs. Interestingly, type I IFN could inhibit viral replication even when added several hours post-infection. In contrast, type III IFN appeared to require a much longer time to establish its antiviral activity and was unable to efficiently protect the organoids after VSV infection has initiated (Fig 1E and 1F).
Importantly, these differences in the kinetics of antiviral activity of type I versus type III IFNs were neither donor nor colon specific as similar results were observed in intestinal ileum-derived organoids derived from different donors (S1 Fig). In addition, the human colon carcinoma-derived cell line T84 (S2 Fig) fully phenocopy the difference in type I versus type III IFN antiviral activity generated by primary mini-gut organoids. Taken together these results demonstrate that while both type I and III IFNs can promote similar antiviral states into target cells, they do so with distinct kinetics. The cytokine-induced antiviral state is promoted faster by type I IFN compared to type III IFN.
To understand how type I and type III IFNs promote an antiviral state in primary IECs but with different kinetics, we analyzed the magnitude of ISG expression over time upon IFN treatment. Colon organoids were treated with increasing concentrations of either type I or type III IFN and the expression levels of two representative ISGs (IFIT1 and Viperin) were assayed at different times post-IFN treatment. Results revealed that type I IFN ultimately leads to a significantly higher induction of both IFIT1 and Viperin compared to type III IFN (Fig 2A and 2B). This difference in the magnitude of ISG stimulation was independent of the duration of IFN treatment (Fig 2A and 2B). To determine if this pattern of expression applies to other ISGs, we treated colon organoids with either type I or type III IFN over a 24-hour time course, and analyzed the mRNA levels of 132 different ISGs and transcription factors involved in IFN signaling (see complete list of genes and corresponding primers in S1 and S2 Tables) (Fig 2C and 2D). Differential expression analysis revealed that both type I and type III IFNs induce almost the same set of ISGs and that most of the genes significantly induced by type III IFN were also induced by type I IFN (Fig 2C). However, similar to IFIT1 and Viperin (Fig 2A and 2B), we found that the magnitude of ISG expression was greater for type I IFN compared to type III IFN (Fig 2D). Similar results were found in the immortalized colon carcinoma-derived T84 cells (S3A–S3C Fig).
To address whether there is any correlation between the different antiviral protection kinetics conferred by type I and III IFNs (Fig 1) and the kinetics of ISG expression, we analyzed the temporal expression of ISGs upon IFN treatment of human colon organoids. Hierarchical clustering analysis of all ISGs up-regulated upon type I or type III IFN treatment defined four distinct expression profiles based on the time of their maximum induction (Fig 3A–3C). Group 1 are ISGs whose expression peaks 3 hours post-IFN treatment. The expression of ISGs in group 2 and 3 peaks at 6 and 12 hours post-treatment, respectively. Group 4 corresponds to ISGs with a continuous increase in expression over time (Fig 3A and 3B). Under type I IFN treatment, ISGs were nearly equally distributed in all four expression groups (Fig 3A, 3C and 3D). By contrast, although the same ISGs were induced by type III IFN, they almost all belong to the expression group 4, being expressed later after IFN treatment (Fig 3B–3D). In line with the primary mini-gut organoids, T84 cells presented similar differences in the kinetics of ISGs expression (S3D Fig).
We next wanted to control that our observed differences in the kinetics of ISGs expression induced by both cytokines were independent of IFN concentration. Colon organoids were treated with increasing amounts of type I or type III IFNs and the transcriptional up-regulation of representative ISGs belonging to each of the expression profile groups (group 1–4) was measured over time (Fig 4). Consistent with our previous results, the temporal expression patterns of each representative ISGs were independent of the IFN concentration and the ISG expression kinetic signature was specific to each IFN (Fig 4). Complementarily, to address whether the observed differences between type I and type III IFNs were not due to the lower affinity of type III IFN for its receptor compared to type I IFN, we employed the high affinity variant of type III IFN (H11-IFNλ3) [32] to monitor the kinetics of ISG expression. Results show that cells treated with the higher affinity H11-IFNλ3 display a higher magnitude of ISG expression but their kinetics of expression were unchanged (S4 Fig). Altogether, our results strongly suggest that although both type I and type III IFNs induce a similar set of ISGs in hIECs, type III IFN induces globally a lower amplitude and a delayed ISG expression compared to type I IFN.
Our data show remarkable differences in the magnitude and kinetics of ISGs induced by type I versus type III IFN (Figs 2 and 3 and S3 Fig), and in the subsequent induction of a differential antiviral state (Fig 1 and S1 and S2 Figs). To investigate the mechanisms underlying these differences, we used data-driven mathematical modeling and model selection. We considered three mechanistic causes for the observed differential signaling: (1) Receptor abundance (different number of IFNLR compared to IFNAR complexes); (2) Receptor regulation (different rates of activation and/or inactivation of IFNLR compared to IFNAR complexes); (3) STAT activation (different rates of STAT activation by type I and type III IFNs). We devised corresponding mathematical models describing the dynamics of receptor activation and inactivation, STAT1/2 phosphorylation and STAT-dependent activation of ISG expression (Fig 5A). The models were implemented as systems of ordinary differential equations (S3 Table) and fitted to the time-resolved data for the prototypical ISG, Viperin, measured with different doses of type I or type III IFNs and with the high affinity H11-IFNλ3. We ranked the models according to the Akaike information criterion corrected for small sample size (AICc), which evaluates the goodness of fit and, at the same time, penalizes the number of fit parameters (for more details see Materials and Methods). Throughout, we allowed different receptor abundance, but this difference alone could not account for the different signaling dynamics (Fig 5B; model M1 has negligible support by the data, as quantified by the small AICc weight, which is a weight of evidence for the respective model). Interestingly, in addition to receptor abundance, the best-fitting model (M3) has also different rates of activation and inactivation of IFNLR and IFNAR complexes. However, alternative models with different rates of STAT activation and/or ISG expression have good performance (M2 and M4, respectively). Therefore, the modeling indicates that differential ISG activation by type I and type III IFNs is likely due to different abundance of the respective receptors and cell-intrinsic differences in how the signals from bound receptors are processed.
The best-fitting model (M3) accounted for the dose-response and the different Viperin expression kinetics triggered by type I, type III and the high affinity H11-IFNλ3 in T84 cells, group 3 and group 4 expression kinetics, respectively (Fig 5C and 5D). The different kinetics of the IFN responses–fast and transient for type I IFN vs slower and sustained for type III IFN–are predicted to be largely due to receptor inactivation, which is faster for IFNAR than for IFNLR complex (S5A–S5C Fig). Interestingly, the model shows that at low IFN concentrations, Viperin is induced almost equally by both IFNs whereas at higher concentrations, type I IFN induces Viperin more strongly (Fig 5E). These dose-dependent features agree with our experimental data (S3B Fig, right panel).
Next, we tested the pivotal impact of receptor expression on ISG induction that was indicated by our model. Specifically, the model predicts that an increase in IFNAR1 or IFNLR1 level will increase the amplitude of ISG induction while preserving the specific kinetic profiles elicited by the two types of IFNs (S5D and S5E Fig). To experimentally validate the model predictions, IFNAR1 and IFNLR1 were overexpressed in T84 cells. Overexpression of the respective IFN receptor chain was confirmed by reverse quantitative PCR (S6 Fig). To ensure the functionality of both IFN receptors, IFNAR1 or IFNLR1 were expressed in our previously characterized knockout T84 cell lines deficient for either the IFN alpha receptor 1 (IFNAR1-/-) or the IFN lambda receptor 1 (IFNLR1-/-) (S7A and S7E Fig) [24]. Our results show that overexpression of IFNAR1 in our IFNAR1-/- T84 cells (IFNAR1-/-+rIFNAR1) restores their antiviral activity, their ability to phosphorylate STAT1 and induce the production of the ISGs IFIT1 and Viperin in the presence of type I IFN (S7B–S7D Fig). Similarly, although IFNLR1-/- cells were insensitive to type III IFN treatment, overexpression of IFNLR1 (IFNLR1-/-+rIFNLR1) restored their antiviral activity, pSTAT1 and ISG induction after addition of type III IFN (S7F–S7H Fig). These results demonstrate the functionality of both IFN receptors and validate our overexpression approach as a means to increase IFNAR1 and IFNLR1 levels at the cell surface.
Wild-type T84 cells overexpressing type I IFN receptor (WT+rIFNAR1) were treated with increasing concentrations of type I IFN. Our results showed elevated levels of STAT1 phosphorylation and ISG induction in response to stimulation with type I IFN compared to wild-type cells (Fig 6A–6D). Importantly, the response of T84 cells overexpressing type I IFN receptor to type III IFN remained unchanged, indicating a selective enhancement of the type I IFN signaling pathway. Similarly, overexpression of type III IFN receptor (WT+rIFNLR1) shows a significant increase in phosphorylated STAT1 and ISG expression compared to wild-type cells upon type III IFN stimulation, while no difference was observed upon type I IFN treatment (Fig 6E–6H). Altogether, our experimental data are consistent with the modeling predictions and confirm the crucial impact of surface receptor levels for regulating the magnitude of type I and III IFN response.
We next addressed whether this increase of ISG expression in cells overexpressing either the type I or type III IFN receptor was associated with an improved antiviral activity. Wild-type T84 cells overexpressing type I IFN receptor (WT+rIFNAR1) were treated with type I IFN at different time points prior to infection with VSV-Luc virus and their antiviral activity was compared to wild-type T84 cells. Our results showed that the potency and the kinetics of the antiviral activity of cells overexpressing type I IFN receptor does not present any significant change upon type I IFN treatment (S8A Fig). Similarly, there is no difference in the antiviral activity when cells overexpressing type I IFN receptor were treated with type I IFN at different time points post-infection (S8B Fig). However, overexpression of type III IFN receptor (WT+rIFNLR1) shows a modest but significant enhancement in type III IFN antiviral potency in the earlier time points of pre-treatment (between 30 minutes and 2 hours) compared to wild-type cells upon type III IFN stimulation (S8G Fig), while they responded similarly to wild-type cells upon type I IFN treatment (S8E Fig). Consistent with this, cells overexpressing type III IFN receptor are more protected than wild-type cells when type III IFN was added post-infection for the early time points (S8H Fig).
Finally, to experimentally validate the limited impact of the IFN receptors abundance on the kinetic profile of ISG expression, as predicted by the model (S5D and S5E Fig), wild-type cells overexpressing either of the IFN receptors were treated with increasing doses of type I or type III IFNs and the expression of a representative ISG belonging to each of the expression profile groups (group 1–4, Fig 3) was analyzed over time (Fig 7A–7D). The experimental data show that the amplitude of ISG expression was dependent on both the dose of IFNs used to stimulate the cells and on the expression levels of the IFN receptors (Fig 7A–7D). Importantly, the kinetic profile of ISG expression was similar between WT cells and cells overexpressing the IFNAR1 (WT+rIFNAR1), independent of the applied IFN type I dose (Fig 7A–7D left panel). Similarly, wild-type cells overexpressing the IFNLR1 (WT+rIFNLR1) showed no change in the kinetic profile of ISG induction upon type III IFN stimulation (Fig 7A–7D right panel). Moreover, we found that the model reproduced the kinetic dose-response data when the IFNAR1 and IFNLR1 expression levels were increased ~2.6 and ~1.5 times, respectively, while all other parameters were held constant (S9 Fig). Indeed, we found that IFNAR1 overexpression was stronger than IFNLR1 overexpression, as judged by the transcript levels (S6B and S6C Fig), with the ratio being consistent with the model prediction (S9D Fig and S6B and S6C Fig).
To directly correlate ISG expression kinetics and amplitude with the expression level of the type III IFN receptor, we thought of overexpressing an IFNLR1 tagged with the GFP fluorescent protein (IFNLR-GFP) in human IECs. To control the functionality of the GFP tagged receptor, the IFNLR1-GFP construct was overexpressed in the human embryonic kidney cell line 293 HEK, which normally elicit a very limited response upon type III IFN treatment. Quantitative RT-PCR revealed that 293 HEK cells overexpressing IFNLR1-GFP produced significantly more ISGs upon type III IFN treatment compared to WT 293 HEK cells or 293 HEK cells expression GFP alone (S10 Fig). Wild-type T84 cells overexpressing the IFNLR1-GFP (WT+rIFNLR1-GFP) were treated with type III IFN over time and cells were sorted by flow cytometry based on their level of IFNLR1-GFP expression (no GFP expressing (neg), or low and high GFP expressing cells) (Fig 8A). The induction of a representative ISG belonging to each of the expression profile groups (group 1–4, Fig 3) was measured over time in each sorted population (negative, low and high, Fig 8B). As anticipated, WT cells overexpressing the IFNLR1-GFP chain show stronger ISG expression compared to WT cells and the magnitude of the ISG induction correlates with the relative levels of IFNLR1 expression (Fig 8B). However, the kinetic profiles of the ISGs upon type III IFN stimulation were not affected by the differential expression levels of the IFNLR1 chain (Fig 8B).
Altogether, our results demonstrate that type I and type III IFNs both induce an antiviral state in hIECs but with different kinetics. We could show that although both cytokines induce similar ISGs, type III IFN does it with slower kinetics and lower amplitude of individual ISG expression compared to type I IFN. Importantly, coupling mathematical modeling of both type I and type III IFN-mediated signaling and overexpression of functional IFN receptors approaches allowed us to demonstrate that these kinetic differences in type I and type III IFN ISG expression are not due to different expression level of the respective IFN receptors but are intrinsic to type I and type III IFN signaling pathways.
In this work, we have for the first time, performed a parallel study of the role of type I and III IFN in human mini-gut organoids and IEC lines. Our results demonstrate that type I and III IFNs are unique in their magnitude and kinetics of ISG induction. Type I IFN signaling is characterized by relatively strong expression of ISGs and confers to cells a fast-antiviral protection. On the contrary, the slow acting type III IFN mediated antiviral protection is characterized by a weak induction of ISGs in a delayed manner compared to type I IFN. Our results are in line with previous studies which also demonstrated that type III IFN is less potent than its type I IFN counterpart [5,21,23,33,34]. Additionally, we have confirmed that the delayed ISG induction seen upon type III IFN treatment of hepatocytes [21,23,25,26] is not tissue specific but likely represents a global pattern of action of this cytokine in cells expressing the type III IFN receptor (i.e. human epithelial cells). In other words, the different kinetics of ISG expression induced by type I and type III IFNs are specific to each IFN signaling pathways.
In the current work, we have employed, a data-driven mathematical modeling approach to explain signal transduction kinetic differences downstream type I and type III IFN receptors. While type I IFN-mediated signaling has been previously modeled [35,36], type III IFN has not. Our model predicted that the receptor levels directly influence the magnitude of ISG expression however, the kinetics of ISG expression appear to be intrinsic to each IFN-signaling pathway and is largely preserved under receptor overexpression. This prediction was experimentally validated by studying the response of wild-type and IFN receptor overexpressing cells to different doses of IFN (Fig 7A–7C and Fig 8). This suggests that the kinetic differences in the ISG induction are intrinsic to each IFN signaling pathway. We propose that these phenotypic differences reflect functional differences, which are important for mounting a well-tailored antiviral innate immune response at mucosal surfaces where type III IFN receptors are expressed.
Both type I and III IFNs have unique and independent receptors which are structurally unrelated. These receptors are likely expressed at different levels on individual cells and their relative expression to each other might also be cell type specific. To address whether the unique ISG and antiviral expression kinetics shown by each IFN were not due to differences in their expression levels, we overexpressed into cells functional type I (rIFNAR1) and type III IFN (rIFNLR1) receptors. Our results from IFNAR1 overexpressing cells (Figs 6 and 7) are in line with previous studies showing a direct relationship between the surface levels of type I IFN receptors and the magnitude of ISG induction [37,38]. Interestingly, we could demonstrate a similar relationship when overexpressing IFNLR1 (Figs 6 and 7) which was also associated with an increase of type III IFN antiviral potency (S8 Fig). These findings are in agreement with previous experiments which show that overexpression of IFNLR1 in cells which normally do not express this IFN receptor rescues both type III IFN-mediated signaling and IFN-mediated antiviral protection [5,28]. Our IFN receptor overexpression approach demonstrates that the observed differences in ISG expression kinetics are not the results of different levels of receptors at the cell surface but is likely specific to each signal transduction pathway. Apart from the expression levels of IFN receptors, lower binding affinity towards their respective receptors could be an alternative explanation for the differential potencies of both type I and type III IFNs. Multiple studies have tried to affect the binding affinity of type I IFNs with their receptors however, results suggest that wild-type IFNs exert their antiviral activities already at maximum potency. Modifications leading to an increased affinity for their receptors do not lead to improvement of antiviral potency [32,38–41]. To address whether the weaker activity of type III IFN could be the result of its weaker affinity for its receptor, Mendoza et al, engineered a variant of type III IFN with higher-affinity for its receptor (H11-IFNλ3). They showed increased IFN signaling and antiviral activity in comparison with wild-type IFNλ3. However, the engineered variant of IFNλ3 was still acting with weaker efficacy compared to type I IFNs [32]. By exploiting the high affinity variant H11-IFNλ3, we could also show a significant increase of the amplitude of ISG expression but importantly, the kinetics of ISG expressions were not altered (S4 Fig).
Our results indicate a model were inherent temporal differences exist between type I and type III IFNs signaling. These differences are not the result of differential surface expression of the receptors but is the result of distinct signaling cascades from the receptors to the nucleus or within regulatory mechanisms of gene expressions.
While few studies have focused on the endocytosis and inactivation of IFNAR1, there is no information about how these processes occur for IFNLR1. It has been shown that the ternary IFNAR complex is internalized by clathrin mediated endocytosis [42] and that upon type I IFN stimulation, IFNAR1 is rapidly endocytosed and routed for lysosomal degradation [43,44], whereas IFNAR2 can be recycled back to the cell surface or degraded [45]. Our data-driven mathematical modeling approach suggests a different kinetics of receptor activation/inactivation between both IFNs (Fig 5B and S5A Fig). Therefore, further studies investigating trafficking of IFNLR1 will be important and may show that subtle changes in the time course of receptors internalization, recycling or degradation can have profound effect on kinetics of IFN activity. Apart from receptor internalization and degradation, several molecular mechanisms leading to IFN receptor inactivation have been described, such as de-phosphorylation [46,47], or by negatively targeting the interaction of IFNAR1 with downstream signaling elements of the JAK/STAT signaling, for instance ubiquitin-specific protease USP18, and members of the suppressor of cytokine signaling protein (SOCS) family. In particular, the inhibitory role of SOCS1 in type I IFN signaling has been demonstrated in a number of previous studies, where they have shown that SOCS1 associates with TyK2 and blocks its interaction with IFNAR1 [48]. USP18 has also been shown as an important negative regulator of type I IFN signaling with a dual role acting as isopeptidase which removes the ubiquitin like-ISG15 from target proteins [49] and as a competitor of JAK1 for binding to IFNAR2 [50]. Although, limited information is available for negative regulators of the IFNLR receptor complex, the specific contribution of USP18 or SOCS in inhibition of type I versus type III IFN mediated signaling has been addressed in recent studies. In particular, it has been showed that both type I and III IFNs (IFNα, IFNβ and IFNλ1, λ2, λ3 and λ4) induced the expression of USP18, SOCS1 and SOCS3 [51–57] and overexpression of all these negative regulators inhibited both IFNα and IFNλ1 mediated JAK-STAT signaling [54,56] suggesting that at ‘‘supraphysiological” expression levels all the inhibitors can block type I and type III mediated JAK-STAT signaling [56]. Additionally, it has been shown that USP18 is induced later and that its level increased over time, correlating with the long lasting refractories of IFNα signaling [51,52,56]. In our study we observed also a later peak of induction of USP18 at 12h or 24h upon type I or type III IFNs, respectively. In line with the above-mentioned studies we also observed rapid and transient induction of SOCS1 upon type I IFN treatment and sustained induction upon type III IFN stimulation. However, further investigation is required to determine the correlation of the kinetics of induction of these negative regulators with the ISGs induction in type I versus type III IFN treatment in human IECs.
In the canonical type I and III IFN signaling pathway the next downstream players from the IFN receptors are the JAKs, STAT1, STAT2 and IRF9, which are all regulated on the level of expression and activation. Our own observations and previous studies could not explain the major differences in the kinetics of type I versus type III IFNs activity by focusing on the time course of phosphorylation of STATs [21,25]. However, given that alternative modifications of STATs (e.g. phosphorylation on alternative residues, acetylation, methylation and sumoylation patterns) have been proposed to contribute to the activity of type I IFNs [26,58–60] it might be possible that new modifiers of STAT activity may determine the kinetic pattern of action of type I versus type III IFNs. In addition, apart from the JAK/STAT axis, there is accumulating evidence which correlates ISG transcription upon IFN treatment with a plethora of JAK-STAT independent pathways, such as members of the CRK [61–63] and MAPKinase family [24,28,64–66], which might also temporally coordinate IFNs kinetic profile of action. Apart from the differences in the signaling cascade of type I versus type III IFNs, an explanation for their differential kinetics of action might stem from the physiology of the different cell types. For example, in a recent study Bhushal et al. reported that polarization of mouse intestinal epithelial cells eliminates the kinetic differences between type I and type III IFNs, by accelerating type III IFN responses [33,67].
Several studies describing the transcriptional activities of both type I and type III IFNs have reported that very similar sets of ISGs are produced upon both type I and III IFN stimulation [12,17,21,22,25,28] while only few ISGs appear to be predominantly expressed upon type III IFN treatment in murine IECs [67]. We believe that there are several functional advantages for adopting a lower and slower activity, like the profile of action of type III IFN, in the antiviral protection of epithelial tissues. The differences in the temporal expression of ISGs could create unique antiviral environments for each IFN. Many ISGs function as pro-inflammatory factors [30,68]. By stimulating ISGs production in high magnitude, an excessive amount of antiviral and pro-inflammatory signals could be produced which on the one hand will eliminate efficiently viral spreading but on the other hand may cause local exacerbated inflammation and irreversible tissue damage, leading to chronic inflammation in mucosal surfaces.
In addition, the expression of different functional groups of ISGs at early and at late time points (Fig 3) might allow cells to create two distinct phases within the antiviral response. At early time points, minimum levels of ISGs may act to protect the host against viral infection. Antiviral ISGs will be responsible for fighting the pathogens and pro-inflammatory ISGs will stimulate members of the adaptive immune system to assist the antiviral protection. At later time points the produced ISGs, may be involved in anti-inflammatory processes, such as resolving of inflammation and tissue healing and repair [66,69]. To exert this anti-inflammatory role, ISGs may need to be produced in higher levels, as they might act more paracrine and spread through the tissue to balance again the tissue homeostasis after the viral attack. In conclusion, we propose that type III IFN-mediated signaling is not only set to act predominantly at epithelium surfaces due to the restriction of its receptor but also is specifically tailored to mount a distinct immune state compared to other IFNs which is critical for mucosal surfaces that face the challenge.
Commercially available primary antibodies were mouse monoclonal antibodies recognizing beta-Actin (Sigma #A5441), phospho STAT1 and STAT1 (BD Transductions #612233 and #610115, respectively). Anti-mouse (GE Healthcare #NA934V), coupled with horseradish peroxidase was used as secondary antibody for Western blot at a 1:5000 dilution. Human recombinant IFN-beta1a (IFNβ) was obtained from Biomol (#86421). Recombinant human IFNλ1 (IL-29) (#300-02L) and IFNλ2 (IL28A) (#300-2K) were purchased from Peprotech and IFNλ3 (IL-28B) from Cell signaling (#8796). High affinity engineered IFNλ3 variant (H11) and wild type IFNλ3 were produced as described in [32]. The IFN concentrations used to treat the cells are stated in the main text and in the figure legends.
T84 human colon carcinoma cells (ATCC CCL-248) were maintained in a 50:50 mixture of Dulbecco’s modified Eagle’s medium (DMEM) and F12 (GibCo) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin (GibCo). SKCO15 cells were maintained in DMEM with 10% fetal bovine serum, 1% penicillin/streptomycin, 15mM HEPES and 1% NEAA (Non-Essential Amino Acids). Mini-gut organoids were harvested and maintained as described earlier [24]. VSV-Luc was used as previously described [24].
Human colon tissue was received from colon and small intestine resection from the University Hospital Heidelberg. This study was carried out in accordance with the recommendations of “University Hospital Heidelberg” with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. All samples were received and maintained in an anonymized manner. The protocol was approved by the “Ethic commission of University Hospital Heidelberg” under the approved study protocol S-443/2017.
RNA was harvested from cells using NuceloSpin RNA extraction kit (Macherey-Nagel) as per manufacturer’s instructions. cDNA was made using iSCRIPT reverse transcriptase (BioRad) from 200ng of total RNA as per manufacturer’s instructions. qRT-PCR was performed using SsoAdvanced SYBR green (BioRad) as per manufacturer’s instructions, TBP and HPRT1 were used as normalizing genes.
Colon organoids and T84 cells were treated with 2000 RU/ml of type I IFN (β) or 100 ng/ml of each type III IFN (λ1,2 and 3). Total RNA was isolated at 3, 6, 12 and 24h post-treatment as described above. For the gene expression analysis of interferon stimulated genes (ISGs), qRT-PCR was performed using the predesigned 384-well assay of type I IFN response for use with SYBR Green assaying the expression of 87 ISGs (Biorad # 10034592). The expression of 45 additional ISGs and transcriptional factors was analyzed by qRT-PCR with primer sets obtained as previously described [27]. The complete gene list monitored in this study and the primers used to amplify each gene is available in S1 and S2 Tables. Differential expression analysis of each treatment was performed by comparing the baseline expression of genes in an untreated control at each time point. Only genes which were either induced or reduced more than 2-fold in any of the samples were considered to be significantly regulated. These genes were either analyzed using scatterplots or visualized by a heatmap after sorting the fold change of expression in response to type I IFN (β) in decreasing order. For the T84 cells all fold change values above 20 and below 0.05 were replaced with 20 and 0.05 respectively. For the organoids, the fold change values above 800 and below 1/800 were replaced with 800 and 1/800. This data adaptation was done to center the heatmap around 0 (white) and to avoid errors in logarithmic calculations. When visualizing the expression peaks, only the highest value is shown per time point for each gene. All analyses were performed using R version 3.3.0 and 3.3.3 including the packages gplots and ggplot2.
At time of harvest, media was removed, cells were rinsed one time with 1X PBS and lysed with 1X RIPA buffer (150 mM sodium chloride, 1.0% Triton X-100, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulphate (SDS), 50 mM Tris, pH 8.0 with phosphatase and protease inhibitors (Sigma-Aldrich)) for 20mins at 4°C. Lysates were collected and equal protein amounts were separated by SDS-PAGE and blotted onto a PVDF membrane by wet-blotting. Membranes were blocked with 5% milk or 5% BSA, when the phospho STAT1 antibody is used, in TBS containing 0.1% Tween 20 (TBS-T) for one hour at room temperature. Primary antibodies were diluted in blocking buffer and incubated overnight at 4°C. Membranes were washed 4X in TBS-T for 15mins at RT. Secondary antibodies were diluted in blocking buffer and incubated at RT for 1h with rocking. Membranes were washed 4X in TBS-T for 15mins at RT. HRP detection reagent (GE Healthcare) was mixed 1:1 and incubated at RT for 5mins. Membranes were exposed to film and developed.
Colon organoids and T84 cells were seeded in a white F-bottom 96-well plate. Samples were pre-treated prior to infection or treated post-infection as indicated with increasing concentrations of type I or type III IFNs. VSV-Luc was added to the wells and the infection was allowed to proceed for 8hrs. At the end of the infection, media was removed, samples were washed 1X with PBS and lysed with Cell Lysis Buffer (Promega) at RT for 20 mins. A 1:1 dilution of Steady Glo (Promega) and Lysis Buffer were added to the samples and incubated at RT for 15 mins. Luminescence was read using an Omega Luminometer.
Fluorescence-activated cell sorting (FACS) was performed on FACSMelody Cell Sorter (BD Biosciences). DAPI was added for nuclear staining. Data were processed using FlowJo 10.0.5.
Knockout of IFNAR1 and IFNLR1 in T84 cells were achieved by using the CRISPR/Cas9 system as described earlier [24]. For back-compensation of the IFN receptor KO cell lines and for generation of wild-type T84 cells overexpressing the IFNAR1 and IFNLR1, plasmids containing the cDNA of IFNAR1 and IFNLR1 were obtained from a gateway compatible ORF bank (pENTRY221-IFNAR1) and from GE Healthcare (pCR_XL_TOPO_IFNLR1, #MHS6278-213246004), respectively. The IFNLR1-GFP construct (pC1-HsIFNLR1-GFP) was generated using the following cloning strategy. A mammalian expression plasmid producing a N-terminal EGFP-tagged extracellular domain of IFNLR1 (EGFP-IFNLR1) was generated as follows: cDNA corresponding to this open reading from was generated synthetically (GeneArt, Life Technologies) and subsequently sub-cloned directly into the pC1 expression plasmid (Promega) backbone. Specifically, monomeric EGFP was introduced between the signal peptide sequence and the remaining glycoprotein flanked by three alanine residues at its amino terminus and a short glycine-serine linker sequence of N-AAASGSGS-C at its carboxyl terminus. Tri-alanine flanking allowed facile incorporation of restriction enzyme sites (Not1 and SacII) allowing removal or swapping of EGFP tag. Sequences available on request. Caspase-cleavage resistant IFNAR1 and IFNLR1 were generated using the Quick Change II XL site directed mutagenesis kit (Agilent Technologies, Germany), following manufacturer’s instructions. Point mutations were controlled by plasmid sequencing. The expression vectors were generated by inserting the respective constructs into the lentiviral vector pDest GW35 by using the Gateway cloning technology (Life Technologies, Germany) according to manufacturer’s instructions. Lentiviruses were produced as previously described [24], and T84 cells were transduced two times using concentrated stocks of lentiviral particles encoding the cleavage resistant IFNAR1 and IFNLR1. 36 hours post-transduction, transduced cells were selected for using blasticidin.
The mathematical model was implemented in terms of ordinary differential equations (ODEs) in MATLAB 2016b (S3 Table). The numerical simulations were conducted using the CVODES, a module from SUNDIALS numerical simulation package, in the MATLB environment. The model was initially set to a steady state condition and most of the initial conditions were set (S4 Table). Only, the IFNLR efficacy factor was estimated using time-resolved ISG expression data that we measured with different doses of type I IFN (β) or III IFN (λ1−3). All of the ISG expression data for the IFNAR1 and IFNLR1 overexpression experiments were reproduced only by fitting new initial values of IFNAR1 and IFNLR1 (S5 Table).
Parameter estimation was conducted by minimizing the weighted nonlinear least squares,
wSSR=∑i=1N(1σi2)∑j=1M(ysimulation_i,j−yobserved_i,j)2,
of model simulations versus data points, j = 1, …, M, of different experiments, i = 1, …, N. The variance, σi2, of every time-resolved experimental data was used as a weighting factor for fitting the corresponding data. The variance was calculated by multiplying the respective mean value with the average coefficient of variation of the experimental data.
To assess the uncertainty in the estimated parameter values, we used the profile-likelihood method [70]. In this method, the parameter confidence bounds are calculated based on their contribution to the likelihoods, or in another word, the objective function (wSSR). This computational approach is conducted in a stepwise manner. In every step, the respective parameter is fixed at a new value distant from the optimum estimated one. Then, the new maximum likelihood is calculated (wSSRmin(θ)). Using this approach, we can calculate the profile of the maximum likelihoods over different values of the considered parameter. Then a threshold, Δα,
Δχ2=wSSRmin(θ)−wSSRmin(θoptimum),
{θ|Δχ2<Δα},
is used to define the confidence bounds for the respective parameter. The threshold, Δα, is the α quantile of the chi-squared distribution.
To investigate the effect of the parameter uncertainty on model predictions we calculated approximate 95% confidence bands, as explained in Seber and Wild [71].
Approx95%confidencebands=ysimulated±tinvN−Pα∙MSE∙(1+S∙(S∙S)−1∙S)12
where “tinvN−Pα” is the α quantile of student's t distribution, “N” is the number of data points and “P” is the number of estimated model parameters, “MSE” is the mean standard error and “S" is the sensitivity matrix of the respective simulated observable.
To select the most parsimonious model, the simplest model with good predictive power, from the ensemble of the four alternative models of the ISG response to type I versus type III interferon, we used the Akaike information criterion corrected for small sample size (AICc). After fitting the models to the experimental data, we calculate the AICc score for every model. AICc is calculated as:
AICc=n(ln(2π∙wSSRn)+1)+2k+2k(k+1)n−k−1,
where n is the number of data points used to fit the model, k is the number of estimated parameters of the respective model, and wSSR is the minimum weighted sum of squared residuals for the respective model. The model with the minimum AICc value is selected as the most parsimonious model from the ensemble of alternative models. In order to compare the selected model with other models, we calculate both ΔAICc, the difference between the AICc value of the models with the minimum AICc value from the ensemble of the models, and the AICc weight (w i). The Akaike weight is a weight of evidence for the respective model and is calculated as:
wi=exp(−12ΔAICci)∑r=1Mexp(−12ΔAICcr).
|
10.1371/journal.pgen.1006261 | The Non-coding Mammary Carcinoma Susceptibility Locus, Mcs5c, Regulates Pappa Expression via Age-Specific Chromatin Folding and Allele-Dependent DNA Methylation | In understanding the etiology of breast cancer, the contributions of both genetic and environmental risk factors are further complicated by the impact of breast developmental stage. Specifically, the time period ranging from childhood to young adulthood represents a critical developmental window in a woman’s life when she is more susceptible to environmental hazards that may affect future breast cancer risk. Although the effects of environmental exposures during particular developmental Windows of Susceptibility (WOS) are well documented, the genetic mechanisms governing these interactions are largely unknown. Functional characterization of the Mammary Carcinoma Susceptibility 5c, Mcs5c, congenic rat model of breast cancer at various stages of mammary gland development was conducted to gain insight into the interplay between genetic risk factors and WOS. Using quantitative real-time PCR, chromosome conformation capture, and bisulfite pyrosequencing we have found that Mcs5c acts within the mammary gland to regulate expression of the neighboring gene Pappa during a critical mammary developmental time period in the rat, corresponding to the human young adult WOS. Pappa has been shown to positively regulate the IGF signaling pathway, which is required for proper mammary gland/breast development and is of increasing interest in breast cancer pathogenesis. Mcs5c-mediated regulation of Pappa appears to occur through age-dependent and mammary gland-specific chromatin looping, as well as genotype-dependent CpG island shore methylation. This represents, to our knowledge, the first insight into cellular mechanisms underlying the WOS phenomenon and demonstrates the influence developmental stage can have on risk locus functionality. Additionally, this work represents a novel model for further investigation into how environmental factors, together with genetic factors, modulate breast cancer risk in the context of breast developmental stage.
| A woman’s lifetime risk of developing breast cancer is affected by both genetic and environmental risk factors that can be further exacerbated by breast developmental stage. Time periods conferring increased risk are referred to as Windows of Susceptibility (WOS) and, generally speaking, the molecular mechanisms responsible for their effect on breast cancer risk are unknown. Our work presented here on the characterization of the rat Mammary Carcinoma Susceptibility 5c, Mcs5c, locus has identified a region within Mcs5c that interacts with the neighboring gene, Pappa, in an age-dependent manner to influence gene expression via genotype-dependent DNA methylation. Importantly, Mcs5c-mediated gene regulation occurs specifically within a WOS, and these finding represent the first identified molecular mechanisms by which a WOS influences the ability of a locus to affect mammary/breast cancer risk. This work highlights the importance developmental stage can have on genetic risk factor function, and we anticipate that the Mcs5c locus will serve as a model for future studies on WOS in combination with genetic and environmental risk factors.
| In the United States, breast cancer is the most frequently diagnosed cancer and second leading cause of cancer death among women [1]. Its etiology is complex, consisting of the interaction of both genetic and environmental risk factors whose contribution to overall risk can vary depending on the developmental context of the individual. In general, time periods in which women are more susceptible to initiating events affecting their long term breast cancer risk are broadly referred to as Windows of Susceptibility (WOS) [2]. In humans, the best documentation of a WOS can be found in studies of radiation exposure in women. Women exposed to radiation between 0 and 30 years of age during either the atomic bombings of Japan or for the treatment of Hodgkin’s lymphoma had an increased risk of developing breast cancer later in life compared to women >30 years of age at time of exposure [3,4]. This time period, therefore, represents one of the WOS, and encompasses ages spanning childhood, adolescence, and young adulthood in women. Animals studies performed in rats to model the human WOS phenomenon [5] further suggest the existence of at least two mechanistically distinct susceptibility windows within the larger human WOS, namely, the sexually immature WOS (iWOS) and the adolescent WOS (aWOS). This division of the WOS is most evident in work by Ariazi et al. [6] on a carcinogen-inducible model of breast cancer, where administration to developmentally immature (3 week) and adolescent-aged (7 week) rats resulted in differential carcinoma development depending on age of administration and the carcinogen used. Additionally, although over 80 genetic loci affecting breast cancer susceptibility have been identified in human genome-wide association studies (GWAS) [summarized in 7], their function in relation to developmental stages has not been characterized. In general, while the effects of window specific exposures are well documented, the cellular mechanisms responsible for their function and governing their interactions with environmental and genetic risk factors are poorly understood.
To begin to understand the complex interactions between WOS, genetics, and the environment, we turned to a comparative genomics approach, utilizing a rat model of breast cancer. The rat is an excellent model for this type of study, as not only does its mammary gland and mammary tumor development mimic that of the human condition [8], but, as previously mentioned, it too displays the WOS phenomenon [5,6]. Additionally, inbred rat strains vary in their susceptibility to carcinogen-induced mammary cancer, allowing for the identification of genetic susceptibility loci through quantitative trait loci (QTL) analysis. This approach was applied in our lab, utilizing the mammary cancer resistant Wistar-Kyoto (WKy) and susceptible Wistar-Furth (WF) inbred rat strains resulting in the identification and subsequent fine-mapping of the Mammary Carcinoma Susceptibility 5c, Mcs5c, locus [9–11]. Mcs5c maps to a 170kb region located in a large gene desert on rat chromosome 5 that shares homology with mice and humans (Fig 1). In both chemical carcinogen and oncogene-induced models of mammary cancer, congenic lines homozygous for the resistant WKy Mcs5c allele showed an approximately 50% reduction in carcinoma number compared to susceptible WF-homozygous controls [11].
Using the Mcs5c locus as a model, we sought to examine the interaction between a genetic risk factor and WOS. We have characterized an 8.5kb temporal control element (TCE) within Mcs5c affecting the expression of neighboring gene Pregnancy-associated plasma protein A, Pappa, in a genotype-dependent manner in mammary epithelial cells (MECs). The function of the Pappa/PAPP-A protein makes it an attractive candidate for involvement in both the WOS phenomenon and breast cancer development. PAPP-A is a protease that acts to positively regulate bioavailability and signaling of the Insulin-like growth factors, IGFs, through the cleavage of IGF binding proteins 2, 4 and 5, IGFBP2/4/5 [12–15]. The specific role of PAPP-A in normal breast development has not been studied, but the IGF-I pathway, in general, is an essential component of breast/mammary gland development, as evident by the severe mammary gland defects of Igf-I and Igf-I receptor (Igf1r) knockout mice [16–18]. The role of IGF-I in breast cancer development is supported by numerous studies which associate the IGF-I signaling pathway with breast cancer initiation and progression [19]. Indeed, in transgenic mice, overexpression of IGF-I in the mammary gland resulted in increased susceptibility and decreased latency to spontaneous and carcinogen-induced mammary adenocarcinomas [20]. Limited studies of PAPP-A function in cancer have demonstrated that increased PAPP-A activity enhanced tumor growth in ovarian and lung cancer cell lines [21,22], and inhibition of its proteolytic function reduced tumor growth in a murine mammary cancer cell line [23]. Furthermore, TCGA data [24] found PAPP-A to be altered in 6% of invasive breast carcinomas, with amplification/mRNA upregulation identified as the most common genetic alterations, and found co-amplification of neighboring loci, encompassing the homologous MCS5C locus, occurring in approximately 1–2% of cases (accessed via www.cbioportal.org; [25,26]).
In this study, we have identified age-specific differences in Mcs5c activity which support the existence of mechanistically distinct susceptibility windows. We have functionally characterized the non-coding Mcs5c locus, finding that it acts during the aWOS to regulate Pappa expression through age-dependent chromatin looping and genotype-dependent DNA methylation. To our knowledge, this study represents the first identification of a molecular mechanism underlying the aWOS phenomenon and highlights the ability of developmental age to influence the activity of a susceptibility locus.
To determine if Mcs5c exerts its effect on carcinoma multiplicity via the mammary gland, transplant experiments were performed. Donor mammary gland tissue from either the Mcs5c resistant 5C-27 line or a Mcs5c susceptible control line was transplanted onto the interscapular fat pad of recipient rats from both genotypes, creating four donor-recipient groups. This direct transplant design allowed for the detection of mammary gland-host interactions and did not result in differential tissue rejection rates, as the lines are isogenenic except at the Mcs5c locus. Transplant tissue rejection rates were not statistically significant between transplant groups consisting of donors and recipients with the same genotype versus groups with different genotypes (Chi-squared test, X = .10, df = 1, p-value = 0.75). Results from the mammary gland transplant experiment are shown in Fig 2. Resistant and susceptible rats receiving resistant donor tissue had a transplant site carcinoma incidence of 21% and 27%, respectively (n = 76, 49), while resistant and susceptible rats receiving susceptible tissue had incidences of 42% and 38%, respectively (n = 69, 39). Recipient rats of either genotype that received susceptible donor tissue had higher transplant site carcinoma incidences than those that received tissue from resistant rats. In this way, the carcinoma phenotype was dependent on the donor tissue genotype and was not influenced by the recipient’s genotype, suggesting that Mcs5c acts within the mammary gland. Indeed, logistic regression analysis found a statistically significant donor effect (p-value = 0.0043; recipient effect p-value = 0.825). Thus, it was concluded that Mcs5c acts in a mammary gland autonomous manner to influence carcinoma multiplicity.
Quantitative real-time PCR (qPCR) was used to investigate expression levels of nearby genes in mammary epithelial cells (MECs) of Mcs5c resistant and susceptible rats at 4–12 weeks of age. This age range was chosen as it captures multiple mammary gland developmental windows, including the iWOS (4 weeks), aWOS (6–9 weeks), and adult (12 weeks) time periods. Pappa, located over 517kb away from Mcs5c, was found to be differentially expressed in MECs in an age-dependent manner (Fig 3). In general, Pappa expression levels were dynamic in Mcs5c susceptible MECs during development, while Mcs5c resistant expression remained relatively steady over time. Compared to Mcs5c resistant rats, Pappa expression was increased in susceptible rats by 43% at 6 weeks (Mann-Whitney U test, p-value = 0.015, n = 13 and 15, respectively), 14% at 7 weeks (p-value = 0.05, n = 23 and 19), and 31% at 9 weeks (p-value = 0.0003, n = 23 and 18). Differential expression disappeared by 12 weeks of age (n = 9 and 18), at which point the mammary gland is fully developed and rats are past the aWOS stage [5]. Expression trends were reversed in 4 week old rats, with susceptible animals showing a sharp decrease in expression relative to resistant rats (p-value = 8e-5, n = 9 and 8, respectively). Mcs5c, therefore, appears to functioning during both the iWOS and aWOS. Unfortunately, we were unable to obtain robust antibodies for analysis of Pappa protein levels in mammary gland tissue. Differential expression in MECs was not observed for neighboring genes Tenascin C, Tnc, and Tumor Necrosis Factor (Ligand) Superfamily, Member 15, Tnfsf15, during the aWOS. However, differential expression of Tnfsf15 was observed in 4 week-old, immature MECs, highlighting the complexity and age-specific nature of Mcs5c locus activity (S1 Fig).
Genotype dependent differential expression seen in MECs led to the hypothesis that Mcs5c contained a long-distance acting regulatory element influencing Pappa expression. Such a relationship could be mediated by a physical association between the two regions, resulting in the looping out of intervening DNA sequence. Chromosome conformation capture (3C) was used to identify such an interaction. To create 3C templates, MECs were isolated from the mammary glands of Mcs5c resistant and susceptible animals at 4, 6, 7, and 12 weeks of age. Two fixed bait regions located at the Pappa locus were chosen for extensive analysis of potential interactions with Mcs5c. These regions, P3-1 and P4-1, span approximately 2.4kb and 2kb in size, respectively, with P3-1 encompassing Pappa exon one and a conserved CpG island, and P4-1 falling within the first intron (Fig 4A and 4C). These two regions were chosen for analysis as their degree of sequence conservation suggested that they may be functionally relevant in transcriptional regulation of the Pappa gene (Fig 4A). Bait region P3-1 was negative for any interaction with Mcs5c at 4, 7, and 12 weeks of age (S2 Fig). Conversely, 3C analysis using bait region P4-1 revealed an 8.5kb region within Mcs5c that displayed a high relative interaction frequency (IF) in 6 and 7 week templates, indicative of a physical interaction between the two regions occurring over a distance of 590kb (Fig 4D). 4 and 12 week templates had a much lower IF at this -590 region, leading to the formation of two distinct, age-dependent interaction groups displaying either a strong (6 and 7 week) or weak (4 and 12 week) IF. The difference in IF for these two groups was statistically significant (Mann-Whitney U test, p-value = 1.02e-10, n = 27 and 38 biological replicates, respectively). For all ages, there was no difference in IF between genotypes, indicating that the interaction is age-dependent but not genotype-dependent. We will therefore refer to the -590 looping region of Mcs5c as the temporal control element (TCE; chr5:84,428,694–84,437,192; RGSC 5.0/rn5). Three additional Pappa bait regions were tested for interactions with the Mcs5c TCE at 4 and 6 weeks of age (S2 Fig). Two of these regions, P4-1A and P4-2, were negative, while the more proximal P3-3 region displayed an aWOS-specific looping interaction that mimicked the TCE/P4-1 interaction. This indicates that the Mcs5c TCE may utilize a more complex looping scheme to facilitate Pappa regulation, and defines the TCE as a functionally important region within Mcs5c. To determine if these interactions are also tissue-specific, 3C profiles were analyzed from 4 and 7 week colon epithelial cells and 7 week liver hepatocytes from Mcs5c resistant rats. The Mcs5c TCE did not interact with P4-1 (Fig 4E) or P3-3 (S2 Fig) in these tissues, implying that the interactions between Pappa and the Mcs5c TCE are tissue-specific in addition to age-dependent. Sequencing of the resistant WKy and susceptible WF TCE alleles revealed 10 variants between the two (S7 Table), and although our 3C results showed that age-specific looping occurs independent of genotype, we speculate that one or more variants may be involved in genotype-dependent expression differences observed during this time period.
CpG island (CGI) shores are regions located approximately 2kb away from CGIs, and have increasingly been identified as the sites of tissue specific differential methylation associated with gene expression changes [27]. The Pappa looping fragment, P4-1, resides in a CGI shore region (Fig 5A). As this region is a target site of Mcs5c TCE looping, we hypothesized that Mcs5c may affect Pappa expression through an epigenetic mechanism targeted to the P4-1 fragment. Methylation levels for 12 CG dinucleotides within and proximal to P4-1 were examined in MECs of Mcs5c resistant and susceptible rats at 4, 6, 7, 9 and 12 weeks of age using custom designed pyrosequencing assays (Fig 5A). Selection of these timepoints allowed for the examination of methylation patterns before, during, and after the aWOS. In general, methylation levels were dynamic across this region, with sites 2–4 consistently displaying the lowest methylation levels (average = 13% methylated) and sites 9–12 displaying the highest levels (average = 68% methylated) (S4 Table). Additionally, there appeared to be few age-specific differences in methylation levels for animals within the aWOS, therefore data for 6, 7, and 9 week old rats were combined within genotypes. Of the 12 sites examined, 6 showed statistically significant genotype-dependent differences in methylation levels after adjusting for multiple comparisons (Mann-Whitney U-test with Bonferroni correction). The percent change in methylation levels along with p-values are shown in Table 1. All statistically significant, genotype-dependent methylation differences occurred during the aWOS and were directionally identical, with methylation levels decreased in Mcs5c susceptible MECs. The percent decrease in methylation levels ranged from 5.0%– 22.7%. Additionally, a number of other sites displayed a similar trend, although these differences were not significant after Bonferroni correction. At ages outside of the aWOS, there were no statistically significant genotype-dependent differences in methylation, although sites 1 and 2 displayed a non-significant trend of increased methylation in Mcs5c susceptible MECs at the 4 week time point. We also investigated the methylation state of the Pappa CGI using 2 pre-made pyrosequencing assays (Fig 5A). Methylation levels for both assays were assessed in 4 week old animals, while one assay was examined at the remaining timepoints. For all CGI assays and timepoints, there were no genotype-dependent differences in methylation levels and, in general, the Pappa CGI is hypomethylated at all ages, with site specific methylation levels ranging from 0.16% - 8.41% (S5 Table). The observation of decreased shore methylation and increased Pappa expression in Mcs5c susceptible MECs strongly supports the canonical role of DNA methylation in gene regulation, that is, that the two are negatively correlated. Indeed, for 6 week MECs, for which we had both DNA and RNA samples, Pappa expression was negatively correlated with the average methylation percentage of the 6 significant shore sites (Fig 5B; Pearson correlation coefficient, R, = -0.67, n = 18, p-value = 0.0023). By contrast, no correlation was observed between Pappa expression and the average methylation percentage of the CGI-2 assay sites (Fig 5C; Pearson correlation coefficient, R, = 0.16, n = 18, p-value = 0.52). The identification of genotype-dependent methylation differences during the aWOS suggests that Mcs5c facilitates genotype-dependent Pappa expression differences observed during this time period through epigenetic modification of the Pappa CGI shore.
In an effort to causally tie the Mcs5c TCE to Pappa expression and CGI shore methylation, the entire 8.5kb region was targeted for deletion in the rat mammary carcinoma cell line, LA7. Two CRISPR guides were used to target the region, and clones were screened via PCR across the cut site, with validation by sequencing (S1 Table). We were unable to identify a clone with all copies of the TCE removed, despite much effort. This was likely due to the aneuploid nature of LA7 cells, and mutations incurred at CRISPR guide target regions (S2 & S3 Tables). Copy number analysis of 9 positive CRISPR edited clones showed that we were able to delete a majority of TCE copies, reducing the copy number by 3.5-fold across all clones (Fig 6A). 3C analysis of positive clones indicated that removal of multiple TCE copies resulted in decreased TCE/P4-1 looping, but did not alter TCE/P3-3 looping, which remained consistent with WT levels (Fig 6B). Interestingly, this suggests that the looping mechanisms responsible for these interactions are functionally distinct. Expression analysis revealed a significant reduction in Pappa expression in CRISPR clones compared to wild-type LA7 cells, with Pappa decreased 4-fold across all clones (Fig 6C). A Pearson correlation coefficient was computed to determine the relationship between Pappa expression and TCE copy number, and a positive correlation between the two was observed (Fig 6D; R = 0.6245, n = 13, p-value = 0.0225). Conversely, no change in Tnc or Tnfsf15 expression was observed with TCE knockdown, and expression levels were not correlated to TCE copy number (S3 Fig). These data support our hypothesis that Mcs5c contains a long-range regulatory element, and emphasizes the functionality of the TCE/P4-1 chromatin loop to Pappa gene expression.
Our in vivo analysis highlighted the importance of Pappa CGI shore methylation to Pappa expression, and we sought to verify this relationship in our in vitro model as well. Treatment of wild-type LA7 cells with the DNA methylation inhibitor 5-aza-2’-deoxycytidine (5-aza-dC) resulted in a 15-fold increase in Pappa expression (Fig 6E), indicating that DNA methylation plays a role in Pappa regulation. To more specifically address the relationship between TCE/P4-1 looping and Pappa methylation, CGI and CGI shore methylation levels were analyzed in wild-type LA7 cells and CRISPR edited clones (S10 Table). Two CGI shore sites showed statistically significant differences in methylation, with a 3.9% decrease and a 25% increase in methylation levels observed in CRISPR clones at sites 5 site 12, respectively (S10 Table and Fig 6F). Methylation changes at site 12 were more pronounced, and were investigated further. We found site 12 methylation levels to negatively correlate with TCE copy number (Fig 6G; R = -0.8034, n = 13, p-value = 0.0009). Site 12 methylation levels were also negatively correlated with Pappa expression (Fig 6H; R = -0.6022, n = 17, p-value = 0.011), mimicking the observed in vivo relationship. Altogether, these data suggest a functional chain of events whereby the TCE, via the TCE/P4-1 loop, affects Pappa CGI shore methylation levels which then, in turn, affect Pappa expression.
Previous work on Mcs5c had fine-mapped the locus to a 170kb non-coding region on rat chromosome 5. This locus resulted in an approximately 50% decrease in both chemical carcinogen and oncogene-induced mammary carcinoma development when homozygous for the resistant WKy allele. The gene Tnc was initially identified as a possible target of Mcs5c activity, with genotype-dependent differential expression observed in the thymus and ovaries exclusively following carcinogen exposure [11]. However, in this study, we have shown that the Mcs5c locus affects carcinoma multiplicity in a mammary gland autonomous manner (Fig 2). This suggests that the previous non-mammary gland expression differences observed following carcinogen exposure do not play a role in carcinoma initiation, and are either irrelevant or secondary to initial carcinoma development that is dependent on mammary gland intrinsic factors. While these hypotheses warrant further investigation, a reevaluation of gene expression within the mammary gland was conducted, revealing genotype-dependent and age-specific differential expression of Pappa in mammary epithelial cells (MECs; Fig 3). Specifically, Mcs5c susceptible MECs from 6 to 9 week old rats showed increased expression of Pappa compared to Mcs5c resistant rats. Importantly, the 6 to 9 week age range encompasses a time period of rapid mammary gland development and maturity, falling within the aWOS [5].
We hypothesized that Pappa expression changes were mediated by a regulatory element within Mcs5c, and our experimental results support this hypothesis, identifying a complex set of mechanisms underlying Mcs5c-mediated regulation of Pappa. Through 3C experiments, we have identified a region within Mcs5c, the temporal control element (TCE), that physically interacts with the Pappa locus at two regions, P4-1 and P3-3, in an aWOS- and MEC-specific manner over distances of 590kb and 580kb, respectively (Fig 4 and S2 Fig). The importance of the TCE/P4-1 long-range looping interaction to Pappa expression was demonstrated in vitro, where removal of TCE copies resulted in a reduction of TCE/P4-1, but not TCE/P3-3, looping, and correlated with decreased Pappa expression (Fig 6). These data indicate that the two observed TCE chromatin interactions are functionally distinct, and demonstrates a strong positive regulatory relationship between the Mcs5c TCE and Pappa expression, which appears to be dependent on TCE/P4-1 chromatin looping. The Mcs5c TCE/Pappa P4-1 interaction, therefore, represents another example of a long-distance acting regulatory region, akin to those identified for the Shh [28] and Sox9 [29] genes, as well as the previously characterized Mcs1a locus [30].
Importantly, as looping occurs in a genotype-independent manner, additional mechanisms must be responsible for the differential expression observed between Mcs5c resistant and susceptible rats. With the intronic P4-1 looping region falling in a CGI shore, DNA methylation of this region became a mechanistic candidate to explain observed expression differences. The importance of differentially methylated CGI shores to gene expression was first highlighted by Irizarry and colleagues in 2009 [27]. Since then, many studies have shown an association between differentially methylated shore regions and gene expression changes [31–38]. Our study identified 6 CG dinucleotides within and proximal to the P4-1 looping region that were differentially methylated between Mcs5c resistant and susceptible MECs (Table 1). Significant methylation differences were observed exclusively during the aWOS, and a negative correlation between shore methylation levels and Pappa expression strongly suggest that DNA methylation plays a role in differential Pappa expression (Fig 5B). This correlation was recapitulated in our in vitro model, where differential shore methylation was also negatively correlated with TCE copy number (Fig 6G and 6H). As copy number acts as an indicator of TCE/P4-1 looping frequency in this model, this suggests a functional relationship between looping and shore methylation, where the TCE/P4-1 loop acts to facilitate differential methylation that, in turn, regulates Pappa expression. Given the inherent difficulties of modeling an age-dependent phenomenon in vitro, these results must be interpreted cautiously; however, we feel that the similarities between our in vitro and in vivo results indicate that these mechanisms are robust and functionally relevant to Mcs5c-mediated Pappa regulation.
Overall, we have identified two mechanisms associated with Mcs5c regulation of Pappa expression during the aWOS, chromatin looping and DNA methylation. Our in vitro experiments have indicated the importance of the TCE/P4-1 loop for Pappa expression and shore methylation; however, in vivo analyses have shown that these actions are mechanistically distinct, as Pappa expression and differential methylation, but not looping, are genotype-dependent. An unresolved issue is precisely how Mcs5c is mediating these activities. We hypothesize that the genotype-independent TCE/P4-1 loop serves to facilitate the recruitment of transcription factors, cofactors, and/or methyltransferases that act separately or together to directly regulate Pappa methylation and expression during the aWOS. The binding of these regulatory factors would be affected by one or more variants within the TCE without affecting chromatin looping. Wright et al. [39] identified a similar interaction at the c-MYC locus, where an enhancer-associated SNP affected transcription factor binding without altering chromatin structure. Sequencing of the resistant WKy and susceptible WF Mcs5c TCE alleles (chr5:84,428,694–84,437,192; RGSC 5.0/rn5) has revealed 10 candidate polymorphisms (S7 Table) for future investigation of their effect on protein binding and subsequent Pappa expression and methylation changes.
Mcs5c activity during the aWOS stands in stark contrast to that observed during the iWOS (4 weeks). Specifically, differential Pappa expression during the iWOS is reversed compared to the aWOS (Fig 2), TCE/Pappa looping is lacking (Fig 4D and S2 Fig), and there are no statistically significant CGI shore methylation differences (Table 1). These data indicate that the regulatory actions of Mcs5c are dependent on developmental context, a phenomenon observed at other regulatory regions, most notably the β-globin locus control region [40]. Age-specific differences in Pappa expression, looping, and methylation could be explained by interactions with proteins specific to these developmental time points. Identifying proteomic differences between the immature and adolescent mammary gland will be crucial in understanding the players driving window-specific mechanistic differences in Mcs5c activity. We hypothesize that age-specific protein expression results in an alternative looping interaction between Mcs5c and Pappa during the iWOS. Differential expression of Tnfsf15 exclusively at the 4 week time point (S1 Fig) indicates that Mcs5c may exhibit a more complex chromatin interaction during the iWOS, regulating multiple genes simultaneously. Additionally, a trend towards increased methylation in Mcs5c susceptible MECs is functionally consistent with the reduction of Pappa expression observed during this time period. It is possible that these sites are indicative of significant methylation differences occurring at sites not examined in this study, both at the Pappa locus as well as Tnfsf15, and shore methylation may still, therefore, be relevant to Mcs5c activity during the iWOS.
Altogether, we have functionally characterized the Mcs5c locus, finding that it acts via two distinct mechanisms to influence Pappa expression in an age-dependent manner during a well-characterized breast cancer WOS (Fig 7). This work highlights the importance of characterizing genetic risk factors in the context of developmental windows of susceptibility (G x WOS), and emphasizes the complex interaction between genetic, environmental, and age-specific risk factors. Mcs5c susceptible rats showed increased expression of Pappa in MECs and an increased susceptibility to carcinogen-induced mammary carcinogenesis, supporting a protective benefit of reduced Pappa levels during adolescent development. Decreased levels of Pappa in the developing mammary gland would result in reduced Igf-I bioavailability through a reduction in Igfbp cleavage [12]. Given that the Igf-I signaling pathway acts to promote proliferation and inhibit apoptosis during mammary gland development [41], it is therefore likely that a reduction in free Igf-I would reduce the proliferative index of MECs. As the effects of many environmental mutagens, such as radiation and chemical carcinogens, are dependent on interactions with the DNA of proliferating cells [42], this would result in fewer targets for mutagenesis, and represents one possible method by which reduced Pappa expression during the aWOS may result in a mammary carcinoma resistant phenotype. Understanding the mechanisms behind G x WOS interactions and how environmental risk factors influence these interactions will play a crucial role in breast cancer risk assessment, and in the identification of targets and strategies for cancer prevention in young women. There is growing concern over the impact adolescent exposure to a broad range of environmental factors may have on long-term breast cancer risk [43]. Our work has demonstrated a functional relationship between genetic risk factor activity and developmental stage, and it is likely that environmental risk factors may further confound such interactions. Indeed, CGI shore methylation has been found to be affected by environmental factors such as ELS and diet [32,33,44,45]. We believe that the Mcs5c locus will serve as a robust model to study how environmental factors affect breast cancer risk by influencing G x WOS interactions, and may encourage the characterization of other such cancer susceptibility loci in this context.
Congenic rat lines were maintained in an AAALAC-accredited facility as previously described [11]. All protocols were approved by the University of Wisconsin–Madison School of Medicine and Public Health Animal Care and Use Committee. Congenic rat lines are defined as having the resistant Wistar-Kyoto (WKy) Mcs5c allele introgressed on a susceptible Wistar-Furth (WF) background. The resistant congenic line used in this study, 5C-27, is WKy-homozygous for a genomic region that includes the entirety of the Mcs5c locus (Fig 1) [11]. Susceptible control animals are WF-homozygous at the Mcs5c locus.
Mcs5c WKy-homozygous congenic rats from line 5C-27 were used as resistant donors and recipients (Mcs5c resistant), and Mcs5c WF-homozygous rats were used as susceptible controls (Mcs5c susceptible). Abdominal and inguinal mammary glands were collected from female donor rats aged 30–35 days old, scissor minced, and split into four equal volumes. One volume was then grafted onto the interscapular white fat pad of four different 30–35 day old female recipient rats. Three weeks after transplantation (51–56 days), recipients were administered the chemical carcinogen 7,12-dimethylbenz(a)anthracene (DMBA), as a single oral dose dissolved in sesame oil at 65 mg/kg of body weight to induce mammary carcinoma formation. At 15 weeks post-DMBA, animals were removed from the study and carcinomas present at the transplant site greater than 3x3mm were counted. Generally, recipient rats developed ≤1 carcinoma at the transplant site, so incidence values were used as opposed to multiplicity. Fat pads were whole mounted and stained with aluminum carmine to verify transplant mammary gland growth. Four transplant groups were studied, with resistant 5C-27 or susceptible donor glands transplanted into both resistant and susceptible recipients (R->R, R->S, S->R, S->S). The tissue rejection rate for transplant groups consisting of donors and recipients with the same genotype (R->R, S->S) was compared to the rejection rate for groups with differing donor and recipient genotypes (R->S, S->R) via a Chi-squared test. The effect of donor and recipient genotype on carcinoma incidence, converted to a binary response value, was analyzed using logistic regression with two independent variables (donor genotype and recipient genotype) and no interaction term.
For all experiments, mammary epithelial cell (MEC) isolation began with fresh mammary glands (abdominal and inguinal, with lymph nodes removed) that were finely minced and digested for 2 hours at 37°C in 10 mL of GIBCO Dulbecco’s modified Eagle’s medium/F12 (DMEM/F12; Life Technologies) containing 0.01 g/mL of type III collagenase (Worthington). Cell pellets were collected by centrifugation and resuspended in 5 mL DMEM/F12. The suspension was loaded onto a 40μm nylon filter to eliminate stromal cells and collect mammary ductal fragments, consisting of an enriched MEC population. DNA was isolated from cells via the DNeasy Blood and Tissue kit (Qiagen). To isolate RNA, cells were homogenized in TRI Reagent (Ambion), followed by RNA extraction using the MagMAX-96 for Microarrays Total RNA kit (Ambion). LA7 cells used for downstream analysis were collected via treatment with 0.25% trypsin/EDTA (Life Technologies). RNA was extracted using the RNeasy Mini Kit (Qiagen) and DNA was extracted using the DNeasy Blood and Tissue kit (Qiagen).
MECs were collected at 4–12 weeks of age from female Mcs5c resistant and susceptible control rats. RNA was isolated as described above. For in vivo and in vitro expression analysis, cDNA was prepared from 1–2μg total RNA using Superscript II reverse transcriptase (Invitrogen). Gene expression was quantified using pre-designed or custom made TaqMan qPCR assays (Pappa, Rn01458295_m1, FAM; Tbp, Rn01455646_m1, VIC; Tnc, probe-FAM 5’ CGAGAGCTGTGATTAGA 3’, primers 5’ GGCTGTCAGAAGGCCAGATG 3’ and 5’ TGCCATGAAGGGATTTGAAGA 3’; Tnfsf15, Rn00595596_m1, FAM) and run on an ABI Prism 7900HT (Applied Biosystems). Tbp was chosen as the reference gene as its expression has been found to be relatively stable across a variety of rat tissues and during different stages of the estrous cycle in the mammary gland [46]. cDNA was diluted 1:4 or 1:8 and run using reaction conditions described previously [11]. Transcript quantities were calculated as described in Smits et al. [30], using a standard curve method to calculate Ct values and extrapolate quantity values. Sample measurements are an average of 3–4 technical replicates and data were analyzed using SDS software version 2.2.2 (Applied Biosystems).
Sample templates were prepared from MECs, colon epithelial cells, liver hepatocytes, and LA7 cells. MECs were isolated from 4, 6, 7, and 12 week old Mcs5c resistant and susceptible rats and the resulting cell suspension was diluted in PBS and fixed via the addition of 1.5% formaldehyde. Colon epithelial cells were isolated from 4 and 7 week old resistant rats, processed as described in Whitehead et al. [47], and fixed in formaldehyde. To isolate hepatocytes, the livers of 7 week old resistant rats were digested via cannulation of the portal vein and blanching of the liver using a pre-warmed solution of HBSS (Gibco) + 0.5mM EGTA followed by digestion via pre-warmed DMEM-low glucose (Gibco) + 1000CDU/mL Collagenase type IV (Worthington). Digested livers were collected in DMEM/F12 + 10% FBS on ice and cells dispersed manually. The suspension was filtered through a 100μm nylon filter and the filtrate spun for 2 minutes at 50xg. Supernatant was removed and cell pellets were washed until media became clear, followed by fixation in formaldehyde. Bacterial artificial chromosomes (BACs) encompassing the rat Mcs5c and Pappa promoter regions (CH230-433D12, CH230-498D4, CH230-256M9, and CH230-244C7) were ordered from Children’s Hospital Oakland Research Institute (CHORI) and used as positive control templates. Subsequent template preparation for all cell types and for BAC controls continued as described in detail in Smits et al. [48]. The restriction enzyme used was BglII. Chromatin interactions were detected via PCR, with bait primers located at the Pappa gene tested against Mcs5c primers spanning the entire locus (Fig 4A and 4B). Primer sequences are listed in S6 Table. Reaction components were 1X Herculase reaction buffer, 0.2mM dNTPs, 0.4μM primers, 0.3μμL Herculase Enhanced polymerase (5U/μL, Agilent) in a total volume of 25μμL. The amount of DNA template to add and optimal annealing temperatures were determined empirically. PCR reactions were performed using the following cycling conditions: 95°C for 1 min, 36 cycles of 95°C for 30 s, Ta for 30 s, 72°C for 20 s, followed by a final extension of 72°C for 8 min. Reactions were analyzed by agarose gel electrophoresis and visualized by ethidium bromide staining. Band intensities were quantified using ImageQuant software (GE Healthcare). A relative interaction frequency (IF) was calculated by dividing the band intensity of the sample templates by that of the BAC control.
Sequencing of the 8.5kb Mcs5c looping region (TCE; chr5:84,428,694–84,437,192; RGSC 5.0/rn5) identified in 3C experiments was performed on MEC DNA from Mcs5c resistant and susceptible rats to assess polymorphisms between the WKy and WF alleles. Sequencing primers are listed in S7 Table. Traditional Sanger sequencing was performed at the University of Wisconsin–Madison Biotechnology Center DNA sequencing facility as described in Smits et al. [30].
The rat mammary carcinoma cell line, LA7, was obtained from the American Type Culture Collection and maintained in DMEM/F12 supplemented with 100 IU/mL penicillin, 100 μg/mL streptomycin (Life Technologies), 5% FBS (HyClone), and 0.005mg/mL insulin. Gene expression analysis proceeded as described above, and copy number analysis was performed via SYBR Green qPCR (Life Technologies). For 5-aza-2’-deoxycytidine (5-aza-dC; Sigma) experiments, cells were treated with 1μM 5-aza-dC for 48hrs followed by cell collection and processing. For quantification of Mcs5c TCE copies, a primer set located within the CRISPR targeted region was used (5’ CAATCACGTTCACTGTGGGT 3’ and 5’ TCACCTCACACTACCCCATG 3’) and as a control region, a primer set located within the non-targeted Pappa gene was used (5’ TCCTCCTGACCACTCTGAGA 3’ and 5’ CCCTACAAACAGCAGAGGGA 3’). The CRISPR-Cas9 plasmid pSpCas9(BB)-2A-Puro (PX459) was provided by Dr. Feng Zhang (Addgene plasmid #48139) [49]. Guide sequences were designed using the CRISPR Design Tool (http://crispr.mit.edu) and phosphorylated and annealed guide oligos were inserted into the PX459 plasmid via a combination digestion/ligation reaction. 100ng PX459 plasmid was mixed with 2μL of oligos (diluted 1:250), 1μL Fast Digest BbsI (Thermo Scientific), 1X Fast Digest Buffer, 1mM DTT, 1mM ATP, and 1500 units T7 ligase (New England BioLabs) and incubated in a thermocycler for 5 minutes at 37°C followed by 5 minutes at 23°C for a total of 6 cycles. The resulting reaction was then treated with Exonuclease V (NEB) according to the manufacturer’s protocol. The product was transformed into competent cells, and colonies expanded and verified via sequencing of the guide insertion site. LA7 cells were transfected with two CRISPR guides flanking the 8.5kb target region. Transfection was performed via electroporation using a Nucleofector II Device and Amaxa Cell Line Nucleofector Kit V (Lonza), according to the manufacturer’s instructions. Stable clones were isolated following puromycin selection, and clonal colonies were expanded. Removal of the targeted region was determined via PCR screening and sequencing. Primers used to create guides and screen clones are listed in S8 Table.
DNA was isolated from wild-type LA7 and CRISPR edited cells as well as Mcs5c resistant and susceptible MECs at 4, 6, 7, 9, and 12 weeks of age. Bisulfite conversion was carried out on 500ng of DNA using the EZ DNA Methylation-Lightening kit (Zymo Research), according to the manufacturer’s instructions. Four primer sets were designed to amplify the 12 CpG sites of interest within the Pappa CpG island (CGI) shore. Their sequences, along with the sequencing primers used for pyrosequencing, are listed in S9 Table. Optimal amounts of template DNA, MgCl2, primers, and annealing temperature were experimentally determined (S9 Table). In general, PCR reactions were performed using the following cycling conditions: 95°C for 5 min, 50 cycles of 95°C for 15 s, Ta for 30 s, 72°C for 30 s, followed by a final extension of 72°C for 5 min. 15μL of PCR product was used for pyrosequencing on a PyroMark MD instrument (Qiagen), with 2–3 technical replicates per sample. Data were analyzed using PyroMark CpG software (v 1.0; Qiagen). For analysis of the Pappa CGI, 2 pre-made assays were obtained from Qiagen (CGI-1, Rn_D3ZNQ7_01_PM; CGI-2, Rn_D3ZNQ7_02_PM), with PCR conditions following manufacturer’s recommendations. Both pre-made assay CGI-1 and CGI-2 amplified 5 CpG sites within the Pappa CGI, for a total of 10 sites in the island examined. For statistical analysis of methylation differences, the non-parametric Mann-Whitney U test was used, with a Bonferroni correction applied for multiple comparisons.
|
10.1371/journal.pntd.0002806 | Methods to Determine the Transcriptomes of Trypanosomes in Mixtures with Mammalian Cells: The Effects of Parasite Purification and Selective cDNA Amplification | Patterns of gene expression in cultured Trypanosoma brucei bloodstream and procyclic forms have been extensively characterized, and some comparisons have been made with trypanosomes grown to high parasitaemias in laboratory rodents. We do not know, however, to what extent these transcriptomes resemble those in infected Tsetse flies - or in humans or cattle, where parasitaemias are substantially lower. For clinical and field samples it is difficult to characterize parasite gene expression because of the large excess of host cell RNA. We have here examined two potential solutions to this problem for bloodstream form trypanosomes, assaying transcriptomes by high throughput cDNA sequencing (RNASeq). We first purified the parasites from blood of infected rats. We found that a red blood cell lysis procedure affected the transcriptome substantially more than purification using a DEAE cellulose column, but that too introduced significant distortions and variability. As an alternative, we specifically amplified parasite sequences from a mixture containing a 1000-fold excess of human RNA. We first purified polyadenylated RNA, then made trypanosome-specific cDNA by priming with a spliced leader primer. Finally, the cDNA was amplified using nested primers. The amplification procedure was able to produce samples in which 20% of sequence reads mapped to the trypanosome genome. Synthesis of the second cDNA strand with a spliced leader primer, followed by amplification, is sufficiently reproducible to allow comparison of different samples so long as they are all treated in the same way. However, SL priming distorted the abundances of the cDNA products and definitely cannot be used, by itself, to measure absolute mRNA levels. The amplification method might be suitable for clinical samples with low parasitaemias, and could also be adapted for other Kinetoplastids and to samples from infected vectors.
| Most experiments on African trypanosomes - including those designed to look for new drugs - have studied parasites either from culture, or from laboratory rodents. We are interested in comparing these parasites that grow in man and domestic animals, where the parasites generally have different nutrient concentrations and much lower parasitaemias than in experimental models. The most accessible way to make the comparison is to measure the amounts of mRNAs. In this paper we describe how methods that are used to purify the parasites from human cells can change the relative amounts of mRNA. We also describe a method to examine RNA from relatively small numbers of parasites that are mixed with host cells.
| African trypanosomes live in several niches - in the blood, tissue fluids and brain of patients with second stage sleeping sickness, and in various parts of the Tsetse fly digestive tract. Nearly all experiments on African trypanosomes, however - including those designed to look for new drugs - have studied parasites either from culture, or from laboratory rodents with high parasitaemias. Drugs have to kill parasites that are living in humans or ruminants. These parasites are very difficult to study because parasitaemias are low, so we have no idea whether their metabolism is really the same as that of parasites in culture. The ideal way to assess differences between these parasites and the standard lab models would be to characterise their proteomes or metabolomes, but the numbers of parasites are insufficient. The amounts of RNA are however sufficient, for transcriptome analysis, especially since amplification techniques are available.
The most sensitive method to characterise transcriptomes is to make cDNA by reverse transcription, followed by second strand synthesis and analysis of the products by high throughput sequencing [1]. To ensure coverage that is independent of the transcript length, the mRNA is fragmented prior to cDNA synthesis [2]. To create a good library for sequencing, 50–100 ng of mRNA is generally needed. When less is available, the cDNA can be amplified. One approach is to include tags at the ends of the cDNA, and use those sequences for polymerase chain reaction amplification or linear amplification [3], [4], [5]. An alternative is to include a promoter for a bacteriophage polymerase in the cDNA primer [3], [5], [6]. Once double-stranded cDNA is made, it has the promoter at one end, and this can be used as a template for further RNA synthesis by the bacteriophage polymerase. The RNA produced in this way is once again converted to cDNA, which is then sequenced. The various amplification techniques can be qualitatively very good for transcript detection, but when tested quantitatively, they were found to cause considerable distortions in measured transcript levels [3], [5].
Several groups have obtained trypanosome transcriptomes using RNASeq [7], including quantitation of full-length developmentally regulated mRNAs [8]. Trypanosome mRNAs are very unusual because each bears, at the 5′-end, a 39 nt sequence called the spliced leader. To map the splice sites, the spliced leader sequence can be used as a primer for second strand synthesis on cDNA. This “spliced leader tagging” approach has been used both to map splice sites [9], [10], [11] and to measure changes in splice site usage in different life cycle stages [9], [10]. In addition, it has been used to assess the effects of a gene knock-out in Trypanosoma cruzi [12].
The study of the transcriptomes of trypanosomes from natural infections is made difficult not only by the low parasite numbers, but also by the fact that the samples contain such an excess of host cells. For all but very high parasitaemias, a purification step will be necessary in order to obtain sufficient parasite sequence reads. One option is to purify the trypanosomes away from the host cells before RNA is prepared. The alternative is to select trypanosome sequences from the RNA mixture before starting the high-throughput sequencing procedure: the spliced leader makes this possible. We here describe an optimized spliced leader priming protocol, and compare the effects of both trypanosome purification and spliced leader priming on the final transcriptome data.
The animal work was approved by the Makerere University animal care committee, clearance number VAB/REC/13/080. Uganda has no law governing experimentation on animals, so the Makerere committee follows EU guidelines.
To test purification methods, approximately 5000 T. b. rhodesiense strain Tbr729 parasites were injected into each rat. Blood was harvested at peak parasitaemia (approximately 4×106 cells/ml) within 5 days post infection.
For the amplification, Trypanosoma brucei Lister 427 were cultured in loosely capped flat-bottomed T-flasks in an incubator at 37°C, 5% CO2, in a humid atmosphere. The cell densities were maintained between 2–10×105 cells/ml in supplemented HMI-9 medium.
To obtain buffy coats, infected rat blood in an EDTA tube was centrifuged at 1200×g for 10 min at room temperature (25°C). (EDTA acts as an anticoagulant.) The buffy coat was pipetted off and centrifuged at 3400×g for 2 min.
For column chromatography [13] 20 ml packed slurry of DEAE cellulose resin (Sigma) pre-equilibrated in phosphate saline glucose (PSG) buffer, pH 8, was set up in a 30 ml syringe. The column was equilibrated with 3 column volumes of PSG and thereafter loaded with the anti-coagulated rat blood (approximately 5 ml) at room temperature. The trypanosomes were eluted with PSG into a 15 ml tube and then centrifuged for 10 min at 850×g at room temperature. The procedure took approximately 30–40 min and this would be similar for standard 5 ml samples of human blood.
Trypanosomes in infected rat blood were also purified by hemolysis followed by centrifugation. Briefly, to one volume of infected rat blood was added 3 volumes of Erythrocyte lysis buffer (Qiagen), then the two were gently mixed and incubated at room temperature (25°C) for 7 min before centrifugation at 850×g for 10 min (total duration approximately 20 min). The content of the Qiagen solution is not known but most common procedures involve the use of isotonic ammonium chloride.
In each case, the trypanosome pellet was immediately resuspended in Trifast reagent and then frozen. Subsequently RNA was extracted according to the manufacturer's instructions. The total RNA concentration was then determined using the Qubit 2.0 fluorometer (Invitrogen).
A detailed step-by-step protocol for the entire procedure is provided in the Supplementary material (Supplementary File S1).
Total RNA was incubated with oligo (dT)-cellulose (Amersham) then the resin was washed before elution of poly(A)+ RNA with water. The RNA was ethanol precipitated and resuspended in water. RNA was then precipitated at −20°C overnight after addition of 30 µl of 5M sodium acetate, 3 µl Glycogen (10 mg/ml), 900 µl of 95% ethanol. The precipitated RNA was then recovered by centrifugation (top speed in the microfuge for 20 min) and resuspended in 6 µl of water.
For all preparations subject to spliced leader priming, cDNA synthesis was done using the Superscript III kit (Invitrogen) according to the manufacturer's instructions. Primers were T3-promoter (5′GCGCGAAATTAACCCTCACTAAAGGGAGA 3′)-tagged oligo-dT (T3(dT)24) and random nanomer (T3N9). The synthesized cDNA was purified using the RNAeasy mini-elute kit (Qiagen).
The second strand cDNA was synthesised using 0.05pM SL-4 (5′GATCTACAGTTTCTGTACTAT3′) primer, Phusion polymerase and buffer (Finnzymes). The ds cDNA was visualized by the incorporation of 25 µCi of alpha (32P)-dCTP in the reaction, resolved on an 8% polyacrylamide-urea gel, exposed to a Phosphor-imager screen and analyzed using the Fuji FLA7000. Unlabeled ds cDNA was either sent for sequencing (T samples) or amplified (TH samples).
The ds cDNA was also analysed by in-vitro transcription using the T3 polymerase reaction mix (MAXIscript, Ambion Inc.). The reaction was carried out using 12 µL of the purified ds cDNA in a mix consisting of 0.5 mM of ATP, CTP, GTP, 50 µCi alpha (32P)-UTP, 1× transcription buffer, T3 enzyme mix in a final volume of 20 µl and incubated at 37°C for 10 min. The reaction mix was then treated with DNase1, 1 µl, and continued incubation at 37°C for a further 15 min. The synthesized RNA was then resolved on an 8% Urea acrylamide gel, exposed to a Phosphor-imager screen and analyzed by the Fuji FLA7000.
Trypanosome cDNA sequences were PCR amplified using a T3-promoter primer (5′GCGCGAAATTAACCCTCACTAAAGGGAGA3′) and a 20mer SL primer (5′ACAGTTTCTGTACTATATTG3′). The cDNA was purified using the QIAquick PCR purification kit (Qiagen).
The cDNA samples were first fragmented with the Covaris S2 system (Covaris), using the AFA microTube at an Intensity 5, 10% Duty Cycles with 200 cycles per burst for 90 seconds. A quality check on 1 µl of the fragmented sample was done on the BioAnalyzer 2100 (Agilent Technologies) using the High Sensitivity chip. The library was then prepared using the NEBNext Ultra DNA Library Prep Kit for Illumina (New England BioLabs Inc.). Libraries were run on either the Illumina GAII or HiSeq 2000 system for single-end 76 bp or 50 bp reads respectively. Library preparation from poly(A)+ RNA was carried out using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England BioLabs Inc.); and similarly run on the HiSeq 2000 system. All the samples were multiplexed.
The sequence reads were aligned to the reference T. brucei brucei TREU 927 genome with Bowtie using the following parameters: –a –best -–strata –v2 –m14 [14]. We extracted all reads that mapped to the annotated mature RNAs (for the amplification experiment) or to coding regions (for the purification experiments). Annotated mature mRNAs in the database do not always include both untranslated regions. For those that did not align, we extracted reads containing sequences of the spliced leader (SL tags) or poly(A) and the T3 promoter, trimmed them, and assigned them to an open reading frame based on their positioning within annotated gene coordinates [15]. The minimum sequence length trimmed was 5 nt. The processing and sorting of the aligned reads was carried out using SAMtools [16] and the read alignment to the genome visualized by Artemis [17], [18].
DESeq was used to identify differentially expressed genes, with a cutoff p-adjusted value of 0.05. Functional category enrichment was carried out using Fisher exact test in R. The heat map was generated in R. Reads per million were calculated using the unique gene list of Siegel et al [8], which excludes all but one gene copy of multigene families. Coding sequence lengths were extracted from TritrypDB. The poly(A)+ mRNA dataset was that from [19], in which reads were mapped to coding regions only.
To plot the average read density along genes, we selected around 1000 short genes of 200–1000b in length, and a similar number of long ones >2.5 kb in length, that had a read coverage of at least 1 on average. PERL scripts were written to extract reads aligning to a window of 0.5% of each gene, which were then normalized to the total read count for the gene, and an average was plotted against the relative location along 1 kb of standardized gene length. Routine calculations were done using Microsoft Excel. Before the calculation of correlation coefficients, the data were log2 transformed. If this is not done, a few very abundant mRNAs excessively influence the result.
To test the effects of trypanosome purification from blood, we compared the two available methods - DEAE cellulose column purification [13] and red cell lysis - with simple centrifugation followed by taking the buffy coat (Table 1). The red cell lysis method does not remove lymphocytes, so would not be suitable at low parasitaemias. The DEAE column does remove a large proportion of the lymphocytes [13], but it takes longer than erythrocyte lysis. Both methods give up to ten times higher trypanosome yields than taking the buffy coat, since many trypanosomes are in the erythrocyte layer after blood centrifugation. However, the buffy coat method does not involve harsh treatment of the parasite, and thus we used it as a reference for the effect of purification methods on the transcriptome.
We infected 9 rats with Trypanosoma rhodesiense and harvested cells at peak parasitaemia. In all cases, to try to mimic optimal field conditions, an individual animal was bled and the cells prepared as rapidly as possible. For three rats, we made RNA from the buffy coat. For a further three, the trypanosomes were purified using DEAE cellulose. For the remaining three, the erythrocytes were lysed. In each case, the cell pellet was resuspended in Trifast reagent. The triplicate samples were then poly(A) selected, randomly sheared and subjected to RNASeq (Table 1 and Supplementary Table S1). Figure 1 shows the reproducibility of the results for each method. For each open reading frame, we calculated the number of mapping reads. We then extracted the data for individual unique open reading frames (ORFs) in order to eliminate multi-copy ORFs, and calculated the numbers of reads in each unique ORF per million total reads (reads per million, RPM). Correlation coefficients were calculated on log-transformed data, since the count distributions are heavily biased towards lower values. Results from the buffy coat samples (BC) showed correlation coefficients of >0.98 for all three pairwise comparisons (Figure 1A, B), whereas both DEAE purification (DE, Figure 1C,D) and red cell lysis (RL, Figure 1E,F) resulted in more variation (coefficients of 0.80–0.94, and 0.94–0.98 respectively). We have not compared the buffy coat results with those for cultured trypanosomes because we do not have results for cultures of this trypanosome strain. Relative standard deviations (the standard deviation divided by the mean) for individual genes did not correlate with either the mRNA abundance or the half-life (Supplementary Figure S1, A–F) suggesting that neither of these parameters was directly linked to the variation seen.
Next, we compared the transcriptome profiles across the methods (Figure 2). For the DEAE-purified cells results correlated well (Figure 2A); no differentially expressed genes were detected between the two methods, partly because they seem to be quite similar, but also because of the high variability between the DEAE replicates. (For criteria used to detect significant differences, see Methods section.) From the slope of the linear regression line (1.3), DEAE purification tended to decrease the abundances of mRNAs that already were present in low copy numbers in the buffy coat parasites.
Purification by red cell lysis (Figure 2B) caused significant changes in the expression of 60% of genes (see Methods for details). Some functional categories were particularly affected: for example, increases in mRNAs encoding ribosomal proteins, chaperones and mitochondrial metabolic proteins, and decreases in mRNAs encoding protein kinases and cytoskeletal proteins (Supplementary Figure S2). We also looked to see whether mRNAs from some genes were “lost” during purification, considering only genes giving at least 10 reads per million from buffy coat trypanosomes. In the DEAE samples, there were around 30 that fell below 10 reads per million, while after red cell lysis the number lost varied from 43 to 204.
We conclude that if purification is necessary, DEAE cellulose is the method of choice, but some variability will be introduced.
For many clinical samples, the numbers of trypanosomes are likely to be too low to allow transcriptome analysis without prior amplification [20], [21], [22], [23], [24](Supplementary Table S2). Even if numbers are adequate, purification may not always be possible under field conditions. We therefore tested methods to amplify parasite mRNA from mixtures of trypanosome and human RNA. To mimic a sample containing 2.5×106 lymphocytes and 104 parasites, we made mixtures containing 5 ng of total T. brucei RNA and 5 µg of HeLa cell RNA (Supplementary Table S2; Figure 3A, B).
We first tested methods for cDNA synthesis. The first strand was made using reverse transcriptase primed with a T3 promoter primer bearing, at the 3′ end, (T)24 or nine random nucleotides (Figure 3C). To make a T. brucei-specific second strand, we tested primers representing various segments of the spliced leader (Figure 3D, Supplementary Figure S3A). Synthesis was measured by including (32P)-dCTP in the reaction, and specificity was tested with a negative control of HeLa RNA alone as template. Initial results showed that it was possible to see specific priming only with purified poly(A)+ trypanosome RNA (not shown). Not unexpectedly, selectivity for trypanosome mRNA was strongly affected by the primer sequence and the annealing temperature. Examples of the results for three primers are shown in Supplementary Figure S3B. When we used a primer that included the entire SL (panel SL), the smear from HeLa RNA alone (lane 3) was the same as that obtained when trypanosome RNA was included (lanes 1 and 2), implying insufficient selectivity for trypanosome mRNA. Using shorter, 20mer SL primers (panels SL20 and SL-2), selectivity was seen: the smears for the samples with trypanosome RNA (lanes 5, 6, 9 and 10) gave stronger signals than those for HeLa RNA alone (lanes 7 and 11). The primer concentration was also critical (Supplementary Figure S3C): for example, using 0.5 pM of primer SL-4, there was no difference with or without trypanosome RNA (lanes 1 & 2) whereas with 0.05 pM primer SL-4, five times more product was obtained if trypanosome RNA was present (lanes 3 & 4).
Next, we tested amplification methods and conditions. We attempted linear amplification using T3 polymerase (for explanation, see Introduction) but yields were poor, so we decided to concentrate on PCR. We tested a variety of combinations of spliced leader primers with a T3 promoter primer, with different annealing temperatures and numbers of PCR cycles (Figure 3E). The most specific result was obtained using a “nested” PCR strategy. After cDNA synthesis, the second strand was primed with the 20mer primer SL-4, which terminates 4 nt upstream of the 3′-end of the spliced leader. With authentic trypanosome mRNAs, the first four nt synthesised will be ATTG - the last 4 nt of the spliced leader. This should be absent in most mis-primed HeLa cDNAs. Amplification was then done with the T3 primer and a 20mer primer (“SL20”) that terminated at the end of the spliced leader. This primer combination will only work if the first four nt after the SL-4 primer really were ATTG, providing an additional level of specificity. PCR results using the optimal primer combination and conditions (see Methods) are illustrated in Supplementary Figure S3D: with this combination, we usually obtained at least twice as much PCR product with the trypanosome-HeLa mix as with HeLa RNA alone.
We now analysed the effects of spliced leader priming and amplification on the measured transcriptome. The cDNAs obtained by spliced leader priming alone, using poly(A)+ RNA as substrate (T in subsequent figures and tables) or using 1000∶1 HeLa mixtures as described above (TH in the figures and tables) was sheared, Illumina adaptors were added, and the resulting fragments amplified and sequenced according to the standard Illumina protocol (Figure 3F–I). In a first experiment duplicate RNA preparations were sequenced using the Illumina GAII sequencer. Later, two more replicates were analysed using the Illumina HiSeq 2000, with duplicate random-shear controls sequenced in the same way. All reads were aligned to the trypanosome genome. Statistics for these experiments are shown in Table 2 and results are in Supplementary Table S3. In the two replicates in first amplification experiment, TH1 and TH2, only about 2%, , of amplified reads aligned to the trypanosome genome (Table 2), which is roughly 20-fold enrichment. In the second experiment (replicates TH3 and TH4), enrichment was 10 times better, with 20% of reads aligning (Table 2). It is possible that the use of HiSeq technology in the second experiment increased the percentage of mapped reads, but variations in the amplification efficiency could also have contributed. We extracted all reads that mapped to the predicted mature RNAs available in TritrypDB (CDS). In addition, we extracted reads containing the spliced leader (SL tags) or poly(A) and the T3 promoter, and assigned them to an open reading frame based on their positions in the genome. Most analyses were done with all reads added together.
Results from technical replicates using spliced leader priming alone, with no amplification, correlated well with each other although as usual, reproducibility was poor for ORFs with low read counts (Figure 4A, B). Spliced leader tagging methods described previously were also reproducible [10]. Although the correlation between the replicates in the first amplification experiment (TH1 and TH2) was reasonable (Figure 4C), the scatter for many genes was unacceptable and over 1700 genes were not represented at all. This is almost certainly due to the very low number of mapped reads (Table 2), and indicates that data should only be used if several million mapped reads are obtained. Results from the second amplification experiment (replicates TH3 and TH4, Figure 4D) correlated well, with less scatter at low RPM values compared to the first experiment because the number of mapped reads was roughly 10-fold higher (Table 2). Results from amplified cDNA correlated to some extent with those from unamplified: Figure 4E shows the results for all four experiments, but the correlation was the same when only samples TH3, TH4, T3 and T4 were considered.
Future discussion will focus mainly on the second experiment (samples T3, T4, TH3 and TH4), because the number of genes with sufficient mapped reads to estimate the mRNA abundance was acceptable for the amplified samples. When randomly sheared mRNA is subjected to RNASeq, the reads should be distributed evenly throughout the open reading frame [2]. We expected spliced leader priming to give 5′ bias [11] even though the cDNA was fragmented before library construction. Figure 5A shows the profiles of read densities along a standardized gene length of 1 kb. For short genes (between 200 bp–1 kb) the reads were generally uniformly distributed with a slight peak at the 5′ end for both amplified and unamplified samples, compared to a randomly sheared sample. Genes of 1–3 kb gave similar results (not shown) but the sample size was much larger than from the short and long genes, which could have concealed many variations. In contrast, there was a big bias towards the 5′ ends for long genes (>2.5 kb); in the case of the un-amplified sample, there was another peak at the 3′ end (Figure 5A), even though random primers, as well as oligo d(T), were used for cDNA priming. We also noticed a double peak at the 5′-ends of long genes, which was not so pronounced for shorter ones. The first peak corresponded to reads with the SL primer and was limited in length to ∼30 nt: this is the read length of 50 nt minus the 20mer SL sequence used for priming. Perhaps fragmentation was not so frequent near the ends and most end fragments carried the SL sequence, hence the drop between the two peaks.
Figure 5B shows alignments for four genes with different patterns. For the highly abundant alpha tubulin mRNA Tb927.1.2340 in the spliced leader primed samples (central panel - T, unamplified), we found a strong peak near the 5′ end but otherwise, fairly uniform reads throughout. After amplification (top panel), distortions arose. In contrast, for Tb927.4.4220, a gene of similar ORF length but with much lower mRNA abundance and longer 5′-UTR, the spliced-leader primed reads were restricted to the 5′-UTR. RNA from this gene would hardly be detected if only CDS reads were analysed. For the tiny ORF Tb927.8.5260, a strong 5′ peak was in the spliced leader primed sample but not the random shear control, and a rather similar pattern was seen for the longer gene Tb927.10.8710.
For randomly sheared mRNAs, the read counts increase with open reading frame length. The mRNA abundance is therefore proportional not to the read counts, but to the read density, which is expressed as reads per million reads per kb of open reading frame (RPKM). As expected, when we calculated the RPKM for the spliced leader primed samples we detected a clear bias against longer transcripts (Supplementary Figure S4 C–F) illustrating what we observed above in the average read density profiles for long genes.
We now quantitatively compared the results from spliced leader primed RNAs with those from classical RNA-Seq of randomly-sheared RNA. The latter is currently the standard method for mRNA quantitation, although it too is likely to have some biases. Since we already knew that our spliced-leader-primed reads were heavily 5′ biased, we first corrected the classical RNASeq data for mRNA length (see Figure 6 legend). In all graphs most mRNAs formed a dense cloud with little apparent correlation, while a small minority was abundant in both datasets (Figure 6). Results for the total counts (Figure 6A, B) correlated rather better than those for SL tags (Figure 6C,D). Correlations for each gene excluding SL and poly(A)/T3 tags were similar (not shown), which is not surprising because there were about six times more gene reads than SL tags (Table 2). Since our amplified and unamplified spliced leader-primed results correlated better with each other (Figure 4F) than with the classical RNASeq, the major cause of the difference between the T and TH samples and randomly sheared pure RNA must have been the spliced leader priming. We found no GC content bias from amplification.
Nilsson et al. have previously used spliced-leader priming followed by RNA Seq to measure mRNAs in purified trypanosomes [10]. Their method differed somewhat from ours. For example, they used a second-strand primer that covered the entire spliced leader; we could not do this because the long primer gave too much background priming on HeLa cDNA (Supplementary Figure S3B). Nevertheless, when they compared their SL tag counts with those obtained from total (not SL-primed) cDNA, the dot plots looked very similar to ours (compare Figure 6E with Supplementary Figure S9 in [10]). This suggests that in their experiments also, SL priming introduced distortions.
We conclude that synthesis of the second cDNA strand with a spliced leader primer, followed by amplification, is sufficiently reproducible to allow comparison of different samples so long as they are all treated in the same way. However, SL priming gives results that differ substantially from those obtained using the standard protocol with randomly sheared mRNA.
To analyse the transcriptomes of field samples, the choice of the most appropriate method will depend on the number of parasites present relative to host cells, and the total number of parasites in the sample. A decision can be made on a sample-to-sample basis if parasitaemias and lymphocyte counts can be checked prior to sampling.
We first assume an experimental scenario with a parasitaemia of 4×105/ml and a lymphocyte count of 4×106/ml. This could arise in T. b. rhodesiense patient samples. An average lymphocyte yields 5 times more mRNA than a trypanosome (see Supplementary Table S3). Use of whole blood gives maximum trypanosome yield with minimal handling. If RNA were prepared from whole blood, and sequenced on a single lane, approximately 1 million reads might map unambiguously to trypanosome coding regions (Supplementary Table S2) (This assumes that globin mRNA is removed prior to library preparation.) 1 million reads is just about the lowest threshold for transcriptome analysis: most coding regions would map around 50 reads [23]. This would also yield the total lymphocyte transcriptome as a by-product, although this will give limited information since it will come from a complex mixture of cell types. For samples with parasite∶lymphocyte ratios of at least 1∶10, the variability that ensues from low read counts has to be weighed against the variability of results, and the decrease in low-abundance mRNAs detected, that are introduced by column purification. An expensive alternative would be to apply the sequencing library to several lanes in order to get more reads.
If parasite∶lymphocyte ratios are lower than 1∶10, either the parasites must be purified, or the parasite mRNA must be specifically amplified. Standard Illumina library preparations work best with at least 50 ng of mRNA sample, which can be obtained from about 2×106 trypanosomes (Supplementary Table S2). If that number is available, column purification is a good option although the results will be less reproducible than for unpurified samples. We already saw variations under ideal lab conditions, but in a field lab purification times might be longer and the temperature could be higher, both of which could have deleterious effects. Samples with fewer parasites could be pooled prior to RNA preparation, which might to some extent compensate for the variability introduced by column purification.
For samples containing less than about 5×105 parasites, amplification of some sort will be unavoidable. We were able to achieve up to 200-fold enrichment of trypanosome sequences from a 1000∶1 HeLa∶trypanosome mixture. This gave ample coverage of the transcriptome starting with only 5 ng of trypanosome RNA. The reads obtained were not directly proportional to mRNA abundance, but were instead strongly influenced by the spliced leader priming and also, to a lesser extent, by the PCR amplification. Despite this, the method was reproducible, especially when 200-fold enrichment was achieved. This indicates that the distortions that were seen must have been due to sequence-specific effects such as blockage of reverse transcriptase by secondary structures, PCR bias towards shorter amplicons, and long 5′-untranslated regions. Because the results from the method are reproducible, data from different samples can be compared. Ideally, all samples should be treated in the same way, with at least 3 replicates per condition; under those conditions relative mRNA levels for each gene could be calculated and differentially expressed genes detected. Alternatively, the results could be converted into approximate “real” mRNA abundances using correction factors. These would need to be measured in individual labs, since they will be affected by the sequencing platform, the precise amplification and library building conditions, and the database version used for the alignments.
Spliced leader priming of mixed human/trypanosome poly(A)+ mRNA has the advantage of starting with quite large RNA amounts, which are easy to handle. The alternative, for samples containing at least 104 parasites, would be to purify the parasites using a column, then make the library using a method or kit designed for picogram amounts of mRNA. There have been relatively few independent evaluations of these amplification protocols for transcriptome quantitation, but all of them cause method-specific distortions. The consensus so far is that if amplification is necessary, the results for different samples can only be compared if the same method is used throughout ([3], [4], [5] and http://openwetware.org/wiki/BioMicroCenter:RNAseq).
Our method for spliced leader priming and amplification of trypanosome mRNA should be suitable for any sample containing at least 104 trypanosomes, and could be particularly useful if conditions do not allow immediate and rapid column chromatography. In principle, samples from T. b. gambiense sleeping sickness patients could be analysed. The same principle could be applied for any other organism in which the mRNAs carry a spliced leader: our results show, though, that careful optimisation of the spliced leader primers, ratios and annealing temperature will be necessary. Once this is achieved, the method might be used to analyse gene expression in samples from Trypanosoma congolense or Trypanosoma vivax infected cattle, for intracellular kinetoplastids, or samples from infected vectors.
|
10.1371/journal.pgen.1002053 | Does Positive Selection Drive Transcription Factor Binding Site
Turnover? A Test with Drosophila Cis-Regulatory Modules | Transcription factor binding site(s) (TFBS) gain and loss (i.e., turnover) is a
well-documented feature of cis-regulatory module (CRM) evolution, yet little
attention has been paid to the evolutionary force(s) driving this turnover
process. The predominant view, motivated by its widespread occurrence,
emphasizes the importance of compensatory mutation and genetic drift. Positive
selection, in contrast, although it has been invoked in specific instances of
adaptive gene expression evolution, has not been considered as a general
alternative to neutral compensatory evolution. In this study we evaluate the two
hypotheses by analyzing patterns of single nucleotide polymorphism in the TFBS
of well-characterized CRM in two closely related Drosophila species,
Drosophila melanogaster and Drosophila
simulans. An important feature of the analysis is classification of
TFBS mutations according to the direction of their predicted effect on binding
affinity, which allows gains and losses to be evaluated independently along the
two phylogenetic lineages. The observed patterns of polymorphism and divergence
are not compatible with neutral evolution for either class of mutations.
Instead, multiple lines of evidence are consistent with contributions of
positive selection to TFBS gain and loss as well as purifying selection in its
maintenance. In discussion, we propose a model to reconcile the finding of
selection driving TFBS turnover with constrained CRM function over long
evolutionary time.
| Transcription factor binding sites (TFBS) turnover (i.e. lineage-specific gain
and loss) is a well-documented phenomenon in eukaryote cis-regulatory modules
(CRM). The wide spread of the phenomenon and the appearance of conserved
expression patterns for diverged orthologous CRM led to the standing view that
the observed gain and loss of TFBS were functionally and selectively neutral. To
the contrary, genome-wide population genetics analyses have unequivocally
identified signatures of positive selection acting in noncoding regions in
general, and particularly in 5′ and 3′ untranscribed regions of
genes. To specifically test the neutral versus selection hypotheses for the TFBS
turnover process, we analyzed natural variation patterns within and between two
closely related Drosophila species. We found the patterns of
divergence and polymorphism for two types of mutations—those inferred to
increase or decrease the binding affinity respectively—are not compatible
with a neutral hypothesis. Instead, multiple lines of evidence suggested that
positive selection has contributed to gain as well as loss of TFBS in the two
lineages, with purifying selection maintaining existing TFBS in the population.
Spacer sequences also showed signatures of negative and positive selection. We
proposed a model of CRM evolution to reconcile the finding of frequent adaptive
changes with constraints on long-term evolution.
| Gene expression in eukaryotes is generally controlled by transcriptional enhancers,
also called cis-regulatory modules (CRM), which are short regions in the genome
consisting of a cluster of transcription factor binding sites (TFBS) spaced by
intervening sequences (spacers). Individual TFBS have been shown repeatedly to be
required for CRM function, yet surprisingly they evolve rapidly and are frequently
gained and lost in evolution, attributes that have been demonstrated for a large
number of CRM and transcription factors [1]–[5]. These observations pose a
challenge to understanding the forces driving the process, especially in cases where
CRM function has been preserved despite sequence and structural divergence [6]–[8].
The gain or loss of a TFBS is unlikely to be functionally irrelevant, as repeatedly
shown in TFBS knockout experiments [9]–[11], and also demonstrated for the evolved differences
between two species by a chimeric enhancer study [12]. One possibility for
reconciling conservation of CRM function with rapid TFBS turnover is to assume that
each loss of a TFBS is precisely balanced by the simultaneous gain of a cognate TFBS
elsewhere in the CRM, a process we will call compensatory evolution [13]. The idea draws
on a model first proposed by Kimura [14], where he considers a pair of tightly linked mutant genes
that are individually deleterious but in combination restore wildtype function. As
applied to TFBS, the gain of a novel site on an allele carrying a mutation that
decreases the quality of an existing binding site can offset the mutants fitness
cost, creating a selectively neutral double-mutant allele. Binding site turnover -
fixation of the double mutant allele - is achieved entirely by genetic drift, thus
preserving both CRM function and population fitness. Recently, a theoretical model
of this compensatory turnover process was developed to ask about the feasibility of
compensatory evolution for TFBS [15]. With plausible assumptions about the mutation rate,
population size and selection coefficient on the individual mutations, a completely
neutral model cannot achieve a high enough level of turnover to explain
Drosophila CRM evolution (as exemplified by
eve stripe 2 enhancer), whereas a model that assumes the double
mutant to be more fit than the wildtype does.
This theoretical finding raises the prospects for positive selection being an
important driving force of TFBS gain and loss. Instances of directional selection
have been documented in cases where a novel regulatory regime is favored [16]. Functional
evolution of a transcription factor (TF) can also drive adaptive co-evolution of its
TFBS [17]–[19]. Broad-scale
studies in noncoding regions and promoters of genes have identified signatures of
both selective constraint and positive selection in fruitfly and human [20]–[24]. However,
only a small number of population genetics studies have been carried out to
specifically test this hypothesis with TFBS or CRM, and because they focus on a
single TF or CRM, they have low statistical power to distinguish between neutrality
and selection [13]. The generality of the conclusions reached in these studies
is also not established [25], [26].
Several different approaches have been designed to detect and quantify selection in
the system. One of them has been to consider the genome-wide ensemble of TFBS as
evolving at mutation-selection balance, with the fitness of each instance of TFBS
being strictly determined by its binding energy [4], [27], [28]. This approach proves useful
in studying the strength of selective constraints on functional TFBS. However, the
assumption of a unidirectional fitness function, i.e. selection always favors
affinity-increasing mutations and against affinity-decreasing ones, could be
violated if the loss of a TFBS were favored or gain (or strengthening) of a TFBS is
deleterious. Another approach calculates the sum of mutational effects in TFBS on
binding affinity and compares it to the expectation under a no-selection model [29]. A higher than
expected sum could imply selective removal of affinity-decreasing mutations and
therefore the action of purifying selection. Applying this approach to two of the
CRM also included in this study, the author provided evidence for purifying
selection acting to preserve the functional TFBS in the anterior
Bicoid-dependent hunchback enhancer and the
even-skipped stripe 2 enhancer. This test can also be used to
detect positive selection, although its power is limited due to the mixed signal
with purifying selection, which is expected to be dominant in most cases.
In this study, patterns of polymorphism and divergence are investigated in a pair of
closely related Drosophila species, D.
melanogaster (mel) and D. simulans
(sim). The short evolutionary distance between the two species
ensures unambiguous alignment for noncoding sequences and also allows one to capture
the potentially rapid dynamics of TFBS gain and loss. A notable challenge in
studying TFBS turnover is assembling a high quality set of TFBS that are precisely
defined and contain few false positives. Large numbers of potential TFBS can be
identified by methods involving genome-wide scans, such as computational prediction
or ChIP, but these approaches generally include a large fraction of false positives,
thus reducing their attractiveness for investigating the mechanisms of binding site
turnover (see Discussion). Instead, we chose to
investigate a curated set of high-confidence TFBS identified by DNaseI footprint in
well-studied D. melanogaster CRM. Short footprint regions usually
contain only a single TFBS motif, which, in most cases, could be perfectly aligned
with the other species to allow identification of single nucleotide differences
within and between the species. Each of these differences, in turn, was evaluated
for the predicted magnitude and direction of effect on TF binding energy. The
neutral and selection models generate distinguishable predictions in both divergence
to polymorphism ratios and in the site frequency spectra. Analysis of these patterns
reveal evidence for purifying selection against affinity-decreasing mutations
segregating in the population, while multiple lines of evidence indicate positive
selection for both gains and losses of TFBS. These empirical findings challenge the
prevailing view of neutral compensatory turnover, and have important implications
for understanding CRM functional evolution. In the course of the analysis, we also
identified and modeled a potential ascertainment that can impact population genetics
studies of genomic features that have been identified only in a reference sequence
such as TFBS.
Our analysis focuses on single nucleotide polymorphism (SNP) and divergence in 645
experimentally identified TFBS for 30 transcription factors in 118 autosomal CRM
(Table
S1), all annotated in REDfly [30]. These 645 TFBS represent the complete set for which we
could obtain unambiguous alignment of both within- and between-species sequences
without insertion or deletion. We used position
weight matrices (PWM) both to
identify TFBS within footprints and to predict the magnitude of binding energy
differences among variant alleles. Our bioinformatic and experimental validations
showed that the PWM used in this study provide reliable and unbiased estimates for
the direction of binding affinity change in both mel and
sim (Materials and
Methods).
Single nucleotide changes within or between mel and
sim were polarized with outgroup sequences from D.
sechellia, D. yakuba and D. erecta
using PAML (Materials and Methods). Each
derived mutation, therefore, could be categorized with respect to species lineage
and to direction of binding affinity change.
Binding sites for an individual TF or a single CRM usually had too few counts of
single nucleotide polymorphism or fixed differences to allow informative
statistical analysis. Furthermore, the breadth of the turnover phenomenon across
almost all investigated TF and CRM suggests a common underlying evolutionary
mechanism [5],
[7], [8], [18], [31]. We
therefore considered pooling observations from across TFs and CRM. To see if the
evolutionary rates in different TFs binding sites are sufficiently uniform, we
measured sequence divergence between mel and
sim for the 30 TF. After accounting for sample sizes, no
significant departure from the average rate is detected by a binomial test
(Figure 1). Moreover,
the pooling approach should be conservative in deriving a general pattern with
respect to among TF variations.
We then estimated percent loss and gain of TFBS on the mel and
sim lineages. For each of the 645 footprint TFBS, a PWM
score was calculated for each occurrence
() in the alignment of mel,
sim and the inferred mel-sim ancestor, by
taking the log2 ratio of the probability of a sequence under the functional
motif distribution versus that under the genomic background distribution (Material and Methods). Using
as a cutoff, approximately 2% of all footprint
sites were found to be present in mel only and may represent
mel specific gains; and about 2.5% were present in
the inferred ancestor (and mel) but lost in
sim. A set of empirical cutoffs were determined for each TF
based on the range of PWM scores among its footprint sites, which produced
similar results (Table S2). Consistent with the sequence divergence patterns, gain
and loss of TFBS appear to be a general pattern across TF and CRM. A total
turnover rate of 4.5% between mel and
sim is similar to a previous finding of 5% for a
single TF Zeste [5].
We observed approximately equal numbers of gains versus losses in our dataset,
although the distribution of these events is asymmetric on the two lineages (16
losses, 0 gain along the sim lineage versus 12 gains, 0 losses
along the mel lineage). This is not unexpected, given that all
footprint TFBS were identified as being present in mel and the
dataset doesn't include sim-specific TFBS. We predicted
that identification of TFBS by computational methods would produce a more even
pattern of gains and losses in both lineages. We tested this prediction for
three TF (Hb,Bcd,Kr) using a stringent cutoff procedure and for each TF we found
a similar total number of predicted binding sites in the two lineages (Text S1;
Figure
S1). We thus rejected the (unlikely) possibility that there has been
a large-scale evolutionary gain of TFBS in mel and loss in
sim.
Gain and loss of TFBS may be subject to distinct evolutionary forces. To
investigate them separately, we assigned each mutation within a footprint TFBS
in mel or sim to either affinity-increasing or
affinity-decreasing group based on PWM score difference between the ancestral
and the derived mutation (Materials and
Methods). Bioinformatic and experimental investigation showed that
this PWM-based procedure for inferring the direction of binding affinity change
is reliable when PWM predicted magnitude of change is not too small (Materials and Methods, Figure S2
and Figure
S3). We established a threshold corresponding to a PWM score
difference of one, i.e. at least two-fold change in the likelihood ratio between
a motif or background distribution, in order to minimize the chance for
mis-assignment. Varying this threshold between zero and two do not affect the
results qualitatively.
We employed two approaches to investigate evolutionary forces acting on affinity
increasing and decreasing changes. One approach is based on contrasting
polymorphism and divergence patterns in a McDonald-Kreitman (MK) test framework
[32].
Positive selection is expected to inflate substitution relative to polymorphism
while negative selection will have the reverse but weaker effect [33]. We used
synonymous changes in the target genes for the CRM as a proxy for a neutrally
evolving class. Following established practices, we further classified each
synonymous change as according to its expected impact on codon bias –
No-Change, Preferred-to-Unpreferred, or Unpreferred-to-Preferred – and
used the No-Change class as the neutral reference. The second approach
investigates the site frequency spectrum of TFBS polymorphism to make inferences
about selective pressures acting more recently on binding sites.
The fact that all footprints were identified in mel impacts the
analysis in two ways. First, gains of TFBS can be observed in
mel but not losses, while the reverse is true in
sim. Therefore, even though similar processes are most
likely operating in both species, our evolutionary analysis of binding site gain
will focus on changes in the mel lineage, whereas losses will
be restricted to changes in the sim lineage.
Second, affinity-decreasing and affinity-increasing mutations have the potential
to differ in detectability as a footprint site in mel. This
arises because mutations in TFBS were sampled conditioned on the TFBS being
detected in mel and affinity-changing mutations in
mel, in turn, have the potential to affect the
detectability of the TFBS. Depending on whether the derived mutation is
affinity-increasing or affinity-decreasing, two distinct biases are introduced
in the expected neutral frequency spectrum (Figure S4).
Given that the dataset consists only of TFBS that are detectable by
footprinting, we assume that the high-affinity allele will always be detectable.
Consider the possible situation in which the low-affinity allele is not
detectable as a footprint: if the derived mutation is affinity-decreasing, the
probability of detecting the TFBS will change inversely with the mutant allele
frequency; conversely, if the derived mutation is affinity-increasing, the
probability of detection will increase with the mutant allele frequency.
Substitutions may be viewed as a special instance of a segregating mutation and
treated similarly.
This effect of ascertainment on neutral expectations for the MK test and the site
frequency spectrum can be modeled analytically (Text S2);
there is no ascertainment if both alleles are equally detectable as footprints.
To incorporate uncertainty in the detectability of the low-affinity allele, the
model incorporates a parameter, f, which specifies the
probability that the weaker affinity allele will not be detected in the
footprint assay. While f is likely to be greater than 0, it is
unlikely to be close to 1 because footprint sites are degenerate and span a
range of affinities. Under the conservative assumption that the lowest affinity
among the footprint sites is the detection limit, we estimate
for the 30 TF (Text S2), indicating that the majority of
TFBS changes will be detectable.
In the following sections, we first present our analysis of polymorphism and
divergence in mel, focusing on the forces acting to either
maintain functional TFBS or to create new ones. We then turn to
sim, focusing on TFBS loss. Finally, we analyze the spacer
sequences between TFBS in both species.
For each class of change we summarized the data in the MK table by calculating
the ratio, . The presence of weakly deleterious mutations can mask
signatures of positive selection, and if removed can improve the power of the
test [34]. Since
most deleterious mutations will be at low frequencies, using 15% as a
frequency cutoff has been shown to achieve most of the benefits of a more
sophisticated model incorporating the distribution of deleterious effects [35]. We
applied this cutoff and denote the ratio of substitutions to common polymorphism
by . Under this procedure, is significantly
higher for nonsynonymous changes than for the synonymous No-Change class (Figure 2A), consistent with
previous findings of positive selection driving amino acid substitutions in
Drosophila
[36].
To delineate the effect of ascertainment from that of selection for the
affinity-increasing and affinity-decreasing mutations, we compared the observed
to the expected neutral ratios under the ascertainment
with different values (Text S2).
For affinity-decreasing mutations in mel, the difference from
the synonymous No-Change class is not statistically significant, even in the
absence of ascertainment bias (Figure 3A green, Figure
2A). This seems to suggest only neutral or deleterious mutations are
present for this class and therefore no positive selection is involved. The
validity of this conclusion can be questioned, however, because any affinity
decreasing substitutions in mel that led to the loss of a site
will not be included in the data while our correction for the ascertainment only
accounts for neutral changes but not a potential adaptive excess. Thus,
rejection of the neutral model in favor of positive selection is not possible
for affinity-decreasing mutations in the mel lineage. However,
this test is possible for the sim lineage (reported in the next
section), where the loss of a TFBS is observable.
For affinity-increasing mutations no amount of ascertainment under our model can
account for the observed relative excess of substitutions (Figure 3A red). We further reasoned that the
ascertainment effect should be weaker or non-existent for TFBS with an
ancestrally strong binding affinity, which would be identified with or without
the affinity-increasing mutations. We therefore investigated whether the excess
of affinity-increasing substitutions differed if TFBS changes were grouped
according to the strength of the inferred ancestral binding affinity. We found a
consistently larger ratio, i.e. an
excess of substitutions, across the entire range of inferred ancestral binding
affinity classes compared to the No-Change class, including binding sites with
the strongest ancestral binding affinity (Figure 3B). These results collectively
suggested that positive selection has contributed to the fixation of
affinity-increasing changes.
To further investigate evolutionary forces acting on the segregating mutations in
TFBS in the population, we utilized the site frequency spectrum, for which we
generated the neutral expectations for affinity-increasing and
affinity-decreasing mutations separately under ascertainment, with
or (corresponding to
no bias or complete bias, respectively). For affinity-decreasing mutations, with
the ascertainment expected to shift the frequency spectrum to lower frequency
classes (Figure 4A, blue
versus grey bar), the observed spectrum is shifted in that direction but is even
more extremely so than the complete bias expectation (Figure 4A, orange versus blue). Since
is clearly an overestimate (compared to our estimate of
), this strongly suggests that forces other than
ascertainment must have shaped this pattern. Both a recent selective sweep and
population growth can produce an excess of rare variants and one or both
mechanisms may be acting in this system, as is suggested by our finding that
synonymous changes also show a relative excess of low frequency mutations (Figure
S5B). However, as we compared the site frequency spectrum of the
affinity-decreasing mutations to that of synonymous sites (corrected for
ascertainment), we found the former is again more significantly shifted than the
latter (Figure
S6). Thus we suggest that the observed frequency spectrum is
consistent with on-going purifying selection against affinity decrease in
functional TFBS. The observed frequency spectrum for affinity-increasing
mutations lies between the two expectations and the differences are not
significant from either one, a possible consequence of the small sample size (15
observed affinity-increasing polymorphisms) (Figure 4B). Thus, while positive selection is
indicated on the basis of the MK test, inference cannot be made about on-going
selection for affinity-increasing mutations.
Patterns of polymorphism and divergence in sim are not
influenced by the ascertainment because the identification of TFBS in
mel is independent of the effect of mutations fixed or
segregating in sim. However, the inclusion of binding sites
gained in mel may confound the analysis as their orthologous
sequences in sim may have evolved under less or different kinds
of selective constraints. We thus restricted the analysis to footprint TFBS
predicted to be present in the mel-sim common ancestor, where
we found a significant excess of substitutions for the affinity-decreasing
mutations compared to the synonymous No-Change class (Figure 2B, Fisher's Exact Test
). Statistical significance of this pattern is robust to
the cutoff for excluding binding sites gained in mel (Table S3).
A relative excess of substitutions might also be a consequence of factors other
than selection, such as systematic differences in the genealogical histories of
CRM versus synonymous sites. However, these factors seem unlikely to be the
cause of this type of departure from neutrality in these two species (Kohn and
Wu 2004). Therefore we consider positive selection a more plausible
explanation.
We also compared the ratio between affinity-decreasing and affinity-increasing
mutations in polymorphism to the expected ratio of the two classes in the
mutational input, i.e. the probability for a new mutation to be one of the two
classes (Materials and Methods). Briefly,
the expected ratio was obtained by considering all possible mutations in each of
the 645 footprint TFBS and their predicted effects on binding affinity the same
way as we did before. Assuming polymorphism for both classes were neutral, we
expected similar ratios, whereas the observed results showed a significant
deficit of affinity-decreasing polymorphism relative to affinity-increasing
polymorphism (Table 1),
which may suggest that among new mutations, affinity-decreasing ones are more
likely to be deleterious, a result consistent with our finding based on
frequency spectrum in mel. A similar approach has been applied
before, using the sum of (individual
mutation's effect on binding affinity predicted by PWM) within a CRM
instead of counts of mutations in binary classes [29]. There the author also found
evidence for purifying selection against affinity-decreasing mutations. The
finding of both on-going purifying selection and potentially positive selection
acting is not dissimilar to patterns found in nonsynonymous changes [36]. We reserve
for the Discussion section the attempt to
reconcile the adaptive loss of TFBS, as observed between the two species, with
on-going purifying selection against affinity-decreasing new mutations.
In both mel and sim we found a significant
excess of substitutions in spacer sequences, indicative of positive selection in
these intervals (Figure 2).
Also, the frequency spectrum for this class is strongly shifted towards lower
frequencies (Figure S5E, Tajima's D = −1.09),
indicative of on-going purifying selection. The implication of these results is
that spacer sequences might contain many unidentified functional elements, for
example, TFBS for known or uncharacterized transcription factors, or perhaps
other structural features not yet understood.
To summarize, analysis of TFBS changes in mel indicates on-going
purifying selection against affinity-decreasing polymorphism in the population,
and positive selection for affinity-increasing substitutions. In
sim, the analysis of affinity-decreasing changes indicates
a significant, and potentially adaptive excess of substitutions that contributes
to binding site loss. Spacer sequences between footprint TFBS in these
well-characterized CRM also exhibit patterns of polymorphism and divergence
consistent with both functional constraint and adaptive evolution.
Natural selection, both positive and negative, has been shown to act throughout
noncoding regions of the Drosophila genome [21], [22], albeit with varying
intensities [23].
Against this backdrop of ubiquitous selection in noncoding DNA, should it be
surprising to find signatures of positive selection in Drosophila
TFBS? We think not. More surprising perhaps is the incompatibility of this finding
with the model of neutral compensatory binding site turnover, a simple and appealing
mechanism that allows for both rapid binding site turnover and functional stasis of
CRM activity. But as explained below, there are good reasons to doubt whether a
strictly neutral compensatory process can actually generate rapid TFBS turnover in
Drosophila, even with its favorably large population size.
Positive selection, in contrast, can drive arbitrarily fast rates of binding site
turnover; the question is whether it can also allow for functional stasis of CRM
activity. Below, we first discuss the strengths and limits of our analysis and then
we describe properties of gene regulatory networks that can promote adaptive binding
site turnover and yet also constrain the function of CRM.
One challenge in investigating cis-evolution is the proper alignment of noncoding
sequences. To minimize this potential problem, we specifically selected a pair of
closely related sibling species, D. melanogaster and D.
simulans for investigation. Sequence divergence between the two species
in noncoding regions ranges only between 5% and 8% [37], which
allowed us to accurately identify single nucleotide differences from unambiguous
alignments of binding sites (those with alignment gaps were excluded from the
analysis). Working with closely related sequences also provided accurate inference
of ancestral states, and thus the direction of mutational change along the
phylogeny, as well as minimized trans-cis co-evolution.
Independently, Bradley et al also recommended mel
and sim for measuring binding site divergence based on these same
issues arising in their analysis of divergence between two more distantly related
species [31].
Another challenge in studying TFBS turnover is the establishment of a TFBS dataset
consisting of biologically functional sites, a difficult task due to both the high
false positive rate in binding site prediction (even in ChIP bound regions) and the
difficulty in validating the biological functionality of individual binding sites.
While many genome-wide datasets for TFBS are becoming available, several properties
of the Drosophila DNase I footprint dataset made it the one of
choice for use in this study. First, the in vitro footprint
experiments were applied not to anonymous noncoding regions but rather to specific
sequences that had been identified with in vivo reporter assays as
containing a CRM. Furthermore, the transcription factors assayed for each CRM were
also chosen based on prior genetic evidence for their involvement in the regulation
of the CRM. For both of these reasons, subsequent experimental analysis of
Drosophila footprint sites has invariably validated their
functionality [38]–[43]. This experimental sampling of footprint site
functionality is unique among available TFBS datasets, and provides evidence for a
low false positive rate. In contrast, a recent attempt to combine known CRM, ChIP
bound regions, and PWM prediction to obtain a genome-wide TFBS dataset estimated
false positive rate [4]. Although the footprint sites were
identified in lab strains particular to each individual experiment, we provided
reasonings and evidence why the annotation is applicable to natural populations
(Text
S3). In particular, we constructed phylogenetic trees based on the genomic
sequences containing the CRM we studied for natural population lines as well as a
representative lab strain (the genome sequence reference strain), which shows that
the later is indistinguishable from the rest (Figure S8).
This also suggests the lab strains were not genetically divergent from the natural
population.
Genome-wide studies have identified signals of both positive and negative selection
in noncoding sequences in Drosophila, but not the biological or
functional basis for this selection. In this study, we distinguished mutations in
the footprint sites by their functional impact – either increase or decrease
the binding affinity of the corresponding TF – and observed different patterns
of polymorphism and divergence between the two classes. For example, we found that
affinity-decreasing mutations are on average more deleterious among new mutations
than affinity-increasing ones, as revealed by a comparison of the ratio between the
two classes in polymorphism with the expectation from mutational input. Such
distinctions were not observed when mutations were grouped in other ways irrelevant
to the function of TFBS (for example, mutations in the first half of the motif
versus the second half). For these reasons we think the evidence supports our
specific model of selection acting on binding site gain and loss as opposed to an
unidentified functionality in noncoding sequences in general. The mechanism of
selection we described here for well-annotated TFBS could in principle be acting
more broadly across noncoding regions inasmuch as noncoding DNA is often associated
with proteins binding.
Our ability to correctly categorize mutations into affinity-increasing or
affinity-decreasing categories hinges on the accuracy of PWM predicted affinity
differences. To investigate this issue, we employed a state-of-the-art microfluidics
technique, MITOMI [44], to experimentally measure the binding affinity
differences for naturally occurring mutations in hunchback and
bcd binding sites. To our knowledge, this is the first time
that accurate measurements have been made on population-level variants in TFBS. We
found that PWM scores correctly predicted the measured direction of affinity change
for 21/25 mutations investigated. Of the four mutations that PWM predicted the wrong
direction, three have effect sizes predicted to be close to zero. The PWM-based
procedure, therefore, may not be accurate for small predicted differences in binding
affinity. Taking these results into consideration, we employed a binary
classification of mutations with PWM differences exceeding a threshold requirement
rather than using quantitative predictions of all PWM score differences as a basis
for our analysis.
Another potential issue concerns applying mel derived PWM to score
sim TFBS binding affinity. Transcription factor protein
evolution between the two species, if it occurred, could lead to underestimation of
binding affinity in sim, although the effect should be similarly
applied to both substitutions and polymorphism and thus is not expected to cause a
relative excess of the former as observed in the sim data.
Nevertheless, we show two lines of arguments that suggest this is not the case in
our study: first, for the 30 TF whose binding sites we investigated, the DNA
bindings domains and other functionally annotated domains are completely conserved
except for one biochemically conservative amino acid difference (Asp/Glu) in
Dorsals RHD domain (Table S4). Although differences exist in other
parts of the proteins, it has been shown that DNA binding domain may singly
determine the sequence specificity of the protein [44], [45]. Second, if what we identified
as affinity-decreasing mutations in sim reflected on-going
adaptations to a slightly different motif, we would expect, but did not find, a
consistent pattern in the position and kind of nucleotide changes for a TF (data not
shown). To further support this argument, we derived PWM using MEME from the
mel footprint sites as well as their aligned sequences in
sim. As shown in Figure S7, our classification of binding site
differences did not differ between using either the mel PWM or the
sim PWM, contrary to what would be expected if TF sequence
specificity had evolved between the two species. Therefore we consider it very
unlikely for the 30 TF included in this study to have undergone significant
evolution in their sequence specificity. In addition, because the SELEX derived PWM
produce consistent results with the footprint derived ones (Figure S3), we
can also rule out the possibility of over-optimization of the PWM inducing a
sequence preference for mel over sim.
Finally, in the course of the analysis, we identified and modeled an ascertainment
bias caused by the identification of footprint sites exclusively in a single strain
of mel, and the possibility that sequence changes in the same
species can lead to creation or destruction of the footprint feature (as described
in the Results section). Many other genomic
features such as miRNA binding sites and recombination hotspots can also satisfy
these two criteria. As new studies attempt comparative evolutionary studies of
genomic features often identified in a single reference sequence, we expect this
problem to become more common and, therefore, to require greater attention. If not
properly accounted for, this form of ascertainment can lead to false rejection of
the neutral hypothesis. The analytical model of ascertainment under neutrality we
developed here should be applicable to population genetic and evolutionary analysis
of many different structural features of genomes.
Our population genetics analysis identified three major forces in TFBS evolution.
First, we found functional TFBS were selectively maintained in the population by
purifying selection, as revealed by a frequency spectrum skewed towards rare
variants for affinity-decreasing polymorphism in mel and a
significantly reduced proportion of affinity-decreasing polymorphism compared to
mutational input in sim. These results are consistent with previous
findings of selective constraints on functional TFBS. Mustonen and Lässig
estimated that the average selection coefficient to maintain TFBS in bacteria and
yeast genomes are on the order of
[28], [46], and a
similar estimate has been obtained for Drosophila
[4]. The substitution
rate with is expected to be less than 0.05% of the neutral rate
in a population with a size as large as Drosophila (Equation
B6.4.1, [47]). This means TFBS loss is unlikely to happen through fixation
of deleterious mutations (0.2 losses expected for 645 footprint TFBS versus 16
inferred in sim). We can think of only three mechanisms by which
TFBS loss can occur at an appreciable rate: (1) there is loss of constraint; (2) a
pair of tightly linked compensatory mutations creates an effectively neutral allele;
or 3) positive selection drives the loss of TFBS. Our second finding – a
significant excess of substitutions compared to the neutral class for
affinity-decreasing mutations in sim – is consistent only
with positive selection for TFBS loss. Occasional adaptive loss of a TFBS is not
inconsistent with more ubiquitous selection to maintain binding sites [28], and has been
suggested to account for the evolution of fermentation pathways in yeast [16].
Our third finding is positive selection contributing to the gain of TFBS, as revealed
by a significant excess of substitutions for affinity-increasing mutations in
mel. Collectively, the three findings indicate that natural
selection is extensively involved in the maintenance, gain, and loss of TFBS. This
conclusion challenges the prevailing view of a neutral TFBS turnover process [4], [13].
We think that a selectionist interpretation of the turnover process is plausible for
several reasons. First, the assumption of CRM functional stasis, which is the main
argument for the neutral (i.e., compensatory) view, is not well supported
experimentally. Reporter transgene assays, in particular, are limited in their
quantitative resolution, and yet even in these studies, repeatable differences were
found between orthologous CRM [7]. A functional rescue experiment is potentially more
sensitive than a reporter transgene assay. As applied to the Drosophila
even-skipped stripe 2 enhancer, it demonstrated clear functional
differences between CRM that were previously believed to have the same spatial
pattern of expression [48].
Second, compensatory neutral evolution cannot account for the patterns of variation
observed in this study. According to this model, affinity-decreasing mutations
should in general be deleterious but occasionally become “effectively”
neutral when a second compensatory mutation occurs in the CRM of the mutant allele.
A mixture of deleterious and compensatory mutations, even if the latter is common,
may bring patterns of polymorphism and divergence close to a neutral scenario, but
cannot produce a signature of positive selection as observed for both classes of
mutations in our analysis. In addition, analytical modeling of the compensatory
evolution of TFBS finds that the waiting time for a turnover event is long if
complete neutrality of the compensating mutations is assumed [15]. To shorten the waiting time
to be compatible with the Drosophila TFBS turnover rate, the
parameterization of the model requires that the double mutant allele have higher
fitness than the non-mutant allele, making it a directional selection model. This
supercompensatory scenario could produce signatures of positive selection both for
binding site gain and loss, the latter occurring because the fixation of a
deleterious mutation in an existing TFBS will have the appearance of being
positively selected as it hitchhikes to fixation on the selectively favored allele.
However, this scenario is biologically unrealistic, as it requires the second
mutation (the gain of a TFBS) to be positively selected only on the background of
the first mutation.
As an alternative, consider the following model of positive selection on CRM
structure/function. We propose that for CRM with large numbers of interacting
partners, the network of cis- and trans-factors will inevitably be constantly
evolving – due to both direct selective pressures imposed on the CRM or
indirect effects caused by adaptations in other components of the network. For
example, egg length variations between and within Drosophila
species have been studied as potentially adaptive traits; if egg length evolves,
genes such as eve whose expression pattern need to scale with the
embryo may need to change its CRM to adapt to the new context [49]. This constant flux of change, we
propose, imposes continual selection pressure for CRM function within the network to
co-evolve and change. This “moving target” hypothesis finds support in
an analytical study, which shows that fluctuating selection may be common in
Drosophila, with changes in the sign of selection coefficient
occurring at nearly the rate of neutral evolution [50]. Adaptive substitutions could
therefore occur before selection switches its sign again, since positively selected
mutations fix at rates much higher than the neutral mutation rate.
At the same time, the high connectivity in the regulatory network implies pleiotropic
effects while the essentiality of genes controlled by the network may call for
accurate regulation, both suggesting that the net change in CRM function will be
highly constrained (Figure 5A).
Under this conceptual model, functionally significant change will be possible on
short evolutionary timescales, but will remain within constrained bounds over longer
timescales. This feature of the model would account for adaptive gain and loss of
TFBS in CRM, and could explain the strongly non-linear relationship between function
and sequence evolution as exemplified by the Drosophila eve stripe
2 enhancer [7],
[8]. Moreover,
it provides an explanation for the finding of a non-clocklike evolutionary pattern:
sequences from D. pseudoobscura rescues a mel eve
stripe 2 enhancer deficiency almost as well as the native mel
enhancer and substantially better than ones from much more closely related species
([48], Figure 5B).
In conclusion, our findings provide empirical evidence for positive natural selection
acting in CRM and TFBS evolution. We suggest that CRM are not as functionally static
as commonly believed, but rather may experience frequent adaptation through binding
site turnover, even though there may be constraints on net change over longer
evolutionary time.
REDfly [30] is a
database of manually curated CRM and TFBS obtained from the literature from
which we chose 118 non-overlapping autosomal CRM for investigation (Table S1).
They regulate 81 target genes and contain binding sites for 82 TF. The 118 CRM
range in size from 65 bp to 4.3 kb (median = 515 bp) and
contain between 1 to 64 DNase I footprint sites
(median = 4). From the set of 82 TF, we identified a subset
of 30 with more than 10 footprint sites represented in the dataset and with
carefully constructed Position Weight Matrices [51]. In each footprint region
plus five flanking bases on each end, we applied the appropriate position weight
matrix to identify the highest scoring match as the core motif for the TFBS
(referred to as TFBS in the text). We only included those TFBS for which the
alignment between mel and sim sequences
contain no insertions or deletions (including both fixed or polymorphic sites).
As a result, a total of 645 TFBS for these 30 TF were included for analysis.
For each of the 118 CRM (coordinates in dm3 of D. melanogaster
reference genome listed in Table S1), we downloaded pre-aligned MAF
blocks from UCSC genome browser for D. melanogaster
(mel), D. simulans (sim),
D. sechellia (sec), and two outgroup
species, D. yakuba (yak) and D.
erecta (ere). D. sechellia is a
sister species to D. simulans and is included to compensate for
the low sequence completeness in the reference sim genome. We
then used the baseml module in PAML 4.4c [52] to reconstruct the ancestral
sequences from the alignments. Following analysis involving polarized changes
were done either using a single ancestral sequence for mel and
sim determined by the most probable ancestral state (A,C,G
or T) at each position, or summing over the posterior probabilities of all four
possible states (full Bayesian approach). The two methods produced essentially
the same results and therefore we only presented results using the most probable
ancestral state. A maximum parsimony method was also investigated and was found
to produce consistent results.
For polymorphism analysis, alignments for the same 118 CRM regions were obtained
of a population sample of 162 D. melanogaster lines (http://www.hgsc.bcm.tmc.edu/projects/dgrp/) and six D.
simulans lines (http://www.dpgp.org/). We also
compiled the genome sequences of 150 coding regions corresponding to the target
genes of the CRM listed in REDfly, for the purpose of compiling synonymous and
nonsynonymous changes. For these data, we used codeml module in PAML 4.4c to
reconstruct the ancestral sequence states following otherwise the same procedure
as described above for CRM regions.
PWM for 30 TF (Antp, Deaf1, Dfd, Kr, Mad, Trl, Ubx, Abd-A, Ap, Bcd, Br-Z1, Br-Z2,
Br-Z3, Brk, Cad, Dl, En, Eve, Hb, Kni, Ovo, Pan, Prd, Slbo, Tin, Tll, Twi, Vvl,
Z, Zen) were obtained from [51]. This set represents all the TF for which Down
et al. identified a single best motif for the REDfly
footprint sites. For comparison, we also constructed five PWM (Hb, Bcd, Kr, Prd,
Twi) from SELEX (Systematic Evolution of Ligands by EXponential enrichment) data
(kindly provided by Mark Biggin). We ran MEME [53] with parameters “-evt
0.01 -dna -nmotifs 3 -minw A -maxw B -nostatus -mod zoops -revcomp text”
on different selection rounds of the SELEX data. The best PWM was chosen based
on the MEME score, percentage of footprint sites recovered and a penalty for the
number of additional matches predicted in addition to the footprint sites (Table
S5).
Consider a mutation at the position in a
binding site motif involving a change from nucleotide
to
( take values 1–4, corresponding to the nucleotides
ACGT). We calculated , where
is the PWM matrix of size
. According to previous theories, the PWM score is
proportional to the physical discrimination energy of the protein to the
sequence and therefore the above calculation may be used to infer the direction
and magnitude of binding energy change due to a mutation [54].
To evaluate the accuracy of the PWM-based inference, we experimentally measured
the binding energy change of observed mutations in Hb binding sites, using a
state-of-the-art microfluidics device that has high sensitivity for relatively
weak molecular interactions (MITOMI, [44]). The experiments were
performed as described in Maerkl et al.
[44].
Sixty-four oligonucleotides were synthesized to test 25 SNP in Hb footprint
sites and their combination in cases of multiple SNPs in a single TFBS between
mel and sim. Data were analyzed in GenePix
6.0, R, and Prism 5.0. We found that the PWM we used correctly predicted the
direction of change in 21/25 cases (Figure S2). Three of the four disagreements
had a predicted PWM score change close to or
smaller than one, which indicates that PWM may not be accurate when its
predicted binding energy differences are small. To minimize the chance of
misassigning the direction of binding energy change to a mutation, we set a
threshold corresponding to a PWM score difference of one, and classified
mutations within (smaller in absolute value) that bound as uncertain. The
conclusions are robust to the setpoint of the threshold (for example, Table S3).
We also compared the PWM derived by Down et al. to the five PWM
derived from SELEX data: 97% (33/34) of mutations in the TFBS were
consistently classified after excluding nine mutations with small predicted
effects by either PWM (Figure S3).
To examine the extent of binding sites gain and loss between the two species, we
calculated PWM scores for each of the
645 footprint TFBS ( from 1 to 645) in
orthologous sequences in mel, sim or the
inferred mel-sim ancestor (j from 1 to 3), using patser v3e (by
Gerald Z. Hertz, 2002). To determine whether a sequence is a binding site or
not, we established two sets of cutoffs for PWM scores. First, we used PWM score
, corresponding to the sequence being more likely from a
binding site distribution than from a background distribution. For the second we
used a set of TF-specific cutoff values chosen by first ranking all footprint
sites of a TF by their PWM scores in descending order and then taking the
80% quantile value. The two cutoff set produced similar results (Table
S2).
To test whether the mel-derived PWM might be over-optimized so
that they would favor mel over sim sequences
independent of the binding affinity differences, we ran MEME on both
mel footprint sites for three TF (Hb, Bcd, Trl) and their
sim orthologous sequences with the same parameters. The two
set of ÒorthologousÓ PWM were then applied to score the observed
variations in the TFBS of the three TF for comparison (Figure
S7).
We attempted to estimate the probability for a random new mutation to be
affinity-increasing () or
affinity-decreasing () by examining all
possible mutations that can occur on the inferred ancestral sequence of mel and
sim for the 645 footprint TFBS. At the site in a TFBS for
TF x, the probabilities are calculated as:(1)(2)(3)where is the original
nucleotide and varies among the
three possible mutations. is the position
weight matrix for TF x of size . These values were
then summed across all 645 TFBS and divided by the total number of nucleotides
involved. Mutation matrix is derived from
polymorphism of the 4-fold degenerate sites of 9,628 genes in D.
simulans
[55].
For the generalized MK test, we counted the number of fixed and segregating sites
for different functional categories in both mel and
sim lineages. In sim, we required at least
two of the six alleles to be non-missing for a site to be included in the
analysis. For coding regions, synonymous sites were further classified into
No-Change, Preferred-to-Unpreferred and Unpreferred-to-Preferred, following
[22].
Polymorphism and divergence sites in both coding and CRM regions were counted
using perl scripts adapted from Polymorphorama (Peter Andolfatto, Doris
Bachtrog, 2009).
Following the suggestion of [34], we considered only common polymorphism (derived
allele frequency 15%) in the
generalized MK test to alleviate the problem caused by negatively selected
mutations in detecting positive selection. For each mutation category, we
compared the substitution-to-polymorphism ratio to the synonymous No-Change
class using Fisher's Exact Test. Two-sided p-values are reported.
Site frequency spectrum (mel only): Next-generation sequencing
data produce variable coverage. To estimate the site frequency spectrum, for
each variable site (TFBS, coding and spacers) with a coverage greater than or
equal to 150 (maximum is 162) we randomly chose 150 and combined the counts for
each frequency class (from 1/150 to 149/150).
|
10.1371/journal.pntd.0004551 | Is Dengue Vector Control Deficient in Effectiveness or Evidence?: Systematic Review and Meta-analysis | Although a vaccine could be available as early as 2016, vector control remains the primary approach used to prevent dengue, the most common and widespread arbovirus of humans worldwide. We reviewed the evidence for effectiveness of vector control methods in reducing its transmission.
Studies of any design published since 1980 were included if they evaluated method(s) targeting Aedes aegypti or Ae. albopictus for at least 3 months. Primary outcome was dengue incidence. Following Cochrane and PRISMA Group guidelines, database searches yielded 960 reports, and 41 were eligible for inclusion, with 19 providing data for meta-analysis. Study duration ranged from 5 months to 10 years. Studies evaluating multiple tools/approaches (23 records) were more common than single methods, while environmental management was the most common method (19 studies). Only 9/41 reports were randomized controlled trials (RCTs). Two out of 19 studies evaluating dengue incidence were RCTs, and neither reported any statistically significant impact. No RCTs evaluated effectiveness of insecticide space-spraying (fogging) against dengue. Based on meta-analyses, house screening significantly reduced dengue risk, OR 0.22 (95% CI 0.05–0.93, p = 0.04), as did combining community-based environmental management and water container covers, OR 0.22 (95% CI 0.15–0.32, p<0.0001). Indoor residual spraying (IRS) did not impact significantly on infection risk (OR 0.67; 95% CI 0.22–2.11; p = 0.50). Skin repellents, insecticide-treated bed nets or traps had no effect (p>0.5), but insecticide aerosols (OR 2.03; 95% CI 1.44–2.86) and mosquito coils (OR 1.44; 95% CI 1.09–1.91) were associated with higher dengue risk (p = 0.01). Although 23/41 studies examined the impact of insecticide-based tools, only 9 evaluated the insecticide susceptibility status of the target vector population during the study.
This review and meta-analysis demonstrate the remarkable paucity of reliable evidence for the effectiveness of any dengue vector control method. Standardised studies of higher quality to evaluate and compare methods must be prioritised to optimise cost-effective dengue prevention.
| Dengue fever has increased dramatically over the past 50 years and today is the most widespread mosquito-borne arboviral disease, affecting nearly half the world’s population in 128 countries. Until the arrival of a vaccine, control of its Aedes vectors has been the only method to prevent dengue infection. With dengue outbreaks occurring at increasing frequency and intensity, we undertook a systematic review and meta-analysis of the literature, to evaluate the evidence for effectiveness of vector control strategies currently available. Forty-one studies (from 5 months to 10 years duration) were included in the review. Most studies investigated combinations of approaches but only 9 studies were randomized controlled trials (RCTs). Remarkably, no RCTs evaluated effectiveness against dengue of insecticide space-spraying (outdoor fogging), the main response to dengue outbreaks used worldwide. Nevertheless, there was limited evidence indicating that house screening and to a lesser extent, community-based environmental management with water container covers could reduce risk of dengue infection. However, skin repellents, bed nets and mosquito traps had no effect while insecticide aerosols and mosquito coils were associated with higher dengue risk. However, the quality of the few studies eligible for inclusion was poor overall, and the evidence base is very weak, compromising the knowledge base for making recommendations on delivery of appropriate and effective control. Given this paucity of reliable evidence, standardised studies of higher quality must now be a priority.
| Dengue is a viral infection transmitted between humans by Aedes mosquitoes. With an estimated 390 million dengue infections occurring every year, and almost half the world’s population exposed to infection with dengue viruses, it is the most widespread mosquito-borne arboviral disease today, affecting 128 countries worldwide [1–3]. The dramatic increase in dengue over the past 50 years can be attributed to a number of factors, ranging from increased urbanization, in-country and international population movement, erratic water supplies and ineffective or unsustainable vector control [4, 5]. The human and economic cost of frequent dengue outbreaks is high, though current Figs are almost certainly underestimates [6–9]. Dengue is showing signs of emergence in more temperate latitudes [10–13] and is a potential threat to many of the international mass-gatherings that are a feature the modern era, such as the FIFA World Cup and the Olympics, or religious gatherings like the Hajj, although their contribution to global spread has never been proven [14, 15].
Until recent advances in vaccine development [16–17], and the approval and potential availability of the first product in 2016 [18], dengue has been unique among the major vector-borne diseases, in that prevention from infection could only be attempted by reducing or eliminating bites by infected vector mosquitoes [19, 20].
Dengue viruses are transmitted primarily by Aedes aegypti, a cosmotropical mosquito that thrives in urban environments. It is highly anthropophilic and breeds in small bodies of fresh water, most commonly in the numerous containers found around the home, ranging from water storage drums and overhead tanks to bottles, buckets and discarded waste items [4]. Between blood feeding and oviposition, adult female mosquitoes rest within or close to human dwellings [19]. A second vector, Aedes albopictus, was originally confined to Asia, but in recent decades has expanded its global range and contributed to the spread of the chikungunya virus, as well as dengue [21–24].
Control of dengue vectors can be directed against the immature aquatic stages (larvae and pupae) or the adult mosquitoes, with a number of methods available for each approach. Described in detail elsewhere [19, 25], they can be grouped according to whether they target the vector directly (i.e. aim to kill mosquitoes using insecticides or natural enemies or prevent them from biting using repellents) or indirectly (e.g. environmental modification or sanitation improvements that reduce potential larval development sites, or house improvements that prevent mosquito entry). Some approaches require skilled staff and/or dedicated resources (e.g. specialised spraying equipment, insecticides, transport) in order to be delivered effectively in a vertical approach. For others, affected communities, empowered through education and advocacy, can mobilize and mount effective control operations relatively independently via horizontal or community-based efforts. Hence, space-spraying and larviciding require trained personnel to deliver potentially toxic insecticides using specialized equipment and are dependent on vertical municipality-driven programs. In contrast, reductions in potential larval development sites can be achieved with householders and communities taking responsibility, supported by education and social mobilization [19].
In dengue-affected communities worldwide, immature vector populations are targeted through the reduction or elimination of potential larval development sites, typically by collection of purposeless or discarded containers in ‘clean-up’ or environmental management campaigns; functional or useful sites are either covered (water storage containers), drained (gutters or channels) or treated with an appropriate insecticide (usually referred to as ‘larviciding’) or biological control agent (predatory copepods or fish). Identification of, and targeted action towards, ‘productive’ container types (i.e. those that are assessed as contributing the greatest burden of pupae, relative to other containers in the area) can potentially enable more cost-effective larval control [26,27].
The typical response to dengue outbreaks is to target the adult mosquito population by space-spraying or fogging with insecticide, delivered outside or inside the home, with the aim of severely reducing the vector population at the time of delivery. This method is not designed to deliver persistent insecticide residues on treated surfaces and if the outbreak continues, it must be repeated at intervals that coincide with the vector life cycle [19].
Previously, Erlanger et al. (2008) [28] reviewed data on the effectiveness on vector indices of all vector control methods and concluded that integrated vector control was the most effective, while environmental management had minimal impact. Notably, the evidence for impact of outdoor space spraying was limited, though only 1 of the studies included was less than 30 years old (dated from 2015). Two subsequent reviews [29, 30] focused on peri-domestic space spraying and concluded that there was no evidence to support its use in dengue outbreak control, either as a standalone intervention or in combination with other interventions. Horstick et al. (2010) [31] also found no evidence for a demonstrable effect of vector control on entomological indices and identified specific weaknesses in funding, management, staffing and community engagement, all of which conspired to lower operational standards and ultimately restrict any likelihood of success. Recent reviews have examined the evidence for the effectiveness of individual methods, including copepods, fish and temephos [32–34].
Today, dengue outbreaks occur at an increasing frequency and intensity in affected communities worldwide and the need for evidence-based selection of the most appropriate interventions has never been greater. What are the best currently available dengue vector control tools, as measured by their impact on dengue infections, and not simply on vector populations? Are previous dengue control failures the result of low operational and management strategies, or are the available tools simply not effective? What evidence exists to provide a basis for evaluating dengue vector control today? To answer these questions and to provide guidance on the most effective strategies currently available to combat dengue, we report here on a systematic review and meta-analysis of the evidence.
To systematically review randomized and non-randomized studies to evaluate the evidence of the effectiveness of vector control interventions in reducing a) Aedes sp. vector indices and b) human DENV infection and/or disease. The original search was conducted in April 2012 and updated in December 2013 and on 10th January 2015.
Table 1 displays the eligibility criteria. Only studies that presented data for a minimum duration of 3 months were included (regardless of the frequency of treatments undertaken within that period), as this was considered the minimum period required to demonstrate a sustained impact on the vector population and/or impact on dengue transmission. In addition, only studies published since 1980 were considered eligible for inclusion, for a number of reasons. The period after 1980 saw the expansion in urban populations worldwide, notably in the less developed countries where the ratio of populations in urban and rural regions began to change dramatically [35,36]. This also was the beginning of the ‘globalization’ era, as characterized by steep increases in trans-national and international movement of humans and merchandise, and the time when all four dengue serotypes were reported in every continent, leading to an increase in the frequency and magnitude of dengue outbreaks [5,37,38]. We are familiar with the achievements prior to 1970, such as the ambitious yellow fever programs when Aedes aegypti populations were significantly diminished, and indeed eliminated from many cities and large geographic areas throughout Latin America [1,4,5,39]. On balance, it was concluded that the control tools available before the 1980s (e.g. the highly persistent insecticide DDT) and the settings in which they were carried out, were not pertinent to the challenge of dengue control in urban environments of the 21st century, based on the significant logistical, sociological and epidemiological changes, and the rise in insecticide resistance in vector populations [40,41] that have occurred in many of those countries during the past 35 years.
The primary outcome was dengue incidence (any reported case data; clinical or lab-confirmed/ serologically positive cases); secondary outcomes were a range of vector indices: Breteau Index (BI), House Index (HI), Container Index (CI), tank positivity, number of mosquito adults, pupae per person index (PPPI), presence of Aedes immatures and ovitrap positivity rates.
All methods were pre-specified in the review protocol. PRISMA Group guidelines were followed as standard methodologies [42,43].
The databases WHOLIS, MEDLINE, EMBASE, LILACS and Science Citation Index were searched using the Medical Subject Heading (MeSH) “dengue” followed by the Boolean operator “and” combined with the following ‘free text’ terms “epidemic” and further combined in succession with: ‘threshold’ ‘sentinel’ ‘early warning’ ‘case management’ ‘vector control’ ‘DDSS’ ‘space spraying’ ‘indoor residual spraying’ ‘fogging’ ‘integrated vector management’ ‘IVM’ ‘source reduction’ ‘container’ ‘larvicide’ ‘repellent’ ‘insecticide’ ‘adulticide’ ‘fumigant’ ‘aerial spraying’ ‘dengue decision support system’. The reference list of each of the included studies was also searched, and ‘‘grey literature” (cited unpublished documents) were sought by communication with authors. No limits were placed on year of publication status or language.
Search results were imported into EndNote (EndNote X5, Build 7473). LRB and PJM independently assessed the title and abstract of each record (or the corresponding full article) retrieved by the search for eligibility; any discrepancies were discussed. The full article was retrieved for each eligible study. The study’s investigators were contacted if eligibility was unclear, additional data were unpublished or the article was inaccessible. Each article was scrutinized to detect multiple publications from the same trial; such publications were included as a single study.
LRB and PJM independently extracted data according to an agreed checklist and differences were discussed. Trial characteristics and risk of bias information were extracted along with outcome data (S1 Table). For each randomized controlled trial, we extracted the number of individuals randomized and the number of individuals analysed for each treatment group. For dichotomous outcomes, we extracted the number of individuals experiencing the event in each treatment group for each study. For continuous outcomes we extracted means and standard deviations (where presented) or medians, interquartile ranges, and ranges. When such data were not reported, we extracted narrative information and tabulated results. For non-randomized studies, we extracted measures of effect, as well as treatment group data.
Using a pre-piloted form, LRB and PJM independently assessed risk of bias and discussed any differences (S2 and S3 Tables). For randomized controlled trials we used the Cochrane risk of bias tool and addressed: random sequence generation; allocation concealment; blinding; incomplete outcome data, selective outcome reporting, and other biases [43]. For each component of each trial, a judgment of high, low, or unclear risk of bias was made and the rationale for the judgment was given (S2 Table and S1 Fig). For non-randomized studies, LRB and PJM used the Quality Assessment Tool for Quantitative Studies [44] (S3 Table). This ensured that each study could be ranked according to inherent study design limitations, which included but were not limited to, bias, confounding and blinding.
Analyses were performed in Review Manager (RevMan Version 5.2. Copenhagen: The Nordic Cochrane Centre, 2012). We extracted the measure of effect and CI from the study reports. Where possible, we stratified analyses by intervention, outcome, measures of effect and study design. For multi-arm trials, data from numerous intervention groups were pooled. We calculated trial-level results (i.e. MD, RR or OR and standard error [SE]) and pooled them using random-effects inverse-variance meta-analysis to account for large variability present between studies. Results were visualised in forest plots. Sub-group analyses were used to stratify studies that used different and/ or combination interventions.
Heterogeneity was assessed using the I2 test statistic, the chi-squared test (P<0.01 indicated possible significance) and by visual inspection of the forest plots to identify overlapping confidence intervals. Studies that could not be visualised in forest plots were presented in tables.
When heterogeneity was detected, possible causes were explored using subgroup analyses and predefined covariates.
Subgroup analyses were planned to explore potential sources of heterogeneity (i.e. effects of seasonality, mosquito species, duration of intervention, coverage), but analyses were not carried out because of the low number of studies available for analysis. For the same reason, sensitivity analyses that excluded studies with a high risk of bias were pre-planned to assess the robustness of results, but were not carried out. Hence, the planned funnel plots were not constructed to explore possible publication biases.
A total of 960 potentially relevant studies were identified using systematic searches of the databases, grey literature and their cited reference lists and 19 more were identified from other sources (Fig 1). After removing duplicates, 582 citations were screened, of which 480 were excluded. The full texts of the remaining 102 records were assessed and 61 articles were excluded.
The reasons for exclusion were: incomplete outcome data (18 studies); study was a review, a non-peer reviewed report or a mathematical model (14 studies); no intervention was carried out (eight studies); undefined or inadequate dengue case definition (three studies); intervention or outbreak duration was less than 3 months (10 studies); study included only one required outcome (three studies); study preceded 1980 (three studies); time series data collection not reported (two studies).
Forty-one studies were included in the review [45–85] (S1 Table), nineteen of which reported sufficient data for inclusion in meta-analyses [46–48, 52, 54, 55, 58, 59, 66, 69, 73, 74, 76, 77, 80–83, 85].
The main characteristics of included studies are summarised in S4 Table. Of the 41 included studies, geographic study locations comprised: SE Asia (n = 11) or Central America (10), South Asia (8), Australasia (4), South America (5) and North America (3). All studies were published between 1986 and 2014, and 2009 was the median year of publication.
Grouped by study design, the studies comprised: 9 randomised controlled trials (i.e. 7 cluster-randomized and 2 randomized controlled trials) and 32 non-randomised studies (i.e. 8 controlled trials, 7 longitudinal studies, 4 interrupted time series studies, 5 before and after studies, 2 observational studies, 1 case-control study, 1 cross sectional study, 1 retrospective observational study, 1 ecological study and 2 models) (S4 Table).
Vertical and community-led interventions were used exclusively in 20 and 10 studies respectively, and 11 studies used a combination of both. Combination interventions (23 studies) were more common than single interventions (18 studies). Study duration ranged from 5 months to 10 years; 16 studies were less than 1 year, 12 took place over 1–3 years and 7 studies were 8 or more years in duration.
Fig 2 (top) summarises the frequency of vector control tools by study design. The most frequently evaluated interventions were clean-up programs (n = 19), of which 4 were cluster randomised controlled trials. Outdoor fogging (9), education (11), larviciding (7) water jar covers (7) also were the subject of multiple studies.
All studies presented data on Aedes aegypti; four presented additional data on Aedes albopictus (S4 Table). Nineteen studies reported dengue incidence, 17 studies reported BI, 16 studies reported HI, 11 studies reported CI, 1 study reported tank positivity, 3 studies reported number of mosquito adults, 6 studies reported pupal indices, 3 studies reported ovitrap data.
Fig 2 (bottom) summarises the reported reduction in outcome at a statistically significant level (p<0.05). Of note was the observation that in studies where it was an outcome, dengue incidence was not reduced in either of 2 randomised study designs, although 8 of 14 studies with other experimental designs reported a statistically significant reduction.
Nineteen studies [46–48, 52, 54, 55, 58, 59, 66, 69, 73, 74, 76, 77, 80–83, 85] provided sufficient data to allow their inclusion in meta-analyses. The results of those analyses are presented here stratified by reported outcome, either the impact on dengue incidence or on vector indices.
The dramatic growth in dengue over the past 35 years has been a remarkable epidemiological event and, as evidenced by its continued global spread, a challenge for which the public health community was not prepared. It is not surprising that 24 of the 41 studies included in this review were published in the past 7 years, reflecting the increase in attention and resources devoted to devising effective control strategies as recognition of the dengue pandemic grew. However, the fact that the global increase in focus on dengue control generated so few studies performed at a standard required for inclusion in this review, indicates that the magnitude of the response to the dengue pandemic has not been sufficient. Moreover, most of these studies investigated the impact of interventions on dengue vector indices alone, rather than dengue incidence. This also is discouraging, as the limitations of the Stegomyia larval indices, primarily their poor correlation with dengue transmission, are well known [86]. Finally, the inadequacy of the response to global dengue threat is demonstrated by the identification of thirteen studies that measured the impact of vector control on dengue incidence in the past 35 years, and that only six of these were suitable for inclusion in a meta-analysis. Simply stated, we do not have a clear understanding of which of the currently available interventions actually work, where or when they succeed or might work best, and the reasons why they succeed or fail.
Nowhere is the inadequacy more apparent than in the absence of appropriately designed trials to evaluate insecticide fogging or space-spraying for the prevention of dengue transmission. Although space spraying is the standard public health response to a dengue outbreak worldwide, and is recommended by WHO for this purpose [19], our study revealed the scant evidence available from studies to evaluate this method sufficiently. Earlier reviews also noted this serious omission from the literature published before 1980 [29,31]. Remarkably, no randomised controlled trials have been undertaken to evaluate the effectiveness of space-spraying or fogging to reduce dengue transmission or dengue incidence, anywhere in the past 35 years. We identified only one study [74] suitable for inclusion in a meta-analysis that demonstrated a significant impact of outdoor fogging on dengue vector populations.
Without adequate evidence, it is impossible to determine how effective space-spraying programs, whether indoor or outdoor, have been. It may be the case that outdoor fogging has the potential to impact on dengue vector populations sufficiently to impact transmission, but the minimum treatment frequency and geographic area requiring treatment remain unknown. The most encouraging report comes from a recent longitudinal study analysing twelve years of data from the city of Iquitos in Peru [84], which concluded that dengue cases could be reduced if intensive city-wide space-spraying (outdoor fogging) was conducted early in the transmission season. Given the cost implications of delivering a similar scale treatment in an even larger city, possibly with the need to do so in advance of an outbreak crisis, further studies to demonstrate the potential benefits are essential.
Of those that could be assessed adequately, the method with the most evidence supporting effectiveness in preventing dengue transmission was house screening. Data from cross-sectional [52] and case-control studies [59] in Australia, and a case-control study in Taiwan (69) were included in a meta-analysis that indicated a significant protective effect of window and door screens on dengue transmission as detected by serology (ELISA or HIA (haemagglutination inhibition assay)) (Fig 3). Although the weaker study designs limited the power of this result, the results are encouraging. Aedes aegypti exhibit predominantly indoor resting and blood feeding behaviour (termed endophagic and endophilic behaviour, respectively)[87], and barriers to access would be expected to impact on this species. Malaria vector mosquitoes and other arthropods of medical importance are also active indoors and can be targeted in the same way, increasing the likelihood of perception of benefit and adoption by householders. “Mosquito-proofing” houses was first considered over a century ago, and its potential as a sustainable and effective tool for malaria control has been evaluated in randomized controlled trials in recent years [88–90]. New investigations of screening for dengue prevention are also underway. Recent studies in a high-risk dengue setting in Mexico reported that window and door screens were a popular and widely-adopted intervention that significantly reduced domestic infestations of Aedes aegypti [91, 92]. House screening is not included in the current WHO dengue guidelines, but given its potential and wide ranging benefits, it is a strong candidate for randomised controlled trials to evaluate its effectiveness in preventing dengue.
Two observational studies reported on the impact of indoor residual spraying IRS, with contradictory results and while one of these reported a positive significant reduction in the odds of (secondary) incidence [66], the second study reported an insignificant increase [69]. Consequently, the pooled odds ratio showed no statistically significant effect between intervention and control groups. While indoor residual spraying can target Aedes aegypti, such methods have rarely been used, nor are currently recommended [19, 93, 94]. Yet IRS is already used widely to control a number of other vector-borne diseases in various settings worldwide and, as it allows the delivery of a range of different insecticide classes, it can be an important tool for managing insecticide resistance [95–98]. The possibility that existing IRS programs might be expanded with minimal change to include dengue is an attractive prospect.
Probably the most widespread practices to suppress dengue vector populations are clean-up campaigns, typically community-driven and in tandem with education and health promotional campaigns as well as numerous additional approaches. Efforts promoting environmental and peri-domestic clean-up to reduce vector larval development sites have been routine practices in many dengue-endemic localities for decades and as shown in Fig 2, they were the most common intervention evaluated in the reviewed studies. However, clean-up campaigns were evaluated only as one element within multiple interventions or they continued to be promoted as a background across all the arms within a study. Thus, source reduction or clean-up campaigns were applied in some way in 20 studies but were associated with interventions ranging from fogging or water container covers targeting adult mosquitoes to larviciding and copepods for control of immatures (S3 Table). Hence it is not possible to dissect their specific contribution to reducing vector populations or their impact on dengue transmission. Of these, the strongest evidence (Fig 3) was from Cuba [58] where results indicated that community working groups (CWGs), initially set up some years earlier, in a preceding study [71] promoting environmental management, conversion of garbage zones into gardens, water pipe repairs and the use of water container covers not only reduced vector indices, but also impacted dengue transmission, significantly more than the routine A. aegypti control programme. Although WHO recommends community participation as an essential element of sustainable dengue prevention [99], there is little evidence that it can impact on dengue transmission [100]. A number of randomised controlled studies have demonstrated significant impacts on vector indices [47, 48, 73, 83, 101](Fig 5) even though the methods of intervention varied considerably between the studies. Results from a cluster randomised controlled trial in Nicaragua and Mexico [102] reported reductions in dengue sero-conversion rates and self-reported dengue cases as well as vector indices, following community mobilisation to deliver pesticide-free vector control. Clearly further evidence is needed. It remains to be determined how best practice is defined in any setting (i.e. which tools or methods the community should employ), and what coverage is necessary in order to not simply reduce mosquito indices, but to impact on dengue virus transmission.
The use of fish and crustaceans as biological control agents that prey on or compete with the immature vector stages may have potential in certain contexts, but we identified only three studies that evaluated copepods (aquatic Crustaceans) [78, 79, 103]. In all cases, the crustaceans were used together with clean-up programs, obscuring the impact of each method, and none of the reports provided sufficient data to be included in a meta-analyses. Consistent with earlier specific reviews [32, 34], there remains little evidence to suggest that biological control has widespread potential.
A substantial number of reports demonstrated impacts on vector indices of insecticide-treated materials (ITMs), deployed as window or door curtains [54, 75, 77, 82, 104, 105], although they were effective only where houses with fewer and smaller windows and doors [75–77, 104] and where coverage of the intervention was particularly high [77]. Hence, in the meta-analyses, no significant impact on vector populations was indicated and the heterogeneity between the studies was high (Fig 4). Effects on dengue incidence of ITMs used as vertical window or door screens or as horizontal covers for water containers, need to be quantified in locations and contexts where housing conditions indicate suitability. ITMs, used as curtains hung or fixed tightly across external windows and doors, function in a similar way to mesh screens, and potentially could provide enough protection without the need for insecticide, as suggested by a study in Mexico, where ITMs reduced vector populations even though the targeted population was highly resistant to the insecticide used [90].
There was no evidence of any impact on dengue infection risk by insecticide-treated bed nets [52,69], mosquito traps [69, 81] or mosquito repellents [52]. Ongoing studies are investigating a range of novel trap designs for Aedes spp. surveillance and control [106–108] but to date, evidence of traps preventing any mosquito-borne disease remains elusive. Both opinion and evidence are weighed against the use of skin repellents for prevention of vector-borne diseases [109], and attention has moved towards a new generation of spatial repellents, to be deployed within or close to houses to prevent mosquito entry, possibly in combination with attractant lethal traps in what is termed a ‘push-pull’ strategy [108, 110].
The significant negative associations found between the use of insecticide aerosols [52] and mosquito coils [52,69] and higher odds of dengue incidence have a number of possible explanations. These tools may have been purchased in response to an actual increase in mosquito numbers, or a dengue case in the home or a neighbour’s house, during a period of dengue transmission. Alternately, householders using aerosols or coils may have relied solely on these anti-mosquito devices and not have adopted any other more effective preventative measures.
Approaches involving the use of genetically modified (GM) mosquitoes or the intracellular symbiont Wolbachia [111] are recent advances in insect control and only one field trial, demonstrating impact on the vector population only [85], was included in this review. An increase in the numbers of reports from ongoing new trials can be expected, although the use of GM mosquitoes for dengue control will have to confront or overcome additional regulatory or ethical challenges and requirements prior to field tests and eventual deployment [112–116].
Regarding trials of methods that require the use of insecticides, we noted that while 23/41 studies examined the impact of insecticide-based tools, only 9 of these cited recent information on insecticide resistance or referred to an evaluation of the susceptibility status of the target vector population at any stage of the study. Resistance to DDT, pyrethroids and other insecticides has been documented widely in dengue vectors, and continues to emerge, potentially impacting on intervention effectiveness [40, 117–119]. Clearly, insecticide susceptibility testing must be an integral part of any trial where insecticide-based interventions are under evaluation, as recommended by the World Health Organisation [4].
Today, there is a widespread perception that Aedes aegypti control ‘has failed’ or that existing methods will not reduce dengue transmission, and that this is why we should abandon existing approaches and invest in or pursue alternative strategies [111, 120, 121]. As we have shown in this review and meta-analysis, this is incorrect. In reality, there is very little reliable evidence from appropriately designed trials to reach a conclusion about any of the control methods available. That this also applies to insecticide space-spraying or fogging illustrates clearly the urgent need for such fundamental trials.
Care in designing studies is critical. Randomized controlled trials are the most robust design for evaluating the effectiveness of any intervention [122]. In our review, only eight of the nineteen reports included in the meta-analysis (7 CRCTs, 1 RCT) were randomised, none of which reported a significant impact on dengue incidence. In contrast, eight other studies that reported a positive reduction in dengue incidence at p<0.05, were not derived from randomised controlled trials, but from weaker experimental designs (see Fig 3). Weakness in the designs of trials investigating vector control tools have been recognised, and expert guidance, identification of challenges and pitfalls and clear recommendations for improvement are available [123,124].
Also apparent from this review is the large number of studies investigating impacts on the vector population alone, with no measures of the effectiveness of the intervention on dengue transmission. We recognise that detecting dengue viruses or confirming current, recent or historic dengue infections are not simple routine or inexpensive tasks, requiring skills and equipment that are not available without considerable investment. However, without this additional investment, the value of many studies that are limited to evaluating impacts on the vector alone is seriously reduced. Demonstration of impact on vector populations is achievable and often reported but is no guarantee that an intervention will translate into a reduction in dengue transmission [125, 126]. This is particularly true for dengue, where the complex relationship between vector abundance, virus transmission and human infection rates are far from clear [86,127,128].
As well as their role in dengue transmission, Aedes aegypti is the main urban vector of yellow fever in Africa and South America, and this species and Aedes albopictus variously are vectors of the Chikungunya and Zika viruses, two emerging human pathogens that constitute a new global threat [129–132]. Despite the fears surrounding these threats, the urge to respond must be tempered by reality, and based on sound evidence. In the large urban zones where these vectors proliferate, to simply continue to use what has always been used, for that reason alone, or to pursue new approaches without sound supporting evidence would be wrong, and potentially a profligate waste of resources. Hence, there is an argument for instituting a global independent advisory body to guide decisions regarding the selection of approaches and tools for control or prevention of infections transmitted by urban Aedes sp. vector populations, and the design of appropriate multi-centre trials to evaluate their effectiveness. With this in mind, we hope that the findings of this review and meta-analysis will contribute to the sound evidence base on which that approach would be founded.
|
10.1371/journal.ppat.1005410 | A20 Deficiency in Lung Epithelial Cells Protects against Influenza A Virus Infection | A20 negatively regulates multiple inflammatory signalling pathways. We here addressed the role of A20 in club cells (also known as Clara cells) of the bronchial epithelium in their response to influenza A virus infection. Club cells provide a niche for influenza virus replication, but little is known about the functions of these cells in antiviral immunity. Using airway epithelial cell-specific A20 knockout (A20AEC-KO) mice, we show that A20 in club cells critically controls innate immune responses upon TNF or double stranded RNA stimulation. Surprisingly, A20AEC-KO mice are better protected against influenza A virus challenge than their wild type littermates. This phenotype is not due to decreased viral replication. Instead host innate and adaptive immune responses and lung damage are reduced in A20AEC-KO mice. These attenuated responses correlate with a dampened cytotoxic T cell (CTL) response at later stages during infection, indicating that A20AEC-KO mice are better equipped to tolerate Influenza A virus infection. Expression of the chemokine CCL2 (also named MCP-1) is particularly suppressed in the lungs of A20AEC-KO mice during later stages of infection. When A20AEC-KO mice were treated with recombinant CCL2 the protective effect was abrogated demonstrating the crucial contribution of this chemokine to the protection of A20AEC-KO mice to Influenza A virus infection. Taken together, we propose a mechanism of action by which A20 expression in club cells controls inflammation and antiviral CTL responses in response to influenza virus infection.
| Influenza viruses are a major public health threat. Each year, the typical seasonal flu epidemic affects millions of people with sometimes fatal outcomes, especially in high risk groups such as young children and elderly. The sporadic pandemic outbreaks can have even more disastrous consequences. The protein A20 is an important negative regulator of antiviral immune responses. We show that the specific deletion of A20 in bronchial epithelial cells improves the protection against influenza virus infections. This increased protection correlates with a dampened pulmonary cytotoxic T cell response and a strongly suppressed expression of the chemokine CCL2 during later stages of infection.
| Disease outcome upon exposure to a certain pathogen relies on the capacity of the host to resist and tolerate the infection [1]. Resistance protects the host by suppressing pathogen replication and promoting clearance of the pathogen, a process that is mostly mediated by the innate and adaptive immune system. Tolerance refers to the ability to improve disease outcome without affecting pathogen burden and by limiting tissue damage. An overactive immune response can negatively impact on the disease by causing severe tissue damage [2]. Immunopathology is an important contributor to death during exposure to highly virulent strains of influenza A such as the 1918 H1N1 virus or highly pathogenic avian H5N1 and H7N1 viruses. The mechanisms contributing to immune pathology during flu virus infection have been well documented, and both innate and adaptive immunity seems to be involved [3–6]. However, the exact molecular mechanisms regulating these processes are not well understood.
Detection of Influenza A by the innate immune system occurs by at least three different mechanisms [7]. Firstly, the cytosolic receptor RIG-I detects 5’-triphosporylated influenza virus genome segments [8,9]. In the absence of the viral non-structural protein 1 (NS1), RIG-I induces a strong antiviral type-I interferon response [10]. Secondly, Toll-like receptors such as TLR3 and TLR7 detect virus-associated RNA molecules. TLR7 is mainly employed by IFN producing plasmacytoid dendritic cells, which produce large amounts of type-I IFN upon infection with influenza virus [11,12]. TLR3, which recognizes double stranded RNA of yet undefined origin, has been shown to influence disease outcome following influenza virus infection [13–16]. Thirdly, the NOD-like receptor family member NLRP3 senses multiple influenza virus-associated stimuli, including increased acidification of the cytoplasm mediated by the viroporin M2, leading to the activation of caspase 1 and the release of the cytokines interleukin-1β (IL-1β) and IL-18 [17–19].
A20 (TNF alpha-induced protein 3 or TNFAIP3) is a key player in the termination of inflammation, and has been shown to regulate these innate signalling pathways [19–22]. We previously showed that A20 in macrophages critically suppresses influenza virus-induced innate immune responses and mice deficient in A20 in myeloid cells are protected against influenza A virus infection. This protective effect is mediated by an enhanced innate immune response and a better clearance of the virus [21].
Epithelial cells of the respiratory epithelium are the primary target cells of human influenza viruses and main producers of infectious viral progeny [23]. Very little is known about the physiological contribution of these cells to antiviral immunity. Epithelial cells have long been considered as passive mediators in immunity, functioning primarily as physico-chemical barriers preventing invading pathogens from entering the submucosal layers or the respiratory system. It has become increasingly evident that epithelial cells also maintain important effector functions directing both innate and adaptive immunity crucial for efficient antiviral responses [24]. Respiratory epithelium actively contributes to pulmonary homeostasis [25], immunity against viruses [26] and influenza induced immunopathology [27]. To study how A20 in epithelial cells influences influenza A disease progression, we generated mice lacking A20 specifically in bronchial epithelial cells (also known as club cells or Clara cells). We found that these mice are protected against influenza A virus infection. This protection does not result from an improved viral clearance or increased immune resistance to the virus, but correlates with a dampened pulmonary CTL response and a strongly suppressed expression of the chemokine CCL2 during later stages of infection.
We studied the role of A20 in airway epithelial cells by crossing conditional A20 knockout mice (A20FL/FL, S1A Fig) [28] with double transgenic animals carrying a reverse tetracycline transactivator controlled by the rat CCSP promoter (CCSP-rTA) and a Cre recombinase under control of the (TetO)7CMV operator [29,30]. This generated A20FL/FL/CCSP-rTA/(tetO)7-Cre triple transgenic offspring, hereafter referred to as A20AEC-KO mice. Treatment of these mice with doxycycline enabled temporally controlled inactivation of the A20 gene specifically in club cells (also known as Clara cells) [31], which constitute most of the epithelial cells found in the proximal airways of the mouse respiratory tract [32]. To ensure deletion of A20 in club cells at all times, breeding pairs and offspring were continuously maintained on a doxycycline diet. A20AEC-KO mice were born at Mendelian ratios and displayed no developmental defects or signs of spontaneous pulmonary inflammation.
Southern blot analysis on genomic DNA extracted from total lungs of A20AEC-KO mice and control wild-type (A20WT) littermates showed Cre-mediated recombination of the A20FL/FL allele only in lungs, but not in spleens, of A20AEC-KO mice (S1B Fig). Since club cells represent a minor fraction of total lung tissue, only partial recombination was observed in total lung tissue from A20AEC-KO mice. Semi-quantitative PCR analysis on genomic DNA isolated from various tissues of A20AEC-KO mice, confirmed the deletion of A20 specifically in lung tissue (S1C Fig). As A20 is expressed at low levels in most cell types including airway epithelial cells, we induced its expression by intratracheal instillation of LPS, a TLR4 agonist and known inducer of A20 in lung epithelium [33]. Analysis of A20 expression of Club cells purified from the lungs of A20AEC-KO mice by Western blot showed the absence of A20 at the protein level (S1D Fig). Absence of A20 was finally confirmed by immunohistochemistry on lung tissue isolated from A20AEC-KO mice (S1E Fig).
A20 is a negative regulator of multiple signalling pathways induced by inflammatory stimuli including TNF and TLR ligands [34,35]. We tested if A20 also regulates these pathways in club cells by intratracheal administration of TNF and poly(I:C), an agonist of the TLR3, RIG-I and MDA5 signalling receptors [36,37]. We found increased numbers of neutrophils and monocytes recruited into the bronchoalveolar space of A20AEC-KO mice compared to control littermates upon engagement of these receptors (Fig 1A and 1C). Also significantly higher levels of IL-6, CXCL1 (KC), CCL2 (MCP-1), but not TNF were detected in the bronchoalveolar lavage (BAL) fluid of mice lacking A20 in club cells (Fig 1B and 1D). These results demonstrate that A20 expression in club cells negatively controls inflammatory responses following exposure of the airway to TNF and poly(I:C). These data are in accordance with previously published results showing a similar regulatory role for A20 in various other cell types [34].
Since poly(I:C) is often used as a stimulus that mimics RNA virus infection, we next investigated the in vivo role of A20 in club cells in a mouse model of influenza A virus (IAV) infection. We used the H3N2 mouse adapted influenza A strain X-47 in most of the experiments [38]. Immunohistochemical staining of the club cell marker CCSP and the influenza A ion channel protein M2, revealed that X-47 infected CCSP expressing club cells, along with alveolar epithelial cells (S2A and S2B Fig). This allowed us to directly study the in vivo role of A20 in the primary target cells of influenza A virus. A20AEC-KO had a clear survival advantage compared to their control littermates following infection with a dose of X-47 virus that proved lethal for most wild type mice (Fig 2A). Upon infection with a sublethal dose of X-47, A20AEC-KO mice displayed less morbidity and weight loss at later stages (> 7 days) of infection (Fig 2B). Challenge with A/Puerto Rico/8/34 (PR8, H1N1 subtype) influenza virus confirmed the reduced susceptibility of A20AEC-KO mice to IAV infection (S3 Fig). This difference was also evident by histological analysis of lung tissue from X-47-infected mice (Fig 2C and S4A Fig) and by quantification of the total protein content in BAL fluid as a measure of lung damage and vascular leakage (Fig 2D). We did not detect significant differences in viral titres in the lungs of A20AEC-KO and A20WT mice at day 2 and day 5 post-infection (Fig 2E). These data indicate that the disease protection in A20AEC-KO mice is likely not a consequence of an enhanced intrinsic capacity of A20 deficient club cells to inhibit viral replication. The levels of IFNα and IFNβ, cytokines with potent antiviral activity, and the IFN-inducible chemokine CXCL10, were comparable in BAL fluid isolated from A20AEC-KO and A20WT animals at different time points post-infection, suggesting that the type-I IFN response is not differentially regulated (S4B Fig). More importantly, we could not detect virus in the lungs of both A20AEC-KO and A20WT mice at day 8 post infection (Fig 2E), which is the time during infection where both groups of mice differ in morbidity. Finally, since A20 is known to confer protection against cell death in multiple cell types (Catrysse et al., Trends Immunol., 2014), we assessed if A20 deficiency in club cells could sensitize these cells to apoptosis after IAV infection. In none of the assays performed, however, we could detect any significant difference in cell death between A20AEC-KO lung samples and control samples (S5 Fig).
The absence of A20 in club cells thus significantly improves disease outcome upon influenza A virus infection without altering viral clearance or type-I IFN responses. These data indicate that A20AEC-KO mice are protected from influenza A virus infection by a mechanism that involves increased tolerance, rather than increased antiviral resistance.
The late onset (> 7 days) of protection to influenza A of A20AEC-KO mice suggests that the effect is driven by the adaptive immune system. Although viral clearance and host survival critically depend on the recruitment of virus-specific CD8+ cytotoxic T cells (CTL) to the lung, influenza A-associated pulmonary immunopathology can be inflicted by an excessive antiviral CTL response. CTLs are major producers of TNF during influenza A virus infection which is known to contribute to immunopathology [5,39,40]. Analysis of CD8+ T cells by in vivo intracellular cytokine staining showed that A20AEC-KO mice displayed reduced numbers of Granzyme B, IFNγ and TNF expressing activated (CD62Llo) CD8+ T cells in the brochoalveolar space and to a lesser extend in the lung tissue at day 8 post infection with X-47 virus (Fig 3A and 3B). NP-pentamer staining of CD8+ T cells from the mediastinal lymph node, spleen and BAL revealed that there was no significant difference between control and A20AEC-KO mice (Fig 3C). This suggests that the extent of influenza antigen-specific CD8+ T cell priming is comparable in the two mouse strains but that their activation in the lung compartment is different. A20AEC-KO mice also showed reduced levels of IFNγ and TNF protein levels in the BAL fluid (Fig 3D). Upon entering the lung, effector CD8+ T cells are programmed to produce high levels of the anti-inflammatory cytokine IL-10, which is an important mechanism to reduce immunopathology [41]. Measurements of IL-10 in BAL fluid showed that also IL-10 levels are lower in A20AEC-KO mice (Fig 3D). This indicates that the reduced CTL response in A20AEC-KO mice is caused by a local signal from the pulmonary environment instead of a skewing of CD8+ T cells towards an anti-inflammatory phenotype.
A decreased CTL response can be the result of less effective priming of naïve T cells in the lung draining lymph nodes. Activated antigen-loaded dendritic cells travelling from the lung to the lung draining lymph nodes dictate the outcome of the CTL response [42]. However, we found no significant differences in the accumulation of CD11b-, CD11b+ or inflammatory dendritic cells (iDC) in the lung draining mediastinal lymph nodes of A20AEC-KO mice compared to control mice (S6A Fig). Furthermore, in vitro re-stimulation with NP peptide of single cell suspensions prepared from spleens of A20WT and A20AEC-KO mice infected with X-47 showed similar numbers of NP-specific CD8+ T cells as measured by IFNγ producing activated CD8+ T cells. In contrast, based on this restimulation assay, less NP-specific T cells responded to peptide re-stimulation by inducing IFNγ expression in the lungs of A20AEC-KO mice (S6B Fig). Loss of A20 in club cells did not affect antiviral humoral immunity as no significant differences in the levels of X-47 neutralizing antibodies were observed in sera from challenged A20AEC-KO mice compared to wild-type mice (S6C Fig).
Together, these data indicate that selective deletion of A20 in club cells suppresses the pulmonary CTL response against influenza A virus leading to a decreased production of cytotoxic CTL effector cytokines such as TNF.
We next addressed whether A20 deficiency in club cells and associated suppression of pulmonary CTL responses affects the recruitment of innate immune cells to the lungs. The number of recruited monocytes, neutrophils and macrophages in BAL was measured at different time points after infection with X-47 virus. At late time points post-infection (> 5 days), the timing at which A20AEC-KO mice show protection from influenza compared to control mice (Fig 2), recruitment of monocytes and neutrophils was significantly lower in A20AEC-KO compared to wild type mice (Fig 4A). CD11b- macrophages (“resident” macrophages) constituted the predominant macrophage population in the lung parenchyma of unchallenged mice and at early time points post-infection, while the number of CD11b+ macrophages (“recruited” macrophages) increased, starting around day 5 and peaking around day 8 post-infection (Fig 4B). The lungs of A20AEC-KO mice contained more CD11b- resident macrophages and alveolar macrophages compared to control littermates at peak CTL response, 8 days post-infection, and this difference sustained at later time points (Fig 4A and 4B).
Recruitment of macrophages to lungs is dependent on the chemokine CCL2 (MCP-1) [43–45]. In agreement with the elevated levels of recruited monocytes and macrophages at day 8 post-infection in control A20WT mice, higher levels of CCL2 could be detected in BAL fluid of these mice compared to A20AEC-KO littermates at this stage (Figs 4C and S7A). Other chemokines, such as KC (CXCL1) and Rantes (CCL5), were also differentially expressed in A20AEC-KO and control A20WT mice at day 8 post-infection, although at lower levels (S7A Fig). Immunohistochemical analysis of X-47-infected lung tissue confirmed enhanced CCL2 staining in CCSP+ lung epithelial cells from A20WT mice compared to A20AEC-KO mice (S7B Fig). To assess the importance of reduced CCL2 levels in A20AEC-KO mice for their protective phenotype upon influenza virus infection, we administered recombinant mouse CCL2 (rCCL2) to A20AEC-KO mice at day 6 post-infection. In contrast to control (PBS) treated A20AEC-KO mice, which again show protection from the virus challenge compared to wild-type littermates, rCCL2 treated A20AEC-KO littermates mice are sensitized to infection and no longer show differences compared to PBS or cCCL2 treated wild-type mice (Fig 4D). Together, these results show that the protection of A20AEC-KO mice from infection results from a reduced CCL2-dependent recruitment of innate immune cells to the lungs of infected mice.
A20 is an essential negative regulator of NF-κB signaling, and A20 deficient mice die prematurely due to massive multi-organ inflammation triggered by infiltrating intestinal bacteria [46,47]. We showed in this study that specific deletion of A20 in respiratory epithelial cells protects mice from Influenza A virus-induced morbidity and lethality. Viral clearance and the production of the antiviral cytokines IFNα and IFNβ was similar in A20AEC-KO and wild-type mice, in agreement with literature stating that epithelial cells are not the primary producers of type-I IFNs upon respiratory virus infection [48]. Interestingly, although the initial recruitment of innate cells and CTLs into the lungs of A20AEC-KO mice was sufficient to clear the virus by 8 days post infection, monocyte recruitment and the local CTL response in the lung were markedly reduced during later stages of infection. This was rather surprising since A20 is characterized as a negative regulator of the antiviral immune response [49]. Indeed we could confirm such a role for A20 in a surrogate viral infection model using intratracheal instillation of the double stranded RNA mimic poly(I:C) showing hyperactive immune responses in AEC-specific A20 knockout mice.
Protection of A20AEC-KO mice against influenza virus infection correlated with reduced recruitment of monocytes and CD11b+ macrophages to the lungs. In line with this, the levels of the monocyte-recruiting chemokine CCL2 were much lower in lungs of A20AEC-KO mice than in wild type mice on days eight and twelve after infection. It has been reported that administration of an MCP-1/CCL2 blocking antibody can reduce mortality and morbidity following influenza A virus infection [50]. Increased CCL2 levels have been reported in patients that had been infected with H7N9 virus [51]. CCL2 contributes to tissue immunopathology following influenza virus infection by its pro-inflammatory effects on macrophages and monocytes [52]. Recruitment of inflammatory macrophages in a manner dependent on the CCL2 receptor CCR2 contributes substantially to lung damage [53]. In line with this, we found that intranasal instillation of CCL2 in A20AEC-KO mice rendered these mice as susceptible to disease caused by influenza A virus as wild type mice. In aggregate, the reduced levels of CCL2 in A20AEC-KO mice could explain their increased tolerance to the infection. Yet after exposure of A20AEC-KO mice to double stranded RNA, a surrogate viral PAMP, we observed increased levels of CCL2 compared to wild type controls. The latter observation is in line with the regulation of CCL2 expression by NF-κB [54]. So what caused the reduced levels of CCL2 in the A20AEC-KO mice at later time points after influenza virus infection? It was remarkable that reduced disease, lower CCL2 levels and decreased CD8+ T cell activation coincided. CCL2 expression can be induced by IFNγ [55]. Therefore, the reduced numbers of IFNγ-producing CD8+ T cells could explain the lower CCL2 levels and monocyte infiltration in the A20AEC-KO mice. Similarly trans-presentation of TNF by influenza HA specific CD8+ T cells to lung epithelial cells has been shown to induce strong expression of CCL2 leading to extensive lung injury due to infiltrating monocytes. Increased CCL2 release by club cells might in addition recruit monocyte-derived inflammatory dendritic cells into the lungs, which could sustain local CTL responses and inflicting further damage to the lung [56,57]. Pro-inflammatory cytokine expression by club cells contributes to immunopathology during later stages of influenza infection [27]. Our study shows that expression of A20 in these cells might prevent lung damage by the host’s immune system. The exact molecular mechanism of how the specific deletion of A20 in club cells leads to increased tolerance to influenza infection is not understood at this moment and is subject for future research.
In conclusion, these data show that loss of A20 in respiratory epithelium can protect mice following Influenza A virus infection. In agreement with previous results, showing that deletion of A20 in myeloid cells also protects from Influenza [21], these data suggest that inhibiting A20 expression, for example by local administration of interfering RNAs, might be promising as a new therapeutic strategy to control disease caused by influenza A virus infection.
All experiments on mice were conducted according to the national (Belgian Law 14/08/1986 and 22/12/2003, Belgian Royal Decree 06/04/2010) and European (EU Directives 2010/63/EU, 86/609/EEG) animal regulations. Animal protocols were approved by the Ethics Committee of Ghent University (permit number LA1400091, approval ID 2010/001). All efforts were made to reduce suffering of animals. Before procedures mice were anesthetised by intraperitoneal (i.p.) injection of ketamine and xylazin.
Conditional A20 knockout mice harbouring two LoxP sequences flanking exon 4 and exon 5 (A20FL/FL) were generated as previously described [28] and were crossed with CCSP-rTA/(tetO)7-Cre mice to specifically delete A20 in airway epithelial cells (AEC) (provided by Dr. J. Whitsett, Cincinnati Children’s Hospital, USA) [29,58]. A20FL/FL/CCSP-rTA/(tetO)7-Cre triple transgenic offspring were fed doxycycline-containing food pellets to delete A20 expression (625 mg/kg, Special Diet Services). All experiments were performed on age- and sex- matched littermates. All mice used in this study carried the A20FL/FL and CCSP-rTA alleles. Those expressing the (TetO)7 allele were termed A20AEC-KO and those lacking this allele were considered as wild type controls (A20WT). Mice were housed in individually ventilated cages at the VIB-UGent Inflammation Research Center (IRC) in a specific pathogen-free animal facility.
After anesthetisation, mice received an intratracheal dose of 50 μg low molecular weight endotoxin-free poly(I:C) (Invivogen) or 0.5 μg recombinant mouse TNF (in house production) in 50 μl PBS. Six or 24 h after instillation mice were sacrificed and broncheoalveolar lavages (BAL) were collected for flow cytometric analysis and cytokine / chemokine analysis.
Mouse adapted influenza A X-47 (H3N2) and PR8 (H1N1) were propagated in MDCK (Madin-Darby canine kidney, ATCC) cells. After anesthetisation, mice were infected intranasally with X-47diluted in 50 μl PBS. For lethal and sublethal infections, mice received 2 X LD50 X-47 and 0.05 X LD50 X-47 or 0.17 X LD50 PR/8, respectively. Mice were euthanized when weight loss exceeded 25% of the initial body weight. Recombinant mouse CCL2 (R&D Systems, endotoxin levels <0.01 EU per μg of protein as measured by the LAL method) was administered intranasally at a dose of 50 μg/kg at day 6 post infection. Pulmonary viral titres were determined by median tissue culture infectious dose (TCID50) determination using MDCK cells. Lungs were homogenized in PBS using a Polytron homogenizer (Kinematica) and ten-fold serial dilutions of cleared lung homogenates were incubated on MDCK cells in DMEM supplemented with trypsin (1 μg/ml), 2 mM L-glutamine, 0.4 mM sodium pyruvate and antibiotics. After 5 days, 0.5% chicken red blood cells (RBC) were added to cell culture supernatant and end-point dilution hemagglutination was monitored. TCID50 titres were calculated according to the method of Reed and Muench [59]. HAI titres in serum of infected mice were determined as follows: serum was treated for 18 h at 37°C with receptor-destroying enzyme (RDE/Cholera filtrate, Sigma Aldrich) to remove sialic acids from serum proteins and prevent nonspecific inhibition of agglutination. RDE was then inactivated by the addition of 0.75% sodium citrate and heating at 56°C for 30 min. To remove sialic acid binding proteins, sera were cleared with 1/10 volume 50% chicken RBC. Titration was done by incubating a two-fold dilution series of sera with 4 HA units of X-47 for 1 h in 96-well U-bottom plate. An equal volume of 0.5% chicken RBC was then added and HAI titres were read after 30 min.
The method for club cell isolation was adapted from [60]. Lungs were inflated with 0.25% trypsin/HBSS and incubated in DMEM at 37°C for 20 min. After trypsin neutralisation with FCS, lungs were cut into small 1 mm3 sections and sequentially filtered without homogenisation through a 40 μm cell strainer. The cells were then placed into a humidified incubator at 37°C for 2 h to remove adherent cell populations (macrophages and fibroblasts). Club cells were specifically pelleted by centrifugation at 30 g for 8 min.
Lungs and mediastinal lymph nodes (MLN) were dissected into small 1 mm3 sections and incubated with collagenase type IV (1 mg/ml, Worthington) and DNase I (100 U/ml, Roche) at 37°C for 30 min. Subsequently, samples were filtered and homogenised through a 40 μm cell strainer. For BAL, tracheas were cannulated and lungs were flushed 4 times with HBSS with 1 mM EDTA. The first ml was treated with EDTA-free protease inhibitor tablets (Roche) and frozen separately for cytokine/chemokine analysis. After treatment with red blood cell lysis buffer (Sigma Aldrich), cells were stained with monoclonal antibodies directed against MHC-II (I-A/I-E) (M5/114.15.2), CD11c (N418) CD8α (53–6.7), F4/80 (BM8), CD62L (MEL-14), Granzyme B (NGZB) B220 (RA3-6B2) from eBiosciences and CD45 (30-F11), CD3ε (500A2), Gr1 (RB6-8C5), Ly6G (1A8), Ly6C (AL-21) CD11c (HL3), CD11b (M1/70), CD4 (GK1.5), CD8α (53–6.7), IFNγ (XMG1.2), TNF (MP6-XT22) and CD16/32 (2.4G2) from BD Pharmingen. Samples were acquired on a LSRII Cytometer (BD Biosciences) and analysis was performed using FACSDiva software (BD Biosciences).
Lungs were inflated with 4% paraformaldehyde. After 1 hour lungs were washed in PBS and embedded in paraffin. 5 μm thick tissue sections were cut from paraffin blocks. For immunohistochemistry, sections were dewaxed, dehydrated and incubated in Dako antigen retrieval solution, brought to boiling temperature and allowed to cool down for 2 hours. Endogenous peroxidase activity was blocked by immersing slides in 3% H2O2 for 5 min. Sections were blocked and permeabilized in 10% goat serum and 1% Triton X-100. Sections were incubated overnight at 4°C in blocking buffer with anti-M2 (in house preparation) and anti-CCSP (Abcam) antibody. Subsequently, slides were incubated with secondary antibody for 1 h (polymer HRP-labelled anti-mouse/rabbit/rat; Vector Laboratories) and peroxidase was detected by diaminobuteric acid (DAB) substrate (Sigma Aldrich). Tissue sections were counterstained with Mayer’s haematoxylin, rehydrated and mounted in Pertex mounting medium (Histolab). For A20 staining, sections were incubated for 60 min at 4°C in blocking buffer with anti-A20 (ProSci) antibody, and subsequently with HRP-labeled anti-rabbit IgG and FITC-labeled anti-HRP. For CCL2/MCP1 staining, sections were incubated for 30 min with anti-CCL2 (Abcam, clone ECE.2) antibody, followed by Cy3-labeled anti-rat antibody.
For in vitro peptide stimulation, 5 x 106 cells isolated from spleen or lungs of animals infected with X-47 were stimulated with different concentrations of NP-peptide (ASNENMETM, JPT peptide solutions). After 18h, brefeldin A was added to the cell culture and after 6 h cells were collected and stained for cell surface markers and IFNγ. In vivo intracellular staining for GrB, IFNγ and TNF from freshly isolated BAL and lungs was performed as described previously [61]. Mice were treated intranasally with 50 μg brefeldin A (Sigma Aldrich) 6 h before tissue collection and single cell suspensions from BAL and lungs were prepared in the presence of 3 μg/ml brefeldin A. Cells were stained for cell surface markers and intracellular molecules and fixed and permeabilized in Cytofix / Cytoperm (BD biosciences) according to manufacturer’s instructions. Activated CD8+ T cells were identified as CD3ε+, CD8α+ and CD62Llo. Live/Dead fixable aqua dead cell stain (Molecular Probes) was used to discriminate live from dead cells.
Spleens, mediastinal lymph nodes and bronchoalveolar lavage (BAL) fluid were isolated from infected and non-infected mice. Single cell suspensions were prepared in PBS containing 0.5% BSA. After removal of red blood cells by osmotic lysis, cells were stained with Aqua Live/Dead (Life Technologies), anti-CD16/CD32 (clone 2.4G2; Fc Block; Becton Dickinson Biosciences), anti-CD19 (clone 1D3; Becton Dickinson Biosciences), anti-CD3ε (clone 17A2; Becton Dickinson Biosciences), anti-CD8α (clone 53–6,7; Becton Dickinson Biosciences) and H-2Db ASNENMETM pentamer (ProImmune) for 25 minutes at 4°C. Prior to measurements, CountBright Absolute Counting Beads were added to each sample. Cells were measured on a BD LSR II cytometer (BD Biosciences) and analysed using FlowJo software (Tree Star).
Detection of CCL2 (MCP-1), CCL5 (Rantes), CXCL1 (KC), IL-6 and TNF in BAL fluid was performed using Multiplex immunoassay technology (BioRad). IFNγ, IL-10 (eBioscience) IFNβ and IFNα (R&D Systems and Invivogen) protein levels were determined by ELISA.
For Western blotting, cells were lysed at 4°C for 15 min in lysis buffer (200 mM NaCl, 1% NP-40, 10 mM Tris-HCl pH 7.5, 5 mM EDTA and 2 mM DTT) supplemented with protease and phosphatase inhibitors. Cell lysates were subsequently separated by SDS-PAGE and analyzed by western blotting and ECL detection (Perkin Elmer Life Sciences). Immunoblots were revealed with anti-A20 (Santa Cruz), anti-actin (MP Biomedicals) and HRP-linked anti-mouse (GE Healthcare) antibodies. For Southern blotting, genomic DNA was digested with BamHI yielding 6.5- and 13.5-kb fragments for A20 floxed and deleted alleles, respectively. DNA was separated on an agarose gel and transferred to a nitrocellulose membrane. Hybridization was performed with a 32P-labeled probe. A20 specific genomic PCR was performed using the following primers: 5’-CAC AGA GCC TCA GTA TCA TGT-3’, 5’-CCT GTC AAC ATC TCA GAA GG-3’ and 5’ GCA GCT GGA ATC TCT GAA ATC 3’.
Apoptosis was analysed by fluorescence microscopy using an in situ cell death detection kit (Roche). Caspase activity was measured by incubation of 25 μg tissue homogenate with 50 μM acetyl-Asp-Glu-Val-Asp-aminomethylcoumarin (Ac-DEVD-amc) (Peptide Institute, Osaka, Japan) in 150 μl cell-free system buffer (10 mM HEPES–NaOH pH 7.4, 220 mM mannitol, 68 mM sucrose, 2 mM NaCl, 2.5 mM KH2PO4, 0.5 mM EGTA, 2 mM MgCl2, 5 mM pyruvate, 0.1 mM PMSF, 1 mM dithiothreitol). The release of fluorescent 7-amino-4-methylcoumarin was measured for 50 min at 2-min intervals by fluorospectrometry at 360 nm excitation and 480 nm emission wavelength, using a Cytofluor device (PerSeptive Biosystems, Cambridge, MA). The maximal rate of increase in fluorescence was calculated (Δfluorescence/min). Lungs were prepared as described for flow cytometric analysis and cells were stained with antibodies directed against CD326 (EpCAM, eBioscience clone G8.8), CD45 and CD31 (eBioscience, clone 390). Prior to flow cytometric analysis cells were washed and incubated with Annexin V and propidium iodide in Annexin V binding buffer according to manufacturer’s instructions (BD Biosciences).
Data were analysed using GraphPad Prism Software. Results are expressed as the mean ± SEM. Statistical significance between experimental groups was assessed using an unpaired two-sample Student’s t test. Statistical significance of differences between survival rates was analysed by comparing Kaplan-Meier curves using the log-rank test.
|
10.1371/journal.pgen.1000843 | The Scale of Population Structure in Arabidopsis thaliana | The population structure of an organism reflects its evolutionary history and influences its evolutionary trajectory. It constrains the combination of genetic diversity and reveals patterns of past gene flow. Understanding it is a prerequisite for detecting genomic regions under selection, predicting the effect of population disturbances, or modeling gene flow. This paper examines the detailed global population structure of Arabidopsis thaliana. Using a set of 5,707 plants collected from around the globe and genotyped at 149 SNPs, we show that while A. thaliana as a species self-fertilizes 97% of the time, there is considerable variation among local groups. This level of outcrossing greatly limits observed heterozygosity but is sufficient to generate considerable local haplotypic diversity. We also find that in its native Eurasian range A. thaliana exhibits continuous isolation by distance at every geographic scale without natural breaks corresponding to classical notions of populations. By contrast, in North America, where it exists as an exotic species, A. thaliana exhibits little or no population structure at a continental scale but local isolation by distance that extends hundreds of km. This suggests a pattern for the development of isolation by distance that can establish itself shortly after an organism fills a new habitat range. It also raises questions about the general applicability of many standard population genetics models. Any model based on discrete clusters of interchangeable individuals will be an uneasy fit to organisms like A. thaliana which exhibit continuous isolation by distance on many scales.
| Much of the modern field of population genetics is premised on particular models of what an organism's population structure is and how it behaves. The classic models generally start with the idea of a single randomly mating population that has reached an evolutionary equilibrium. Many models relax some of these assumptions, allowing for phenomena such as assortative mating, discrete sub-populations with migration, self-fertilization, and sex-ratio distortion. Virtually all models, however, have as their core premise the notion that there exist classes of exchangeable individuals each of which represents an identical, independent sample from that class' distribution. For certain organisms, such as Drosophila melanogaster, these models do an excellent job of describing how populations work. For other organisms, such as humans, these models can be reasonable approximations but require a great deal of care in assembling samples and can begin to break down as sampling becomes locally dense. For the vast majority of organisms the applicability of these models has never been investigated.
| When studying natural populations, reasonable models of isolation, migration, and population growth should be applied to estimate the population structure of an organism [1]. Furthermore, it is also important to understand the way in which a species' population structure has been altered by anthropogenic disturbance. The population structure of domesticated organisms such as corn or rice are clearly drastically influenced by human intervention and provide extreme examples of how demographic processes can influence the genetic diversity and distribution of a species [2]–[6]. There are now few organisms whose habitat range does not coincide with human activity or for whom interference in their population structure is of little concern. The degree of impact humans have - be it on purpose or not - on the population structures of species that are not targets of domestication is unclear.
In this paper we present the results of a large scale study of the global population of Arabidopsis thaliana as an example of a natural organism that, like many others, exists in a predominantly continuous habitat that is much larger than the migration range of any individual, engages in sexual reproduction (with at least some regularity), and exists partially as a human commensal but serves no agricultural purpose.
We analyzed 5,707 plants collected around the globe (Figure 1) with 139 SNPs spread across the genome. These plants cluster into 1,799 different haplogroups with approximately three quarters of those haplogroups consisting of a single unique plant. Some haplogroups are represented by tens, or even hundreds, of individuals (Figures S1, S2, S3). One haplogroup was found over a thousand times across North America and another was found more than 200 times across the United Kingdom. Looking at the distribution of all pairwise genetic distances highlights three types of inter-plant relationships: they can be genetically identical (approximately 3% of all pairs in the sample, mostly pairs within North America), they can be completely unrelated plants given our marker resolution (approximately 85% of pairs in the sample, mostly inter-continental pairs or pairs within Eurasia), or they can show an intermediate degree of relatedness to each other (approximately 12% of pairs in the sample, mostly pairs with North America with very few inter-continental pairs) (Figure 2). Simulations demonstrate that these intermediate relations cannot be explained in a panmictic population and are therefore consistent with a more structured population.
Arabidopsis thaliana frequently reproduces by self-fertilizing and only occasionally outcrosses. The level of heterozygosity in the sample is therefore quite low compared to most organisms that obligately outcross. With self-fertilization and bi-parental inbreeding, we find that 95% of plants having five or fewer heterozygous loci. We estimated outcrossing rate in each field site from the distribution of number of heterozygous markers in each individual. As a whole our sample selfed 97% of the time overall in its recent history with the middle 50% of sites having estimates ranging from 95% to 99%. The estimates were lower in North American sites (Wilcoxon test p-value<0.005) which had an average of a 92% selfing rate and range of the middle 50% from 92% to 96% (Figure 3). Three sites had 0% selfing as their maximum likelihood estimates. These sites included 2, 3, and 5 plants (respectively). While the estimates are robust across loci (bootstrapping gives upper 95% confidence intervals of no more than 10% selfing for any of these sites), the small sample sizes may not be representative of the site as a whole. Most of the material used for this analysis was taken from seeds collected in the field or from mature plants grown under lab conditions from field-collected seed. As such there was a reduced chance for natural selection to influence the heterozygosity of the sample as it may have done had the seeds been allowed to grow to maturity under natural conditions. If inbreeding depression plays a significant role in A. thaliana [7],[8] the heterozygosity of a cohort of mature plants would be expected to be higher than the seed population from which it grows. Under these circumstances the effective selfing rate, the contribution to future gene pools from self-fertilized plants, could be somewhat lower than we estimate here. Differences in sample tissue composition between North American and Eurasian samples may contribute to the difference in estimated selfing rate between the continents.
While this level of selfing is high enough to greatly depress the individual heterozygosity of the sample, it is low enough to thoroughly mix haplotypes whenever two distinct haplotypes find themselves in close proximity. (Figure 4) shows the probability that two plants drawn from a given site are from a different haplogroup. Approximately 1/5th of sites are dominated by a single haplogroup (>80%). This includes nearly half the sites in North America but only 1/8th of Eurasian sites. The polymorphic field sites, however, are often quite variable and comprised of plants with unique haplotypes.
Looking at measures of similarities between pairs of plants as a function of geographic distance we see striking differences in pattern between pairs of Eurasian plants and pairs of North American plants. Figure 5B and Figure 6B show the strong broad trend of decay of genetic similarity with increasing geographic distance across Eurasia. The fraction of differing alleles rises to saturation across the continent and the probability of finding two plants of the same haplogroup becomes negligible beyond 1000 km. Panels A, showing effects of similar scale in North America, show extremely wide-spread haplogroups and little relation between distance and allelic similarity. The entire negative slope of Figure 6A can be explained by the distribution of haplogroups in Figure 5A. Figure 5C and Figure 5D are the same data on a smaller geographic scale. Figure 5D is similar to Figure 5B and show that Eurasia's isolation by distance continues in a smooth manner at this level of resolution. Figure 5C reveals that North American Arabidopsis thaliana does exhibit a measure of isolation by distance at this smaller scale though with a great deal more noise than in Eurasia. Figure 5E and 5F continue this trend at a very fine scale. Both continents exhibit isolation by distance at this level though the pattern is more pronounced in Eurasia.
When a species has established itself across a broad geographic range, migrates relatively slowly, and outcrosses with reasonable frequency, isolation by distance is an inevitable outcome. Every time a new haplotype migrates to a nearby area it recombines with the local haplotypes creating organisms of intermediate relatedness. Occasional long-distance migration events may have only weak effects on this continuum, as crossing and back-crossing with local haplotypes would dilute the impact. Aggressively invading haplotypes and selective sweeps can, however, strongly disrupt this process. Both can allow individual haplotypes to spread over much greater distances before being broken apart by the locally established haplotype pools. This is consistent with the pattern that has previously been identified in smaller studies of Arabidopsis thaliana within regions of Europe and Asia [9],[10].
A species newly introduced to a region is expected to have a different pattern. As the species spreads across its new range its migration events bring it to previously unoccupied areas. Without established local haplotypes there is no recombination, no intermediate genotypes are formed, and single, un-recombined haplotypes can spread uninterrupted over great distances. As the new range becomes filled with the species, however, isolation by distance will begin to establish itself, first on very local scales and gradually spreading out as recombination creates geographically unique haplotypes and migration and recombination between occupied areas blends them together. These patterns are consistent with our observations. In Eurasia, where Arabidopsis thaliana has flourished for thousands of years, it has established a strong gradient of isolation by distance. In North America, which has been colonized in the last three hundred years [11], haplotypes are spread across the entire continent but weak isolation by distance is emerging, particularly over shorter distances.
Arabidopsis thaliana is often a human commensal in both North America and Eurasia. The largest difference between its natural history on the two continents is that it has existed across Eurasia for thousands of years and in North America for only a couple of centuries. Human disturbance does not appear to have radically altered its natural population structure in Eurasia and the results suggest that the disturbance in North America is transitory and that a natural form of isolation by distance will emerge over time. This suggests that for organisms like Arabidopsis thaliana human disturbance only has a particularly large effect on population structure when established local populations are small or absent, or when an entire local gene-pool is replaced by artificial migrants. Otherwise, even moderate human disturbance can be swamped out by natural processes.
This kind of continuous isolation by distance is a type of population structure that the field of population genetics is poorly equipped to deal with. While there are several exceptions [12]–[16], most of population genetics theory is premised on the existence of discrete populations of exchangeable individuals. Even the modern field of landscape genetics [17]–[18] is focused on finding discrete regions within continuous habitats that behave like classic populations. Organisms like Arabidopsis thaliana, however, do not fit such models. With continuous geographic variation the probability of observing a particular set of alleles in an organism depends on the unique location of that organism and the alleles at the next closest organism are expected to have been drawn from a slightly different distribution. Sufficiently fine-scaled lattices of stepping-stone models may approximate many of the important features of this kind of structure, but it is not straightforward to determine the appropriate scale and having too coarse a scale may quickly degrade the numerical results (particularly for populations not at equilibrium) [19]. Hierarchical models are particularly inappropriate. The migration rate is low compared to the outcrossing rate, which very quickly (on a scale generally less than a kilometer) creates a geographic blend of alleles and extremely rich pools of local haplotypes. There is no bifurcating process to be uncovered (Figure S4, Figure S5, Text S1). To accurately estimate effective population size, gene flow, recombination, and natural selection in populations exhibiting continuous variation it will be necessary to reexamine the often over-looked theory of spatial genetics and develop new methods. A recent review of the subject [20] suggests several promising approaches.
For researchers using Arabidopsis thaliana as a model organism for ecological and evolutionary studies this paper provides several lessons and raises several new questions. One important point is that it is necessary to recognize that both genotype and environment are expected to vary spatially. Any study of local adaptation or gene by environment interaction should expect to find correlations between genotypes and environments simply through spatial correlation. Study design and analysis must take this into account and show that similarities between plants separated by a given distance within environments are greater than those at similar distances but between environments. Another point is that in terms of genetic diversity, Arabidopsis thaliana needs to be thought of as a sexually reproducing species: the difference between outcrossing and highly selfing organisms is quantitative rather than qualitative. Each plant in the wild may contain multiple hybrid siliques. While the vast majority of individual seeds are self-fertilized, the outcrossing rate is sufficient to introduce considerable genetic recombination after just a few generations. This will help make natural samples of Arabidopsis thaliana a powerful research subject for genome-wide association studies and linkage mapping [21], but create difficulties in reconstructing even fairly recent phylogeographic events such as the colonization of North America (let alone older events such as the re-colonization of Eurasia after the most recent ice age). Future studies using higher-density marker sets will have considerably more power to address these questions.
The collection is described in detail at http://arabidopsis.usc.edu/Accession/. It contains 4756 new accessions and 1201 accessions obtained from the Arabidopsis Biological Resource Center (ABRC) as a leaf from a single reference plant such that the distributed seed matches the genotype in this study. The collection spans 42 countries and four continents.
Genomic DNA was isolated using Puregene 96-well DNA purification kit (Gentra Systems) with the modified protocol [22]. All DNA samples were normalized to 10 ng/ul, and then genotyped using The Sequenom MassArray (compact) system at Sequenom (San Diego, CA) and University of Chicago DNA sequencing facility (Chicago, IL) with 149 SNPs. The primer sequences of the 149 SNPs and their physical and relative genetic distances are listed on the web (http://borevitzlab.uchicago.edu/resources/molecular-resources/snp-markers). They were selected from loci exhibiting minor allele frequencies between 25 and 30% in a set of globally-distributed DNA alignments [23] using MSQT [24].
Samples were removed if they contained excess missing genotype calls (>50 of 149) as this indicates poor quality of the genomic DNA or contamination. Information from ten SNP assays was removed due to excess missing genotypes or heterozygous calls (>25% of sample) which is often an indicator of poorly performing genotype assays. Haplogroups containing common lab strains Col, Ler, Ws2, and Nd were also removed to limit the chances of contamination. Multiple samples of each were found and at suspiciously broad global distributions.
Each plant was assigned to a single unique haplogroup. All plants in a haplogroup have haplotypes that are potentially identical given the number of SNPs genotyped and the accuracy of the SNP genotyping. Clusters are defined by a modified QT-clustering [25] algorithm. The distance function between two haplotypes is derived from the binomial probability of finding the observed number or more of marker mismatches between them given the number of observed markers. The first haplogroup is defined by finding the central haplotype around which it is possible to form the largest haplogroup. Haplotypes are proposed in order of their distance from the central haplotype and are included if their distance is less than 0.05 times the current size of the cluster. Once the largest haplogroup is defined it is removed from the sample and the next largest haplogroup is defined. This is iterated until every plant has been placed in a haplogroup. Heterozygous markers were treated as missing data.
To simulate the distribution of pairwise fraction of non-matching alleles we simulated a sample of 10,000 haplotypes. For each marker in each haplotype an allele was taken from the corresponding site of an observed haplotype randomly chosen with replacement. The simulation adjusted for production of identical haplogroups was done in the same manner, however only one representative of each haplogroup was included in the random sampling.
Selfing rates were estimated for 88 field sites with 8 from North America. These are all the sites for which the genotyped tissues were from plants that were from plants grown from field-collected seed (1820) or mature field-grown plants (219, all from North America) and for which there were at least two haplogroups present. Estimates were derived from the inbreeding coefficient FIS[26] in each field site as implemented in [27] http://lewis.eeb.uconn.edu/lewishome/software.html. The selfing rate is calculated as 2/(1/FIS+1). This relationship between FIS and the selfing rate assumes that outcrossing occurs uniformly across individuals within field sites and that the populations have reached equilibrium with respect to allele frequencies and heterozygosity. To the extent that mating is structured by within-field site geography our estimates will be slightly inflated from the true values.
|
10.1371/journal.ppat.1002021 | A Viral Satellite RNA Induces Yellow Symptoms on Tobacco by Targeting
a Gene Involved in Chlorophyll Biosynthesis using the RNA Silencing
Machinery | Symptoms on virus-infected plants are often very specific to the given virus. The
molecular mechanisms involved in viral symptom induction have been extensively
studied, but are still poorly understood. Cucumber mosaic virus
(CMV) Y satellite RNA (Y-sat) is a non-coding subviral RNA and modifies the
typical symptom induced by CMV in specific hosts; Y-sat causes a bright yellow
mosaic on its natural host Nicotiana tabacum. The Y-sat-induced
yellow mosaic failed to develop in the infected Arabidopsis and
tomato plants suggesting a very specific interaction between Y-sat and its host.
In this study, we revealed that Y-sat produces specific short interfering RNAs
(siRNAs), which interfere with a host gene, thus inducing the specific symptom.
We found that the mRNA of tobacco magnesium protoporphyrin chelatase subunit I
(ChlI, the key gene involved in chlorophyll synthesis) had
a 22-nt sequence that was complementary to the Y-sat sequence, including four
G-U pairs, and that the Y-sat-derived siRNAs in the virus-infected plant
downregulate the mRNA of ChlI by targeting the complementary
sequence. ChlI mRNA was also downregulated in the transgenic
lines that express Y-sat inverted repeats. Strikingly, modifying the Y-sat
sequence in order to restore the 22-nt complementarity to
Arabidopsis and tomato ChlI mRNA resulted
in yellowing symptoms in Y-sat-infected Arabidopsis and tomato,
respectively. In 5′-RACE experiments, the ChlI transcript
was cleaved at the expected middle position of the 22-nt complementary sequence.
In GFP sensor experiments using agroinfiltration, we further demonstrated that
Y-sat specifically targeted the sensor mRNA containing the 22-nt complementary
sequence of ChlI. Our findings provide direct evidence that the
identified siRNAs derived from viral satellite RNA directly modulate the viral
disease symptom by RNA silencing-based regulation of a host gene.
| Cucumber mosaic virus (CMV) Y satellite RNA (Y-sat) is an
interesting subviral RNA because it changes the green mosaic induced by CMV into
a bright yellow mosaic in Nicotiana tabacum. The molecular
basis underlying the induction of symptoms by viruses is not well understood,
and this Y-sat-mediated modification of symptoms has been a long-standing
mystery. In this study, we discovered the molecular mechanism involved in the
Y-sat-induced yellowing. First, we showed that transgenic N.
benthamiana plants that expressed the inverted-repeat
sequence of Y-sat also developed a yellow phenotype, similar to the
Y-sat-infected plants. Then, we found that tobacco magnesium protoporphyrin
chelatase subunit I gene (ChlI, the key gene involved in
chlorophyll synthesis) was downregulated in the transgenic plants and in the
Y-sat-infected plants. We then identified a 22-nt long sequence that is
complementary to the Y-sat including four G-U pairs in the ChlI
mRNA. Finally, we demonstrated that a short interfering RNA (siRNA) derived from
Y-sat specifically targeted and downregulated the ChlI mRNA,
thus impairing the chlorophyll biosynthesis pathway. This discovery of the
molecular basis of the symptom modification induced by Y-sat is the first
demonstration that a subviral RNA can induce disease symptoms by regulating host
gene expression through the RNA silencing machinery.
| Plants infected with viruses often display various symptoms, which can be very
specific to given viruses. Despite past efforts, the molecular bases underlying
virus-induced diseases symptoms are still poorly understood. Subviral non-coding RNA
molecules such as satellite RNAs (satRNAs) or defective interfering (DI) RNAs are
often associated with plant viruses and can modify the symptoms induced by helper
viruses [1], [2], [3]. Because such
subviral RNAs dramatically modify the symptoms induced by helper viruses, they are
potential tools for gaining insights into the molecular mechanisms of symptom
development.
SatRNAs of Cucumber mosaic virus (CMV) are dependent on helper
viruses for their replication and encapsidation and often attenuate the disease
symptoms induced by CMV. Specifically, Y-satellite RNA (Y-sat) modifies the symptoms
and exacerbates the pathogenicity of CMV in specific hosts; Y-sat induces a bright
yellowing of leaves of Nicotiana tabacum (the natural host) and
other related species (i.e., N. benthamiana),
which is yellower than a typical chlorosis, whereas it induces systemic necrosis on
tomato [4],
[5], [6], [7]. The sequence
domains on Y-sat, which are responsible for the symptom induction, have been
identified in our previous and several other reports [6], [7], [8], [9], [10]. We also suggested that a
single, nuclear-encoded, incompletely dominant gene in tobacco controls the
Y-sat-mediated yellowing in tobacco plants [11], but no such host genes have
ever been shown to be involved in the symptom modification nor has the molecular
mechanism been reported. An attractive model based on RNA silencing has been
suggested [2], [12], but the solid
experimental data are still needed.
RNA silencing is a conserved, sequence-specific gene regulation system, which has an
essential role in development and maintenance of genome integrity. RNA silencing
relies on short RNA (sRNA) molecules (21–24 nt), which are the key mediators
of RNA silencing-related pathways in almost all eukaryotic organisms [13], [14], [15]. In plants,
similar to other eukaryotic organisms, there are two main classes of sRNAs:
microRNAs (miRNAs) and short interfering RNAs (siRNAs), but the latter class
contains several different types [16], [17]. These sRNAs are produced from double-stranded RNA
(dsRNA) or from folded structures by Dicer-like (DCL) proteins and guide Argonaute
(AGO) proteins to target cognate RNA or DNA sequences [13], [18]. In higher plants, RNA silencing
also operates as an adaptive inducible antiviral defense mechanism. As a
counter-defense strategy, plant viruses have evolved viral suppressors of RNA
silencing (VSRs) [19] that interfere with the RNA silencing pathway at
different steps by binding to viral siRNA and/or dsRNAs or directly interacting with
AGO1 [20], [21].
Subviral RNAs such as satRNA and DI RNA of tombusvirus have been also used to
understand the roles of RNA silencing in viral replication and in symptom
development. The DI RNA-induced RNA silencing response is known to control the level
of helper virus, facilitating the long-term co-existence of the host and the viral
pathogen [20],
[22], [23], [24]. In addition,
progress in understanding plant antiviral RNA silencing has revealed cross
relationships between RNA silencing and viral pathogenicity. Recent studies suggest
the possibility that virus-derived siRNA (vsiRNA) could mediate virus–host
interactions through a shared sequence identity with the host mRNA, resulting in
silencing of the host genes and subsequent viral symptom development. A few
interactions between host mRNAs and vsiRNAs that resulted in the vsiRNA-guided
cleavages of host mRNAs have been experimentally shown [25], [26], although their roles in the
virus–host interaction have not been determined to date.
Magnesium (Mg)-chelatase is the key enzyme in chlorophyll biosynthesis, and three
subunits (ChlI, ChlH and ChlD) of the tobacco magnesium protoporphyrin chelatase are
required for the proper function of the enzyme [27]. Indeed, tobacco plants
defective for ChlI have the yellow phenotype [28], suggesting that
chlorophyll biosynthesis is impaired. The same yellow phenotype was observed when
the ChlI gene of tobacco or cotton was targeted by virus-induced
gene silencing (VIGS) [29], [30], [31]. Furthermore, an Arabidopsis mutant
defective for ChlI also had pale-green to yellow leaves [32]. Importantly,
the plants defective in the function of the Mg-chelatase enzyme had a very similar
yellow phenotype to plants infected with CMV and Y-sat. Thus, these results raised
the possibility that the ChlI is downregulated by Y-sat in the
virus-infected plants.
In this study, we show that transgenic N.
benthamiana plants develop a yellow phenotype when expressing
the inverted-repeat sequence of Y-sat, similar to the symptoms of the Y-sat-infected
plants. Moreover, we provided evidence that Y-sat targets the ChlI
gene using the host RNA silencing machinery in such a way that Y-sat-derived siRNAs
efficiently downregulate ChlI mRNA through RNA silencing-mediated
cleavage. Our findings strongly suggest that this yellow phenotype is the result of
a disorder in chlorophyll synthesis caused by the downregulation of the
ChlI gene.
To identify host genes involved in the Y-sat-induced symptom modification, we
created transgenic N. benthamiana plants that express the Y-sat
sequence, expecting the yellow phenotype to be induced without CMV as a helper
virus. We have used this strategy to avoid any effect of virus replication on
host gene expression, because virus infection itself has been shown to regulate
the expression of numerous genes [33]. We first created
transgenic plants that expressed the Y-sat sequence either in the sense or
antisense orientation, but these transgenic plants failed to have any phenotypic
changes (data not shown). However, when the Y-sat inverted-repeat (IR)
sequence-expressing cassette (Figure 1A) was introduced into N.
benthamiana plants, we observed that the transgenic
N. benthamiana lines (16c:YsatIR) had a
yellow phenotype (Figure
1B), although the yellow phenotype was less pronounced in the 16c:YsatIR
lines than in the Y-sat-replicating system. Of four transgenic lines that we
obtained, two had phenotypes with distinct yellowing; line 1 had vein yellowing,
and line 2 had a yellow mosaic. No yellow phenotype was observed on the
N. benthamiana that expressed dsRNA of GUS
(16c:GUSIR), demonstrating that the expression of dsRNA of an unrelated sequence
in the same Y-sat IR transformation cassette does not cause a yellow symptom
(Figure 1B and 1C). We
also confirmed the lack of viral contamination in the 16c:YsatIR lines by RT-PCR
using primers that are specific to CMV genomes (data not shown).
To identify putative plant genes responsible for the yellow phenotype, we carried
out microarray analyses of RNA extracted from the 16c:YsatIR plants (Text S1).
In 16c:YsatIR plants, 134 genes were significantly downregulated to levels that
are at least 40% lower than in their wild-type counterparts
(N. benthamiana 16c) (Table S1).
Among them, 31 genes were actually involved in chlorophyll biosynthesis and
chloroplast biogenesis (Table S1), further supporting the hypothesis
that the yellow phenotype could be the result of downregulation of the host
gene(s) involved in the biosynthesis pathway of chloroplast pigments. Indeed,
proteome analyses showed that several chloroplast-related proteins, such as
RuBisCo small subunit, RuBisCo activase and glyceraldehyde-3-phosphate
dehydrogenase were significantly affected in 16c:YsatIR plants (Text S1,
Figure
S1). More interestingly, the mobility of the RuBisCo small subunits
was shifted in a two-dimensional gel (Figure S1), indicating that the proteins had
been modified. All together, these results suggest that the expression of
chloroplast-related genes and subsequent synthesis of proteins were altered in
the 16c:YsatIR plants.
When we aligned the sequences of the 31 genes involved in chlorophyll
biosynthesis and chloroplast biogenesis identified by microarray analysis with
the Y-sat sequence, we found a high degree of sequence complementarity (22 nt in
a row including four G-U pairs) between the yellow-inducing domain of Y-sat
[7], [34] and the
tobacco magnesium (Mg) protoporphyrin chelatase subunit I
(ChlI) gene (accession AF014053). Because ChlI is a component
of the primary enzyme that catalyzes the first step in chlorophyll synthesis via
the tetrapyrrole biosynthesis pathway [32], this evidence encouraged us
to clone and sequence the ChlI gene of N.
benthamiana. We then found that both the
ChlI genes from N.
tabacum and N.
benthamiana had the 22-nt sequence complementary to the
Y-sat sequence (Figure 2A).
Hereafter, we called the 22-nt complementary sequence for the
ChlI gene and the Y-sat sequence as the yellow region (YR)
and satellite yellow region (SYR), respectively (Figure 2A). We then examined the mRNA levels
of the ChlI gene by Northern blot analysis and quantitative
real-time RT-PCRs in 16c:YsatIR and Y-sat-infected N.
benthamiana plants. The outputs of these analyses showed
that the ChlI mRNA was markedly downregulated in both plants
(Figure 2B and C) and
confirmed the results of the microarray analysis. To confirm that the
downregulation of the ChlI mRNA was due to the satRNA itself,
we further conducted a quantitative real-time RT-PCR using RNAs from
N. benthamiana protoplasts transfected
with the dsRNA of Y-sat. As controls, we transfected protoplasts with dsRNA of
three other CMV satRNAs; S19-sat, T73-sat [35] and CM-sat [36]. These
satRNAs are different from Y-sat in the corresponding SYR sequences and do not
induce any yellow phenotypes in tobacco plants [35]. As shown in Figure 2D, the
ChlI mRNA level was lower in protoplasts treated with dsRNA
of Y-sat than in those treated with dsRNA of the other satRNAs. In addition, the
mRNA level of another chloroplast-related gene, CAB3, decreased
in the Y-sat dsRNA-treated protoplasts (Figure 2D), confirming our findings from the
microarray analysis. In the proteome analysis, many chloroplast-related proteins
were affected in the transgenic 16c plants expressing Y-sat dsRNA; thus, it is
conceivable that the down-regulation of the ChlI gene caused a
decrease in other chloroplast-related genes expression in the Y-sat
dsRNA-treated protoplasts.
We next examined whether silencing of the ChlI gene using VIGS
can induce similar yellow symptoms in the absence of Y-sat. The 150-bp of
ChlI (817 to 966) was inserted into the two CMV vectors,
CMV-A1 and CMV-H1; CMV-A1 lacks the C-terminal one-third of the intact 2b
protein [37],
while CMV-H1 vector lacks the entire 2b protein [38] (Figure 3A). In the VIGS experiments, we used
a pseudorecombinant virus that contains RNA components derived from RNA1 and
RNA3 of CMV strain L to avoid the severe mosaic symptoms induced by CMV-Y.
N. benthamiana plants infected with either of the viral
vectors had systemic yellow symptoms similar to those induced by the replicating
Y-sat in the presence of the helper virus (Figure 3B). Although CMV-H1:ChlI150 induced
the yellowing more slowly than CMV-A1:ChlI150 in the early stage of infection,
the results of quantitative real-time RT-PCR confirmed that the
ChlI mRNA was downregulated in both CMV-A1:ChlI150- and
CMV-H1:ChlI150-infected N. benthamiana plants compared to
control plants infected with one of the empty vectors (Figure 3C). Using enzyme-linked immunosorbent
assay (ELISA), we confirmed that both pseudorecombinant viruses carrying the
inserted ChlI sequence replicated and accumulated to a similar
level in the systemic leaves at 14 days post-inoculation (dpi) (Figure 3D).
The ChlI genes of pepper, tomato and Arabidopsis
thaliana were obtained from the gene database, and the 22-nt
complementary sequences of the ChlI genes and Y-sat were
aligned (Figure 4A). Pepper
has the same YR sequence in the ChlI gene as those of tobacco
and N. benthamiana. Conversely, several
mismatches were found in the case of the tomato ChlI and
Arabidopsis ChlI (ChlI1 and
ChlI2) genes (Figure 4A). We next examined whether the Y-sat can induce yellow
symptoms on pepper, tomato and Arabidopsis plants. As expected,
infected pepper plants developed bright yellow symptoms (Figure 4B, right plant), whereas tomato
plants did not (Figure 4C,
right plant). By site-directed mutagenesis of the SYR, we generated three Y-sat
derivatives having the 22-nt continuous sequence complementary to the
corresponding YRs of tomato ChlI gene, Arabidopsis
ChlI1 and ChlI2 genes (Y-sat-Tom, Y-sat-Ara1 and
Y-sat-Ara2, respectively) (Figure
4A). When tomato plants were inoculated with the Y-sat mut-Tom and
the helper virus, yellow symptoms appeared at 10 dpi (Figure 4C, left plant). However, some of the
introduced mutations in individual plants had reverted to the original
nucleotides at 21 dpi. Notably, the Y-sat mut-Tom did not induce yellow symptoms
in N. benthamiana (Figure 4D, left plant). Similarly, when
Arabidopsis plants were infected with CMV-Y and Y-sat
mut-Ara1, yellow symptoms appeared (Figure 4E, right plant). On the other hand, Y-sat mut-Ara2 did not
induce yellowing (data not shown). The last observation is consistent with the
previous studies by Huang and Li [32], who reported that ChlI2 of
Arabidopsis has lower functionality than
ChlI1 due to a reduced level of expression. In addition,
like Y-sat mut-Tom, Y-sat mut-Ara1 did not induce yellowing in
N. benthamiana (Figure 4F). Quantitative real-time RT-PCRs
confirmed that the mRNA levels of the ChlI gene in the Y-sat
mutants-infected N. benthamiana plants were not downregulated,
unlike in the Y-sat-infected plant (Figure 4G). There were little differences in satRNA or viral
accumulation between Y-sat-infected- and Y-sat mut-Ara1-infected leaves of
N. benthamiana (Figure 4H and 4I), confirming that the Y-sat
mutant was replicated to a level similar to that of the original Y-sat in the
systemic leaves of N. benthamiana. These results, all together,
strongly suggest that a specific interaction between Y-sat and the
ChlI host gene is involved in development of the yellow
symptom.
Because Y-sat and the host ChlI gene seemed to have a specific
interaction through their sequence complementarity, we then examined the
possible involvement of RNA silencing in the Y-sat-mediated yellow phenotype.
First, we tested whether the Y-sat-derived siRNAs can be hybridized and detected
by ChlI mRNA probe. As shown in Figure S2,
sense siRNAs from Y-sat in both Y-sat-infected and 16c:YsatIR plants were
clearly detected in Northern blots using the ChlI sense RNA
probe. On the other hand, we failed to detect antisense siRNAs from Y-sat by
Northern blots using the ChlI antisense RNA probe. This result
seems reasonable because the YR and SYR sequences do not share complementarity
in the antisense orientation (Figure S2). In addition, we also detected
siRNAs derived from Y-sat mut-Ara1 using the Arabidopsis ChlI1
sense RNA in Northern blots. As shown in Figure S3, 351-bp Arabidopsis
ChlI1 sense RNA probe, which contains the 22-nt sequence
complementary to Y-sat mut-Ara1, detected the siRNAs of Y-sat mut-Ara1 in the
Arabidopsis leaves infected with CMV and Y-sat mut-Ara1.
Assuming that the yellow symptoms are the result of post-transcriptional RNA
silencing of host genes directed by Y-sat specific sequences, we further
analyzed Y-sat-derived siRNAs profile to find whether Y-sat siRNAs targeting the
ChlI mRNA accumulate in the Y-sat-infected plants. We thus
conducted small RNA deep sequencing to map the small RNAs on the Y-sat sequence.
As the result, Y-sat-derived siRNAs covered almost the entire Y-sat sequence,
and the majority of Y-sat siRNAs accumulated in the sense orientation in the
Y-sat-infected plants. In addition, 21-nt and 22-nt siRNAs were abundant among
the Y-sat small RNAs populations (Figure 5A). Y-sat-derived siRNAs in both sense and antisense
orientation were non-uniformly distributed along the sequence with a few small
RNA-generating hot spots (Figure
5A). Abundant siRNAs were accumulated from the regions around
positions 100, 180, 211 and 280 on the Y-sat. Northern hybridization confirmed
that the most abundant siRNAs were generated from the region at positions
1–200 as opposed to 201–369 (Figure S4). Furthermore, we found abundant
siRNAs homologous to the SYR (Figure 5B). The accumulation of siRNAs corresponding to SYR in
16c:YsatIR and Ysat-infected plans was confirmed by Northern hybridization using
LNA probes specific to SYR of Y-sat (Figure 5C). In deep-sequencing analysis, we
also identified the ChlI siRNAs in the Y-sat-infected tissues
although the amounts were not very high (Figure S5). The profile of the
ChlI siRNAs revealed a very unique feature; all siRNAs
derived from ChlI were generated only from the 3′ region
downstream of the cleavage site as described below.
To clarify whether the ChlI mRNA is cleaved in the
Y-sat-infected plant, we analyzed the 5′ ends of the cleaved mRNA products
with a 5′-RACE assay. Sequencing of the 5′-RACE products revealed
two distinct cleavage sites in the YR of the ChlI mRNA. Almost
all identified cleavage sites were mapped at the middle position in YR (between
890 and 891), which agrees with the expected cleavage site(s) driven by the
21-nt and 22-nt siRNAs (Figure
6A). To verify that Y-sat can direct sequence-specific cleavage, we
created a GFP sensor construct in which the 3′ non-coding region contained
the 22-nt YR sequence (Figure
6B). The construct was delivered by agroinfiltration into
Y-sat-infected N. benthamiana leaves that had bright yellow
symptoms (Figure 6B). GFP
accumulation was monitored using UV light after agroinfiltration. As shown in
Figure 6B, GFP
fluorescence was reduced in the Y-sat-infected tissues, and this observation was
supported by the results of quantitative real-time RT-PCR of the
GFP mRNA (Figure 6C). The accumulation of GFP protein was also reduced in the
Y-sat-infected tissues (Figure
6D). These results clearly demonstrated that the 22-nt YR sequence in
the sensor mRNA was sufficient for the sequence-specific downregulation of
GFP-YR mRNA in Y-sat-infected tissues.
Plant RNA silencing has often been implicated as a molecular mechanism for symptom
induction caused by viruses or viral subviral agents. Viral suppressors of RNA
silencing (VSRs) are able to compromise the endogenous RNA silencing pathways [19], [20], and these
virus-encoded silencing suppressors have also been identified as pathogenicity
determinants. Indeed, virus-induced developmental abnormalities are often explained
by the interference of virus-encoded VSRs with host miRNAs involved in the
developmental processes [39], [40]. However, no explanation for specific symptoms caused by
VSRs has ever been confirmed nor has any report explained the molecular basis for a
specific viral symptom including yellowing and necrosis. In recent studies, host
mRNAs were identified as potential targets of siRNAs and miRNAs in virus-infected
tissues, and several have been proved to be downregulated [25], [26]. For example, Moissiard and Voinnet
[25]
demonstrated that the RCC1 gene in Arabidopsis
infected with Cauliflower mosaic virus (CaMV) was downregulated by
virus-derived siRNAs, but contrary to expectations, the decrease in gene expression
did not affect either viral accumulation or symptoms. It is, in fact, quite
difficult to clarify the relationship between such small RNAs and viral
pathogenicity although the idea that host gene silencing against a particular gene
might contribute to the specific expression of symptoms is very attractive.
In the present study, we have shown that siRNAs derived from Y-sat induced bright
yellow mosaics on tobacco by specifically targeting mRNA of the host
ChlI gene, resulting in the inhibition of chlorophyll
biosynthesis. Here we provide several lines of evidence that Y-sat-induced bright
yellow mosaics are the outcome of specific interference between the pathogen-derived
siRNAs and a host gene. First, the 22-nt long region of Y-sat (SYR) produces
specific siRNAs that were complementary, including four G-U pairs, to the 22-nt long
region of tobacco ChlI mRNA (YR). Second, the ChlI
mRNA could detect Y-sat-derived siRNAs in Northern blots. Third, 5′-RACE
experiments revealed that the ChlI mRNA was cleaved exactly in the
expected middle of the YR. Fourth, the levels of the ChlI
transcript significantly decreased in both Y-sat-infected plants and the transgenic
plants expressing Y-sat dsRNA. Fifth, the Y-sat mutants that had the modified SYR to
either Arabidopsis ChlI1 mRNA or tomato ChlI mRNA
were able to induce yellow symptoms in these host plants. In contrast, these
modified Y-sat lost the ability to induce yellow symptoms on tobacco. Sixth, the GFP
sensor construct carrying the YR sequence was specifically targeted in
Y-sat-infected plants. Considering all these results, we propose a model that
explains that the Y-sat-mediated yellow symptom results from the cleavage of host
ChlI mRNA by RNA silencing machinery (Figure 7).
In deep-sequencing analysis, we found abundant Y-sat-derived siRNAs in the
Y-sat-infected N. benthamiana. Furthermore, we noticed that the
ChlI-derived siRNAs also accumulated in the Y-sat-infected
tissues although the amounts were not very high. The profile of the
ChlI siRNAs was very unique because all siRNAs derived from
ChlI were generated only from the 3′ region downstream of
the cleavage site (Figure S5). Importantly, spread of RNA silencing beyond the targeting
site in endogenous plant genes has not been shown [30], [41], [42], except for
trans-acting siRNAs [43]. Whether secondary siRNAs can be
generated from the ChlI mRNA after vsiRNA-directed cleavage, and
whether such secondary siRNAs are involved in the downregulation of the
ChlI gene still need careful studies.
Here we propose that Y-sat caused the yellow symptoms on tobacco by directing
post-transcriptional RNA silencing against the ChlI mRNA. However,
yellow symptoms appeared much brighter in Y-sat-infected plants than in 16c:YsatIR
plants (Figure 1B). With regard
to the observation, the amount of Y-sat-derived siRNAs in 16c:YsatIR plants was
lower than in Y-sat-infected plants (Figure 5C), probably leading to different yellow phenotype between
16c:YsatIR plants and Y-sat-infected plants. Indeed, the level of the
ChlI transcript analyzed by the Northern blot was higher in the
16c:YsatIR plants than in the Y-sat-infected plants (Figure 2B). Alternatively, as suggested by Du et
al. [44], Y-sat siRNAs
from secondary structures (T-shaped hairpins) may predominate over the Y-sat siRNAs
generated from perfect dsRNA forms. Thus it is likely that RNA silencing against
ChlI and subsequent yellow phenotype can vary depending on the
qualities and amounts of siRNAs derived from satRNA.
In conclusion, we discovered the molecular basis of the symptom modifications induced
by Y-sat: the involvement of RNA silencing mechanism in the pathogenicity of Y-sat.
But the molecular mechanism underlying the synergistic and/or antagonistic
interaction between satRNAs, helper viruses and host plants still remain to be
explored. In addition, the origin(s) of satRNAs, their evolutionary strategy and
biological significance have long been intriguing topics. Since the original
isolation of Y-sat in Japan more than 30 years ago [4], no other satRNAs that induce
yellow mosaics on tobacco have been isolated in the world, suggesting that Y-sat is
a rare satRNA that specifically induces yellow mosaics on tobacco. We have observed
that Y-sat cannot compete with other similar size satRNAs [35], and thus Y-sat may survive
through a different strategy from other satRNAs; the Y-sat-induced yellowing of
leaves, which could preferentially attract aphids (the vectors of CMV and its
satRNAs), may have favored the transfer of CMV that harbors Y-sat during the its
evolutionary history.
Nicotiana benthamiana, Capsicum annuum,
Solanum lycopersicum and Arabidopsis
thaliana were used as host plants for the analysis.
Nicotiana benthamiana line 16c having a single copy of the
GFP transgene [45] was obtained from Dr. D. Baulcombe (Sainsbury
Laboratory, UK) and was also used for the analysis. All plants were grown in a
plant growth room with a 16-h light/8-h dark at 24°C and 50% relative
humidity.
Transgenic N. benthamiana lines expressing the inverted repeat
(IR) of Y-sat were generated by transforming N. benthamiana 16c
with the binary vector pIG121-Hm carrying the IR of Y-sat under the CaMV 35S
promoter. In the sense and antisense orientations, the 317-bp (53 to 369) Y-sat
sequence (GenBank accession D00542) was inserted in the pJM007 vector [46], then the
inverted repeat (IR)-expressing cassette was transferred to a Ti-plasmid vector,
pIG121-Hm. The Ti plasmid vector containing the IR (1004 nt) of the GUS sequence
(GUS-IR) was previously constructed [47].
CMV strain Y (CMV-Y) was used as a helper virus for satellite RNA. To induce gene
silencing to the ChlI gene, we used two CMV-based vectors,
CMV-A1 and CMV-H1. CMV-A1 and CMV-H1 are derived from RNA2 of CMV-Y, and CMV-A1
lacks the C-terminal one-third of the intact 2b protein as a consequence of
introducing a multiple cloning site [37], while CMV-H1 vector lacks
the entire 2b protein [38]. The 150-bp of the ChlI gene (817 to
966) was inserted into the CMV vectors to create CMV-A1:ChlI150 and
CMV-H1:ChlI150, respectively. To avoid severe mosaic symptom induction by CMV-Y,
we used a pseudorecombinant virus that contains RNA components derived from RNA1
and RNA3 of CMV strain L together with RNA2 of the vector. Each plasmid
containing a full-length cDNA clone of RNA1 to RNA3 was transcribed in vitro
after linearization with a restriction enzyme [37]. Infectious viruses were
then created by mixing transcripts of RNAs 1 to 3. For virus propagation, leaves
of 4-week-old plants of N. benthamiana were
dusted with carborundum and rub-inoculated with the RNA transcripts. For
inoculation of tomato plants, leaves of young plants were rub-inoculated with
the sap from virus-infected tissues of N.
benthamiana. Successful systemic infection with the virus
containing the full insert sequence was confirmed by RT-PCRs. Viral accumulation
was examined by conventional ELISA [48] using the antibodies raised
against the CMV CP.
Total RNAs were extracted by either a conventional phenol/chloroform method [47] or a method
using Trizol reagent (Invitrogen) following the manufacturer's
instructions. The N. benthamiana ChlI clone
including the entire ORF was amplified by RT-PCR using the primer pair designed
from the tobacco ChlI sequence (5′-GCTCTAGAATGGCTTCACTACTAGGAAC-3′ for forward
primer, 5′-GCCCAAGCTTAGGCGAAAACCTCATAAAATTTC-3′ for
reverse primer). Quantitative real-time RT-PCR was performed essentially as
described before [37]. Primers for quantitative real-time RT-PCR for the
N. benthamiana ChlI gene were as follows:
5′-CTTATTGGTTCGGGTAATCCTG-3′ for forward primer
and 5′-GCTGAGTCGATTTGGTTCTG-3′ for reverse primer.
The N. benthamiana actin gene was amplified using
5′-GCGGGAAATTGTTAGGGATGT-3′ for forward primer
and 5′-CCATCAGGCAGCTCGTAGCT-3′ for reverse primer
and used for data normalization. Northern blot hybridization was performed
essentially as previously described [49]. Specific probe for the
ChlI gene was generated by PCR with the PCR DIG Probe
Synthesis Kit (Roche Diagnostics) to amplify the 371 bp (634 to 1004) of
3′-terminal regions of the ChlI gene using the primer
pair ChlI-634F (5′-GAGCCTGGTCTTCTTGCTAAAGC-3′) and ChlI-1004R
(5′-GCTGAGTCGATTTGGTTCTG-3′). In the Northern
blots of the small RNAs corresponding to the 22-nt complementary sequence region
(satellite yellow region, SYR), the SYRs were detected by using
32P-labeled locked nucleic acid (LNA) oligonucleotide probes
described previously [50].
The ChlI mRNA cleavage sites were analyzed by modified
RNA-ligase mediated 5′-RACE [51]. Total RNA (10 µg) was
purified using the MicroPoly(A) Purist Kit (Ambion), then the fractionated
Poly(A)+ mRNA was ligated to the GeneRacer RNA Oligo adaptor
using the GeneRacer Kit (Invitrogen). Ligated RNAs were reverse transcribed
using the gene-specific reverse primer for the ChlI gene,
ChlI-1004R (5′-GCTGAGTCGATTTGGTTCTG-3′). The 5′end of
the cDNA was then amplified by PCR using the GeneRacer 5′ primer and the
gene-specific reverse primer used for the reverse transcription for the first
PCR. The GeneRacer 5′ nested primer was also used for the subsequent
nested PCR. The amplified product from the nested PCR was excised from
1.2% agarose gel and cloned into pGEM-T Easy (Promega) for
sequencing.
Protoplasts were prepared from leaves of N.
benthamiana as described before [52]. The dsRNA of four satRNAs
(Y-sat, S19-sat and T73-sat [35] and CM-sat [36]) were used. DsRNA of satRNA
was prepared by in vitro transcription using a PCR-amplified fragment containing
the T7 promoter sequence as described previously [52]. The prepared protoplasts
were transfected with the satRNA dsRNAs (2 µg) in a PEG–calcium
solution as described [52] and then incubated for 20 h. Total RNA was extracted
from the harvested protoplasts with Trizol reagent (Invitrogen), and the mRNA
levels of the ChlI and CAB gene were measured
by quantitative real-time RT-PCR (mean ± SE;
n = 3). Primers for quantitative real-time
RT-PCR for the CAB gene were 5′-CGGCCGATCCAGAAACTTT-3′ for forward primer
and 5′-GCCCATCTGCAGTGAATAACC-3′ for reverse
primer.
Total RNA was extracted from CMV and Y-sat-infected N.
benthamiana plants. Small RNAs were isolated essentially as
described [49]
and submitted to Hokkaido System Science (Sapporo, Japan), where deep-sequencing
analysis was performed on an Illumina Genome Analyzer using the standard
protocol of the manufacturer. The 18–45-nt small RNA reads were extracted
from raw reads and aligned with the Y-sat sequence using the program SOAP [53] to search for
perfectly matched sequences.
The GFP-YR sensor gene was inserted between the BamHI and
SacI sites in the pBE2113 vector. The Ti-plasmid construct
was then introduced into Agrobacterium tumefaciens KYRT1
strain, which was supplied by Dr. G. B. Collins (University of Kentucky, USA).
Agrobacterium infiltration was carried out essentially as described [49].
Total proteins were extracted from the sample tissues by grinding in Laemmli
buffer, separated by SDS-PAGE, and /transferred onto a PVDF membrane (Immobilon,
Millipore). Anti-GFP antibodies were purchased from Roche and used at a
1∶1000 dilution. For immunostaining, an alkaline phosphatase-conjugated
goat anti-rabbit antibody was added to the blots at a 1∶3000 dilution
followed by colorimetric development with BCIP and NBT.
|
10.1371/journal.pcbi.1005483 | Nucleotide-time alignment for molecular recorders | Using a DNA polymerase to record intracellular calcium levels has been proposed as a novel neural recording technique, promising massive-scale, single-cell resolution monitoring of large portions of the brain. This technique relies on local storage of neural activity in strands of DNA, followed by offline analysis of that DNA. In simple implementations of this scheme, the time when each nucleotide was written cannot be determined directly by post-hoc DNA sequencing; the timing data must be estimated instead. Here, we use a Dynamic Time Warping-based algorithm to perform this estimation, exploiting correlations between neural activity and observed experimental variables to translate DNA-based signals to an estimate of neural activity over time. This algorithm improves the parallelizability of traditional Dynamic Time Warping, allowing several-fold increases in computation speed. The algorithm also provides a solution to several critical problems with the molecular recording paradigm: determining recording start times and coping with DNA polymerase pausing. The algorithm can generally locate DNA-based records to within <10% of a recording window, allowing for the estimation of unobserved incorporation times and latent neural tunings. We apply our technique to an in silico motor control neuroscience experiment, using the algorithm to estimate both timings of DNA-based data and the directional tuning of motor cortical cells during a center-out reaching task. We also use this algorithm to explore the impact of polymerase characteristics on system performance, determining the precision of a molecular recorder as a function of its kinetic and error-generating properties. We find useful ranges of properties for DNA polymerase-based recorders, providing guidance for future protein engineering attempts. This work demonstrates a useful general extension to dynamic alignment algorithms, as well as direct applications of that extension toward the development of molecular recorders, providing a necessary stepping stone for future biological work.
| This work demonstrates a necessary computational tool for the development and implementation of molecular recorders, a promising potential technique for massive-scale neuroscience. Molecular recorders use proteins to encode levels of a substance we want to measure (e.g. calcium in neural applications) as detectable changes in a linear cellular structure, e.g. misincorporations in a strand of DNA, or fluorescent proteins traveling down a microtubule. This encoding represents levels of the measured substance over time, much like a ticker tape represents information linearly on a strip of paper. The unique intracellular nature of this approach promises a significant scaling advantage over current techniques. The molecular recording approach suffers a particular drawback involving timing: unlike most methods of recording signals, in simple molecular recording systems we do not observe when each data point was recorded. This timing information is almost always required in order to make associations between our recorded data and the rest of the experiment. In this work, we propose a method to estimate the timing of these data points using easily-observable experimental measurements. We demonstrate the application of this method in a simulated neuroscience paradigm, investigate the effect of experimental design on this method, and determine protein properties that would be desirable in molecular recorders. These findings are useful both as a computational proof-of-concept, and as guidelines for current efforts to engineer proteins for molecular recording.
| As we seek to understand complex questions in neuroscience, we are increasingly interested in the feasibility of massive-scale methods for neural recording [1–5]. One such proposed method is molecular recording, which uses engineered DNA polymerases (DNAPs) to encode information about neural activity onto a newly synthesized DNA strand, such that the position in the DNA sequence corresponds to the order and approximate timing of recorded events [6–8]. Rather than reading out neural activity from an electrode or photodiode during an experiment, molecular recorders would store neural activity intracellularly. This information would not be read out in real-time, but post-hoc using high-throughput DNA sequencing. The recording DNAPs could be genetically encoded and selectively expressed in neurons, allowing us to obtain activity records from large populations of cells. DNAP-based recording techniques promise an inherently ultrahigh-scale neural recording technique, building off of advances in biotechnology and computational power. However, significant hurdles remain in realizing such a technology.
While molecular recorders promise massive-scale neural recording, they do not inherently provide all the data obtained using current recording techniques. With current techniques, e.g. electrical or optical recording, data about the timing of each sample is recorded alongside the desired recording. With DNAP-based recorders, we sample data using DNA sequencing, which occurs after an experiment has concluded. That is, without any inherent clocking mechanisms, the output from molecular recorders lacks any explicit timing information about what it recorded. Without timing information, recorded neural activity cannot be interpreted in the context of other signals observed during experiments, e.g. movement or delivered stimulus. The central problem here is that we do not know which nucleotides were written at which times, i.e. we cannot link our representation of neural activity to things we observe in the outside world. Thus, the timing of data from molecular recorders must be inferred or estimated before it can be useful to understand the brain.
Due to the stochasticity inherent in DNAP activity (or that of any protein), it is difficult to predict when a nucleotide was incorporated de novo. Uncertainty in timing estimates result in uncertainty about the underlying signal; without timing information, signal estimates become highly inaccurate, providing at most a few seconds of reliable recording under common conditions [7]. However, if we observe experimental data that should be correlated with neural activity during our experiments, we can generate predictions of what possible patterns of neural activity we might observe given that data. This, in turn, can provide some information about the timing of nucleotide incorporations: if we see a particular pattern of activity in our DNA-based record, the DNA was likely written by a neuron whose tuning would generate a similar activity pattern in response to the experimental variables we observe, and at a time where the neuron would have generated that pattern. If we enumerate the ways in which we believe a neuron could respond to the observed experimental variables in question, we can search for the most-likely response given the DNA-based record we observe. It is worth stating that this type of approach is not model-free, and there are many situations where this assumption of a tuning model is inappropriate, i.e. in areas of the nervous system that we either model poorly or do not know what form a model would take. However, in areas where we have reliable modeling approaches or seek to evaluate particular models, a model-based approach may be able to provide considerable insight.
One way to utilize these models to estimate timing is the one we use here: generate predictions of neural activity with known timing using observed experimental variables, then find the globally most-similar alignment between those predictions and our recorded data. This class of alignment problems is frequently found in the time series analysis domain, e.g. in speech or signature recognition [9–11]. Dynamic time warping (DTW) is an efficient solution to this class of alignment problems, determining the optimal alignment between the template and signal using dynamic programming principles. With a probabilistic interpretation, DTW allows us to infer the most likely timing of a signal with respect to a given template, as well as determining the most likely template from a set of possible templates [12]. These properties make DTW-class algorithms uniquely suited for the determination of signal timings for molecular recorders.
Given that we are interested in applying this algorithm to massive-scale datasets, we are immediately interested in algorithms that can efficiently harness large-scale computing resources. As DTW is a dynamic algorithm, with successive steps depending on previous calculations, it is difficult to apply asynchronous computing approaches, at least on an algorithm level. Thus many, though not all, parallel approaches to DTW have largely focused on task-level parallelism rather than parallelizing cost computation [13–18]. As a result, for computationally-intensive individual alignments, it tends to be difficult to fully utilize the massively parallel computing resources that are becoming more common. A highly-parallelized dynamic alignment algorithm would be useful for a number of reasons.
Here we describe a parallelized step-pattern variant of DTW with applications to the analysis of molecular recorder output. We demonstrate the algorithm’s ability to accurately determine incorporation times for single DNA strands generated by a simulated molecular recorder, compensating for the timing issues inherent in protein-based molecular recorders. We demonstrate the utility of this algorithm in practice through simulated neuroscience applications, and use simple simulated experiments to explore how DNAP parameters such as speed and error rate affect the accuracy of our timing estimates. Through proposal and application of this algorithm, we present findings relevant for current biological research into molecular recording.
Our algorithm solves a problem central to interpreting molecular recorder output in the context of neural recording: it aligns a single DNA-based record to an estimate of neural activity. We evaluate the local likelihood of each nucleotide being written at any time within some recording window given some assumed neural and DNAP properties. Then, using a dynamic programming-based technique, we attempt to find a global alignment given the local likelihoods and a prior defined by the DNAP kinetics. This algorithm is similar in structure to Dynamic Time Warping, utilizing a modified step pattern that reflects certain biological realities (See Algorithm Methods,S1 Fig). The step pattern limits the possible search space by enforcing these constraints: 1) nucleotides cannot be aligned to the same time point, 2) nucleotides can only be aligned to one time point, and 3) there can be a variable amount of time between incorporation of two adjacent nucleotides. We weight the potential options from this step pattern so that alignments made more likely by DNAP kinetics are favored. Notably, this approach enables significant algorithm parallelism, emerging from the constraint that nucleotides can only be aligned to one time point. As there are no dependencies between possible alignments of a given nucleotide, we can calculate the costs of all possible alignments of a given nucleotide concurrently.
In order to demonstrate the utility of this algorithm, we apply our technique to simulated output of molecular recorders (Fig 1A), demonstrating various aspects of algorithm performance as well as exploring the ability of DNAPs to encode neural information. The general experimental pipeline consists of four parts: (1) simulation of a molecular recording experiment (Fig 1B and 1C), (2) alignment of single recorder outputs to a set of time-indexed expected DNAP error rates, which represent potential neural tunings to observed experimental covariates, (3) selection of a template that best matches the molecular recorder output (Fig 1D), and (4) inference of neural parameters using time-aligned DNA-based signals (see Methods).
We simulate a biologically-inspired generative model with several parts: (1) an explicit parameterized model of how neural activity either depends on a stimulus or results in observed behavior (Neural Tuning), (2) how this neural activity modulates DNAP error rate, via Ca2+ concentration or other mechanisms (DNAP Tuning), and (3) a probabilistic description of DNAP kinetic properties, e.g. incorporation rate and pausing (DNAP Kinetics). This generative model can be parametrized using existing knowledge about neural and polymerase properties where known. In this paper, we use DNAPs with optimistic DNAP error tuning, i.e. maximum error rates higher than many DNAPs with incorporation rates suitable for recording, but with otherwise-realistic properties [19–21]. We also assume knowledge of these system characteristics (apart from neural tuning) in order to parametrize the alignment algorithm.
Given simulated DNA output and a time-varying input to the system, we iterate over potential neural tunings to find a tuning that provides an alignment most consistent with the observed DNA-based signal. We then use this maximum a posteriori alignment to generate a time-indexed DNA signal, and use this signal to infer neural parameters. We evaluate algorithm performance both by accuracy of timing estimation, i.e. how many seconds estimated incorporation times differ from true incorporation times on average, and accuracy of inferred neural tuning parameters, i.e. how the estimated behavior of a neuron differs from the true neural behavior. Specifically, to evaluate accuracy of timing estimates, we examine the root-mean square deviation (RMSD) between the estimated timings and the true incorporation times for a given alignment.
There is a highly non-linear relationship between alignment “success” and timing accuracy, as nearby alignments do not necessarily have similar likelihoods. Thus, we provide both a mean and median value for timing accuracy when those values differ by a large amount. To evaluate tuning accuracy, we estimate tuning parameters from the aligned DNA data and examine the distance between the algorithm-estimated parameters and those derived directly from the recorded neural data, which we treat as ground-truth for these studies.
Before exploring algorithm applications, it is worth exploring the performance implications of this approach. It bears mentioning again that, while they do not calculate the same cost function, our algorithm and traditional DTW are closely related; both are dynamic programming algorithms with effective worst-case complexity of O(NT) where N and T are the lengths of the two inputs being aligned. As we have mentioned, our algorithm has significant differences in implementation that allow it to be substantially parallelized; this allows for substantial performance increases using modern computing devices (See Algorithm Methods). While a naïve implementation of our algorithm performs more slowly than traditional DTW for a given set of inputs, parallelized implementations substantially outperform traditional DTW (S1 Fig). We observe up to a 16x speedup over traditional DTW when using a GPU-based implementation of our algorithm on a personal computer, and up to a 5x speedup when using a CPU-based implementation.
The feasibility of a “ticker tape” DNA-based recording scheme depends heavily on the properties of the DNAP used. For instance, the length of records (in base pairs) influences how much information is contained about neural activity, and thus impacts algorithm performance. Similarly, the speed, pausing, and fidelity properties of the DNAP used influence the information about neural activity contained in a DNA-based record [7]. Here, we look to determine the effect of these properties on the accuracy of our algorithm, and thus the expected performance of a molecular recording setup. Determining these effects allow us to form guidelines as to what kinds of DNAPs would be required for successful recording and alignment.
We use an entirely-simulated experiment here, i.e. we fully know the tuning linking stimulus to neural activity. This allows us to isolate the effects of DNAP properties on alignment from the effects of inaccurate neural activity estimates. We simulate a neuron with a linear response to an artificial stimulus; we deliver random levels of stimulus in 5s blocks over the course of 2000s (~30 minute recording window), and simulate the neuron’s spiking activity and intracellular calcium. We then simulate the output of a molecular recording system during that time period. We then align the molecular recorder output to the true stimulus signal. Using this simulation, we can focus on error induced by the DNAP and alignment algorithm in isolation.
We aim to estimate nucleotide incorporation timings, as well as the strength of the neuron’s tuning to the stimulus, i.e. the slope of the neuron’s tuning curve. The best alignments possible under this scheme have timing error up to the size of the stimulus features (5s); alignments with timing error less than this are generally considered to be accurate. Error with respect to tuning parameter is presented as a proportion of the true parameter. Except for the DNAP parameter being varied, the simulated DNAPs are identical (~100 Hz, mean pause duration of 2s; see Methods).
As record length increases, finding a randomly generated pattern that resembles the record becomes less likely, and alignment to a unique site should become easier. However, from a biological perspective, generating longer sequences may be more difficult, requiring polymerases with specialized properties, e.g. high processivity, high activity, or strand-displacement activity. Thus, it is useful to know minimal record lengths for successful alignment. When we increase record length in our simulations, we indeed find a resulting decreasing timing error. Generally, we find that records with length longer than 2.5K basepairs align with <5s median timing error (Fig 2A and 2B). Interestingly, we find that slope estimation is relatively constant regardless of record length, suggesting that, while record length is crucial to timing estimation, information about neural tuning in the record is not necessarily absent in shorter records (Fig 2C).
DNAP speed effectively changes the sampling rate of our system; if we have a slow DNAP, we can record for longer periods of time for a given strand length, but also record less information about any given interval. If we are interested in longer time-scale phenomena (e.g. environmental sensing, medical diagnostics) [22], we may wish to use slow DNAPs. However, due to the low sampling rate, we may not be able to recover useful information about timing and tuning in a neural paradigm. In our simulated stimulation paradigm, we find that slower DNAPs in fact increase timing accuracy (Fig 2D). However, median timing error stays relatively constant as speed decreases, implying that slow DNAPs simply decrease the amount of extreme timing errors we observe (Fig 2E). This runs parallel to our observations about record length; aligning to a longer time-indexed template is easier than aligning to a short one. However, our accuracy in determining tuning parameters decreases as we use slower DNAPs (Fig 2F). This indicates that we should, in general, be using fast DNAPs if we are interested in recovering tunings [19]. Meanwhile, slower DNAPs can provide longer records for a given strand length at the expense of diluting the information they carry about underlying phenomena.
Another property of DNAPs that can affect the quality of recordings is the transfer function relating analyte (e.g. calcium) concentration to error rate, f(·). We have modeled f(·) as a sigmoid with three parameters:
f(C)=Rmax⋅11+exp[b(C−C0)]
(1.1)
where C0 denotes the [Ca2+] that leads to half-maximum error rate, b denotes the steepness of the response curve, and Rmax denotes the maximum error rate of the DNAP. When selecting (or engineering) DNAPs to record with, we will need to optimize over these parameters. Here, we analyze DNAPs with varying transfer function slopes b, i.e. varying sensitivities to [Ca2+], ranging from step-like DNAPs to DNAPs with a wide dynamic range. We find that DNAPs with moderate sensitivities to [Ca2+] provide the most accurate timings, while both step-like and overly shallow transfer functions decrease alignment accuracy (Fig 2G and 2H). We find similar results for parameter estimation (Fig 2I), where appropriately-sloped DNAP tunings provide better estimates of neural parameters than DNAPs that are either too insensitive (low |b|) or too step-like (high |b|) with respect to [Ca2+]. This adds evidence to an assumption many investigating molecular recording techniques have been working under: DNAPs will have to be tailored in order to achieve optimal recording of even simple signals.
We are also interested in how the maximum error rate Rmax affects alignment accuracy. This is of particular interest from a biological perspective: many natural DNAPs with incorporation rates suitable for high-resolution recording have low error rates. It is useful to understand what minimal error rates would be feasible for molecular recorders, as well as examine system performance as Rmax scales. Here, we consider DNAPs that have near-zero error rates at low [Ca2+], and increase to some maximum error rate Rmax under high [Ca2+] conditions. We find that alignment accuracy increases as maximum error rate increases (Fig 2J and 2K), as expected. Interestingly, we find that parameter estimation is relatively insensitive to Rmax. Again, this seems to suggest that while timing accuracy tends to degrade with unfavorable DNAP parameters, molecular recorder output tends to retain information about underlying neural tuning.
Here, we demonstrate the feasibility of molecular recorders in a conventional neuroscience experimental paradigm. We analyze single-unit neural data recorded from M1 and pre-motor cortex during a center-out reaching task in a rhesus macaque, estimating the preferred movement directions of recorded neurons (data obtained from the DREAM reaching experiment database, see Flint 2012 for details [23–25]). We use the recorded spikes as the basis for simulated calcium transients and molecular recorder output. We also generate a set of estimates of neural activity from the kinematic data recorded during the task, with estimates representing velocity-tuned neurons with preferred directions distributed uniformly on [0, 2π]. Here, we use eight activity estimates as alignment templates. We apply our alignment algorithm to this data, aligning the molecular recorder output to each of the estimates, then selecting the maximum-likelihood alignment. The result, an estimated mapping of nucleotides to time, allows us to generate tuning curves for the recorded neurons. From this, we can estimate neural tuning parameters and infer how neural activity is modulated with respect to the recorded kinematics (details in Methods). The alignments here encounter alignment- and DNAP-based error, as in the previous section, but also encounter biology-based error when estimating neural activity from kinematic data. Thus, these experiments serve as an estimate of molecular recorder performance in a real-world scenario.
Using a plausible set of DNAP parameters (~100 Hz incorporation rate, mean pause duration of 2s, ~17% of time spent paused; see Methods for further details), we find that we are generally able to recover rough timing estimates and accurate tuning parameters from the simulated molecular recording experiment. As an initial demonstration, we examine several neurons that exhibit high firing rates and significant directional tuning (Fig 3A). Under these conditions, we are able to estimate nucleotide timings to within an average of ~15s (95% confidence intervals for average trial RMSD: [10.0,16.5], [12.1,20.3], and [14.8,22.5] seconds, Fig 3B). While timing accuracy is lower than desired, particularly for experiments that require sub-second precision using current techniques, these alignments still allow us to generate the estimated neural tuning direction θ* with error of ~10% (average errors of 0.5, 0.3, and 0.3 radians, Fig 3C). Median timings are substantially better than average timings across the board (95% confidence intervals for median trial RMSD: [3.8,7.2], [3.1,8.7], and [6.5,13.7] seconds).
Some of the error we encounter when generating alignment estimates may stem from our discrete parametrization of neural tunings. That is, we may not provide an estimate of neural activity similar enough to the true activity in order to generate accurate alignments. We can examine the contributions of this effect to algorithm accuracy by supplying a neural activity estimate generated using the neural tuning estimated from electrophysiology data, the best possible estimate we can provide given a particular model. Indeed, if we supply a neural activity estimate generated using the ground-truth neural preferred direction in our motor control experiment (rather than the 8 naïve preferred directions), we substantially reduce both timing error and error in θ* (S2 Fig). While we do not know the true preferred direction a priori and this kind of analysis could not be performed in practice, this suggests that a large portion of observed error can be attributed to the discrete parametrization of the search space. Increasing the resolution of the search space should improve alignment accuracy at the expense of execution time.
We apply our algorithm to each neuron in the dataset, examining aggregate performance over a population of recorded neurons. We find that the technique has middling performance on the whole dataset, only able to estimate timings to within 24s for 12% of neurons recorded (S3 Fig). If we limit the set of analyzed neurons to those that have substantial reach-modulated activity (model pseudo-R2 > 0.05, firing rate λ > 20 spikes/s), this improves to 47%. We are able to estimate preferred direction to within ±0.2π (±36°) for 39% of the dataset; this improves to 59% of the reach-modulated neurons (S3 Fig). While this filtering does not explain all observed error, it is useful when reconciling the results for individual neurons in Fig 3 with the larger dataset. This improvement upon filtering for active, well-modeled neurons demonstrates two things: 1) this method performs poorly on sparse-firing neurons, and 2) this method performs poorly on neurons that are not well-described by the set of models we consider. Both of these shortcomings are as expected given the algorithm. The former can be addressed by evaluating average neural activity represented by a DNA-based record, which can be done in a naïve, model-free manner. The latter, an inability to align signals that we cannot already model accurately, remains a shortcoming of this approach when attempting the interpretation of molecular recorder output.
We also analyze recording systems with a hypothetical DNAP that exhibits no pausing, but is otherwise identical to the previous DNAPs (see Methods). When examining the same neurons as above, we find drastically decreased timing errors (RMSD 95% CIs of [0.17,0.18], [0.31,0.39], and [0.47,3.0] seconds) and parameter estimation errors (average errors of 0.1, 0.2, and -0.04 radians, S4 Fig). Using these highly optimized DNAPs, we approach the timing resolution that would seem to be useful for high-precision neuroscience experiments, and retain high-accuracy prediction of neural tunings. A conclusion from this analysis is that much of the error we observe with our technique resolves when DNAPs behave more regularly. These results are of particular interest to us because of their biological implications: DNAP pausing generally has both DNAP-based and sequence-dependent components, and can be ablated using sequence context, chemical, or temperature-based means [19,26,27]. This significant improvement in both timing accuracy and parameter estimation suggest that decreasing DNAP pausing through these or other methods could be a useful approach to improve the accuracy of molecular recording systems.
We observe that errors in tuning parameter estimation in our simulated reaching experiments are not always normally distributed; rather, in a number of neurons, there appear to be several preferred directions that alignments converge upon, including peaks at a neuron’s anti-tuned direction (Fig 3C). This effect persists, although less prominently, when using a non-pausing DNAP (S4 Fig). This is useful to consider given the underlying center-out task in our experiment, where subjects reach in a direction then immediately make a reach back to the center, i.e. the opposite direction of the initial reach. It seemed possible that pathologic alignments could arise from this repetitive temporal structure, where alignments to tuned and anti-tuned templates are effectively identical save for a time-lag. Disrupting this structure through appropriate experimental design could lead to improved accuracy.
We generated a dataset composed of shuffled 2-second-long patches of neural and kinetics data such that the temporal structure of the original dataset was disrupted. We find that shuffling the data can both reduce selection of anti-tuned preferred directions (Fig 4A and 4B), as well as decrease overall tuning estimation error (Fig 4C). However, it is important to note that the shuffling scheme we describe here does not improve alignment for all neurons, and can even disrupt alignment of neurons that are otherwise predicted correctly (S5 Fig). While this argues against naïve shuffling as a universal strategy, it further demonstrates the effect of an experiment’s temporal structure on alignment accuracy. These findings suggest that experimental design cognizant of alignment-based analysis can improve robustness to pathologic alignments, and thus the feasibility of molecular recording-type experiments.
We describe an algorithm that generates estimates of nucleotide incorporation times for a molecular recording system, along with estimates of parameters that characterize the underlying recorded system. We improve upon naïve estimates of these values by incorporating observed experimental data along with a probabilistic description of recorder properties. We apply the algorithm to simulated neuroscience experiments, demonstrating the viability of this algorithm (and the general molecular recording scheme) in a number of scenarios. Our findings suggest that single-strand molecular recording is statistically feasible in neuroscience contexts. Further, by introducing experimental information into our estimation techniques, we improve upon previously-understood limits on the technique. Single-strand recording promises to be a useful technique in neuroscience and biology in general for a number of reasons; establishing a statistical framework for the interpretation of those signals is an important step towards the realization of this technology.
This algorithm is computationally novel, as it incorporates dynamic programming, probabilistic inference, and biologic constraints into a single framework. We modify existing DTW approaches to signal alignment, constraining our action space to physiologically possible actions (e.g. two nucleotides cannot be incorporated at the same time), as well as incorporating beliefs about DNAP kinetics. These constraints have a convenient property in that they restrict our action space to a set that can largely be calculated independently, allowing for parallelization of a dynamic algorithm. While the algorithm maintains the same approximate time complexity of traditional DTW (worst-case of O(NT)), its inherent parallelism can lead to dramatically decreased runtime.
Further, while not discussed at length here, if recording start or end times are known, variance of incorporation times scale with N assuming a Poisson-like DNAP. Path-constraint techniques could take advantage of this property, reducing effective worst-case time complexity to O(N12T) and allowing further speed increases [10,28]. These speed improvements are of particular importance due to the inherently large scale of molecular recording: if we want to record from hundreds-of-thousands to millions of neurons, the computational techniques necessary to interpret these signals should scale well.
To this end, there are a number of different biological methods that could be used to explicitly mark the start or end of molecular recorder output, e.g. by delivered analyte pulses or by optogenetic manipulation. These methods could also be used to provide time-coding throughout an experiment, making timing inference substantially easier. Similarly, designing behavioral tasks to modulate neural activity at levels significant enough to be detected, but low enough not to alter behavior, e.g. temporally modulating the brightness of visual stimuli, could be used as an implicit time-coding technique. These experimental methods for encoding timing information into molecular records can work alongside algorithmic alignment methods to improve both timing and parameter inference.
This work also has implications on current work in the biological space. It is useful to understand the effects of DNAPs with different behaviors (e.g. speed, error rate) on the ability to record information, both for our application to molecular recorders, as well as for other approaches that aim to record continuous signals intracellularly. Understanding the general space where recorders work (or fail) is useful not only for determining what kinds of DNAPs we need to find or design, but also for determining which kinds of phenomena might be amenable to study using molecular recorders.
There are many ways in which existing DNAPs already satisfy the requirements necessary for a single-strand biological recorder, e.g. processivity, speed, calcium-sensitive error rates, and pausing kinetics [19,26,29]. The one property that we have not observed in DNAPs is a calcium-sensitive error rate at physiological concentrations [20]. Further, natural DNAPs tend to be either fast or error-prone, but not generally both; the highest error rates we see in high-incorporation-rate DNAPs are at the low end of what we simulate here [21,30]. In order to develop practical molecular recorders, we will both need to understand how to substantially increase DNAP error rates in processive, high-speed DNAPs, as well as develop a scheme to make DNAP error rates calcium-sensitive at physiologically relevant scales. Alternatively, schemes that do not rely on calcium-tuned error rates, but rather modulate other DNAP properties via calcium, may provide an easier way forward.
While many caveats apply to this work, and to the prospect of molecular recorders in general, the results described here are helpful on a number of fronts. On a technical side, we describe a DTW-class algorithm that applies generally to point processes with variable temporal indexing. The algorithm is designed to allow probabilistic interpretation of its output, and can be used to find maximum a posteriori alignments to a set of known templates. We provide a highly-parallelized implementation of this algorithm which leverages advances in asynchronous computing techniques. With respect to molecular recorders, we provide a framework for interpretation of recorder output in the face of uncertain recording times. We also provide guidance to the ongoing research that looks to engineer DNAPs for this kind of recording. Perhaps most importantly, we have shown that, should a DNAP with certain properties be developed, we can provide temporal indexing to its output and capture neural behaviors using a molecular recording approach. While this is purely a simulation study, our work sets constraints and goals for the development of DNAPs for massive-scale neural data recording, and outlines experimental scenarios for their successful use.
This technique is intended to align a DNA-based recording with no temporal indexing to a longer, time-indexed estimation of calcium activity, a template. It assumes the DNA sequence as a binary “error”/”no error” code, then assesses the similarity of that sequence to a discrete-time continuously-valued estimate of neural activity, the template, via alignment. We use a novel DTW-class algorithm to perform this alignment, incorporating beliefs about DNAP kinetics to limit the space of potential actions.
|
10.1371/journal.ppat.1007590 | PML nuclear body-residing proteins sequentially associate with HPV genome after infectious nuclear delivery | Subnuclear promyelocytic leukemia (PML) nuclear bodies (NBs) are targeted by many DNA viruses after nuclear delivery. PML protein is essential for formation of PML NBs. Sp100 and Small Ubiquitin-Like Modifier (SUMO) are also permanently residing within PML NBs. Often, large DNA viruses disassemble and reorganize PML NBs to counteract their intrinsic antiviral activity and support establishment of infection. However, human papillomavirus (HPV) requires PML protein to retain incoming viral DNA in the nucleus for subsequent efficient transcription. In contrast, Sp100 was identified as a restriction factor for HPV. These findings suggested that PML NBs are important regulators of early stages of the HPV life cycle. Nuclear delivery of incoming HPV DNA requires mitosis. Viral particles are retained within membrane-bound transport vesicles throughout mitosis. The viral genome is released from transport vesicles by an unknown mechanism several hours after nuclear envelope reformation. The minor capsid protein L2 mediates intracellular transport by becoming transmembranous in the endocytic compartment. Herein, we tested our hypothesis that PML protein is recruited to incoming viral genome prior to egress from transport vesicles. High-resolution microscopy revealed that PML protein, SUMO-1, and Sp100 are recruited to incoming viral genomes, rather than viral genomes being targeted to preformed PML NBs. Differential immunofluorescent staining suggested that PML protein and SUMO-1 associated with transport vesicles containing viral particles prior to egress, implying that recruitment is likely mediated by L2 protein. In contrast, Sp100 recruitment to HPV-harboring PML NBs occurred after release of viral genomes from transport vesicles. The delayed recruitment of Sp100 is specific for HPV-associated PML NBs. These data suggest that the virus continuously resides within a protective environment until the transport vesicle breaks down in late G1 phase and imply that HPV might modulate PML NB assembly to achieve establishment of infection and the shift to viral maintenance.
| Promyelocytic leukemia (PML) nuclear bodies (NBs) are often targeted and reorganized by DNA viruses to counteract their antiviral activity. Human papillomavirus (HPV) also associates with PML NBs after infectious entry. While PML protein is required for nuclear retention and efficient transcription of incoming HPV genomes, Sp100, another PML NB component, was identified as a restriction factor. HPV virions are delivered to the nucleus during mitosis while continuously residing in membrane-bound transport vesicles. L2 protein directs trafficking via its carboxyl terminus by becoming transmembranous in the endocytic compartment. Herein, we demonstrate that PML protein associates with viral particles still residing in transport vesicles after nuclear delivery, possibly to provide a continuous protective environment after disruption of the membrane bilayer of the transport vesicle. In contrast, Sp100 recruitment is delayed specifically for PML NBs forming around HPV particles, suggesting that HPV transiently modulates PML NB composition. In contrast to large DNA viruses, which encode factors capable of reorganizing PML NBs, HPV seems to take advantage of the disassembly occurring at the onset of mitosis. As such, it utilizes well-established cellular pathways to orchestrate the regulation of viral transcription during the immediate early events of the viral life cycle.
| Promyelocytic leukemia (PML) nuclear bodies (NBs) are highly dynamic nuclear structures that have been associated with numerous cellular processes, including apoptosis, transcriptional regulation, and innate and intrinsic immune responses [1]. While their size and number of residing proteins vary according to the cell condition, the main component of PML NBs is PML protein [2]. It is present in seven isoforms that constitute the main scaffold of PML NBs, with the exception of PML VII that lacks a nuclear localization signal and remains in the cytosol [3]. PML protein is required for the formation and stability of PML NBs, as cells knocked down for PML protein fail to form these structures [4]. SUMOylation of PML protein with SUMO-1 and SUMO-2 is necessary for this process and SUMOylated PML proteins then recruit other PML NB-residing proteins that are either SUMOylated themselves or contain SUMO interacting motifs [2,4,5,6,7]. Transcriptional repressors Sp100 and Daxx are two additional proteins that permanently reside in PML NBs [1]. PML NBs are modified during cell cycle progression [1,8,9]. They disassemble upon the onset of mitosis and PML protein forms large aggregates in the cytosol, also referred to as mitotic accumulations of PML proteins (MAPPs). MAPPs do not contain Sp100 or Daxx but only de-SUMOylated PML protein and are recycled after completion of mitosis. Following nuclear envelope reformation, released PML protein molecules are translocated back into the nucleus to form new PML NBs in the daughter cell nuclei and recruit other proteins.
Despite extensive research on the role of PML NBs, their specific function in the cell remains unclear. However, they have been shown to be involved in innate and intrinsic immunity as both repressors of viral gene expression and coregulators of the type I interferon pathway [10]. Consequently, many DNA viruses, such as herpes simplex 1 (HSV-1), human cytomegalovirus (HCMV), simian virus 40 (SV40) and adenovirus 5 (ADV5), target PML NBs during primary infection and induce the reorganization and degradation of the residing proteins, including PML protein and Sp100 [11–15]. Specifically, HSV-1 and ADV5 encode early immediate proteins, ICP0 and E1A-13S, respectively, which target PML protein isoforms for degradation and thus enhances viral gene expression [15,16]. Similarly, HCMV targets Sp100 for degradation through its immediate early protein IE1 to prevent their transcriptional repression activity and enhance the early stages of infection [14]. It is thought that PML NBs are sensors of DNA/protein complexes and are thus recruited to virally-induced foci [17,18]. In the case of HSV-1 infection, PML NBs have recently been shown to be recruited to incoming HSV-1 genomes following nuclear delivery [18,19]. Furthermore, high-resolution microscopy showed that PML NB-residing proteins engulfed viral genomes shortly after nuclear entry. As SUMOylation and SUMO interaction are critical for the formation and dynamics of PML NBs, HSV-1 ICP0 is thought to target PML protein through SUMO interaction or recognition of their SUMO-1 conjugation motifs [16,20].
Similar to most other DNA viruses, papillomaviruses (PVs) associate with PML NBs at several stages of their life cycle. PV genomes along with minor capsid protein L2 have been observed to associate with PML NBs after infectious delivery into the nucleus of target cells. L2 protein also localizes to PML NBs in natural productive lesions, although transiently, and when over-expressed in cell culture [21–24]. However, while PML NBs restrict gene expression of most viruses, PVs, such as bovine papillomavirus 1 (BPV1) and human papillomavirus (HPV) types 16 and 18, have been shown to require PML protein for efficient transcription [25–27]. Transcription driven by both PV and heterologous promoters delivered by PV particles was repressed in the absence of PML protein, suggesting that PML protein does not function in a promoter-specific manner [25]. In addition, Sp100 was shown to restrict HPV18 transcription and replication [26,28]. These findings suggest that PML NBs may play an important role in the regulation of the PV life cycle.
Following infectious entry, the HPV capsid uncoats within acidified endocytic vesicles. This is facilitated by host cell cyclophilin which allows for the partial dissociation of the major capsid protein L1 from the minor capsid protein L2, which remains in complex with the viral genome [29–31]. While most of the L1 protein appears to be degraded in the late endosome, a subset of L1 protein, likely arranged as capsomeres, remains associated with the viral genome [30–32]. L2 protein assumes a transmembranous configuration, which is promoted by a newly described chaperone function of γ-secretase [33]. A putative transmembrane region spanning from residues 45 to 65 separates a small luminal domain from the large carboxy-terminal region that can interact with cytosolic factors, including the machinery mediating retrograde transport along microtubules (MTs) towards the trans-Golgi network (TGN) [34,35]. Prior to associating with PML NBs, the HPV genome needs to be delivered to the nucleus. Rather than utilizing nuclear pores and the nuclear import machinery, HPV takes advantage of nuclear envelope breakdown during mitosis to gain access to the nucleus [36,37]. HPV-harboring vesicles likely rely on L2 protein, which retains its transmembranous configuration during vesicular mitotic transport, to interact with motor proteins, such as dynein and kinesins, for transport along astral and spindle MTs, respectively [32,38–40]. Surprisingly, the viral genome resides within the transport vesicle in the nucleus for several hours after completion of mitosis and nuclear envelope reformation, resulting in delayed transcription when compared to delivery of naked DNA [25,30].
More recently, we reported that viral pseudogenomes delivered by HPV16 particles were lost after successful nuclear delivery in the absence of PML protein in the spontaneously immortalized HaCaT keratinocytes but not in HPV18 transformed HeLa cells [27]. Viral genome loss in HaCaT cells was prevented by inhibitors of the Jak/Stat signaling axis, although transcription was not restored. These findings pointed towards a protective role of PML protein in the immediate early stages of the HPV life cycle. Thus, we were prompted to pose the following questions: 1) when does the association of viral genome with PML protein and other PML NB-residing proteins occur; 2) which viral factors may play a role mediating this interaction during infection; and 3) whether viral genomes target preformed PML NBs or rather PML NB-residing proteins are recruited to incoming HPV genomes. Given that DNA successfully delivered to the nucleus by HPV particles is lost in the absence of PML protein, we hypothesized that PML protein is recruited to HPV-harboring transport vesicles prior to release from this membrane-bound environment and that likely L2 protein is mediating this association. L2 protein harbors a SUMO conjugation domain on its N-terminus, as well as one highly conserved SUMO interactive motif (SIM) and two additional putative SIMs on its C-terminus, which we hypothesized might be involved in recruitment of PML protein [24,41]. We also hypothesized that the recruitment of Sp100 is delayed and occurs after release of the viral genome from the transport vesicle. Herein, we utilized differential staining of viral pseudogenomes in combination with high-resolution immunofluorescence to determine the spatio-temporal recruitment of PML protein, Sp100, and SUMO-1 to incoming viral genomes during infectious entry and establish an order of events following nuclear delivery of HPV genomes [30,32].
To investigate the order of events following nuclear delivery of viral genomes, we needed to estimate the amount of time that has passed throughout mitosis and interphase. To achieve this, we observed the morphology of the nucleus and the localization of PML protein by immunofluorescent microscopy (Fig 1). During mitosis, PML protein forms MAPPs that are observed around mitotic chromosomes at all stages of mitosis (Fig 1A) [8,9,42]. Using DAPI staining, we define late telophase as cells exhibiting decondensing DNA and reforming nucleus, as well as the presence of PML protein aggregates (Fig 1B). As MAPPs translocate back into the nucleus after nuclear envelope reformation, early interphase cells harbor large cytosolic aggregates of PML protein and a few, typically small, PML protein foci in the newly formed nucleus. The number and size of MAPPs decrease, while the number and size of PML protein foci inside the nucleus increase, throughout interphase. To estimate how much time has elapsed after the completion of mitosis, we counted the number of nucleoli, as it is inversely correlated with time, a method we have used in our previously published work [30,43]. Therefore, we estimated that 7+ nucleoli are present in the nucleus of early interphase cells, while late interphase cells are characterized by 1–6 nucleoli.
Although the association of HPV genomes with PML protein has been known for decades, how it occurs is still unclear [22]. It has been assumed that incoming viral genomes are targeted to preformed PML NBs rapidly after nuclear delivery [22,25], despite the now known dynamics of PML protein [8,9]. However, in HSV-1 infection, PML protein was shown to be recruited to and engulf incoming viral genomes following nuclear entry by high resolution microscopy [19]. Furthermore, our previously published findings suggest that PML protein provides a protective environment for the viral genome [27]. Therefore, we wanted to determine whether PML protein is recruited towards viral genomes or vice versa and to visualize the architecture of this association. We acquired high resolution images of HaCaT cells infected with HPV16 pseudovirions (PsVs) harboring an EdU-labeled pseudogenome after immunofluorescent staining. Images are z-stacks combined with 3D reconstruction for PML protein (green) and EdU-labeled viral pseudogenomes (red) (Fig 2A). While previous confocal microscopy could only show EdU puncta adjacent to PML protein [25,27], high resolution microscopy allows us to observe the structure of PML protein in association with incoming viral genomes in the nucleus of infected cells. As expected, PML protein aggregates in cytosolic MAPPs in mitotic cells, whereas the EdU-labeled pseudogenomes are present throughout the cell with some in the vicinity of mitotic chromosomes as previously reported [30,36]. Majority of EdU puncta did not co-localize with PML protein aggregates during mitosis (for quantification, see Fig 3D), such as metaphase or late telophase, unlike previously reported [44]. However, we observed EdU puncta co-localizing with PML protein foci of different sizes in early interphase cells, which is then engulfed in the later interphase cells. In order to quantify these observations, we measured the distance between the center of EdU puncta and the center of PML protein puncta in early (7+ nucleoli) or late (1–6 nucleoli) interphase and found that PML protein is very closely co-localizing with EdU in late interphase cells, while the distance is very variable and overall greater in early interphase cells (Fig 2B). In addition, as a measurement of engulfment, we calculated the ratio of PML intensity over EdU intensity, each normalized to the area of co-localizing foci (Fig 2C). While the intensity of EdU puncta remained similar throughout, the intensity of PML puncta significantly increased in late interphase as we observe entrapment of EdU signal. Taken together, these data indicate that PML protein targets incoming viral genomes and forms around them, rather than viral genomes being recruited to preformed PML protein structures.
We previously described that incoming viral genomes are lost in cells depleted for PML protein, implying that PML protein provides a protective environment for the viral genomes. If this is the case, we would predict that PML protein accumulates around incoming pseudogenomes before the release from the transport vesicles. To test this assumption, we employed a differential staining technique that has been previously described by our lab [30,32]. HaCaT cells were infected with EdU-labeled PsVs for 24 h. Following fixation, cells were permeabilized with a low concentration of digitonin, which only permeabilizes cholesterol-rich membranes, such as the plasma membrane and endocytic vesicles directly derived from the plasma membrane. Next, the cells were subjected to the Click-iT reaction using AlexaFluor (AF) 555 as reactive dye to stain the viral genome (green). Subsequently, cells were completely permeabilized with Triton X-100 (TX-100) and subjected to another round of Click-iT reaction, this time using AF647 dye to stain the viral genome (red). Only EdU-labeled genomes either present on the cell surface, in early endocytic vesicles, or after egress from the endocytic compartment will be stained with AF555, whereas all genomes will be stained with AF647 after TX-100 permeabilization. As a positive control, we treated cells with TX-100 instead of digitonin prior to the first staining for total permeabilization (S1 Fig, Fig 3B). To control for intracellular membrane integrity, we tested the reactivity of an antibody recognizing a luminal epitope of TGN46 (cyan) in digitonin- and TX-100-treated cells (S1 Fig). We observed that the luminal epitope of TGN46 was recognized in TX-100-treated cells, but not in digitonin-treated cells, suggesting that the plasma membrane but not internal membranes were permeabilized with the low concentration of digitonin. Representative images of infected HaCaT cells in various stages of mitosis demonstrate the inaccessibility of pseudogenomes to AF555 after digitonin but not TX-100 permeabilization (S1 Fig). We combined differential staining of EdU-labeled pseudogenomes with immunofluorescent staining for PML protein (cyan) and quantified the presence of PML protein as a function of genome accessibility in different phases of the cell cycle (Fig 3). Once again, we observed essentially no co-localization of PML protein with EdU in late telophase cells and it was not until after mitosis was completed that viral genomes were shown co-localizing with PML protein (Fig 3A and 3B). In early interphase cells, PML protein co-localized with nuclear-localized EdU puncta that were inaccessible to AF555 after digitonin permeabilization, whereas EdU puncta were accessible to both dyes and co-localized with PML protein in late interphase cells. We quantified the number of single red (inaccessible) or dual green/red (accessible) EdU puncta in mitosis and early (7+ nucleoli) and late (1–6 nucleoli) interphase (Fig 3C). Chromosome-localized EdU puncta were 95% inaccessible in mitosis (5% accessible), and become more accessible over time after completion of mitosis in digitonin-treated cells (45% and 80% accessible EdU in early and late interphase, respectively). In the TX-100-treated control cells, EdU was consistently 95% accessible throughout the cell cycle (Fig 3C). These results are consistent with our published findings suggesting that accessibility of the viral genome is delayed after completion of mitosis [30]. Next, we quantified the number of inaccessible (In) or accessible (Ac) EdU puncta that co-localized with PML protein in mitosis, early interphase (7+ nucleoli), and late interphase (1–6 nucleoli) in digitonin-treated cells (Fig 3D). During mitosis, EdU puncta did not co-localize with PML protein, which was visible as large cytosolic aggregates. In interphase cells, accessible EdU puncta largely co-localized with PML protein (79% and 76% in early and late interphase, respectively). Interestingly, we observed 70% of inaccessible EdU puncta co-localized with PML protein in early interphase. This implies that EdU puncta co-localize with PML protein while the viral genomes are still inaccessible immediately after completion of mitosis and remains associated to a comparable level in late interphase (62%). The differences in PML protein co-localization with inaccessible and accessible EdU puncta in early and late interphase cells were not determined to be statistically significant, thereby suggesting that EdU and PML protein associate in early interphase and remain associated throughout interphase. Taken together, these data suggest that, as PML protein translocates back into the nucleus following completion of mitosis, PML protein is subsequently recruited to incoming viral genomes when still inaccessible and they remain associated as the cell progresses through interphase and viral genomes become accessible.
PML protein is SUMOylated and interacts with and recruits other proteins by non-covalent SUMO interactions, mainly through SUMO-1, which is essential for PML NB formation, stability, and localization [2,4]. Therefore, we sought to examine the recruitment of SUMO-1 to incoming viral genomes. To address this, we performed immunofluorescent staining on EdU-labeled PsV-infected HaCaT cells to detect EdU-labeled viral pseudogenomes (red), PML protein (cyan), and SUMO-1 (green) and acquired high resolution images of z-stacks combined by 3D reconstruction (Fig 4A). We observed very limited detection of SUMO-1 aggregates in cells undergoing mitosis, with little to no co-localization with PML protein; whereas, following the completion of mitosis, SUMO-1 was detected co-localizing with PML protein in the nucleus of interphase cells. In addition, EdU puncta co-localized with these SUMO-1/PML protein foci in the nucleus of interphase cells. SUMO-1 was also observed encompassing the EdU signal in a similar manner as PML protein alone previously was (Fig 2). Next, we investigated SUMO-1 co-localization as a function of viral genome accessibility. We combined immunofluorescent staining of SUMO-1 with the same differential staining technique described in S1 Fig and Fig 3. HaCaT cells were infected with EdU-labeled PsVs and differentially stained to detect inaccessible (red) and accessible (green/red) EdU-labeled pseudogenomes along with SUMO-1 (cyan) in mitotic and interphase cells (Fig 4B). EdU accessibility was quantified in mitosis, early interphase (7+ nucleoli), and late interphase (1–6 nucleoli) in digitonin- and TX-100-treated cells. Just like we observed in Fig 3, we reproduced the same pattern of EdU genome accessibility throughout the cell cycle (Fig 4C). Next, we quantified the number of inaccessible (In) or accessible (Ac) EdU puncta that co-localized with SUMO-1 in mitosis, early interphase (7+ nucleoli), and late interphase (1–6 nucleoli) in digitonin-treated cells (Fig 4D). Not surprisingly, SUMO-1 co-localization with EdU puncta was very similar to PML protein in Fig 3D. EdU puncta did not co-localize with SUMO-1 during mitosis (2% and 1% of inaccessible and accessible EdU, respectively). However, 51% of inaccessible EdU puncta co-localize with SUMO-1 in early interphase and 61% in late interphase. Accessible EdU puncta also largely co-localize with SUMO-1 in early and late interphase (66% and 76%, respectively). Taken together, these data suggest that SUMO-1 is recruited to incoming viral genomes prior to becoming accessible, likely along with PML protein.
Our recently published work suggested that a subset of the L1 protein traffics and is delivered into the nucleus along with the L2/viral genome complex [31]. Our data also indicated that L1 protein resides within the transport vesicle during trafficking and is lost when viral genomes become accessible in the nucleus. Therefore, we hypothesized that PML protein would associate with L1 together with viral genomes in early interphase. To test this, we performed immunofluorescent staining on HaCaT cells infected with EdU-labeled PsVs (Fig 5). Representative images of infected HaCaT cells in mitosis, early, and late interphase stained for PML protein (cyan), EdU-labeled pseudogenomes (red), and L1 protein (green) are displayed (Fig 5A). We quantified the total number of chromosome-associated or nuclear EdU puncta co-localizing with L1 protein in following phases of the cell cycle: mitosis, early interphase (7+ nucleoli), and late interphase (1–6 nucleoli) (Fig 5B). During mitosis, the majority of EdU puncta co-localized with L1 (81%). The L1 signal is still present with EdU in early interphase (56%) but is dramatically reduced in late interphase (30%), which is consistent with our published results [31]. Next, we quantified the number of EdU-L1 puncta co-localizing with PML protein during mitosis, early (7+ nucleoli) and late (1–6 nucleoli) interphase (Fig 5C). Once again, PML protein did not co-localize with EdU puncta during mitosis (0%). However, EdU/L1 signal did co-localize with PML protein in the nucleus of early interphase cells (53%). In late interphase, while EdU puncta remained co-localized with PML protein, L1 signal is lost (17%). Taken together, these data suggest that L1 remains associated with the viral genome as PML protein is recruited in early interphase but is lost in later stages of interphase, which corresponds to when viral genomes become accessible.
In contrast to L1, we had showed that L2 proteins remained with the viral genome for a longer period of time, even after viral genomes become accessible [31]. Logically, we sought to investigate whether L2 remained at PML NBs along with the viral genomes as interphase progresses and performed the same analysis as for L1 protein (Fig 6). Representative images of infected HaCaT cells in mitosis, early, and late interphase stained for PML protein (cyan), EdU-labeled pseudogenomes (red), and L2 protein (green) are displayed (Fig 6A). We quantified the total number of chromosome-associated or nuclear EdU puncta co-localizing with L2 protein in the different phases of the cell cycle (Fig 6B). Although there was a decrease in the amount of L2 co-localized with EdU between mitosis and early interphase, similarly to what was observed with L1, the L2 signal remained constant with about 50% L2 viral genomes after completion of mitosis and was still present in late interphase. Then, we quantified the number of EdU-L2 co-localizing with PML protein in the same conditions as previously described (Fig 6C). As expected, L2 also remained with EdU co-localized with PML protein even in late interphase cells. Taken together, these data suggest that, while L1 is lost when viral genomes become accessible in later stages of interphase, L2 remains associated with the now accessible viral genomes.
Another major component of PML NBs is Sp100. While the presence of PML protein is critical for HPV transcription, Sp100 is known to restrict HPV transcription and replication [26,28]. Therefore, we hypothesized that Sp100 is recruited with a delay compared to the recruitment of PML protein after the viral genomes becomes accessible. To address this, we performed immunofluorescent staining on EdU-labeled PsV-infected HaCaT cells to detect EdU-labeled viral pseudogenomes (red), PML protein (cyan), and Sp100 (green) and acquired high resolution images of several z-stacks combined by 3D reconstruction (Fig 7A) and confocal images (Fig 7B). During mitosis, PML protein formed cytosolic aggregates and Sp100 signal was not detected. In early interphase, Sp100 was detectable but was not seen co-localized with EdU puncta co-localizing with PML protein. During late interphase, both PML protein and Sp100 co-localized with EdU puncta, engulfing the signal in a similar manner as observed in Fig 2B. We quantified the number of Sp100-containing PML foci for the presence or absence of EdU puncta in the same infected cells (Fig 7C). In early interphase cells, 53% of EdU puncta co-localized with PML protein foci containing Sp100, while the rest co-localized with PML protein only. In contrast, in late interphase cells, nearly all EdU puncta co-localized with PML protein and Sp100 (92%). Interestingly, EdU-negative PML/Sp100 foci are nearly as abundant in early and late interphase cells (82% and 89%, respectively). It is important to note that EdU puncta were never seen to co-localize with Sp100 alone. Therefore, our data suggest that Sp100 is recruited with a delay to the viral genome and PML protein. More fascinating, this delay seems to be specific for PML NBs forming around incoming viral genome.
Next, we examined the recruitment of Sp100 with PML NBs as a function of viral genome accessibility using differential staining to distinguish between inaccessible EdU-labeled pseudogenomes (red), accessible pseudogenomes (red/green), and Sp100 (cyan) within cells undergoing mitosis or during early or late interphase (Fig 8A). EdU puncta showed a very reproducible pattern of accessibility as previously observed (Fig 8B). We quantified the number of inaccessible (In) or accessible (Ac) EdU puncta that co-localized with Sp100 in mitosis, early interphase (7+ nucleoli), and late interphase (1–6 nucleoli) in digitonin-treated cells (Fig 8C). Sp100 signal was not detected in mitotic cells. Only a marginal number of inaccessible EdU co-localized with Sp100 during early interphase (26%), which significantly increases in late interphase (56%). Sp100 co-localized with accessible EdU in early interphase and more in late interphase (49% and 72%, respectively). Taken together, these results suggest that Sp100 is recruited to incoming viral genomes and PML protein after viral genomes become accessible.
Other DNA viruses have been shown to target PML protein via SUMO interaction with viral proteins, such as HSV-1 ICP0 and ADV E1A [15,16]. Considering that HPV L2 protein is interacting with cellular factors during trafficking [32,39], we hypothesized that L2 is also interacting with and recruiting PML protein and possibly Sp100. A SUMO conjugation motif (K35), one highly conserved SIM (aa286-289), and two putative SIMs (aa105-109 and aa145-148) have been identified on L2 protein [24,41]. Therefore, we sought to investigate whether at least one or more of these sites played a role in recruiting PML NB proteins to viral genomes. To test this, we generated EdU-labeled PsVs carrying mutations in L2 protein (S1 Table). We performed site-directed mutagenesis on our L2 expression plasmid to disrupt the SUMO conjugation motif with a residue substitution K35R [41]. The L2 mutants disrupting each SIM (SIM 105-9A, SIM 145-8A, SIM 286-9A) have recently been described [24]. All mutant L2 proteins were efficiently incorporated into PsVs and the mutations did not affect PsV binding to the cell surface (S2 Fig). Next, we infected HaCaT cells with WT or L2 mutant EdU-labeled PsVs and performed immunofluorescent staining to detect PML protein (cyan) and viral pseudogenomes (red) (Fig 9A). Although cells were infected with same amounts of viral genome equivalents (vge) and comparable numbers of EdU puncta were detected in whole cells (Fig 9B), the number of EdU puncta present in the nucleus was dramatically decreased in cells infected with L2 mutant PsVs compared to WT (Fig 9C). Consequently, we observed a significant reduction in the number of L2 mutant EdU puncta co-localized with PML protein compared to WT. However, when we normalize the number of EdU co-localized with PML protein to the total number of EdU puncta in the nucleus of infected cells, there is no significant difference between WT and L2 mutants, apart from SIM 286-9A as almost no EdU puncta were observed in the nucleus (Fig 9D). These data imply that L2 mutant PsVs are not delivered into the nucleus as efficiently as WT PsVs. To test this, we examined the association of viral genomes with mitotic microtubules and chromosomes in HaCaT cells infected with WT and L2 mutant EdU-labeled PsVs by immunofluorescent staining to detect EdU (red) and microtubules (white) (Fig 9E). We observed a significant loss of EdU puncta associated with condensed chromosomes during mitosis in cells infected with L2 mutant PsVs compared to WT PsV-infected cells (Fig 9F). Taken together, these findings suggest that the mutations on L2 protein render the PsVs deficient for nuclear delivery, in a similar phenotype to a previously identified mutant R302/5A [30].
PML protein has been shown to be critical for the retention of HPV genomes in the nucleus of host cells and transcription [25,27]. However, the temporal recruitment of PML NB proteins and how they associate with incoming viral genomes is still poorly defined. The findings described herein suggest that PML protein and SUMO-1 are recruited to and assemble around incoming viral genomes after nuclear delivery and completion of mitosis but prior to the genome becomes accessible in the nucleus. Furthermore, L1 protein accompanies the viral genome into the nucleus followed by PML protein recruitment during early interphase, but L1 protein becomes lost as the cell progresses through interphase, while L2 protein remains with the viral genome. The transcriptional repressor Sp100 showed a delayed recruitment to viral genomes after the viral genome becomes accessible. Lastly, we determined that disruptions in the SIMs of L2 protein result in varying degrees of deficient nuclear delivery of incoming viral genomes.
The recruitment of PML protein towards incoming viral genomes rather than viral genomes targeting preformed PML NBs is a critical aspect for understanding how HPV genomes are delivered to the nucleus. During mitosis, HPV-harboring vesicles do not associate with MAPPs, or only what seems to be incidental and only transient. MAPPs are still predominantly cytosolic when viral genomes are delivered to the nucleus. This was specifically brought to light using high resolution microscopy as only rotating the images in the three-dimensional plane could reveal that EdU puncta do not truly co-localize with PML protein aggregates in mitotic cells, whereas confocal microscopy could not always distinguish large PML protein aggregates from EdU puncta [44]. At the moment, we cannot completely rule out that PML protein below our limit of detection is co-localizing with viral genomes. However, it seems unlikely as PML protein-deficient HaCaT and HeLa cells are fully capable of delivering viral genomes to the nucleus [25,27]. We observed that PML protein then translocates into the nucleus in early interphase and targets viral genomes to form PML NBs. Therefore, unlike what was recently suggested by Broniarczyk et al., our findings suggest that PML protein is not involved in nuclear delivery of HPV genomes, but rather starts playing a role in the nucleus, after viral genomes have already been delivered, which confirms other previous findings [25,27,44]. This is further supported by the recent findings by the Boutell group who observed the recruitment of PML protein and other PML NB-residing proteins to HSV-1 incoming genomes after nuclear entry but prior to the initiation of lytic replication [19]. High resolution microscopy also showed the structure of PML protein entrapping HSV-1 genomes similarly to our observation with PML protein engulfing HPV pseudogenomes.
The temporal recruitment and structure of PML protein surrounding viral genomes offer support for our hypothesis that PML protein provides a protective environment for the viral genome against innate and intrinsic immune sensors, rather than an environment favoring transcription as previously suggested [25]. Indeed, our previous findings suggest that incoming PV genomes can be sensed in PML protein-deficient cells and subsequently targeted for degradation [27]. IFI16 (IFN Gamma Inducible Protein 16) has been a major candidate for viral DNA sensing in host cell nucleus and restricts HSV-1 and HPV18 replication and transcription [45]. However, more recently, the repression of HSV-1 replication was shown to occur rapidly after association with PML NBs and independently of IFI16 and induction of ISG (IFN-stimulated gene) expression [19]. However, in the context of HPV infection, knocking down IFI16 did not prevent genome loss in HaCaT cells [27]. Another possible candidate is the Myb-related transcription factor MYPOP that has recently been shown to sense incoming HPV DNA and L2 protein and subsequently inhibit early gene expression [46]. Here, PML protein may compete for binding of such restriction factors to L2 protein to protect the infectious HPV complex and prevent transcriptional repression.
Nevertheless, PML protein seems to protect the viral genome from such a fate. Our previous work demonstrated that the viral genomes are delivered to the nucleus of target cells in a membrane-bound vesicular compartment [30,32]. We have previously shown that the egress from the transport vesicle by an unknown mechanism is slow, resulting in a delay of transcription by four to five hours when compared to a transfection method that requires mitosis for nuclear delivery [30]. The delay strongly suggests that transcriptional activity requires an additional step, which is presumably the egress from the transport vesicles. However, we are aware that a minority (5%) of viral genomes present on mitotic chromosomes is accessible to staining in differentially permeabilized cells and could be the infectious ones. However, we believe that this level of background is most likely due to the extensive processing required for two sequential immunofluorescent stainings. Indeed, in the absence of immunofluorescent processing, viral genome is completely protected from degradation by nucleases when the cells were arrested in mitosis [30]. Furthermore, the work done by the Schelhaas, Campos, DiMaio, and Tsai groups, using a BirA-based system, supports the importance of the transport vesicle for productive infection [47–50]. Therefore, upon nuclear delivery, the viral genome is already presumably protected from DNA sensors within the transport vesicle. The formation of the PML protein structure around the genome-harboring vesicle allows for egress of the viral genome, while remaining protected. We speculate that this step allows for the initiation of transcription responsible for the primary amplification of viral genomes resulting in the establishment of infection [51]. Interestingly, we observed a delayed recruitment of Sp100 to viral genomes when compared to PML protein. This delayed recruitment of Sp100 seems to be specific for HPV-harboring PML NBs and occurs mostly after viral genomes become accessible, although, due to the unknown specificity of our Sp100 antibody, we cannot rule out that low but undetectable levels of Sp100 are present early. It has been demonstrated that Sp100 restricts HPV18 early transcription during establishment of infection [26]. It is attractive to speculate that HPV exploits PML NBs to regulate early transcription, PML protein allowing early transcription to establish infection and delayed Sp100 recruitment helping transition to the maintenance phase. The McBride group also demonstrated that Sp100 does not seem to be involved in the maintenance phase [28]. Additionally, it restricts viral processes in later stages of infection, during the differentiation-induced viral amplification [28]. Our data strongly suggest that L2 protein mediates the recruitment of PML protein to viral genome prior to complete release within the nucleus. It has become clear in recent studies that L2 protein is ultimately lost after orchestrating viral delivery [31]. Therefore, we assume that PML NB association is lost after the next round of mitosis, which would be consistent with the findings by Stepp et al. [26,28]. However, further experimentation is needed to test this assumption and to link PML NB composition and transcriptional activity of incoming viral genome.
Our recent work also focused on the role of the capsid proteins during trafficking and nuclear delivery, more specifically the L1 protein. We have shown that L1 protein remains associated with the viral genome in the nucleus of infected cells [31]. We demonstrated that L1 protein directly interacts with the viral DNA and a transmembranous L2 protein while inside a transport vesicle during trafficking and after nuclear delivery. In addition, reactivity of conformationally-dependent antibodies provided evidence that L1 protein was likely arranged as capsomeres while it accompanies the viral genome to the nucleus. Herein, we observed that L1 protein remains associated with the viral genome within the nucleus and that PML protein is recruited and assembles around it. In late interphase, L1 protein dissociates, timed with release of the viral genome, while the viral genome remains at PML NBs. At this time, there is no evidence to suggest that L1 protein plays a role beyond just incidental trafficking. Therefore, we only refer to the loss of L1 protein as a marker for the point in time that coincides with release of the viral genome in the nucleus.
Considering that L1 protein resides within the lumen of the transport vesicle with the viral genome during trafficking into the nucleus, our lab and others have hypothesized that L2 protein is involved in recruiting PML NB proteins. Indeed, L2 protein is already known to contain many domains involved in interacting with cellular factors to facilitate trafficking [32,38,39]. These domains also include SUMO conjugation and interacting motifs [24,41]. SUMOylation is such a critical step in the formation of PML NBs and the recruitment of additional proteins, therefore it is thought that L2 protein may be responsible for recruiting PML protein and Sp100 via SUMO interactions. The Florin group identified three SIMs on L2 protein, at residues 105–9, 145–8, and 286–9. The latter was found to interact with cellular SUMO-1/2 and to be essential for interaction with PML protein as PsVs carrying a disrupted SIM resulted in a decrease in PML co-localization [24]. However, they also noted that infection with the mutant PsVs exhibited reduced amounts of L2 protein and viral genomes in the nucleus, suggesting the mutation may affect events upstream of PML accumulation, such as nuclear localization and retention. The Schelhaas group also investigated the nuclear delivery of the SIM 286–9 and observed that the mutant PsVs were impaired in interacting with mitotic chromosomes, although the SIM 286–9 mutation had been shown to result in clear nuclear accumulation and loss of PML co-localization after L2 overexpression [24,47]. Herein, we show that all of the mutants used in the study were deficient in nuclear delivery. These mutants had a similar pattern to the nuclear retention mutant, 302/5A, that failed to traffic along spindle microtubules, which also resulted in lower infectivity [30]. L2 protein is very compact and harbors many important domains on its C-terminus [23,29,30,32,38,39,47,52]. Therefore, it is a possibility that by mutating the SIMs, we may have disrupted other domains on L2 protein that are essential for cellular trafficking or nuclear delivery. We also cannot exclude defects in proper assembly and early events of the infectious entry process. Therefore, at this time, we are unable to directly test whether the SIMs are responsible for PML recruitment, although these regions are still of particular interest as they all seem to be important for the delivery of the viral genome to the nucleus.
In summary, we present herein a promising model defining the order of events following nuclear delivery of HPV genomes. We demonstrated that PML protein and SUMO-1 are recruited to and assemble around viral genomes that still reside within the transport vesicle in early interphase cells. As L1 protein accompanies the viral genome to the nucleus, it also localizes at PML protein, but is lost in later stages of interphase along with the transport vesicle and release of the viral genome into the nucleus. Then, Sp100 is recruited to viral genomes and also engulfs them in a PML NB structure. Further studies will be necessary to link HPV transcriptional regulation to PML NB composition. However, our study defines these events, thus providing new insights into the kinetics of the primary HPV infection and how HPV relies on the specific temporal recruitment of various factors necessary to promote infection.
The 293TT cells (a kind gift of Dr. John T. Schiller, Laboratory of Cellular Oncology, National Cancer Institute, Bethesda) used for the generation of pseudovirions were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% fetal bovine serum (FBS), non-essential amino acids, L-Glutamax, and antibiotics. HaCaT cells (purchased from the American Type Culture Collection) used in the infection studies were grown in low glucose DMEM containing 5% FBS and antibiotics.
HPV16 pseudoviruses (PsVs) encapsidating a green fluorescent protein (GFP) expression plasmid (pfwB) were generated in 293TT cells as previously described using puf16L1 and puf16L2HA-3’ [53–56]. The pfwB plasmid was a kind gift from Dr. John Schiller (National Cancer Institute, Bethesda, MD). The L1 and L2 expression plasmids harbor codon-optimized genes and were kindly provided by Dr. Martin Müller (Deutsches Krebsfoschungszentrum, Heidelberg, Germany) [57]. For detection of pseudogenomes by IF staining, pseudogenomes were labeled by supplementing the growth media with 100 μM 5-ethylnyl-2’-deoxyuridine (EdU) at 6 hours post transfection during PsV generation, as previously described [58]. Viral DNA within the virions was isolated using NucleoSpin Blood QuickPure (Machery-Nagel; #740569.250) supplemented with 4 μM Ethylenediamineteraacetic acid (EDTA) and Dithiothreitol (DTT) and genome copy number was quantified by quantitative PCR (qPCR). Capsid composition was verified by western blot analysis. For all experiments, 100 to 300 vge/cell were added.
Point mutation K35R in L2 [41] was generated by site-directed mutagenesis using a pair of complementary PCR primers specific to codon-optimized plasmid puf16L2HA-3’ using the PfuUltra II Hotstart DNA Polymerase 2x Master Mix (Stratagene; #600630) following manufacturer’s protocol. The entire plasmid was amplified during the PCR reaction and the PCR products were digested with DpnI to remove methylated template DNA. The PCR products were transformed and the mutation was confirmed by sequencing (Macrogen). L2 expression plasmids harboring SIM mutations (pCDNA16L2-ΔSIM105-109, pCDNA16L2-ΔSIM145-148, pShell16L1L2-ΔSIM286-289) were a kind gift from Dr. Luise Florin (University of Mainz, Germany) [24]. Mutant L2 expression plasmids were used to generate mutant PsVs as described above. Capsid composition and vge were determined by western blot and qPCR as described above. Same amounts of vge were used in experiments comparing WT and mutant PsVs.
Antibodies used for the IF studies were as follows: rabbit polyclonal antibody (pAb) anti-PML (BETHYL; #A301-167A), mouse monoclonal antibody (mAb) anti-PML (Santa Cruz Biotechnology; #sc-966), rabbit pAb anti-TGN46 (Thermo Scientific; #PA5-23068), rabbit pAb anti-SUMO-1 (abcam; #ab11672), AlexaFluor (AF) 488-conjugated rabbit pAb anti-GFP (Molecular Probes; #A21311), mouse mAb anti-CD147 (Affinity BioReagents; #MA1-19202), mouse mAb anti-alpha-Tubulin (Cell Signaling; #3873S), AF647-conjugated phalloidin (Molecular Probes; #A22287), and goat AF-labeled secondary antibodies (Life Technologies; #A11034, #A21236). Rabbit polyclonal anti-Sp100 antibody was a kind gift from Dr. Hans Will (Heinrich Pette Institute, Hamburg, Germany) [59]. L1-specific mouse mAb 33L1-7 and rabbit pAb K75 were described previously [60,61]. L2-specific mouse mAb 33L2-1 was also previously described [62]. Click-iT EdU Imaging Kit (Molecular Probes; #C10338) was used for Click-iT reactions to label EdU-labeled pseudogenomes. For western blot analysis, L1 was detected with mouse mAb HPV16 312F and L2 with mouse mAb 33L2-1, combined with peroxidase-conjugated AffiniPure pAb goat secondary anti-mouse (Jackson ImmunoResearch; #115-035-003).
HaCaT cells were grown on coverslips at approximately 50% confluency and infected with EdU-labeled PsVs at approximately 106 viral genome equivalents per coverslips for 24 h at 37°C. Cells were fixed with 4% paraformaldehyde (PFA) for 15 min at room temperature, washed with phosphate-buffered saline (PBS; pH 7.5), permeabilized with 0.5% TX-100 in PBS for 5 min at room temperature, washed with PBS, and blocked with 5% normal goat serum (NGS) for 15 min at room temperature. The Click-iT reaction containing AF555 followed for 30 min at room temperature and protected from light to specifically detect EdU-labeled pseudogenomes [58]. After cells were washed with PBS, they were incubated with primary antibodies in 2.5% NGS for 30 min at 37°C in a humidified chamber. After extensive washing with PBS, the cells were incubated with AF-labeled secondary antibodies in 2.5% NGS for 30 min at 37°C in a humidified chamber. After another round of extensive washing with PBS, the cells were mounted in ProLong Gold antifade reagent with DAPI (4’, 6’-diamidino-2-phenyllindole; Invitrogen; #P36931). Confocal images were acquired in single-slices or z-stacks with Nikon A1R confocal microscope using a 100X objective and NIS Elements software. Number of EdU-labeled pseudogenomes in nuclei was quantified in z-stacks spanning the whole nucleus. Results are expressed in average number of EdU or percent of EdU-labeled viral pseudogenome in the nucleus ± standard error of the mean (SEM). High resolution images were acquired with Nikon N-SIM E Super Resolution microscope using 100X objective. Several z-stacks spanning the whole nucleus were acquired and assembled using 3D reconstruction in NIS Elements software. All images from individual experiments were acquired under the same laser power settings and enhanced uniformly in Adobe Photoshop.
HaCaT cells were grown on coverslips to 50% confluency and infected with EdU-labeled PsVs for 24 h at 37°C. Cells were fixed with 4% PFA for 15 min at room temperature, washed with PBS, and selectively permeabilized with 0.625 μ/mL of digitonin in water for 5 min at room temperature, washed with PBS, and blocked with 5% NGS for 15 min at room temperature. Cells were treated with the first Click-iT reaction with AF555 for 30 min at room temperature protected from light. Cells were washed with PBS, permeabilized with 0.5% TX-100 for 5 min at room temperature, and blocked with 5% NGS for 15 min at room temperature. Cells were treated with the second Click-iT reaction with AF647 for 30 min at room temperature protected from light. Cells were washed with PBS and incubated with primary and secondary antibodies to detect protein of interest as described above. Cells were extensively washed and mounted with DAPI. Differential staining of EdU-labeled pseudogenome using sequential Click-iT reactions in selectively permeabilized HaCaT cells was previously described in greater details [30]. A control experiment was performed by treating the cells with 0.5% TX-100 in both permeabilization steps. In a parallel experiment, cells were incubated with primary Ab anti-TGN46 in 2.5% NGS for 30 min at 37°C in a humidified chamber following the first Click-iT reaction. Cells were washed extensively and incubated with secondary AF488-labeled Ab in 2.5% NGS for 30 min at 37°C. This parallel experiment acts as a permeabilization control, as previously described [32]. All images were acquired in z-stacks spanning the whole nucleus with Nikon A1R confocal microscope using a 100X objective and NIS Elements software. All images from individual experiments were acquired under the same laser power settings and enhanced uniformly in Adobe Photoshop. Co-localization of EdU puncta with protein of interest was quantified by counting the number of nuclear EdU puncta as a function of single or double EdU staining and co-localization with protein of interest and expressed as percent co-localization of total nuclear accessible or inaccessible EdU puncta ± SEM.
HaCaT cells were grown on coverslips to 70% confluency and equal number of PsVs were allowed to bind to the cell surface for 1 h at 37°C. Cells were stained as described above without the Click-iT reaction. Instead, conformational L1 protein was detected with K75 Ab. Assay was quantified as pixel sum ratio of L1-specific signal on the cell surface to region of interest (ROI) area and expressed as percent of WT (100%) ± SEM. All images were acquired in single slice through the cell with Nikon A1R confocal microscope using a 100X objective and NIS Elements software. All images from individual experiments were acquired under the same laser power settings and enhanced uniformly in Adobe Photoshop.
|
10.1371/journal.pgen.1002626 | Cell Contact–Dependent Outer Membrane Exchange in Myxobacteria: Genetic Determinants and Mechanism | Biofilms are dense microbial communities. Although widely distributed and medically important, how biofilm cells interact with one another is poorly understood. Recently, we described a novel process whereby myxobacterial biofilm cells exchange their outer membrane (OM) lipoproteins. For the first time we report here the identification of two host proteins, TraAB, required for transfer. These proteins are predicted to localize in the cell envelope; and TraA encodes a distant PA14 lectin-like domain, a cysteine-rich tandem repeat region, and a putative C-terminal protein sorting tag named MYXO-CTERM, while TraB encodes an OmpA-like domain. Importantly, TraAB are required in donors and recipients, suggesting bidirectional transfer. By use of a lipophilic fluorescent dye, we also discovered that OM lipids are exchanged. Similar to lipoproteins, dye transfer requires TraAB function, gliding motility and a structured biofilm. Importantly, OM exchange was found to regulate swarming and development behaviors, suggesting a new role in cell–cell communication. A working model proposes TraA is a cell surface receptor that mediates cell–cell adhesion for OM fusion, in which lipoproteins/lipids are transferred by lateral diffusion. We further hypothesize that cell contact–dependent exchange helps myxobacteria to coordinate their social behaviors.
| All cells interact with their environment, including other cells, to elicit cellular responses. Cell–cell interactions between eukaryotic cells are widely appreciated as large multicellular organisms coordinate cell behaviors for tissue and organ functions. In bacteria cell–cell interactions are not widely appreciated, as these organisms are relatively simple and are often depicted as single-cell entities. However, over the past decade, the concept of bacteria living in microbial communities or biofilms has received broad acceptance as a major lifestyle. As biofilm cells are packed in tight physical contact, there is an opportunity for cell–cell signaling to provide spatial and physiological clues of neighboring cells to elicit cellular responses. Although much has been learned about diffusible signals through quorum sensing, little is known about cell contact–dependent signaling in bacteria. In this report we describe a new mechanism where bacterial cells within structured biofilms form contacts that allow cellular material to be exchanged. This exchange elicits phenotypic changes, including in cell movements and development. We hypothesize that OM exchange involves kin recognition that bestows social benefits to myxobacterial populations.
| Biofilms are ubiquitous in nature. Within these structures microbes adhere to surfaces and each other in dense communities coated by an extracellular matrix. Although biofilms are of great medical and industrial interest [1], little is known about how these cells interact. In some cases, cell-cell contacts likely promote communication and provide spatial cues about neighboring cells to direct biofilm maintenance and maturation [2], [3]. Experimentally, biofilm research is hindered by limited knowledge and approaches to study their cellular dynamics [4]. Recently we described a novel biofilm dependent process whereby myxobacteria exchange their outer membrane (OM) lipoproteins [5], [6]. This transfer process can result in phenotypic changes and may represent a unique mechanism in which biofilm cells communicate. Although OM lipoprotein exchange is an interesting phenomenon, little is known about the mechanism and protein components required for transfer.
Myxobacteria are gram-negative soil dwelling microbes that exhibit complex multicellular behaviors. Central to these behaviors is gliding motility, which powers and coordinates swarm expansion, rippling, predation and fruiting body development on solid surfaces. Myxococcus xanthus has two distinct motility systems called A (adventurous) and S (social) motility, which served as the experimental backdrop for the discovery of OM lipoprotein exchange [7], [8]. S-motility is powered by the retraction of type IV pili adhered to external surfaces, effectively pulling the cell forward [9]. The motor powering A-motility is beginning to be defined and may involve cell surface adhesins that translocate on tracks [10]. Nonmotile mutants (A−S−) thus typically contain two mutations. Of interest here is a small subset of motility mutants that can be complemented extracellularly when mixed with another strain that encodes the corresponding wild-type gene [8], [11]. Historically, this process was called ‘stimulation’ as the recipient mutant transiently gains the ability to glide. Stimulation only involves phenotypic changes; there are no genotypic changes. Of the six stimulatable motility genes (cglB/C/D/E/F and tgl) [7], [8], only two have been previously identified; cglB (A-motility) and tgl (S-motility) [12], [13]. Importantly, both of these genes encode type II signal sequences (SS) for lipoproteins. The mechanism of stimulation was determined to involve cell-to-cell transfer of either the CglB or Tgl lipoproteins from donor to recipient cells, thus restoring missing protein function to the respective mutant [5]. Strikingly, lipoprotein transfer is efficient as recipient cells accumulate approximately equal quantities of proteins as donors [5], [6]. Recently, we described the identification of the cglC/D/E/F genes [14]. These genes encode either a type I or type II signal sequence.
To determine the molecular mechanism of OM lipoprotein exchange (stimulation) we recently defined the cis factor requirement in the cargo protein [6]. Surprisingly, simply a type II signal sequence for OM localization is sufficient for heterologous transfer of the mCherry fluorescent protein. Cytoplasmic or inner membrane reporters were not transferred. Transfer also requires specific cell-cell contacts where motility is apparently required to align biofilm cells [6], [15]. Here, we sought to identify trans or host genetic determinants required for lipoprotein transfer. In a prior study we screened known S-motility mutants for stimulation defects [16]. This resulted in the identification of a subset of pil mutants that were conditionally defective in tgl stimulation. However, these mutants were not further pursued because they are functional for cgl stimulation, and tgl stimulation occurs when cells are mixed on hard agar at low cell densities. This report identifies two gene products universally required for stimulation and lipoprotein transfer. In addition, we provide evidence, for the first time, that myxobacteria exchange their OM lipids, and that this process can regulate swarming and developmental behaviors.
To elucidate the mechanism of lipoprotein transfer we sought to identify mutants defective in stimulation. We reasoned that cgl and tgl stimulation occurs by a common mechanism, whereby OM proteins, and perhaps periplasmic proteins, are transferred from donor to recipient cells that lack a corresponding protein function. To avoid trivial or idiosyncratic mutants associated with particular cgl or tgl genes, we sought mutants universally defective in stimulation of the six known cgl/tgl complementation groups. We initiated these studies by first characterizing select mutants in the Dale Kaiser strain collection, the laboratory in which A- and S-motility and stimulation were discovered [7], [8]. One such mutant (DK396), isolated by Jonathan Hodgkin, appeared to possess the desired phenotype. This strain was isolated by ultraviolet light mutagenesis on an A−S+ (DK1211) strain and then screened for the loss of S-motility (nonmotile A−S−). Serendipitously, this mutant was found to be donor defective for stimulation, a phenotype we verified for all cgl/tgl mutants.
As the donor defect mutation was not known nor easily mapped, the DK396 genome was sequenced to identify the gene of interest. Upon >39X sequence coverage the DK396 genome was compared to the wild-type DK1622 genome to identify DNA changes [17]. Mutations in 20 gene candidates were identified (Table S1). The mutations responsible for the A- and S-motility defects, but not the stimulation defect, were easily found as they were in known motility genes (Table S1; aglT and pilR, respectively) [18], [19]. Based on the severity of the mutations and predictions of gene function and subcellular localization, a prioritized list of 9 gene candidates was chosen. Assuming the phenotype was caused by a loss-of-function mutation, these genes were systematically tested for a role in stimulation by a rapid gene disruption method in a nonmotile donor strain. From these experiments one insertion mutation in mxan_6895 (hereby named traA for transfer) was found to recapitulate the donor defective phenotype observed in DK396. Figure 1 shows that a disruption mutation in traA results in a complete block of stimulation for all the cglB, C, D, E, F and tgl mutants, as indicated by sharp colony edges (Figure 1 row D). The traA+ isogenic control donor stimulates all cgl/tgl mutants for A or S-motility (Figure 1 row C). The degree to which strains were stimulatable varied and only involved partial motility restoration (Figure 1 compare row C to A). From these results it was concluded that traA was universally required for cgl/tgl stimulation.
Next we tested whether TraA was required for SSOM-mCherry transfer [6]. This reporter has a type II SS for OM lipoprotein localization fused to a fluorescent protein. In this assay a nonmotile and non-stimulatable SSOM-mCherry donor was mixed with an A-motile GFP+ labeled recipient. The cell mixture was pipetted onto a TPM agarose pad and motile recipients were allowed to swarm. The swarm edge was then examined by epifluorescence microscopy to determine whether SSOM-mCherry was transferred from the nonmotile donor to motile recipients. As shown in Figure 2C controls, SSOM-mCherry was readily detected in GFP labeled recipient flares [6]. In contrast, an isogenic donor that contained the traA::km disruption exhibited no SSOM-mCherry transfer (Figure 2F). To verify these results we conducted related experiments where the same strains were again mixed and spotted on agar, and after short incubations cells were harvested and microscopically examined on glass slides. Here transfer was directly tested by assessing whether GFP labeled recipients become red. As previously reported, control strains show transfer (Figure 3 left green and red panels, see arrows), where typically >90% of recipients obtain detectable levels of SSOM-mCherry [6]. In contrast, when an isogenic traA− donor was used no SSOM-mCherry transfer was detected (Figure 3 middle merged panel). We further note that replication of this experiment; under similar or different conditions/strain backgrounds, where thousands of cells were evaluated, never resulted in detection of SSOM-mCherry transfer from a traA− donor. We conclude that TraA is required for OM lipoprotein transfer and stimulation.
The traA ORF and the downstream mxan_6898 ORF (locus tag numbers are not consecutive) overlap by four bases, suggesting they form an operon and their gene products may function in the same pathway (Figure 4A). To test this we created an insertion mutation in mxan_6898. This mutant exhibited a complete block in stimulation for all cgl and tgl mutants and was completely defective in SSOM-mCherry transfer (Figures S1 and S2). In addition, markerless in-frame deletions in traA and mxan_6898 were constructed and found to elicit the identical phenotypes reported here. Therefore mxan_6898 was named traB and its gene product is predicted to function in the same pathway as TraA.
The mxan_6894 ORF is located 126 bps upstream of traA, suggesting it is not part of the traAB operon. To test for a possible role in stimulation/transfer an insertion mutation was again created. In contrast to traA and traB, the mxan_6894::km mutant showed no overt defect in simulation or SSOM-mCherry transfer. To test whether the stimulation/transfer defect of DK396 was solely caused by the traA mutation, the selectable mxan_6894::km mutation and the tightly linked traA+ allele were transduced into DK396. All resulting Kmr transductants were fully competent for stimulation, thus the traA mutation in DK396 caused the stimulation/transfer defects found in this strain. Since the mutation in DK396 was a missense substitution (Table S1; 227P→L), we tested whether it caused a dominant-negative phenotype by complementation analysis. Here, the wild-type traAB genes were cloned into a plasmid that directs site specific recombination into the Mx8 phage attachment site. Integration of this plasmid into the DK396 genome restored stimulation to the resulting strain, thus demonstrating the traA227P→L allele was recessive. In addition, this plasmid, which has traAB under the heterologous transcription control of the strong pilA promoter, was introduced into a tra+ strain that contains the SSOM-mCherry reporter. Strikingly, upon microscopic examination this TraAB overexpressing strain was found to dramatically cause cells to adhere to one another in both kinked end-to-end chains and side-by-side contacts (Figure S3). The implication of this observation is discussed below.
Next, we tested whether TraAB plays a role in recipient cells for stimulation/transfer. Since traA and traB mutants are fully motile, one or more of these mutations were introduced into all the cgl/tgl mutants. Importantly, when recipient cells contain a traA or traB mutation and mixed with a Tra+ donor, no stimulation occurred (Figure 1 row E and Figure S2). We conclude that TraAB are required in both donor and recipient cells for stimulation. Next, defects in protein transfer were tested. As described above, when a nonmotile (traA+) donor was mixed with a motile traA− recipient, SSOM-mCherry was not transferred (Figure 2G–2I and Figure 3 right column). We conclude that TraAB are required in donor and recipients for stimulation and lipoprotein transfer.
The finding that OM lipoproteins are efficiently and apparently non-specifically transferred suggests that OM lipids may also be exchanged. To test this, donor cells were stained with a fluorescent lipophilic dye called DiD oil. As shown, DiD specifically stained the cell envelope, which fluoresced red (Figure S8). Importantly, when stained cells were harvested, washed and mixed with GFP labeled recipients in solution, recipients did not fluoresce red, indicating the dye did not freely diffuse between cells. As transfer requires a hard surface, cell-cell contact and motility, we next tested, under these conditions, for DiD transfer [6]. As shown in Figure 6 (left panels), DiD transfer readily occurred to GFP-labeled recipients. As controls, no DiD transfer occurred when isogenic recipients contained a traA mutation or when donor and recipients were both nonmotile (Figure 6, middle and right panels, respectively). In accordance with the above results, TraA was also required in donors, and similarly TraB in donors/recipients, for DiD transfer (Figure S9). These experiments show, similar to SSOM-mCherry transfer (Figure 2 and Figure 3) [6], that lipophilic dye and hence OM lipid, requires a hard surface, cell motility, and TraAB functions in donor and recipient cells for transfer.
As noted above, the traA and traB mutants exhibited no overt defects in A or S-motility, suggesting that OM transfer was not required for motor functions. However, the exchange of OM lipids and proteins involves significant resource sharing between cells and therefore this process must involve physiological consequences. One such phenotypic consequence was the restoration of swarming defects to certain motility mutants (Figure 1). However, extracellular complementation might have little significance between wild-type cells as they contain a full complement of motility proteins. In strain-mixing experiments we discovered that tra+, but not tra− strains, dramatically inhibited swarm expansion when a nonmotile strain was mixed with a motile strain. An example of how a nonmotile strain inhibits swarm expansion of an A+S− strain was illustrated in Figure 7A. In contrast, when identical mixing experiments were done between isogenic traA− strains, swarm expansion occurred (Figure 7B and 7C). As was found for lipoprotein and lipid transfer, the relief of swarm inhibition occurred when the traA mutation was introduced into either the nonmotile or motile strains. However, we note, swarm expansion was consistently more robust when the motile strain, instead of the nonmotile strain, contained the traA mutation (compare Figure 7B to 7C). An identical relief of swarm inhibition was again found when strains instead contained the traB mutation. Similarly, a Tra+ dependence for swarm inhibition of A+S+ motility was found when these strains were instead mixed with a nonmotile strain. In contrast, inhibition of A−S+ motility was minimal. To test whether swarm inhibition was specific to certain motility genes we test a variety of A−S− double mutants, including combinations of dsp/dif, pilA, pilM, pilT, pilQ, stk, aglB, aglR and aglM mutations, and in all cases these nonmotile strains inhibited swarm expansion of A+S− motile strains. We conclude that swarm inhibition was not dependent on specific motility genes, but instead was dependent on TraAB and thus OM exchange.
Macroscopically swarm inhibition was apparent (Figure 7A, 4 day incubation); however swarm inhibition was not absolute as flares were initially observed emerging from inoculation mixtures (Figure 7D, 15 hrs). Microscopically, the number and size of these early emerging flares were reduced compared to traA− mixtures (Figure 7, compare 7D to 7E and 7F). However, over longer incubations, e.g. 4 days, the strain mixtures that were Tra+ failed to swarm farther (Figure 7, compare 7A to 7B and 7C). To investigate this behavior time-lapse microscopy was used to track cell movements. Consistent with the above observations, for the first ≥1 day after plating the A-motile cells exhibited similar cell movements with respect to speed, reversal frequency and percent of cells moving, whether the mixtures contained tra+ or tra− cells. In contrast, by day 2 these same cell mixtures exhibited drastically different behaviors. That is mixtures containing traA+ cells exhibited a complete block in group movements, while isolated cells occasionally exhibited motility that was aberrant (Video S1). In sharp contrast, isogenic strain mixtures with traA− mutations in either the motile or nonmotile strain exhibited robust group and single cell motility (Videos S2 and S3). Swarm inhibition does not appear to depend on a diffusible signal, because when these identical tra+ strains were separated by a membrane (nitrocellulose) or soft agar overlay, no motility inhibition was observed. Hence, we hypothesize that nonmotile cells produce a time dependent (≥2 days) physiological signal that was transferred by OM exchange to motile cells that blocked their motility.
Myxobacteria are noted for the social behaviors and ability to form multicellular fruiting bodies in response to starvation. We thus tested whether Tra plays a role in development. A traA mutation was introduced into a wild-type strain, but no overt defects in fruiting body formation or sporulation was observed. To extend the above swarm inhibition findings, we next tested whether genetically distinct strain mixtures, as found in nature [33], interfered with development in a Tra dependent manner. First, the traA mutation did not significantly alter the ability of A+S− strain to sporulate (Figure 8) [34]. Second, as development is coupled to motility [35], nonmotile strains cannot fruit or sporulate and a traA mutation does not alter this phenotype (Figure 8). Strikingly, however, when the A+S− strain was mixed in a 1∶1 ratio with a nonmotile strain no viable spores were detected (≥6-logs; Figure 8). In contrast, when isogenic strains contained the traA mutation in either strain, the ability of the A-motile strain to sporulate was restored to control levels (Figure 8). Thus similar to swarm inhibition, a nonmotile strain can block development of a motile strain that depends on TraA and hence OM exchange.
To understand the mechanism of lipoprotein exchange we identified mutants universally defective in cgl/tgl stimulation and protein transfer. Interestingly, these TraAB proteins were required in both donor and recipient cells. To our knowledge, this is the first bacterial transfer system where the same gene products are required in both donor and recipient cells. This finding and the ability of M. xanthus cells to rapidly and homogeneously exchange lipoproteins [5], [6] implies that lipoproteins are transferred in a bidirectional manner. A bidirectional transfer mechanism is distinct from known secretion and conjugative systems [36], [37], where proteins or DNA are transferred unidirectionally from donor to recipient cells.
Since OM lipoprotein exchange occurs efficiently and involves a form of bulk transfer [5], [6], we hypothesized that OM lipids may also be exchanged. This hypothesis was supported by the finding that a lipophilic fluorescent dye was readily exchanged between cells. Importantly, transfer of lipophilic dye and hence membrane lipids, have the same stringent requirements in transfer as OM lipoproteins [6]. That is, dye transfer only occurred when cells were motile within structured biofilms; no detectable dye transfer occurred in liquid or between nonmotile (non-stimulatable) cells on a solid surface. In addition, dye transfer required the TraAB proteins in donor and recipient cells. We thus conclude that dye exchange does not occur by diffusion or by diffusible OM vesicles, but instead requires specific cell-cell contacts mediated by cell motility. Based on earlier observations that OM, but not IM, lipoproteins are transferred [6], we surmise that only OM lipids are exchanged bidirectionally. Presumably transfer consists of the outer leaflet lipopolysaccharide (LPS) and the inner leaflet phospholipids. In this respect it is interesting to note that species of Borrelia have been directly observed to fuse their OMs, a process apparently mediated by cell motility [38], and Bacillus subtilis reportedly transfers proteins in biofilms via membrane enclosed nanotubes [39].
Based on sequence, domain architecture and functional similarities to eukaryotic proteins, we propose that TraA serves as a cell surface receptor. In particular, TraA has similarities to the Saccharomyces cerevisiae FLO1 and FLO5 cell surface receptors/adhesions [21], [24], [40] (Figure 4). These FLO proteins have domain architecture consisting of a SS, N-terminal PA14 domain, a central tandem repeat region and a C-terminal protein sorting tag (GPI site; glycosylphosphatidylinositol anchor) for cell surface attachment [26]. Thus, by analogy, we suggest that in TraA the SS serves to transport the protein to the periplasm followed by SS cleavage. The processed N-terminal PA14 domain would serve as a receptor for ligand binding, presumably a glycan. The cysteine-rich tandem repeats could serve as a rigid stalk for PA14 presentation on the cell surface. The MYXO-CTERM motif could function, analogous to a GPI site, in protein sorting to the cell surface. Recent reports suggest the MYXO-CTERM and related C-terminal tags (Figure 5) are widely distributed in bacteria and archaea, where they are proposed to be posttranslationally modified and direct protein sorting to the cell surface [29]–[31]. Although initial attempts to generate TraA antibodies or fluorescent protein fusions were unsuccessful, TraAB overexpression was found to dramatically increase the ability of cells to adhere to one another (Figure S3). This result is consistent with TraA serving as a cell surface adhesin. Furthermore, the identification of the traA227P→L missense mutation within PA14 highlights the importance of this domain for function (Figure 4). We also note that Dictyostelium discoideum, a eukaryotic soil slime mold that exhibits similar multicellular behaviors as M. xanthus [41], produces two secreted signals, called DicA1 (PsiF) and PsiA, whose proteins contain PA14 domains followed by cysteine-rich repeats (Pfam00526) of various lengths that show some resemblance to TIGR04201 [21], [27], [42], [43]. Thus, M. xanthus and other microbes, including eukaryotes, appear to utilize PA14 encoding proteins as extracellular signaling and recognition molecules to mediate social interactions.
Recent bioinformatic analysis suggests gram-negative bacteria encode C-terminal protein sorting tags that function analogously to the well-characterized gram-positive LPXTG/sortase system [29]. In the case of MYXO-CTERM, we postulate that this motif forms a transmembrane α-helix and anchors pre-TraA into the IM [29], [31]. Here the Arg rich C-terminal tail would reside in the cytoplasm, while the remainder of the protein would be in the membrane or periplasm (Figure 5). Thus analogous to lipoprotein processing [44], an acyl transferase could attach a lipid moiety via a thioether bond to the invariant Cys (Figure 5 and Figure S7). Subsequently, an endoprotease would cleave the TIGR03901 motif downstream of the aforementioned Cys residue. Once processed a system analogous to the Lol pathway could transport these proteins to the cell surface.
As the traB gene overlaps in a bicistronic operon with traA (Figure 4A) and mutations in each gene elicit identical phenotypes, suggests that TraAB likely function in the same transport pathway. Since the C-terminal region of TraB contains an OmpA-like domain (Pfam00691), it likely binds non-covalently to the cell wall. The N-terminal region constitutes the majority of this protein (∼400 amino acids) and has no ascribed function (Figure 4), but theoretically could interact with the OM and even traverse the OM to interact with TraA. It is also plausible that TraB may facilitate TraA's localization to the cell surface.
A working model for the mechanism of cell contact-dependent exchange is outlined in Figure 9. First, cell-cell recognition is postulated to be mediated by TraA serving as a cell surface receptor. We suggest that the distant PA14 domain may function in ligand binding to neighboring cell surfaces. Glycans found in LPS or glycoproteins are possible ligands. In a variation of this model TraA may function as a homophilic receptor. Similar to the FLO1 system, a key component of this model involves reciprocal TraA binding by both cells. A ‘donor’ cell was arbitrarily assigned and its OM (mCherry) lipoproteins were symbolized as red lollipops. Upon aligned cell-cell contact and docking the OM membranes of adjoining cells fuse. Although not directly depicted, TraAB may facilitate membrane fusion by bringing OMs into close proximity and perhaps causing local membrane perturbations that help catalyze OM fusion. Membrane fusion may also be facilitated at cell poles where the membranes have high tip curvatures and thus are more fusogenic [45]. Once cells are adhered cell motility could also stress the membrane. Upon OM fusion, lipids and lipoproteins rapidly exchange bidirectionally; a process presumably driven by lateral diffusion. Integral and associated OM proteins are also likely transferred as the CglE and CglF proteins encode type I signal sequences [14]. It is unknown whether soluble periplasmic proteins are transferred. Prior studies clearly indicate inner membrane lipoproteins and cytoplasmic proteins are not transferred [6]. Following fusion cells physically separate, a process likely facilitated by gliding motility.
The exchange of OM lipoproteins has phenotypic consequences to the cell, including complementation of mutational defects (Figure 1). Whether the restoration of mutation defects is ecologically important is unknown; however population heterogeneity within biofilms, especially from an environmental setting are significant [4], and consequently some individuals within a population are less fit. Thus, we hypothesize that the ability to exchange and share the OM proteome allows some individuals to gain fitness and for the population to establish OM homeostasis. In turn, homeostasis may increase population fitness by normalizing intercellular signal output and reception by reducing population heterogeneity. Thus community behaviors, such as swarming and development might be better coordinated. In this respect, our findings that a mixture of nonmotile cells with motile cells inhibits the latter cells from swarming in a TraAB and time dependent manner (Figure 7), suggests these cells are communicating and coordinating their behaviors via OM exchange. Similarly, OM exchange can regulate development behaviors between genetically distinct strains (Figure 8). The use of strain mixtures to study cell-cell interactions in motility and development is ecologically relevant, as diverse M. xanthus isolates are found in close proximity in nature [33], [46]. The mechanism for developmental inhibition by nonmotile cells on motile cells is unknown, but may simply reflect a block in motility (Figure 7) [47]. Alternatively or in addition, OM exchange with nonmotile cells may transmit a signal that blocks development. Currently, we are investigating the nature of these putative signals.
Our results indicate that myxobacteria exchange and thus share a significant amount of their cellular resources. This has led us to hypothesize that cell contact-dependent OM exchange represents a form of cooperative social behavior that may involve kin recognition. A kin recognition mechanism avoids the theoretical and ecologically relevant concern that ‘cheater’ cells could exploit or disrupt this social behavior to obtain resources [48]. This problem is highlighted by observations that environmental M. xanthus populations arise from diverse origins [33], [46]. Thus unlike artificial laboratory settings where multicellular behaviors are typically studied with a single homogenous culture, natural myxobacteria isolates must recognize kin from non-kin cells as they vacillate between single cell and multicellular life.
The data presented here provide three lines of evidence that cell contact-dependent OM exchange involves kin recognition. First, TraAB proteins are required in both ‘donors’ and ‘recipients.’ Thus if one cell does not express TraAB, exchange cannot occur. Second, exchange appears bidirectional, thus both cells are giving and receiving. Therefore, there is no inherent advantage one cell type has over another, unless one cell is starving and has depleted resources. Third, TraA contains a PA14 domain, with features resembling PA14 domains in yeast flocculin proteins involved in kin recognition and social behaviors. More specifically, flo1 and other genes within this group were classified as ‘greenbeard’ genes, which by molecular definition are cell surface receptors that recognize other cells carrying the same gene to provide social preferential treatment [49], [50]. In the case of FLO1 the protein allows yeast cells to enter the protective domain of a floc, where cells are so tightly joined they become deformed. Within flocs cells are protected from environmental stresses and cheater cells (flo1−) cannot enter [40]. In another greenbeard example, the Dictyostelium csA gene, which encodes a homophilic cell surface receptor, plays a discrimination role in partitioning cells to desirable locations within fruiting bodies [51]. Current experiments are testing whether TraA plays such a role.
Bacterial strains and plasmids are listed in Table S2 [52]. M. xanthus was grown at 33°C in CTT medium (1% casitone, 1 mM KH2PO4, 8 mM MgSO4, 10 mM Tris-HCl, pH 7.6) in the dark and when necessary supplemented with kanamycin (Km; 50 µg/ml), oxytetracycline (Tc; 15 µg/ml), or streptomycin (Sm; 600 µg/ml). For ½ CTT, casitone was reduced to 0.5%. On plates, agar concentration was 1.0 or 1.2%. TPM buffer contains 10 mM Tris, 1 mM KH2PO4 and 8 mM MgSO4, pH 7.6. Escherichia coli was grown at 37°C in LB medium and when necessary supplemented with Km (50 µg/ml), ampicillin (100 µg/ml) or Sm (100 µg/ml).
The DK396 genome was sequenced by using Illumina second generation DNA sequencing technology (NCGR, Santa Fe, NM). Sequence reads were aligned and analyzed for mutations against the wild-type DK1622 reference genome within the Alpheus bioinformatic platform [53].
DNA cloning followed routine protocols [54]. Chromosomal and plasmid DNA was isolated with UltraClean Microbial DNA and Mini Plasmid isolation kits (MO BIO Laboratories, Inc.), respectively, as described by the manufacture. All insertion mutations were created by PCR amplification of internal gene fragments with Taq 2X Master Mix (New England BioLabs) followed by direct cloning of products into pCR2.1 TOPO (Invitrogen) and then transformed into DH5α. To overexpress the traAB operon it was fused downstream of the strong pilA promoter with an optimally designed ribosomal binding site [55]. Specifically, the pilA promoter was amplified with Phusion High-Fidelity PCR Master Mix with HF Buffer (New England Biolabs) and cloned into pSWU19 at the EcoRI to XbaI restriction sites [18]. traAB was then similarly amplified and cloned into the XbaI and HindIII sites. Primers are listed in Table S3. Plasmid constructs were confirmed by restriction digestion analysis or DNA sequencing. Verified plasmids were electroporated into M. xanthus and integrated into the genome by homologous recombination with antibiotic selection [34]. To identify the donor defect mutation from DK396, insertion mutations were made in DK6204 [56] or DK8601 A−S− donor strains (Table S1). Mx4 or Mx8 bacteriophages were used for strain construction by generalized transduction [15]. Mutants were verified by phenotypes and molecular methods including PCR and sequencing.
M. xanthus strains were grown to a Klett ∼100 (∼3×108 cfu ml−1), concentrated by centrifugation and resuspended to a calculate Klett of 1000 in TPM buffer. For stimulation, donors and recipients were mixed at a 1∶1 ratio and 3 µl were pipetted onto ½ CTT 1% agar pads containing 3 mM CaCl2 (added after autoclaving) and incubated in a humid chamber for various times. Micrographs were taken with either an Olympus SZX10 stereo microscope (whole colony) or a Nikon E800 phase contrast/fluorescent microscope (colony edge) coupled to digital imaging systems.
A heterologous fluorescent OM lipoprotein reporter, called SSOM-mCherry, was used to monitor protein transfer in live cells [6]. To clearly differentiate recipients from SSOM-mCherry expressing donors, the former cells expressed the green fluorescent protein (GFP). Thus, in general terms, protein transfer was scored as the ability of green cells to become red. Lipoprotein transfer was microscopically determined by mixing donor and recipients (1∶3 or 1∶1 ratios) and either (i) detected as motile recipient flares emerging from inoculum spots with nonmotile donors, or by (ii) harvesting cell mixtures and inspecting single cells on glass slides as previously described [6]. To reduce background fluorescence, the former cells were spotted on a thin TPM agarose (1%) pads prepared on a glass slide.
A sampler kit (Invitrogen; cat# L7781) containing different lipophilic fluorescent dyes were evaluated for M. xanthus OM staining. According to the manufacture these dyes are not transferred from stained to unstained cells. DiD oil (component B; DilC18(5) oil) was chosen for further studies where a Texas Red-4040B (Semrock) filter set was used to visualize staining. Cells were grown to Klett ∼100, harvested by centrifugation and resuspended in TPM buffer to a calculated Klett of 250. To stain cells, 1 µl of dye (1 mg/ml, dissolved in ethanol) was added to 49 µl of cells and incubated for 1 to 2 hrs in the dark at 33°C. Cells were then pelleted by centrifugation, washed with 1 ml TPM and microscopically examined (100× objective). Similar to monitoring SSOM-mCherry transfer, dye transfer was also assayed by mixing stained donors with GFP labeled recipients (1∶1 ratio) and spotted on a ½ CTT 1% agar. After 4 hrs incubation, cells were scraped from the agar surface, washed 2× in 1 ml TPM, placed on glass slide with cover-slip and inspected whether green cells also stained red.
Log phase M. xanthus cultures were concentrated by centrifugation to a calculated Klett of 1000 and pipetted onto TPM starvation agar (four 25 µl spots) and incubated for 5 days at 33°C. Cells and spores were harvested and placed into a tube with 500 µl of TPM buffer, heated at 50°C for 2 hrs and then gently pulse sonicated to disperse spores. Spore suspensions were serial diluted and 10 µl samples spotted on CTT agar. After 7 days of incubation, viable spores were enumerated as CFUs. All developmental assays were done in triplicate and averaged.
PA14 domain analysis and alignments are described in Results and Figure S4. The cysteine-rich repeat of MXAN_6895 was identified by inspection. TIGRFAMs model TIGR04201 was developed by multiple sequence alignment of several repeats, HMM construction, search against a large collection of proteins from prokaryotic reference genomes, and iteratively refined. In proteins identified by TIGR04201 as having at least one copy of the repeat, additional, lower-scoring repeats are confirmed by manual inspection of HMM search results. Completed HMMs were added to the TIGRFAMs database, which uses the HMMER 3.0 software package [57].
A search was undertaken for candidate protein-sorting domains with architectural elements similar to the LPXTG-containing recognition sequence of sortase A [28], the PEP-CTERM putative recognition sequence of exosortase and the PGF-CTERM putative recognition sequence of archaeosortase A [29]. The common architecture was; signature motif, hydrophobic predicted transmembrane helix, cluster of basic residues, positioned at the extreme C-terminus and found in protein regions lacking other homologies. A general purpose classifier, TIGRFAMs [58] HMM TIGR03901, was constructed to model a candidate protein-sorting signal domain approximately thirty-three residues long, with an invariant Cys residue in its signature motif, universal in but restricted to the eight species of Myxococcales among 1460 prokaryotic reference genomes; scoring thresholds give no false-positive in any species. To identify atypically low-scoring instances of the domain in M. xanthus, a species-specific HMM was derived from TIGR03901 by HMM search, inspection of results, realignment, and repetition of the search through several iterations. Extensive biocuration of the similar but shorter GlyGly-CTERM motif found primarily in gammaproteobacteria, modeled by HMM TIGR03501, improved the disambiguation of GlyGly-CTERM (which does not occur in M. xanthus) from MYXO-CTERM.
Approximate atomic coordinates for the PA14Tra structure was automatically generated from the alignment of known PA14 domains (Figure S4, residues 62 to 259). This was done by using a standard two-step template-based modeling protocol. The initial 3-D model was obtained using MODELLER 9.9, with 2XJP (FLO5) as template structure (default parameters, best-scoring of 20 models) [59]. To produce the final model, side-chain atoms were refined using SCWRL4 [60].
|
10.1371/journal.ppat.1002635 | Leishmania Induces Survival, Proliferation and Elevated Cellular dNTP Levels in Human Monocytes Promoting Acceleration of HIV Co-Infection | Leishmaniasis is a parasitic disease that is widely prevalent in many tropical and sub-tropical regions of the world. Infection with Leishmania has been recognized to induce a striking acceleration of Human Immunodeficiency Virus Type 1 (HIV-1) infection in coinfected individuals through as yet incompletely understood mechanisms. Cells of the monocyte/macrophage lineage are the predominant cell types coinfected by both pathogens. Monocytes and macrophages contain extremely low levels of deoxynucleoside triphosphates (dNTPs) due to their lack of cell cycling and S phase, where dNTP biosynthesis is specifically activated. Lentiviruses, such as HIV-1, are unique among retroviruses in their ability to replicate in these non-dividing cells due, at least in part, to their highly efficient reverse transcriptase (RT). Nonetheless, viral replication progresses more efficiently in the setting of higher intracellular dNTP concentrations related to enhanced enzyme kinetics of the viral RT. In the present study, in vitro infection of CD14+ peripheral blood-derived human monocytes with Leishmania major was found to induce differentiation, marked elevation of cellular p53R2 ribonucleotide reductase subunit and R2 subunit expression. The R2 subunit is restricted to the S phase of the cell cycle. Our dNTP assay demonstrated significant elevation of intracellular monocyte-derived macrophages (MDMs) dNTP concentrations in Leishmania-infected cell populations as compared to control cells. Infection of Leishmania-maturated MDMs with a pseudotyped GFP expressing HIV-1 resulted in increased numbers of GFP+ cells in the Leishmania-maturated MDMs as compared to control cells. Interestingly, a sub-population of Leishmania-maturated MDMs was found to have re-entered the cell cycle, as demonstrated by BrdU labeling. In conclusion, Leishmania infection of primary human monocytes promotes the induction of an S phase environment and elevated dNTP levels with notable elevation of HIV-1 expression in the setting of coinfection.
| Leishmaniasis is a parasitic disease that infects several human host immune cells, including neutrophils, monocytes, and macrophages. Moreover, while HIV-1 infects monocytes and macrophages, only the infected macrophages productively release viral progenies. Importantly, patients coinfected with both pathogens progress more rapidly to AIDS. In this study, we examine how Leishmania major changes the cellular environment of monocytes in vitro. We found that Leishmania-infected monocytes actively mature into macrophages in the absence of GM-CSF, and that these cells up-regulate the expression of ribonucleotide reductase, an enzyme that catalyzes the formation of deoxynucleoside triphosphates (dNTPs). We confirmed the elevation of dNTP concentrations using a very sensitive dNTP assay for monocytes and monocyte-maturated macrophages. Collectively, these data support a model in which infection of monocytes with Leishmania elevates the intracellular dNTP pools, which is one of the natural anti-viral blocks to HIV-1 infection in monocytes and macrophages in patients.
| Leishmaniasis has recently been recognized to be both one of the world's most neglected and most important parasitic diseases, threatening an estimated 350 million people worldwide [1], [2]. Surveys have estimated that approximately 12 million people are currently infected with 2 million new cases reported yearly, primarily afflicting the world's poorest populations in some 88 countries [3]. Leishmaniasis is transmitted to humans by the bite of the female Phlebotomine sandfly upon taking a blood meal [4]. Infection results in three basic clinical presentations. Cutaneous and mucocutaneous leishmaniasis are disfiguring and even mutilating diseases, while visceral leishmaniasis (VL) is characterized by fever, massive hepatosplenomegaly, pancytopenia, and a wasting syndrome called Kala-azar, which is nearly uniformly fatal without treatment [5], [6].
Early after the emergence of the global Human Immunodeficiency Virus Type 1 (HIV-1) epidemic, clinicians recognized that reciprocal activation of each pathogen by the other frequently occurred. It was noted, on the one hand, that infection with HIV-1 modifies the natural history of leishmaniasis, leading to 100–2,230 times increase in the risk of developing VL and reducing the likelihood of a therapeutic response [7]–[11]. At the same time, VL was shown to induce activation of latent HIV-1, increase viral load, and cause a striking acceleration in the progression of asymptomatic HIV-1 infection to AIDS that corresponded to a reduction of life expectancy in patients [12]–[15]. Similarly, it was recognized that monocytes and macrophages are the primary cell types coinfected with both HIV-1 and Leishmania. Initial studies demonstrated that Leishmania coinfection reactivated HIV-1 replication in latently infected monocytoid cell lines [16]. Subsequent studies in primary MDMs coinfected with L. infantum and HIV-1 also found enhanced HIV-1 replication associated with increased secretion of the pro-inflammatory cytokines TNF-α, IL-1α, and IL-6. In these experiments, HIV-1 replication, as measured by p24 ELISA, was reduced in the presence of either chemical inhibitors or blocking antibodies to these three cytokines [17].
Human monocytes circulate in the blood and reside in bone marrow and spleen and are generally believed not to proliferate in the steady state [18], [19]. However there is an emerging awareness that human monocytes possess far greater heterogeneity than originally perceived, and subpopulations of monocytes have recently been described that can re-enter the cell cycle in response to both Macrophage- and Granulocyte Macrophage-Colony Stimulating Factors (M-CSF and GM-CSF, respectively) [20]–[22]. Proliferation of these presumably immature peripheral blood monocyte subpopulations has been demonstrated by multiple techniques including uptake of 5-bromo-2′-deoxyuridine (BrdU) and CFSE labeling, leading to this population being termed “proliferative monocytes” [23], [24].
Such cellular proliferative capacity has important implications because cellular dNTP levels correlate directly with the replicative capacity of mammalian cells [25]. Consistent with this observation, a variety of studies, including those from our laboratory, have reported that dNTP levels are consistently higher in dividing versus non-dividing cells [25]–[31]. Among the retroviruses, HIV-1 possesses the unique ability to infect both dividing (activated CD4+ T cells) and non-dividing cells (macrophages). This ability is due, at least in part, to the evolutionary adaptation of its reverse transcriptase (RT) to function under conditions of extremely limited dNTP availability [32]. However, as noted for the replicative capacity of mammalian cells, HIV-1 replication efficiency is also directly correlated with cellular dNTP concentrations and proceeds with far greater efficiency in both tumor cells and PHA-stimulated CD4+ T cells, in which the average dNTP levels are 150–225 times higher than that of non-dividing MDMs [32], [33]. Several recent studies have shown that HIV-2 Vpx protein promotes the degradation of the SAMHD1, a host anti-viral restriction factor [34]–[37]. Recently, SAMHD1 was shown to function as a dNTP hydrolase [38], [39], limiting the cellular dNTP pool and restricting HIV-1 replication in cells of myeloid lineage [40]. Moreover, our recent paper shows a direct connection between SAMHD1 degradation, an increase in dNTP levels and enhanced transduction of HIV-1 in myeloid cells [41].
In the present study, we found that in vitro infection of freshly isolated, undifferentiated CD14+ primary human monocytes with Leishmania consistently led to maturation into macrophages and to higher cell numbers over time as compared to uninfected control cells. In addition to the inhibition of apoptosis previously reported in Leishmania-infected MDMs, we also report the unexpected finding that a sub-population of CD14+ human MDMs proliferate in response to Leishmania, as measured by BrdU incorporation at days 12–14 after infection. As the efficiency of HIV-1 RT DNA synthesis and subsequent viral replication are directly dependent on cellular dNTP concentration, we subsequently employed a highly sensitive single nucleotide incorporation assay that was recently developed in our laboratory to measure cellular dNTP concentration [32], [42], [43]. We found a marked increase in the content of dNTPs in Leishmania-maturated MDMs as compared to uninfected control cells. Consistent with this observation, elevated levels of ribonucleotide reductase (RNR), the rate-limiting enzyme for dNTP synthesis, was also found in Leishmania-maturated MDMs as compared to control cells. Finally, we found significantly enhanced expression and transcription of a GFP-expressing pseudotyped HIV-1 (HIV-1 D3 GFP) in Leishmania-maturated MDMs as compared to control cultures as assayed by FACS analysis of HIV-1 D3 GFP expressing cells and qPCR for 2 LTR-circle copy number.
As noted above, previous studies have suggested a role for Leishmania infection of monocytes causing the induction of pro-inflammatory cytokines as a stimulus to HIV-1 replication in coinfected cells. Our data support a novel model whereby Leishmania infection stimulates monocytes' differentiation and cell division. Consistent with the increased proliferation capacity, Leishmania infection increases cellular dNTP concentrations that facilitate enhanced HIV-1 coinfection.
The effect of Leishmania infection on cell survival of primary human monocytes was examined over a time course of 28 days from eleven individual donors. Preliminary experiments were performed to examine potential effects of both heat-inactivated Leishmania (also applied to monocytes at an MOI = 7) and day 7 conditioned medium from Leishmania-infected monocytes re-applied to freshly isolated monocytes. These experiments demonstrated no significant effects on monocyte cell survival, maturation, or proliferation (data not shown). In parallel experiments, Leishmania labeled with the vital dye PKH showed that at an MOI = 7 virtually all monocytes within the culture became infected (Figure S1). This MOI is well within the range of those previously published [16], [17].
Purified human monocytes were cultured at 1×106 cells/well in 6 well dishes, and three wells from each of three culture conditions were combined and counted: 1) RPMI media with 10% FBS (“control cells”), 2) RPMI media with 10% FBS plus 5 ng/ml human recombinant GM-CSF (“GM-CSF”), or 3) RPMI media with 10% FBS with Leishmania major (MOI = 7) at the time of plating (“Leishmania”). The GM-CSF-treated monocytes differentiate into MDMs and were used as a positive control for all the studies. Medium was changed at day 7 and then weekly, replating any non-adherent cells into their respective wells. As illustrated in Figure 1A, a marked decline in the cell numbers was seen at day 3 after initial plating in all three conditions, though more notably in the control monocytes as compared to either GM-CSF-treated or Leishmania-infected monocytes. Cell numbers fell from 3×106 at day 0 in all three conditions and were consistently lower in control cells as compared to either GM-CSF-treated (positive control) or Leishmania-infected cells at all times tested from day 3 to day 28. Control monocyte numbers declined until day 7, when their numbers stabilized through day 28. Cell numbers for GM-CSF-treated and Leishmania-infected groups remained significantly higher than control monocytes at all time points from day 3 to day 28 (Friedman test; p<0.05).
Next, we examined the different cell populations using light microscopy. The control cells largely retained a small, mostly rounded morphology (Figure 1B) at day 14 as compared to either GM-CSF-treated (Figure 1C; positive control) or Leishmania-maturated MDMs (Figure 1D). For both treatments, the monocytes were larger, more adherent and spread out with some processes, which is characteristic of mature macrophages.
Using FACS analysis, the GM-CSF-treated and Leishmania-maturated MDMs were larger (as assayed by forward scatter) with greater cellular complexity (assayed by side scatter) as compared to control monocytes (Figure S2). These findings were further confirmed and quantitated by FACS analysis of cell surface CD14 expression from six independent donors. This demonstrated a decreased cell surface expression of CD14 (CD14low) in both day 14 GM-CSF-treated and Leishmania-infected MDMs as compared to control monocytes (CD14high), again consistent with monocytes to macrophages maturation in the GM-CSF and Leishmania-infected cultures (Figure S2). FACS analysis for both Annexin V and propidium iodide also showed pronounced reduction in cell death for the Leishmania-infected monocytes compared to uninfected controls (Figure S3) Collectively, these data suggest that Leishmania infection of monocytes leads to less cell death and increased cellular maturation towards a macrophage phenotype compared to control monocytes.
While performing the kinetic studies of Leishmania-infected monocytes, we observed clusters of small cells lying on top of larger, more differentiated appearing macrophages in both the Leishmania-infected and GM-CSF-treated (positive control) cultures but not for the control cell culture. Although, as noted above, human monocytes are generally believed not to proliferate once released from the bone marrow [18], [19], it has been more recently recognized that these cells possess far greater heterogeneity than originally believed and subpopulations of monocytes have been recently described that can re-enter the cell cycle in response to M-CSF and GM-CSF [20]–[22]. Proliferation of these monocyte subpopulations has been demonstrated by multiple techniques including uptake of BrdU and CFSE labeling [23], [24]. Thus, we next asked whether their presence might also be induced in the setting of Leishmania infection. To address this, we did a time-course analysis at days 3, 7, 10, and 14, examining BrdU uptake at 48 hours after treatment for the Leishmania-infected groups [24]. As expected, we detected a few cells that were uniformly BrdU+ (green) and nuclei counterstained with DAPI (blue) (Figure 2A). We detected a progressive increase in the numbers of BrdU+ cells over time, with maximal numbers of BrdU+ cells observed at day 14 of cell culture. Lastly, we co-labeled primary human monocytes with PKH-labeled L. major (orange) and then pulsed with BrdU (Figure 2A, bottom right panel Day 21). BrdU+ nuclei were seen in Leishmania-infected cells suggesting that infection may promote re-entry into the cell cycle for a sub-population of cells. This may be of importance to the dissemination of Leishmania within a host because macrophages are generally considered terminally differentiated, non-dividing cells [19].
We subsequently performed quantitative FACS analyses to compare the percentages of BrdU+ cells. As shown in a representative FACS plot, Figure 2B, a relatively large sub-population of BrdU+ cells was seen in both Leishmania-infected (13.4%) and GM-CSF-treated cells (14.8%) but not in control cells (<1.0%). Figure 2C summarizes results for 48 hour BrdU incorporation for seven independent donors between days 12–14. Leishmania-maturated MDMs demonstrated highly statistically significant (p<0.01) elevations of the percentage of BrdU+ cells as compared to control cells while GM-CSF-maturated MDMs were significantly (p<0.05) higher. We also CSFE-labeled fresh monocytes and found at least one cell division in a small subpopulation of cells for the GM-CSF-treated and Leishmania-infected groups (data not shown). Collectively our results are consistent with previous studies of a proliferative monocyte sub-population that can be stimulated to enter cell division by the related monokine M-CSF [21], [24]. However, of greater relevance is the demonstration that L. major infection of monocytes can induce an S phase environment as assayed here by BrdU incorporation. Whether this promotes cell division in vivo, allowing for greater dissemination of Leishmania, remains unclear.
We employed the highly sensitive HIV-1 RT based assay for measuring cellular dNTP content [32], [42]–[45]. As depicted in Figure 3A, HIV-1 RT is bound to a template/primer complex. HIV-1 RT can extend the primer by one nucleotide, depending on the template nucleotide (N) present at the 5′ end of the template. This assay allows for the determination of differences between cellular extracts for a specific cellular dNTP. Using this assay, we compared the cellular content of dGTP (purine) and dTTP (pyrimidine) for the different treatment groups. Figure 3B shows a representative result for primer extension of dGTP (left panel) and dTTP (right panel). Summary results for nine individual donors are presented in graph form in Figure 3C and are summarized below.
In Figure 3B, left side panel, dGTP levels were assayed at days 7 and 13, while the right side panel shows dTTP analysis for the same days. In lanes 1 for both dGTP and dTTP analysis, no dNTPs were added to the reaction, leading to no extension product of the labeled primer (open arrow). In lanes 2, exogenous dNTPs were added as a positive control to show extension of all primers in the reactions (closed arrow). In lanes 3–8, days 7 and 13 cellular extracts were analyzed. Content of dGTP were notably higher in GM-CSF- and Leishmania-maturated MDMs as compared to untreated control cells at day 7 (lanes 4 and 5) and day 13 (lanes 7 and 8) after treatment. In comparison, dTTP concentrations at day 7 were slightly higher for the GM-CSF-maturated MDMs (lanes 4, positive control) as compared to the control and Leishmania-maturated MDMs. At day 13, we detected much higher dTTP concentrations in the GM-CSF and Leishmania-maturated MDMs at day 13 (lanes 7 and 8) as compared to the control group (lanes 6). These data demonstrate that Leishmania infection can lead to notable increases in cellular dNTP concentrations and this conclusion is fully validated by quantification of the assay results in nine individual donors (Figure 3C). Results for dGTP (Figure 3C, upper panels) demonstrated statistically significant increases for GM-CSF-matured and Leishmania-infected MDMs as compared to control monocytes at day 7; by day 13 dGTP increases were now highly significantly elevated in Leishmania and still significantly elevated in the GM-CSF-maturated MDM groups as compared to controls. The results for dTTP at day 7 (Figure 3C, lower left panel) trended higher in Leishmania-maturated MDMs as compared to monocyte controls but only reached significance in GM-CSF-maturated MDMs. However at day 13, (Figure 3C, lower right panel) Leishmania-maturated MDMs were significantly increased in dTTP concentrations as compared to monocyte controls. These data demonstrate that Leishmania infection of monocytes induces elevation of both purine and pyrimidine concentrations in the host cell. The finding of elevated purine levels is particularly intriguing in light of the fact that Leishmania species are entirely dependent on host cell synthesis for their supply of purine nucleotides [46].
Mammalian RNR is a dimeric enzyme essential for catalyzing the direct reduction of relatively large intracellular pools of ribonucleotides into the corresponding deoxyribonucleotides for DNA synthesis. The catalytic enzyme is a heterodimer, containing two subunits of R1 and either two subunits of R2 or p53R2. Expression of the R2 subunit is strictly limited to the S phase of the cell cycle [47]. As shown in Figure 4A, western blot analyses were done for R2 and p53R2 on cell extracts using freshly isolated monocytes, day 13 GM-CSF or Leishmania-maturated MDMs. As shown in Figure 4B, we quantitated the western blots for four independent donors and found that R2 was significantly (p<0.05) increased in the Leishmania-maturated MDMs over control monocytes. For the p53R2, we found a significant increase in the GM-CSF-treated cells but the increase failed to reach significance for the Leishmania-infected cells when compared to monocytes, which were set to 1. Moreover, the R2 and p53R2 antibodies were specific for human ribonucleotide reductase and did not cross-react with L. major (data not shown). Collectively, these data show that 1) R2 subunit expression, which is S phase linked, is significantly increased upon Leishmania infection, and 2) that infection indirectly leads to an increase in the p53R2 subunit, which is involved in increasing cellular dNTP concentrations in non-dividing cells.
Next, quantitative reverse transcriptase quantitative PCR (qRT-PCR) using Taqman analysis was performed in three individual donors to examine whether the observed increase in RNR R2 subunit and P53R2 protein expression showed transcriptional regulation (Figure 4C). Consistent with the significantly increased protein expression of the RNR R2 subunit seen by western blot, significantly increased transcription was seen in Leishmania-infected monocytes as compared to GM-CSF-treated MDM and control monocytes. It is also possible that these results may be due, at least in part, to an increase in RNR R2 transcript stability. In contrast, increased expression of p53R2 protein likely occurs due to post-transcriptional regulation as no significant elevation of transcription was seen in either the GM-CSF-treated or Leishmania-infected MDMs as compared to control monocytes.
As noted above, cellular dNTP levels serve as a biomarker for the replicative capacity of mammalian cells, a finding corroborated by the presence of consistently higher dNTP levels in dividing cells as compared to non-dividing cells [25]–[30]. HIV-1 replication efficiency is also directly correlated with the cellular dNTP concentration, and we and others have reported that it proceeds with far greater efficiency in tumor cells or PHA-stimulated CD4+ T cells in which the average dNTP level is 150–225 times higher than in non-dividing monocytes/macrophages [32], [33]. Given our findings that Leishmania infection induces both significant elevation of dNTP levels and replication capacity in MDMs, we examined whether transduction of Leishmania-maturated MDMs with a VSV-g pseudotyped HIV-1 vector, designated HIV-1 D3 GFP, resulted in accelerated HIV-1 expression, as determined by GFP expression. Six days after isolation, control cells, GM-CSF maturated MDMs, and PKH-labeled (red) Leishmania-maturated MDMs were transduced in 6-well dishes with equal amounts of HIV-1 D3 GFP vector. We examined cells by bright field and fluorescence microscopy 24 hours later (Figure 5A). HIV-1 D3 GFP expression (“GFP” [green-top 3 panels]) was markedly enhanced, relative to control cells, in both the GM-CSF- and Leishmania-maturated MDMs (upper middle and right-sided panels, respectively), consistent with both strikingly increased intensity and numbers of HIV-1 D3 GFP transduced cells in these two conditions relative to control cells (Figure 5A; top left panel). Only a rare control cell appeared to express HIV-1 D3 GFP. GM-CSF- and Leishmania-maturated MDMs had many more cells expressing GFP as compared to control cells. Leishmania maturated MDMs labeled with PKH showed comparable numbers of GFP+ cells/field as compared to GM-CSF-maturated MDMs (Figure 5A; middle and top right panels). We next quantified the three different groups by FACS analysis (Figure 5B and Table 1). For these studies, the Leishmania were not labeled with PKH dye. As shown in Table 1, cells from four independent donors were examined at 24 and 48 h after the addition of the HIV-1 D3 GFP vector. The percent of GFP+ cells for Leishmania-maturated MDMs were consistently higher as compared to the control cell group with a somewhat weaker trend to higher percentage of GFP+ cells also found in the GM-CSF-maturated MDMs as compared to control cells.
Next, we examined 2LTR circles, an indicator for the completion of DNA synthesis by HIV-1 reverse transcriptase but a failure of the DNA to integrate into the host genome. As shown in Figure 6, the 2LTR circles copy number ratio was significantly higher (*, p<0.05) in the Leishmania maturated MDMs group as compared to control cell group (set to 1.0). The 2LTR circle number ratio for GM-CSF-maturated MDMs group is higher than the controls cells, but did not achieve statistical significance. Collectively, these data indicate that Leishmania infection promotes a pro-HIV-1 environment within the cell, leading to higher dNTP concentrations that allow for more efficient viral infection.
In mutually endemic areas of the world, Leishmania species and HIV-1 primarily co-infect mononuclear phagocytes of infected mammalian hosts. It is widely believed that Leishmania infection found concurrently with HIV-1 induces a state of chronic immune activation leading to subsequent increased HIV-1 viral load and accelerated progression to AIDS [48]. Although the mechanisms underlying this phenomenon are incompletely understood, in vitro studies to date have implicated a variety of Leishmania-induced pro-inflammatory cytokines including TNF-α, IL-1, and IL-6, in stimulating HIV-1 replication in both monocytoid cell lines and macrophages [17], [49]–[52]. For example, the induction of TNF-α is known to activate HIV-1 replication through mechanisms involving transcriptional activation of nuclear factors binding to NF-κB sequences in the HIV-1 LTR [49], while IL-6 and IL-1 appear to promote HIV-1 replication through less well-defined NF-κB-independent transcriptional and post-transcriptional mechanisms [51], [52]. In this study, a novel mechanism is described in which Leishmania infection of HIV-1 infected CD14+ primary human monocytes promotes accelerated HIV-1 expression by induction of MDMs RNR with subsequent elevation of intracellular dNTP concentrations.
This same mechanism could explain numerous previous in vitro and in vivo observations of accelerated HIV-1 replication in AIDS clinical trials for patients treated with GM-CSF [53]. Soon after the recognition that HIV-1 was the etiologic agent of AIDS, it was recognized that physiological stimuli, including GM-CSF, could exert an inductive effect on HIV-1 replication in infected monocytoid cells, though the potential mechanisms for this induction have remained unknown [54]. Most subsequent studies have largely confirmed this original observation [55]–[61], although some others have demonstrated opposite results with the suppression of HIV-1 replication [62], [63]. In vivo, however, the results of four clinical trials using GM-CSF therapy in HIV-1 infected patients not treated with anti-retroviral drugs all demonstrated increased plasma levels of HIV-1 RNA and p24 antigen as compared to control patients [64]–[67]. Most recently, the results of the previous negative in vitro studies, in which treatment with GM-CSF may have lowered HIV-1 replication, may be reconciled: the majority of results showed that up-regulation of viral replication was generally enhanced in GM-CSF-maturated MDMs when grown at low densities, whereas more crowded cultures of MDMs and excessive acidification of the medium led to suppressed viral replication [68].
Although GM-CSF treatment promotes maturation of monocytes into macrophages, which are terminally differentiated, non-dividing cells, there is an emerging awareness that, although human monocytes do not proliferate in the steady state, a proliferative monocyte sub-population exists that can re-enter the cell cycle in response to both GM-CSF and M-CSF [18], [19], [21], [24], [69]. Leishmania-infected monocytes/macrophages have been found to be able to produce a variety of colony stimulating factors, most notably GM-CSF [70]–[72]. Here we confirm that monocyte sub-populations treated with GM-CSF are able to re-enter the cell cycle and show, for the first time, that Leishmania infection promotes an S-phase environment in normally quiescent monocyte sub-populations. Statistically significant elevated percentages of BrdU+ cells were found in Leishmania-infected MDMs compared to uninfected controls (Figure 2C). Further, both monocyte maturation and proliferation occurred through a mechanism independent of GM-CSF as treatment with a high concentration of neutralizing antibody, fully sufficient to block the effects of 5 ng/ml added GM-CSF had no effect on the Leishmania-infected cells (Figure S4). These findings are in accord with newly described rodent data demonstrating local in situ proliferation of tissue macrophages in response to infection with a rodent filarial nematode [73] and a previous study demonstrating in situ proliferation of macrophages in the lungs of hookworm-infected mice [74].
The promotion of monocyte proliferation by both GM-CSF treatment and Leishmania infection has profound implications for monocyte cell biology. We now report the quite novel finding that monocyte proliferation, induced by the presence of GM-CSF and, more potently by infection with L. major, also promotes significantly higher dNTP levels at days 7 and 13 in culture as compared to freshly isolated peripheral blood monocytes. Elevated synthesis of the purine, dGTP in particular, was highly statistically significant in day 13 Leishmania-infected MDMs compared to levels in control monocytes (Figure 3C). In addition, induction of cellular RNR, the enzyme catalyzing the direct reduction of ribonucleotides to their corresponding dNTPs was found to be significantly elevated in the Leishmania-maturated MDMs. Specifically, an approximately 40-fold increase in RNR protein levels was observed in immunoblots of day 13 Leishmania-maturated MDMs versus freshly isolated monocytes using an antibody directed against the R2 subunit of human RNR (Figure 4B). That this induction of RNR R2 is regulated at the transcriptional level is supported by the similarly statistically significant elevation of RNR R2 RNA assayed by qRT-PCR (Figure 4C). These findings are particularly intriguing in that the expression of the R2 subunit is known to be strictly and specifically restricted to the S phase of the cell cycle [47], consistent with the observed induction of cell cycle re-entry in both Leishmania- and GM-CSF-maturated MDMs.
The present demonstration that Leishmania infection of human monocytes induces elevated dNTP concentrations also has far-reaching implications for Leishmania pathogenesis. Unlike their mammalian hosts, Leishmania lack the metabolic machinery needed for purine nucleotide synthesis. They must therefore rely on the host cell production of purines and have evolved an obligatory purine salvage pathway for this purpose [46]. The dimeric enzyme ribonucleotide reductase is the major source of dNTPs in mammalian and other cells, forming them from the far more abundant pool of rNTPs by the removal of the 2′ OH on the ribose sugar moiety [75]. Our finding that Leishmania infection of human monocytes induces MDMs upregulation of RNR (Figure 4B) is fully consistent with the elevated dNTP concentrations noted above and represents an elegant evolutionary adaptation by which Leishmania can salvage necessary host purines (and pyrimidines). A more recent consequence of Leishmania-mediated induction of host RNR and elevated dNTP concentrations has been to provide a highly permissive environment for HIV-1 replication in the setting of co-infection. These findings are especially significant in light of data that HIV-1 proviral DNA synthesis in non-dividing cells is slower than in dividing cells [76], and can be accelerated by experimentally elevating the intracellular dNTP concentration [42]. They may also be of particular relevance in the setting of infection with Leishmania, in which rapid proliferative expansion of local splenic and bone marrow monocyte/macrophage progenitor populations has been described [70]. In this setting, elevated dNTP concentration would also be expected with accompanying enhancement of HIV-1 replication in such dividing cells.
HIV D3 GFP transduction, a model for HIV-1 infection, is also markedly enhanced in these matured cells (Figures 5A, 5B, and Table 1). Both the fluorescent microscopic and flow cytometry results demonstrated substantially increased numbers of HIV-1 D3 GFP+ transduced cells in the setting of Leishmania infection. These findings were further confirmed by a statistically significant elevation of the 2LTR circle copy number ratio in Leishmania infected MDM compared to control monocytes by qPCR (Figure 6).
Our results for MDMs maturated by GM-CSF treatment or infection with Leishmania conform well to the majority of studies showing enhanced HIV-1 replication, most likely due to monocytes maturating into macrophages. This is a critical finding in that we have recently reported that HIV replication efficiencies in a wide variety of relevant cell types, including monocytes and macrophages, is directly related to the relative intra-cellular dNTP concentrations [31], [32]. Thus, the finding of elevated dNTP levels in both GM-CSF- and Leishmania-maturated human MDMs, as compared to both freshly isolated monocytes and untreated control cells, offers a novel mechanism to explain both the present results as well as prior in vitro and in vivo studies that demonstrate accelerated HIV-1 replication in both GM-CSF-treated and Leishmania co-infected patients [55], [67], [77], [78]. These results are consistent with the 200–1500 times decrease in replication competence of wild-type HIV-1 in monocytes as compared to the corresponding differentiated MDMs [33].
The present study represents the first demonstration that Leishmania promotes both maturation and proliferation phenotypes in primary human monocytes. During this process we detected elevated intracellular dNTP pools in Leishmania-infected cells, which allows more efficient replication of intracellular co-infected HIV-1. This observation of enhanced pathogen expression in co-infected target cells may be a more generalized phenomenon. For example, the course of HIV-1 related immunodeficiency is also known to be accelerated by active infection with Mycobacterium tuberculosis (MTB) [79], and in vitro studies have demonstrated that MTB-infection of MDMs subsequently infected with HIV-1 produce increased levels of virus as compared to MDMs uninfected with MTB [80]. In matched CD4+ T cell cohorts, both HIV-1 viral load and heterogeneity are increased by MTB infection. In addition, infection of monocytes/macrophages with two other clinically relevant Mycobacterium was found to enhance HIV-1 replication both in vitro and in situ [81]–[83]. Conversely, patients co-infected with HIV-1 and MTB have altered granulomas within the lung [84]. Also higher bacterial burden was detected for HIV-1 and MTB co-infection of MDMs in vitro [85]. Our data suggests that we are just beginning to understand the synergy between virus and parasite co-infections of human cells.
These experiments used primary human primary monocytes obtained from human buffy coats (New York Blood Services, Long Island, NY). These are pre-existing materials that are publicly available, and there is no subject-identifying information associated with the cells. As such, the use of these samples does not represent human subjects research because: 1) materials were not collected specifically for this study, and 2) we are not able to identify the subjects.
Primary human monocytes were isolated from the peripheral blood buffy coats by positive selection using MACS CD14+ beads as previously described [32]. Three culture condition were used: 1) RPMI 1640 containing 10% FCS and Penicillin/Streptomycin antibiotics without further supplements indicating “control” monocytes, 2) RPMI containing 10% FCS, Pen/Strep antibiotics and 5 ng/ml human recombinant GM-CSF (R&D Systems) indicating “GM-CSF-treated” monocytes, or 3) RPMI 1640 containing 10% FCS, Penicillin/Streptomycin antibiotics and Leishmania major (MOI = 7) indicating “Leishmania-infected” monocytes. Leishmania major promastigotes (strain WHOM/IR/–/173) were grown to stationary phase culture and infectious metacyclic promastigotes were isolated by negative selection using peanut agglutinin [86]. L. major were labeled with 2 µM PKH26 fluorescent cell dye (Sigma) as per manufacturer's protocol.
HIV-1 D3 GFP vector generation: HIV-1 D3 GFP vector encodes the HIV-1 NL4-3 genome with the eGFP gene in place of the HIV-1 nef gene and has a deleted envelope [32]. To generate virus, 293T cells in T225 flasks were transfected with 60 µg pD3-HIV and 10 µg pVSV-g plasmids using 140 µl polyethyenimine (1 mg/ml) in 37 ml DMEM media/flask. At day 1 of HIV-1 production, media was discarded and replaced with fresh DMEM media. At day 2, media was harvested and replaced with fresh DMEM media. The media was centrifuged at 2500 RPM for 7 minutes to remove cellular debris, and then stored at 4°C in T75 flask. Day 3 media was harvested and processed as described for day 2. HIV-1 D3 GFP was concentrated using ultracentrifugation (22K RPM for 2 h in a SW28 rotor). Viral pellets were DNase I digested for 1 h at 37°C. Afterwards, debris was removed by centrifugation (14K for 5 minutes). Sample aliquots were frozen at −80°C until used. Different groups were transduced with HIV-1 D3 GFP and then the samples were analyzed using Accuri C6 flow cytometer monitoring GFP expression at 24 h or 48 h after transduction. Data files were analyzed using FlowJo software (TreeStar).
Nucleotide incorporation assay employs a 19-mer DNA template (3′-CAGGGAGAAGCCCGCGGTN-5′). The N indicates the change in template for detecting a specific dNTP within the cellular extract. The template is annealed to a 5′ end 32P-labeled 18-mer DNA primer (5′-GTCCCTGTTCGGGCGCCA-3′). HIV-1 RT is used for this reaction [87]. 1×106 cells for control monocytes, GM-CSF-treated MDMs, and Leishmania-infected MDMs were collected and lysed with 60% cold methanol. Cellular debris was cleared by 14K centrifugation. Supernatant was dried. Pellet was resuspended in 20 µl reaction buffer (50 mM Tris-HCl, pH 8 and 10 mM MgCl2). Two microliters were used in the primer extension assay.
Forty-eight hours before harvesting, cells were pulsed with 300 µM BrdU. For microscope analysis, media was removed and the 6-well plate was washed once with PBS. Cells were fixed for using 4% paraformaldehyde for 20 minutes and then washed with PBS. Two milliliters of Target Retrieval Solution (Dako) was added and plates were heated in a rice cooker for 15 minutes at 95°C. Afterwards the plates were removed and allowed to cool. Cells were stained with rat anti-BrdU-FITC antibody (AbD Serotec) for 20 minutes at 4°C. Images were captured using a Zeiss microscope. For FACS analysis, on the day of harvest, the free cells were collected while the adherent cells were Trypsin treated for 30 minutes before scraping the 6-well plate. Both free and adherent cell populations were pooled, centrifuged at 1200 RPM for 5 minutes. Supernatant was removed and the cells were fixed using 4% paraformaldehyde for 20 minutes. After fixing, the cells were washed once with PBS. The cells were stored at 4°C until processing for BrdU staining. For BrdU staining, cells were transferred to a 6-well plate containing 2 ml of Target Retrieval Solution and heated in a rice cooker for 15 minutes at 95°C. Afterwards the plates were removed and allowed to cool. Cells were transferred to tubes and cells washed once with PBS. Next the cells were stained with rat anti-BrdU-FITC antibody for 20 minutes at 4°C. The sample data were collected using an Accuri C6 flow cytometer.
Samples were processed in RIPA buffer containing 1 µM DTT, 10 µM PMSF, 10 µl/ml phosphatase inhibitor (Sigma) and 10 µl/ml protease inhibitor (Sigma). The cells were sonicated with 3X, 5 second pulses, to ensure complete lysis. Cellular debris was removed by 15K RPM centrifugation for 10 minutes. Supernatants were stored at −80°C before use. Cell lysates (25 µg) were resolved on an 8% SDS-PAGE gel. Proteins were transferred to a nitrocellulose membrane. The membrane was blocked with 2% non-fat milk in TBST for 1 h, followed by the addition of primary goat anti-R2 antibody (Santa Cruz Biotechnology) and incubation overnight at 4°C. The next day, the membrane was washed (3X, 20 minutes with TBST) followed by staining with donkey anti-goat HRP for 1 h at room temperature. The membrane was washed 3× with TBST and developed using SuperSignal West Femto Kit (Thermo Scientific). The immunoblot was then stripped and re-probed for actin. Images were captured using a BioRad ChemiDoc Imager.
4×106 cells were lysed and RNA prepared using the RNeasy Mini Protocol as per the manufacturers' instructions (Qiagen, Valencia, CA). Pre-mixed Taqman primer/probe sets for RNR R2 and p53R2 were obtained from Life technologies (Cat numbers Hs01072069_gi and Hs00968432_m1, respectively). Template RNA was diluted to 80 ng/µl and 4 µl from each sample, mixed with Express One-Step SuperScript qRT-PCR reagents, was ran in triplicate using an Applied Biosystems 7300 Real Time thermocycler. Data were normalized to GAPDH mRNA.
Genomic extracts were prepared using QuickGene-810 Nucleic Acid Isolation System (FujiFilm Global). The DNA was assayed for 2LTR circles by real time PCR using the following primers: 5′-LTR region — 5′-GTGCCCGTCTGTTGTGTGACT-3′ and 3′LTR region — 5′-CTTGTCTTCTTTGGGAGTGAATTAGC-3′, and the probe 5′-6-carboxylfluorsecein-TCCACACTGACTAAAAGGGTCTGAGGGATCTCT-carboxytetramethylrhodamine-3′ (IDT). All samples were normalized to total DNA. The control samples for each donor were set to 1.0 and 2LTR circle copy number ratio was plotted.
Prism software was used for plotting the data. All the data sets were compared for significant difference using ANOVA analysis (Friedman test).
|
10.1371/journal.pntd.0005256 | Do Two Screening Tools for Chikungunya Virus Infection that were Developed among Younger Population Work Equally as Well in Patients Aged over 65 Years? | Chikungunya is an endemo-epidemic infection, which is still considered as an emerging public health problem. The aim of this study was to evaluate in a 65+ population, the accuracy of two chikungunya screening scores that were developed in younger people.
It was performed in the Martinique University Hospitals from retrospective cases. Patients were 65+, admitted to acute care units, for suspected Chikungunya virus infection (CVI) in 2014, with biological testing using Reverse Transcription Polymerase Chain Reaction. Mayotte tool and Reunion Island tool were also computed. Sensitivity, specificity, positive predictive value, negative predictive value, and Youden’s statistic were calculated.
In all, 687 patients were included, 68% with confirmed CVI, and 32% with laboratory-unconfirmed CVI. Fever (73.1%) and arthralgia (51.4%) were the most frequent symptoms. Sensitivity ranged from 6% (fever+headache) to 49% (fever+polyarthralgia); and Youden’s index ranged from 1% (fever + headache) to 30% (fever+polyarthralgia). PPV and NPV ranged from 70% to 95%, and from 32% to 43%, respectively.
Performances were very poor for both tools, although specificity was good to excellent. Our results suggest that screening scores developed in young population are not accurate in identifying CVI in older people.
| Chikungunya virus is an alpha-virus transmitted by Aedes egyptii or albopictus bites. This infection is still considered as an emerging public health problem. In the acute stage of infection, typical physical signs of Chikungunya virus infection are febrile illness associated with severe and debilitating polyarthralgia affecting the small joints. Several studies have shown that mortality rates increased during the outbreak. Age over 85 years has been shown to be associated with increased mortality, and the mortality rate is higher in 65+ subjects than among younger population. During epidemics, prevalence rates vary from 18% to 48%. Rapid and reliable diagnosis is required especially for frail elderly population. Diagnosis based solely on physical examination may underestimate the magnitude of the epidemic. The systematic use of biological diagnosis during an outbreak is not feasible, especially in low- and middle-income countries. The use of predictive scores would thus be very helpful in this situation.
| Chikungunya virus infection (CVI) is still considered as an emerging public health problem in both tropical and temperate regions [1]. It is usually symptomatic and may have three phases: acute (day (D)1 to D21), post-acute (D21 to D90), and chronic stage (beyond D90) [2, 3]; the latter two are sometimes absent. In the acute stage of infection, typical physical signs and symptoms of CVI are febrile illness associated with severe and debilitating polyarthralgia affecting the small joints. Severe functional disabilities characterise this phase. Other signs that can be observed include myalgia, headaches, or maculo-papular rash. In most cases, symptoms resolve within a few days with symptomatic treatment [2, 3].
Prior to the outbreak of 2005–2006 in Reunion Island (France), CVI was not considered to be life-threatening. Usually, the any cause overall mortality rate from CVI is considered to be low, comparable to that of seasonal influenza [4]. However, several studies have shown that mortality rates increased during the outbreak as compared to the same period in previous years [5–8]. Fatality increases in populations with atypical presentations, and the incidence of such atypical, serious or fatal cases increases with age. Indeed, age over 85 years has been shown to be associated with increased mortality [8], and the mortality rate is five times higher in subjects aged 65 years or older (65+) than among those under 45 years [5]. On Reunion Island, excess mortality concerned mainly people aged 75 years or older (75+) [9, 10]. Several comorbidities as well as increased age are linked with atypical presentation [11].
During epidemics, CVI prevalence rates are not fully known, and vary from 18% to 48% [12–14]. To meet patients’ needs, rapid and reliable diagnosis is required. Patients with CVI should be identified early, and receive appropriate care. Moreover, people with symptoms and signs consistent with CVI but who suffer from another type of disease must be diagnosed rapidly. Management without delay of differential diagnoses is essential. However, establishing a diagnosis of CVI in a simple and reliable way is very challenging. This concern is especially relevant to the frail elderly population. Furthermore, diagnosis based solely on physical examination may underestimate the magnitude of the epidemic [13].
The systematic use of biological diagnosis during an outbreak is not feasible, especially in low- and middle-income countries (e.g. due to lack of access to laboratory testing, difficulties processing samples, delays in the treatment of patients, etc.). The use of predictive scores would thus be very helpful in this situation. During the outbreak in Mayotte and Reunion Island, two predictive scores were developed. Sissoko et al. [15] retrospectively derived a clinical score (Mayotte tool) in a population of children and young adults. This score was based on the pairing of fever with the four most common clinical signs (polyarthralgia, myalgia, headaches, and back pain). More recently, Thiberville et al. [16] established a clinico-biological score (Reunion Island tool) from a population of patients aged 18 to 65 years. The performances of these scores were good, making them useful screening tools. However, they have not been evaluated in the elderly. Thus, we aimed to evaluate diagnostic performances of these two scores in a 65+ population, admitted to acute care units of Martinique University Hospital, with symptoms suggestive of CVI during the epidemic that occurred in 2014.
This was a diagnostic study performed in the University Hospital of Martinique (French West Indies) from retrospective cases.
Eligible patients were aged 65 years or older, admitted to acute care units including the emergency department (ED), for suspected CVI (presence of fever or arthralgia at admission based on Rajapakse et al 2010), from 10 January to 31 December 2014, and who underwent biological testing using Reverse Transcription Polymerase Chain Reaction (RT-PCR). Patients whose clinical and/or biological data were missing in their medical records, as well as those for whom it was not possible to compute either Mayotte tool or Reunion Island tool, were excluded.
We recorded baseline characteristics including age, sex, time since onset of Chikungunya symptoms, as well as presence or absence of the following features: fever, arthralgia (any of the following: knee, ankle, metacarpo-phalangeal joints, wrist, elbow, shoulder girdle, and pelvis), myalgia, digestive or neurological symptoms, and comorbidity burden (assessed using Charlson’s comorbidity index [17]). The Charlson’s comorbidity index measures patient comorbidity using the tenth International Classification of Diseases Diagnoses Codes. Each comorbidity has a weight (from 1 to 6) depending on its severity. The higher the score, the higher is the comorbidity burden. Biological testing included: white cells, neutrophils, lymphocytes, and RT-PCR. All patients included in this study had serum samples tested using RT-PCR with the RealStar® Chikungunya RT-PCR Kit (Altona Diagnostics GmbH, Hamburg, Germany). We considered as confirmed CVI all suspected cases in whom biological confirmation was obtained by positive RT-PCR. The Mayotte tool and Reunion Island tool were calculated for all patients.
The study was performed in accordance with the Declaration of Helsinki, and was approved by the “Commission Nationale de l’Informatique et des Libertés” (CNIL): authorisation number 1898399 v 0. Patient’s data was completely anonymised according to the CNIL requirements. All data was solely accessed and analysed retrospectively from the University Hospital of Martinique.
The sample size was estimated based on the expected precision of sensitivity (Se) and specificity (Sp) confidence intervals. In a previous study [15], the prevalence of symptomatic CVI was 28% (318/1154). For an expected Se and Sp of 90% each, with a precision of 5%, and an alpha error of 5%, the estimated sample size was 192 for Se, and 494 for Sp. Therefore, we planned to include at least 494 patients.
In the acute phase, RT-PCR was considered as the gold standard to identify subjects with or without CVI. Sensitivity (%), specificity (%), positive predictive value (PPV, %), negative predictive value (NPV, %), and Youden’s index (J = Sensitivity (%) + Specificity (%)– 100) were estimated. Youden’s index is a single statistic that captures the performance of tests. Its value ranges from -100% (totally useless test) to 100% (perfect test).
Quantitative variables are described as mean ± standard deviation, and categorical variables as using number and percentage. Baseline characteristics were compared according to RT-PCR results using Student’s t-test (continuous variables) and chi2 test (categorical variables) Statistical analyses were performed using SAS release 9.4 (SAS Institute Inc., Cary, NC, USA).
During the study period, 894 patients were potentially eligible. Among these, 207 were excluded. A flowchart of the study population is shown in Fig 1. Excluded subjects did not significantly differ from subjects included in terms of age (79.0±8.0 vs. 80.4±8.0 years, respectively) or sex (49% vs. 51% women, respectively). In all, 687 patients were considered in the present study. The mean Charlson’s comorbidity score was 1.7±1.9. The average time between onset of symptoms and admission was 1.3±2.3 days.
Clinical and biological characteristics at admission to hospital are presented in Table 1. Fever (73.1%) and arthralgia (51.4%) were the most frequent symptoms. The knee (22.3%), and the ankle (19.1%) were the most frequent sites of arthralgia. For biological characteristics, 77.9% of patients had a neutrophil count< 7500, and 61.3% had a lymphocyte count <1000.
Patients with positive RT-PCR (chik+) CVI (n = 467) and patients with negative RT-PCR (chik-) CVI (n = 220) did not differ significantly with respect to age (80.6±7.8 versus 80.0±8.3, respectively; p = 0.33), sex (female sex 45.9% versus 52.9%, respectively; p = 0.09), or Charlson’s comorbidity score (1.6±1.8 versus 1.7±1.9, respectively; p = 0.73).
Performance indicators of the Mayotte tool and the Reunion Island tool are presented in Table 2. Sensitivity ranged from 6% (for fever+headache) to 49% (for fever+polyarthralgia). Youden’s index ranged from 1% (for fever+headache) to 30% (for fever+polyarthralgia). PPV and NPV ranged from 70% to 95%, and from 32% to 43%, respectively.
Our study shows that the diagnostic performance of two scores to screen for potential CVI, both developed in younger populations, is poor among older patients, as shown by the associated Youden’s index. While the specificity and the PPV of the scores are good to excellent, the sensitivity and NPV are mediocre, not to say poor. The specificity of the Mayotte tool [15] was 81% in our series, which was only slightly lower than the 89% reported in Sissoko’s seminal study. Regarding the Reunion Island tool [16], its specificity in our series was excellent, at 97%, compared to 85% in the original population. Conversely, the sensitivity of both scores was poor in our series; at 49% for the combination of fever plus polyarthralgia (for Mayotte tool), and 23% for Reunion Island tool. The authors of both these scores reported higher sensitivity (80% and 84% respectively). Using the clinical features score to compare three other pairs of symptoms found even lower sensitivity rates. These differences are likely due to the different clinical profiles observed in elderly subjects, which renders the use of scores developed in young populations perilous. In our study, the average age was 80.4 years, with an average comorbidity index of 1.7, underlining the geriatric profile of our population. In the two scores we tested, the average age in the development cohorts were 27.2±16.8 years for the Mayotte tool, and 40.1±12.4 years for the Reunion Island tool. Indeed, Mayotte tool is based on signs of fever plus polyarthralgia, which were present in 83.6% of the chik+ patients. In our series, this pair of symptoms was only observed in 48.6% of chik+ cases. This variation in the clinical profile of elderly subjects has previously been reported by other authors, who suggested that the incidence of atypical, severe or fatal cases increases with age [5]. In the Reunion Island tool developed by Thiberville et al. [16], the presence of fever and polyarthralgia were among the inclusion criteria, and therefore present in 100% of subjects. In our population, these two symptoms were found in 79.4% and 62.5% of chik+ patients, while we observed lymphopenia in 75.3% of chik+ subjects, compared to 79% in Thiberville’s study [16]. The symptom profile observed in our study was less specific, with fewer rheumatological symptoms than usually described in the semiology of CVI [3].
Modifications in clinical presentation in elderly people are frequently observed in general practice [18, 19]. In many cases, the primary complaint is rarely directly related to the precipitating event. This phenomenon has been widely studied, and led to the modelling of clinical presentations in elderly subjects by Fried et al. [19]. Fried’s diagnostic models take account of comorbidities, as well as the influence of functional and psychosocial factors. Indeed, the classical model in which symptoms correspond to those habitually observed in the causal disease is rarely the norm. Frequently, the physician (and/or the patient) may attribute recent symptoms to a known disease, whereas the symptoms may in fact be the result of an acute affection. Fried and colleagues called this the attribution model and facilitating complaint, whereby the concern identified at presentation to medical care was not the major underlying problem. In another model, termed the causal chain model, an elderly subject, often frail with multiple diseases, experiences an acute event that disturbs the patient’s fragile health equilibrium, and subsequently precipitates a chain of complications that may mask the initial events and/or aggravate co-existing diseases. All of these models illustrate the complexity of establishing an accurate diagnosis in this special population, especially using signs that were initially observed in a younger population.
Mediocre or poor sensitivity has major implications for the implementation of adequate treatment of CVI, even though treatment is mainly symptomatic. In older people, the problem is twofold. On one hand, sudden functional disability and loss of autonomy may lead to health complications (falls, dehydration, pressure ulcer, delirium, etc.). On the other hand, CVI may aggravate chronic disorders with possible adverse outcomes. In addition, older people may present atypical signs, which expose them to inadequate patient care due to serial misdiagnoses (differential diagnosis like dengue fever, leptospirosis, or bacterial infection).
The lack of validated tools for use in elderly patients is a common problem in routine care. Although a small number of screening tools or predictive scores have been validated for use in the elderly (e.g. the Mini Nutritional Assessment [20, 21], gait speed [22], or the timed “Up and Go” test [23]), many other instruments are widely used on a daily basis to aid management of elderly populations without robust scientific evidence confirming their clinimetric properties (e.g. the Wells score, or the Short Physical Performance Battery [24, 25].
Our study presents several strengths. Firstly, the sample size is very large, and includes specifically elderly patients (older age and higher comorbidity scores). The number of missing data per variable is also very low (3% at most). This provides a robust basis for results observed. The clinical and biological data were recorded by geriatric medicine and virology physicians from the hospital’s medical informatics system, with cross-checking from the patients’ medical records. Furthermore, confirmation of the diagnosis of CVI was obtained by RT-PCR using the same kits for all the subjects included in the study. Several limitations deserve to be addressed. We did not use serological testing to confirm CVI diagnosis. This could have impact in our results because people who have presented later their infection could have been misdiagnosed when using only RT-PCR. This would be very unlikely as patients for whom delay from onset symptoms to biological testing exceeded 48 hours were excluded from our study. The retrospective nature of the study could have been a limitation. Indeed, it would have been relevant to compare the performances of the Mayotte and Reunion tools in Martinique with the younger population they were developed in before comparing them in older population. Our population could be not representative of the overall elderly cases.
The existing Mayotte tool and Reunion Island tool to predict CVI, developed in populations of younger patients, are not useful for the detection of CVI in 65+ patients. Population ageing and the likely recurrence of other epidemics of this virus justify the development of a specific clinical and/or clinico-biological score for elderly subjects in order to ensure early diagnosis and adequate management.
|
10.1371/journal.pcbi.1003602 | Linear Superposition and Prediction of Bacterial Promoter Activity Dynamics in Complex Conditions | Bacteria often face complex environments. We asked how gene expression in complex conditions relates to expression in simpler conditions. To address this, we obtained accurate promoter activity dynamical measurements on 94 genes in E. coli in environments made up of all possible combinations of four nutrients and stresses. We find that the dynamics across conditions is well described by two principal component curves specific to each promoter. As a result, the promoter activity dynamics in a combination of conditions is a weighted average of the dynamics in each condition alone. The weights tend to sum up to approximately one. This weighted-average property, called linear superposition, allows predicting the promoter activity dynamics in a combination of conditions based on measurements of pairs of conditions. If these findings apply more generally, they can vastly reduce the number of experiments needed to understand how E. coli responds to the combinatorially huge space of possible environments.
| Bacteria face complex conditions in important settings such as our body and in biotechnological applications such as biofuel production. Understanding how bacteria respond to complex conditions is a hard problem: the number of conditions that need to be tested grows exponentially with the number of nutrients, stresses and other factors that make up the environment. To overcome this exponential explosion, we present an approach that allows computing the dynamics of gene expression in a complex condition based on measurements in simple conditions. This is based on the main discovery in this paper: using accurate promoter activity measurements, we find that promoter activity dynamics in a cocktail of media is a weighted average of the dynamics in each medium alone. The weights in the average are constant across time, and can be used to predict the dynamics in arbitrary cocktails based only on measurements on pairs of conditions. Thus, dynamics in complex conditions is, for the vast majority of genes, much simpler than it might have been; this simplicity allows new mathematical formula for accurate prediction in new conditions.
| Bacteria respond to their environment by regulating gene expression [1]–[5]. Gene expression is determined by global factors such as the cell's growth rate and overall transcription and translation capacity [6]–[10], together with specific factors such as transcription regulators that respond to specific signals.
The environments that bacteria encounter are often complex, made up of combinations of many biochemical components and physical parameters. For example, natural habitats of bacteria include the soil [11], [12] and the human gut [13]–[15]. Complex conditions are also of interest in applications such as food science and bioenergy [16]–[20]. It is therefore of interest to understand how cells respond to complex conditions. However, experimental tests run up against a combinatorial explosion problem: in order to test all combinations of N factors, one needs 2N experiments. For example, a food scientist that seeks to test bacterial gene expression in all possible cocktails of 20 ingredients at two possible doses needs more than a million experiments, 220 = 1,048,576 experiments. If four doses are considered, 420∼1012 experiments are needed. Important recent advances on bacterial gene expression made by Gerosa et al [7] and Keren et al [10] do not overcome this concern, because one still needs to measure expression in each combination of conditions. Thus, the search for simplifying principles is important.
One such simplifying principle was suggested in a study of the protein dynamics in human cancer cells in response to drug cocktails [21]. Protein dynamics in a drug combination were well described by weighted averages of the dynamics in the individual drugs. This feature was termed linear superposition (also known as convex combination or weighted average). Furthermore, it was found that measuring dynamics in drug pairs could be used to predict the dynamics in drug triplets and quadruplets. This opens a possibility for avoiding the combinatorial explosion problem: To predict gene expression in all possible combinations of N drugs it is sufficient to measure all N(N-1)/2 pairwise combinations instead of 2N. For example, the response to all combinations of 20 drugs can be well approximated by measurement of the 190 pairwise combinations, rather than over a million combinations. The number of necessary experiments is reduced by more than 5000 fold.
Here, we asked whether the linear superposition principle might apply also to understanding the response of E. coli to combinations of growth conditions. Since we consider the transcriptional response of bacteria to natural stress conditions, rather than the proteomic response of cancer cells to anti-cancer drugs, this study explores this principle in a very different biological context. We used a promoter library to obtain accurate dynamics of 94 promoters as bacteria grew from exponential to stationary phase in all possible combinations of a set of nutrients and stresses. We find that dynamics in a mixture of conditions is, for most genes and conditions, well described as a linear combination – a weighted average – of the dynamics in the individual condition. The weights sum up to approximately one. We also found that part of the reason for this feature is that promoter activity dynamics for each gene seem to be quite limited, and are explained effectively by one or two principal components. Using linear superposition, we employ mathematical formulae that allow predicting the dynamics in cocktails of conditions based on measuring pairs of conditions. This suggests that the combinatorial explosion problem may be circumvented, to understand and predict bacterial responses to complex conditions.
We studied 94 genes and 2 control strains (see Materials and methods), in a 96 well plate format. We chose 94 genes which represent a wide range of biological functions (Table S1), and which have a strong detectable fluorescence signal in a range of growth conditions (more than 2 standard deviation above background).
We measured promoter activity of these genes as a function of time using the E. coli reporter library developed in our lab [22] (Figure 1a). Each reporter strain had a rapidly maturing GFP variant (gfpmut2) under control of a full length intragenic region containing the promoter for the gene of interest, on a low copy plasmid (Figure 1a). Promoter activity was measured as the time derivative of GFP fluorescence accumulation divided by cell density, as described [23]–[25] (Methods). Using this approach, the temporal dynamics of promoter activity can be measured at high accuracy [26]–[28].
We aimed at understanding the promoter activity dynamics in growth media composed of combinations of chemical conditions. For this purpose we chose 4 elementary conditions. Each condition is based on a chemically defined medium, M9+0.2% glucose as the carbon source. In each elementary condition one supplement is added (A) 0.05% casamino acids, (B) 3% ethanol, (C) 10 µM hydrogen peroxide H2O2 (D) 300 mM NaCl salt. In all four conditions, cells reached a similar final optical density (OD), with different growth rates (Table S2).
We studied combinations of these conditions by mixing the appropriate supplements into the standard medium. Thus, condition A+B is standard medium supplemented with 0.05% casamino acids and 3% ethanol (Figure 1b). In total, we studied all four single conditions, all six pairs, all four triplets and the quadruplet A+B+C+D (The different growth rates of all combinations is given in table S2).
In each condition, we measured promoter activity of the 94 genes at an 8 minute resolution, throughout batch culture growth, including exponential growth phase and stationary phase. Depending on the growth rate in a given condition the stationary phase was reached after 8 to 22 hours of growth. Each experiment was repeated on four different days.
We observed that promoter activity dynamics of a given promoter can vary both in shape and in amplitude across different growth conditions. Using principal component analysis we can identify the typical shapes of every promoter across conditions. In figure 2 we show the activity dynamics of fliY in all measured conditions (Figure 2a) and its two principal dynamic curves PC1 and PC2 (Figure 2b). We found that each promoter can be well described by two principal component dynamic curves, which explain 80–99% of its variance (Figure 2c). In more than 93% of the promoters, the two first PCs explain 90% or more of the variance (Figure 2c). Because of the 2PC property, each promoter activity curve is a linear combination of its two PCs to a good approximation. The first two PCs explain much more variance than expected in randomized data (See Figure 1 in Text S1).
About one third (30/94) of the promoters were well explained by one principal component in all measured conditions (Figure 2f). The dynamics of these promoters thus had a rather constant shape in different conditions, and differed only in amplitude. For example one PC explains 98% of the variance in the σ70 activated ribosomal promoter rrnB (Figure 2d,e). The other 2/3 of the promoters, explained well by 2PCs, showed condition-dependent shape changes in their dynamics. The low number of principal component curves needed in order to explain the promoter activity dynamics could be a result of general nonspecific transcription for promoters with only one PC (with only change in amplitude with different growth rates), and could be condition dependent yet limited in number for promoters with two principal component curves.
We find that for 76% of promoter activities, the first PC is highly correlated (R2 above 0.8) with instantaneous growth rate (See Figure 6a,b,c in Text S1). This may relate to a principal component analysis by Bollenbach et al [29] that instead of considering dynamics, considered a single point at exponential growth in response to antibiotic combinations. The first PC correlated with growth rate and the second with drug specific effects. The second PC in our dataset varies more widely in shape between different promoters (See Figure 6d in Text S1).
We now use the 2PC property to understand how promoter activity dynamics in a mixed condition PA+B relate to the dynamics in each supplement alone, PA and PB. Since a promoter can be described as a linear combination of the same 2PCs in any condition, we expect the combined PA+B to be a combination of the one-supplement conditions PA and PB:
Where the best fit weights are wA and wB. To find the best fit weights we aligned the dynamics in conditions A, B and A+B according to a shared axis of generations ( – see Materials and methods and Text S1 Extended methods). Using a generation axis helped compare conditions despite variations in growth rate. We performed linear regression of PAB based on PA and PB. These weights are constant over time. Similarly, dynamics in three and four supplements can be represented as linear combinations of the one supplement dynamic:
We determined the best fit weights wi(1…N) using an error-in-variables linear regression [30] (where wi(1…N) is the weight contributed by condition i which best fits the combined condition 1…N – see Text S1). To measure how similar a linear combination is to the measured combination dynamics we compute the relative fit error between the two (Text S1). Linear combination describes the dynamics well (relative error 10% see Text S1), as expected.
So far, these findings are consistent with the 2PC finding. However, we find information beyond the 2PC property, when we examine the sum of the weights in these equations. We find that the sum of weights wA+wB in each fit is distributed around one (See Figure 2 in Text S1), with a standard deviation of 0.6. The weights are usually positive (76%>−0.05). This means that the linear combination is approximately a weighted average (see also [29]).The same applies to three and four supplement mixtures. We therefore tested a simpler model, named linear superposition, in which the weights are constrained to sum to one, and be positive:
Here the dynamics in the mixture conditions is a linear combination of the individual supplement conditions but with only one free parameter wAB, with the constraint that wAB ranges between 0 to 1.
In most conditions, the linear superposition model gives a better score in describing the data in tests that take model simplicity into account (Akaike information criterion [31], which sums the log likelihood of the model fit and the number of model parameters, see Text S1). The linear superposition model also gave better predictions than a multiplicative superposition model (in which , See Text S1). A representative sample of promoter dynamics and the corresponding linear superposition model is shown in figure 3. The mean fit error is 12%, with 86% showing less than 20% fit error. This compares well with the day-to-day experimental error estimated from 4 day-to-day repeats, with average error of 14% (See Figure 3 in Text S1). A table with the weights and errors for all promoters and conditions is provided in the SI (Table S3).
We also sought conditions where linear superposition does not apply. We found one such condition using the classic diauxic shift experiment [32]–[34]. In this case, bacteria grow on a combination of two sugars, glucose and lactose. They begin to utilize the preferred sugar, glucose, and only when glucose is depleted switch to using the second sugar, lactose. The cells thus delay the production of the lactose utilization system – the lacZ promoter – until glucose concentration becomes low[23]. Then, cells switch to growth on lactose and express lacZ vigorously.
Considering glucose and lactose as conditions X and Y, one does not find that lacZ is a linear combination in the combined condition X+Y. This is because under glucose alone, lacZ is weakly expressed (Figure 4), and under lactose alone it is strongly and constantly expressed (Figure 4). Linear combination would mean a constant expression at some intermediate value. In contrast, in X+Y, lacZ expression is strongly time dependent (Figure 4). Such an effect is expected whenever two conditions interact to regulate genes sequentially [17], [35], rather than simultaneously. Another example we found is the metabolic operon nudC, which showed behavior similar to lacZ, and a poor fit to linear combination (See Figure 4 in Text S1). A table with the weights and errors for all promoters in the diauxic shift is provided in the SI (Table S4).
We note that all of the other 92 genes in our study showed good linear superposition in the diauxie condition. This suggests that linear combination might break down for specific genes where the conditions have a nonlinear, sequential effect or more generally distinct temporal dependence on their dynamics.
We now use linear superposition to predict the dynamics in a combination of conditions given only data on individual-supplement dynamics, and data on pairs (that is, given the weights wi(ij) in pair conditions). Previous work by Wood et al [36], based on a different approach, successfully predicted the growth-inhibitory effect of antibiotic cocktails based on measurement of pairs of drugs. Such predictions are potentially useful because, as discussed in the introduction, it is much easier to measure all pairs than to measure all possible cocktails of N conditions.
The predictions rely on the assumption of linear superposition, specifically that weights sum to one. We apply the formula developed by Geva-Zatrosky et al [21] for predicting protein dynamics in cancer drug cocktails. The formula uses the fact that a combination, say A+B+C, can be treated in three different ways: a mixture of A+B and C, and equivalently as a mixture of A+C and B, and as a mixture of B+C and A. Each of these three possibilities can be described using superposition, and should yield the same result. This provides enough equations to predict the weights needed to calculate the triplet dynamics (See Text S1).
The formula predicts the linear superposition weights in an N-supplement cocktail
The prediction for the weights wi(1…N) based on measurements of the weights in all cocktails of N-1 supplements is [21]:where the superscript (≠j) relates to which supplement is missing in the N-1 cocktail. When only pair data is available, this formula is used iteratively: the triplets are predicted from pair weights, the quadruplet uses these predictions for the triplets weights and so on.
Using this equation, with pair data only, we find good predictions for the promoter dynamics. Representative dynamics and predictions are shown in figure 5. The median relative error between prediction and measurement is 27% for triplets and 34% for the quadruplet (See Figure 5 in Text S1). These prediction errors are about 2 times larger than the day-to-day experimental error. To evaluate the predictive power of this formula we compared it to what one could expect given no additional information. For this purpose, we ‘predicted’ the dynamics for a given promoter in condition X by randomly picking an exemplar from the available set of measured curves for that promoter in all conditions except X. We then averaged the error between these ‘predictions’ and the measurement in condition X. For example, for a given promoter in condition A+B+C, we used the measured curves in all 14 conditions except A+B+C, namely the 4 single conditions (A,B,C,D), 6 pairs, 3 triplets after excluding A+B+C and one quadruplet. We generated 14 errors and compared the average error to the present formula prediction error. Our formula predictions show about 2.3 times less error than the average error for triplet conditions and about 1.5 times less error in the quadruplet condition (Figure 6).
We studied promoter activity dynamics in combinations of conditions by means of fluorescent reporters. We find that almost all promoters and conditions tested show a linear combination property: the dynamics in a combined condition is a linear combination of the dynamics in individual conditions. The weights in the combination tend to sum to one, and thus combinations act as weighted averages of individual conditions, a property called linear superposition. Linear superposition allowed us to predict the dynamics in triplets and quadruplet based on the dynamics in pairs of conditions. This prediction formula offers a way to reduce the combinatorial complexity of understanding complex conditions.
Genes regulated by specific signals that are strongly time dependent in the complex environment, such as lacZ in a diauxic shift experiment (Figure 4), may not display the linear superposition principle. Note that in the diauxie condition, 92 of the 93 other promoters did show linear superposition with good accuracy.
Almost all promoters in this study needed only two principal components to explain their dynamic curves across conditions. This finding is in line with studies on gene expression in a range of organisms [37]–[41]. About one third of the promoters did not show an environmental specific change in the shape of their dynamics and were well explained by only one principal component (Figure 2d–f). It would be interesting to extend this study to investigate the biological meaning of these principal components. It seems that the first PC captures general effects related to the growth [29] (See Figure 6 in Text S1), and the second captures the way that the specific regulation of the promoter changes its first PC dynamics.
The fact that two PCs explain the data well means that promoter activity in a mixed condition can be described as a linear combination of the promoter dynamics in the basic conditions. A further finding is that the sum of weights in this combination is distributed around one. A model of linear superposition, in which weights are constrained to be positive and sum to one, explain the data very well in most conditions. This feature- sum of weights equals one- is crucial to allow predictions of higher order combinations. If the sum of weights was not constrained, one would not have enough equations to predict the weights in a cocktail.
The linear superposition property calls for a biological explanation. One possible framework is the recently suggested finding that when cells compromise between a few tasks, their optimal solution is a gene expression profile that is a weighted average of the optimal profiles for each individual task [42]–[44]. Testing this theory, which is based on a multi-objective compromise between several tasks [45], also known as Pareto optimality, would require understanding the tasks of the cells under the present conditions. Pareto theory points to one possible reason why linear combination might be optimal, which applies in the limit of strong selection under environments which include many combinations of conditions. How linear summation is achieved is a mechanistic question which needs further research. One way that a linear summation can be achieved is when regulatory factors compete over a limiting component - for example: σ70 and σS compete over the RNA polymerase, such that the fraction of σ70-RNApol is equal to 1 minus the fraction of σS-RNApol (here we neglected other σ factors). Therefore, the fraction of transcription allocated to growth (σ70) and survival (σS) genes follows a line in gene expression space [43]. The position on the line is determined by the ratio of the two σ factor concentrations.
It would be interesting to extend this study to other genes, conditions and organisms. It would be important to find conditions where superposition breaks down, as for lacZ in the diauxie conditions described here, to find the limitations of this approach. This approach can be tested also in other levels of cell response, for example one may ask whether linear superposition applies to dynamics of metabolite fluxes [45], [46]. It would be interesting to extend this analysis to situations in which cells show all-or-none patterns of gene expression [35], [47]–[49], and to enhance our understanding of how bacteria compute [50]. If the present approach for predicting dynamics in complex conditions applies more generally, one may attempt to computationally navigate the combinatorial huge space of possible environments, to search for growth conditions with desired gene expression profiles.
All media were based on M9 defined medium (42 mM Na2HPO4, 22 mM KH2PO4, 8.5 mM NaCl, 18.7 mM NH4Cl, 2 mM MgSO4, 0.1 mM CaCl). The media used in this study are: Casamino acids (M9 minimal medium, 0.2% glucose, 0.05% Casamino acids); NaCl (M9 minimal medium, 0.2% glucose, 300 mn NaCl); H2O2 (M9 minimal medium, 0.2% glucose, 10 µM H2O2); Ethanol (M9 minimal medium, 0.2% glucose, 3% ethanol); and all 15 combinations: 4 single conditions, 6 pairs, 4 triplets and one quadruplet. For example Casamino acids+NaCl (M9 minimal medium, 0.2% glucose, 0.05% Casamino acids, 300 mn NaCl). In addition we measured glucose alone (M9 minimal medium,0.2% glucose); Casamino acids with no glucose (M9 minimal medium,0.05% Casamino acids); Low concentration glucose (M9 minimal medium,0.04% glucose); Lactose (M9 minimal medium,0.4% lactose); Lactose with low concentration glucose (M9 minimal medium,0.4% lactose, 0.04% glucose). In each experiment the bacteria were cultivated in the presence 50 µg kanamycin/ml.
GFP levels were measured over time for 96 reporter strains (Table S1), each bearing a green fluorescent protein gene (GFP) optimized for bacteria (gfpmut2) on a low copy plasmid (pSC101 origin). All strains in this study were derivatives of wild type E. coli K12 strain MG1655. Reporter strains were inoculated from frozen stocks and grown over-night on M9 with 0.2% glucose and 0.05% casamino acids for 16 hours in 600 µl high-brim 96-well plate and reached a final OD of ∼0.9. The 96-well plate was covered with breathable sealing films (Excel Scientific Inc.). The 96-well plates were prepared using a robotic liquid handler (FreedomEvo, Tecan Inc). Overnight cultures were diluted 1∶500 into the micro 96-well experiments plates. The final volume of the cultures in each well was 150 µl. A 100 µl layer of mineral oil (Sigma) was added on top to avoid evaporation and contamination, a step which we previously found not to significantly affect growth [25], [28]. Cells were grown in an automated incubator with shaking (6 hz) at 37°C. A robotic arm moved the micro 96-well plates from the incubator-shaker to the plate reader (Infinite F200, Tecan Inc.) and back. Optical density (600 nm) and fluorescence (535 nm) were thus measured periodically at intervals of ∼8 minutes until reaching stationary phase with a final OD of ∼0.15. Since the overnight cultures on high-brim 96-well plate reached a higher final OD equivalent to about 3 extra generations beyond the micro 96-well plates we obtain data for ∼6 generations of growth.
Data was obtained from the plate reader software (Evoware, Tecan) and processed using custom Matlab software. Background fluorescence was subtracted from GFP measurements using a reporter strain bearing promoterless vector U139 for each well. Then, promoter activity was calculated using temporal derivative of GFP computed by finding the slope of a sliding window of 17 data points of GFP fluorescence using regression, divided by the mean OD over this window. Varying window size between 5 and 30 affects curve smoothness but does not change the conclusions of this study.
|
10.1371/journal.pbio.1001635 | The HILDA Complex Coordinates a Conditional Switch in the 3′-Untranslated Region of the VEGFA mRNA | Cell regulatory circuits integrate diverse, and sometimes conflicting, environmental cues to generate appropriate, condition-dependent responses. Here, we elucidate the components and mechanisms driving a protein-directed RNA switch in the 3′UTR of vascular endothelial growth factor (VEGF)-A. We describe a novel HILDA (hypoxia-inducible hnRNP L–DRBP76–hnRNP A2/B1) complex that coordinates a three-element RNA switch, enabling VEGFA mRNA translation during combined hypoxia and inflammation. In addition to binding the CA-rich element (CARE), heterogeneous nuclear ribonucleoprotein (hnRNP) L regulates switch assembly and function. hnRNP L undergoes two previously unrecognized, condition-dependent posttranslational modifications: IFN-γ induces prolyl hydroxylation and von Hippel-Lindau (VHL)-mediated proteasomal degradation, whereas hypoxia stimulates hnRNP L phosphorylation at Tyr359, inducing binding to hnRNP A2/B1, which stabilizes the protein. Also, phospho-hnRNP L recruits DRBP76 (double-stranded RNA binding protein 76) to the 3′UTR, where it binds an adjacent AU-rich stem-loop (AUSL) element, “flipping” the RNA switch by disrupting the GAIT (interferon-gamma-activated inhibitor of translation) element, preventing GAIT complex binding, and driving robust VEGFA mRNA translation. The signal-dependent, HILDA complex coordinates the function of a trio of neighboring RNA elements, thereby regulating translation of VEGFA and potentially other mRNA targets. The VEGFA RNA switch might function to ensure appropriate angiogenesis and tissue oxygenation during conflicting signals from combined inflammation and hypoxia. We propose the VEGFA RNA switch as an archetype for signal-activated, protein-directed, multi-element RNA switches that regulate posttranscriptional gene expression in complex environments.
| Many cells of our body, particularly cells such as monocyte/macrophages involved in host immunity, are exposed to diverse and constantly changing environments. These cells require mechanisms by which they can rapidly respond to multiple, sometimes conflicting, environmental cues to generate appropriate responses. The 3′ untranslated regions (UTRs), i.e. the noncoding tail of messenger RNAs, often contain multiple protein- and RNA-binding elements, thereby making it an ideal setting for receiving multiple such environmental cues, which can then be integrated into a single response that regulates the gene's expression. Monocytic cells exposed to inflammation and hypoxia produce vascular endothelial growth factor (VEGF)-A, a critical factor in blood vessel formation. VEGF-A expression is regulated under these conditions via a complex regulatory mechanism that involves its 3′UTR. Here we show how this regulatory switch works. Inflammation induces formation of a four-protein complex that binds an RNA element present in the VEGFA 3′UTR and blocks translation. Hypoxia, however, triggers the assembly of a completely different three-protein complex (termed “HILDA”) that coordinates the function of three neighboring RNA elements to “flip” the RNA conformation in such a way that prevents the first complex from binding, thereby allowing VEGF-A expression. We hypothesize that the VEGFA switch might function to ensure appropriate angiogenesis and tissue oxygenation when cells are exposed to conflicting signals from combined inflammation and hypoxia conditions.
| Mammalian cells integrate diverse, and sometimes conflicting, environmental signals to generate appropriate, condition-dependent responses. Tissue myeloid cells are exposed to a plethora of stimulatory and inhibitory signals, and thus its integrated response is particularly complex. This task is made more problematical, and possibly more critical, in dynamic, pathological environments. Myeloid cell vascular endothelial growth factor (VEGF)-A is critical for blood vessel formation during development, wound-healing, and tumorigenesis [1]. Hypoxia is possibly the most potent agonist of VEGF-A expression, working at the levels of transcription, mRNA stabilization, and translation [2],[3]. VEGF-A synthesis is induced in monocyte/macrophages activated by pro-inflammatory agonists, including interferon (IFN)-γ and bacterial lipopolysaccharide. Overproduction of VEGF-A can cause excessive neovascularization, blood vessel permeability, and enhanced leukocyte recruitment, all hallmarks of chronic inflammatory conditions, including cancer and atherosclerosis [4]–[6]. Agents that inhibit VEGF-A or its receptor have been applied clinically to successfully limit colorectal and renal cell carcinoma [7]. Positive and negative regulation of VEGF-A expression has been reported in human macrophages in multiple stressed conditions. We have shown that VEGF-A expression in myeloid cells is translationally repressed by the IFN-γ-triggered GAIT (interferon-gamma-activated inhibitor of translation) system [8],[9]. Importantly, under certain pathological conditions, for example within the avascular cores of tumors and in the thickened intima of atherosclerotic lesions, macrophages are simultaneously exposed to both inflammatory cytokines and hypoxia that act concurrently in multiple pathophysiological scenarios to regulate gene expression.
Treatment of human monocytic cells with IFN-γ induces the synthesis of VEGFA mRNA and protein for up to about 12 to 16 h. However, VEGF-A synthesis and secretion are suppressed about 16 h after IFN-γ treatment despite the presence of abundant VEGFA mRNA [10]. Translational silencing of VEGFA and other GAIT targets requires binding of the GAIT complex to its cognate GAIT element in the target mRNA 3′UTR [10]. The GAIT element is a defined 29-nt stem-loop with an internal bulge and unique sequence and structural features. The human GAIT complex is heterotetrameric containing glutamyl-prolyl-tRNA synthetase (EPRS), ribosomal protein L13a, NS1-associated protein–1, and glyceraldehyde 3-phosphate dehydrogenase (GAPDH) [11],[12]. A C-terminus truncated form of EPRS, termed EPRSN1, functions as a dominant-negative regulator of GAIT complex activity and maintains basal expression of VEGF-A [13].
RNA-binding proteins (RBPs) that regulate mRNA stability or translation generally recognize their target mRNAs through structural or sequence-specific elements in the 5′ or 3′UTRs of mature mRNAs. The activity of trans-acting RBPs can be modulated by dosage (in turn regulated by synthesis rate and stability), cellular localization, posttranslational modification, noncoding RNAs, and interacting protein partners. Heteronuclear ribonucleoprotein (hnRNP) L is a key posttranscriptional regulator of VEGF-A expression. Human hnRNP L has three consensus RNA recognition motifs (RRM) [14] and binds CA-rich elements (CARE) in coding and noncoding regions of multiple transcripts [15]. hnRNP L contributes to pre-mRNA splicing [16], mRNA nucleocytoplasmic transport [14], internal ribosomal entry site-mediated translation [17], translational repression [18], and mRNA stabilization [19].
The molecular mechanisms by which signal transduction systems integrate multiple environmental cues into a binary response that determines gene expression remain largely unexplored. We have reported that hnRNP L operates a hypoxia-stimulated, binary conformational RNA switch that overrides IFN-γ-induced GAIT-mediated translational silencing of VEGFA mRNA in human monocytic U937 cells and in primary human peripheral blood monocytes (PBMs) [20]. The proposed switch permits high-level VEGF-A expression under combined inflammatory and hypoxic stress. Here we elucidate the molecular mechanism underlying the IFN-γ- and hypoxia-dependent regulatory RNA switch. The switching mechanism involves condition-dependent posttranslational modification and relocalization of hnRNP L, and subsequent formation of an hnRNP L-containing heterotrimeric complex that stabilizes the VEGFA HSR in a translation-competent conformation.
HnRNP L is an essential component of the RNA switch that blocks GAIT-mediated translational silencing of VEGF-A mRNA, and permits high-level expression of VEGF-A in myeloid cells in the presence of IFN-γ and hypoxia (Figure S1) [20]. To determine whether hnRNP L is sufficient for RNA switch function, the activity of recombinant protein was determined by in vitro translation of luciferase reporter bearing the VEGFA HSR in a wheat germ extract system in the presence of active GAIT complex from IFN-γ-treated U937 cells (Figure 1A). hnRNP L failed to overcome the translational repression suggesting that posttranslational modification of hnRNP L or additional protein factors may be required. Identical results were seen using a rabbit reticulocyte lysate system (not shown). Hypoxia-dependent hnRNP L binding partners were determined by RNA affinity purification using a 30-nt, 5′-biotinylated, extended CARE (CARE-E) from the VEGFA HSR (Figure 1B). To reduce nonspecific binding, lysates from U937 cells incubated under normoxic or hypoxic conditions were pre-cleared with an excess of 5′-biotinylated antisense CARE-E RNA in which CA pairs were mutated to GU. Cleared lysates were incubated with biotinylated, wild-type CARE-E RNA and μMAC magnetic streptavidin microbeads, and applied to a magnetic column. Bound proteins were eluted with salt solution, concentrated, and subjected to SDS-PAGE and Coomassie stain (Figure 1C). Bands enriched in lysates from hypoxia-treated cells were subjected to mass spectrometric analysis, and peptides corresponding to hnRNP L, hnRNP A2/B1, and DRBP76 (nuclear factor 90 or interleukin enhancer binding factor 3) were identified (Table S1). Binding of the proteins to CARE RNA was confirmed by RNA affinity isolation and immunoblot analysis of lysates from hypoxia-treated U937 cells. A hypoxia-inducible complex of hnRNP L, DRBP76, and hnRNP A2/B1 (HILDA) was shown to bind wild-type but not mutant antisense CARE RNA; substantially less binding of the three proteins to CARE RNA was observed in normoxic lysates (Figure 1D).
The formation of an RNA-binding heterotrimeric complex was investigated by co-immunoprecipitation (IP). Lysates from U937 cells and primary human PBM treated with IFN-γ under normoxic or hypoxic conditions were subjected to IP with anti-hnRNP L antibody, and probed with hnRNP A2/B1- and DRBP76-specific antibodies (Figure 1E, left panel). A hypoxia-dependent interaction of hnRNP L with hnRNP A2/B1 and DRBP76 was observed. The interaction between hnRNP L and hnRNP A2/B1 was RNA-independent as shown by the lack of an effect of RNase A treatment. However, the RNase diminished the interaction between hnRNP L and DRBP76, suggesting that the hnRNP L-DRBP76 complex is stabilized by RNA. The expression levels of the three HILDA complex constituents were not altered by hypoxia exposure (Figure 1E, right panel). In vitro GST-pulldown experiments showed that recombinant GST-hnRNP L directly interacted with recombinant hnRNP A2/B1 and DRBP76 (Figure 1F, left panel). In a parallel experiment, GST-hnRNP A2/B1 was found to directly bind hnRNP L but not DRBP76 (Figure 1F, right panel). hnRNP L contains an N-terminal glycine-rich domain, three RNA-binding motifs (RRM1–3), and a proline-rich linker domain connecting RRM2 and RRM3 (Figure 1G, top). Domain mapping experiments revealed that hnRNP A2/B1 binds the proline-rich linker in hnRNP L (Figure 1G, left). In contrast, the RRM3-containing, C-terminal domain of hnRNP L was the binding site for DRBP76 (Figure 1G, right).
EPRS and hnRNP L from IFN-γ-treated U937 cells, in either normoxia or hypoxia, bind in vitro synthesized VEGF-A HSR in a mutually exclusive manner [20]. To provide in vivo evidence of the VEGF-A switch, RNA from cells treated with IFN-γ in the presence of normoxia or hypoxia for 24 h were immunoprecipitated with anti-EPRS and -hnRNP L antibodies and subjected to qRT-PCR using transcript-specific primers. GAIT complex EPRS and HILDA complex hnRNP L recognized and bound VEGFA mRNA following stimulation by IFN-γ under normoxic and hypoxic conditions, respectively, consistent with previous results (Figure 2A) [20]. To determine whether hnRNP A2/B1 or DRBP76 are required for hnRNP L binding to VEGFA mRNA, lysates from cells treated with IFN-γ and hypoxia were subjected to ribonucleoprotein IP (RIP) using anti-hnRNP L antibody, coupled with RT-PCR. hnRNP L interacted with VEGFA mRNA in control transfected cells; however, the interaction was substantially reduced following siRNA-mediated depletion of either hnRNP A2/B1 or DRBP76 (Figure 2B). Similarly, the interaction of hnRNP A2/B1 or DRBP76 with VEGFA mRNA required the presence of the other (Figure S2, left and center panels). Moreover, the interaction of hnRNP A2/B1 and DRBP76 with VEGFA mRNA was abolished following hnRNP L depletion by siRNA-mediated gene silencing (Figure S2, right panels), suggesting that HILDA binding to VEGFA mRNA requires integrity of the entire complex. To begin to understand the roles of the individual protein components in RNA switch activity, their binding sites within the HSR region were mapped by UV-crosslinking. Of the three proteins, only hnRNP L and DRBP76 directly bind the VEGFA HSR. Interestingly, the two interacting proteins bind different regions of the HSR, hnRNP L binds the CARE, whereas DRBP76 binds the AU-rich stem loop (AUSL) (Figure 2C). The less robust binding to the individual ascending (AUSL-A) and descending (AUSL-D) regions of the AUSL suggests that DRBP76 stabilizes the double-stranded AUSL in a conformation that prevents formation of the GAIT element, which overlaps AUSL-A (Figure 1B). We determined the specific DRBP76-binding region by constructing a series of mutations in either AUSL strand. Mutation of M2 (U404UAUAU409 to AAUAUA), but not M1 (A416AUAUA421 to UUAUAU), inactivated the RNA switch of the HSR-bearing reporter RNA, suggesting the upper stem-loop region of the AUSL is critical (Figure 2D and Figure S3A,B). Differences in luciferase activities of the mutant forms were due largely to altered translation as shown by comparable firefly luciferase mRNA levels determined by semi-quantitative RT-PCR (Figure 2D, insert); renilla luciferase mRNA levels were essentially the same for all transfections (not shown). Complementary covariant mutations (M2–M3, A381UAUAA386 to UAUAUU) on the M3 strand opposing M2 were introduced in an attempt to restore function. However, the M2–M3 double mutant failed to recover RNA switch activity, possibly due to disruption of the GAIT element structure by M2 mutation. Thus, we further created complementary mutations of U358UAUAU363 to AAUAUA (M4) to restore the GAIT element structure at the distal 6-bp stem region. RNA switch activity was partially restored in the M2–M3–M4 triple mutant, indicating the stem structure, not the sequence, is critical for DRBP76 activity in the RNA switch. As controls, individual M3 and M4 mutants lacked GAIT-mediated translational silencing activity and RNA switch function. In the VEGFA HSR, the CARE adjoins the GAIT element with not even a single nt separating them (Figure S3A) [20]. To determine the maximum distance between the elements that permits RNA switch activity, we inserted 5- to 25-nt poly(C) spacers between them in an HSR-bearing reporter. Spacers up to 15 nt permitted RNA switch activity, but 20- and 25-nt spacers were inhibitory (Figure 2E and Figure S3A,C), consistent with a distance limit for an effective interaction between the binding proteins hnRNP L and DRBP76. The insertions did not affect mRNA expression of FLuc (Figure 2E, insert) and RLuc (not shown) significantly, indicating that altered translation was responsible for differential Luc activity. Together these results suggest that whereas hnRNP L is responsible for target selectivity, DRBP76, through binding a nearby stem-loop region, has primary responsibility for stabilizing the RNA form lacking the GAIT structural element, thereby suppressing GAIT complex-directed translational silencing (Figure 2F). Knockdown of DRBP76 did not significantly alter VEGFA mRNA half-life, providing additional evidence that DRBP76 influences VEGF-A expression primarily at the level of translation (Figure S3D).
By knockdown and overexpression experiments, we previously reported that hnRNP L is essential for hypoxia-induced switch activity in U937 cells [20]. To test the requirement for the other HILDA components, DRBP76 and hnRNP A2/B1, both were subjected to siRNA-mediated knock-down (hnRNP L knock-down served as positive control) (Figure 3A, top). Cells were treated with IFN-γ and hypoxia for up to 24 h, and lysates tested for their effect on in vitro translation of an HSR-bearing reporter. As seen before, 24-h lysates from IFN-γ-treated normoxic cells inhibited translation of the reporter, but 24-h lysates from hypoxic cells were inactive (Figure 3A, bottom). However, deletion of either DRBP76 or hnRNP A2/B1 dramatically impaired the hypoxia-driven RNA switch to an extent comparable to that of hnRNP L knockdown, and permitted GAIT complex-mediated translation inhibition by 24-h lysates (Figure 3A, bottom). We investigated the effect of these lysates on endogenous gene expression. As before, hypoxia prevented IFN-γ-mediated inhibition of expression of VEGF-A observed at 24 h (Figure 3B). However, siRNA-mediated knock-down of either DRBP76 or hnRNP A2/B1 restored translational inhibition of VEGF-A without significantly altering the steady-state level of VEGFA mRNA (Figure 3B). Polysome profiling was done to verify that the effects on VEGF-A expression were due to altered translation. IFN-γ activation of the GAIT pathway inhibited VEGF-A mRNA translation-initiation [21], and this inhibition was reversed by hypoxia [20]. Indeed, following IFN-γ treatment under hypoxia, knock-down of either hnRNP A2/B1 or DRBP76 induced a dramatic shift of endogenous VEGFA mRNA from translationally active polysome pools to translationally inactive free mRNP pools (Figure 3C and Figure S4).
hnRNP L expression is markedly reduced in normoxic, IFN-γ-treated cells, thereby permitting GAIT complex binding to the VEGFA mRNA and transcript-specific translational silencing [20]. Semiquantitative RT-PCR (Figure 4A) and Northern blot analysis (Figure S5) showed that hnRNP L mRNA expression is unaltered by either hypoxia or IFN-γ treatment for up to 24 h, and that altered hnRNP L expression must be posttranscriptional. hnRNP L half-life was measured in the presence of cycloheximide to inhibit protein synthesis. In nonstressed monocytic cells (normoxia, no IFN-γ) the half-life of hnRNP L is about 12 h (Figure 4B and Figure S6A). The half-life of hnRNP L was shortened to about 4 h by IFN-γ treatment in normoxia; however, hypoxia suppressed the effect of IFN-γ, restoring the half-life to about 12 h (Figure 4C and Figure S6B). As shown previously, the proteasome inhibitor MG132 blocked IFN-γ-mediated hnRNP L degradation, indicating an important role of the ubiquitin/proteasome pathway in regulating hnRNP L expression [20].
To investigate the mechanism underlying IFN-γ-induced hnRNP L degradation, hnRNP L ubiquitination was determined. IFN-γ treatment in the presence of MG132 induced accumulation of a high molecular weight form of hnRNP L consistent with ubiquitination (Figure 4D). Expression of HA-ubiquitin and detection with anti-HA-tag antibody confirmed formation of high molecular weight, ubiquitinated hnRNP L, and exposure to hypoxia dramatically diminished hnRNP L ubiquitination (Figure 4E). We considered the von Hippel-Lindau (VHL)-containing ubiquitin ligase complex as a candidate E3 ubiquitin-protein ligase because of its normoxia-dependent role in regulation. VHL specifically targets proteins, e.g., hypoxia inducible factor (HIF)-1α tagged by O2-dependent prolyl hydroxylation [22]. VHL was shown to interact robustly with hnRNP L, but not with hnRNP A2/B1 or DRBP76, in an IFN-γ-dependent manner (Figure 4F). Also, siRNA-mediated knockdown of VHL markedly reduced hnRNP L polyubiquitination (Figure 4G, left panel) with MG132 treatment, and increased hnRNP L stability following IFN-γ treatment in absence of MG132 (Figure 4G, right panel). However, overexpression of VHL did not affect the stability of hnRNP L or the assembly of the HILDA complex in hypoxia, suggesting that HILDA complex formation might contribute to protection of hnRNP L from VHL-mediated degradation (Figure S7). In an in vitro ubiquitination system reconstituted with exogenous E1 and E2 enzymes and E3 ubiquitin ligase pVHL derived from lysate of 8 h, IFN-γ-treated U937 cells in normoxia further confirmed robust polyubiquitination of hnRNP L (Figure S8). In contrast, cell lysate from hypoxia-treated U937 cells failed to modify hnRNP L. Similar results were obtained with primary human PBM (not shown). These results suggest that proteasomal degradation of hnRNP L in U937 cells and in human PBM is mediated by IFN-γ-triggered ubiquitination by a VHL-containing E3 ubiquitin ligase.
hnRNP L is primarily localized in the nucleus in human monocytic cells but substantially redistributes to the cytoplasm during hypoxia [23]. Fluorescence visualization verified hypoxia-driven cytoplasmic relocalization of hnRNP L, even in the presence of IFN-γ (Figure 5A). Similar hypoxia-stimulated cytoplasmic relocalization of hnRNP L was observed in primary human PBM-derived macrophages induced by macrophage colony stimulating factor (M-CSF) (Figure S9). Immunoblot analysis of cytosolic and nuclear fractions from IFN-γ- and hypoxia-treated cells further confirmed hnRNP L translocation (Figure 5B). Cellular localization of RBPs can be regulated by their phosphorylation state [24]–[26]. Metabolic labeling with 32P-orthophosphate showed that hypoxia induced robust phosphorylation of hnRNP L at 8 h, and the modification was stable for at least 24 h (Figure 5C). Immunoblot analysis of hnRNP L immunoprecipitated from hypoxia-treated cells with phospho-specific antibodies revealed strong phosphorylation at Tyr, but not at Ser or Thr (Figure 5D). A time course experiment showed modest hnRNP L Tyr-phosphorylation after 0.5 h of hypoxia and maximal phosphorylation after 4 h in U937 cells (Figure 5E) and in primary human PBM (not shown). Immunoblot analysis with anti-pTyr antibody showed Tyr-phosphorylated hnRNP L was almost completely restricted to the cytoplasm in hypoxia-treated cells (Figure 5F).
To identify the hypoxia-induced phosphorylation site, hnRNP L was immunoprecipitated from lysates of hypoxia-treated cells, and phospho-sites detected by mass spectrometry. Total coverage with three protease treatments was 84%, but phosphorylation events were not detected (Figure S10). Endogenous hnRNP L in U937 cells was knocked down with siRNA targeting the 3′UTR, and cells transfected with cDNA constructs containing specific, site-directed Tyr-to-Ala mutations at residues in regions not covered by the mass spectrometry analysis. Among the five hnRNP L mutants tested, only Y359A was not phosphorylated in U937 monocytic cells (Figure 5G) and in human PBM (not shown). Tyr359, and the surrounding sequence, is evolutionarily conserved from frogs to humans (Figure 5H), and has been identified as a phospho-site by high-throughput proteomic survey (www.phosphosite.org) in both mouse and human (in addition to Tyr phosphorylation at positions 47, 48, 92, 267, 285, 333, 340, 363, 375, 565, 574, and 576). To determine the role of Tyr359 phosphorylation in hnRNP L localization, cells were transfected with c-Myc-tagged wild-type cDNA or, phospho-dead (Y359A) or phospho-mimetic (Y359D) mutants. Under normoxic conditions, wild-type hnRNP L is primarily localized in the nucleus, but also present in the cytoplasm, as observed previously [23]. In contrast, the Y359A mutant was exclusively in the nucleus, and the Y359D mutant was exclusively cytoplasmic (Figure 5I). Similarly, following IFN-γ stimulation under hypoxia, the Y359A and Y359F hnRNP L mutants were exclusively localized in the nucleus (Figure S11). As a control for specificity, Tyr130 mutants did not partition with the Tyr359 mutants. Cells were transfected with c-Myc-tagged wild-type or mutant hnRNP L, immunoprecipitated with anti-c-Myc antibody, and probed with hnRNP A2/B1 antibody. Y359D exhibited much greater binding to hnRNP A2/B1 compared to wild-type or Y359A mutant hnRNP L (Figure 5J). Remarkably, the Y359D mutant, but not the Y359A mutant or wild-type protein, was completely resistant to IFN-γ-stimulated degradation (Figure 5K). Consistent with the cellular translocation of hnRNP L (Figure 5A), Tyr phosphorylation was induced by IFN-γ treatment in hypoxia (Figure S12). In summary, hypoxia-inducible Tyr359 phosphorylation of hnRNP L facilitates its cytoplasmic relocalization and prevents its degradation.
Because hnRNP A2/B1 does not bind the HSR directly, it is more likely involved in regulation of its binding partner hnRNP L, than in operating the RNA switch itself. We tested the possibility that hnRNP A2/B1 contributes to hypoxia-induced stabilization of hnRNP L. siRNA-mediated knockdown of hnRNP A2/B1 resulted in hnRNP L destabilization following IFN-γ treatment in hypoxia (Figure 6A). In contrast, hnRNP A2/B1 knockdown did not induce DRBP76 degradation (Figure 6B). Also, siRNA-mediated knockdown of DRBP76 did not affect hnRNP L stability (Figure S13). Interestingly, hnRNP L was subject to IFN-γ-dependent Pro hydroxylation as shown by IP followed by probing with anti-hydroxyproline antibody (Figure 6C). Hypoxia prevented the IFN-γ-inducible prolyl hydroxylation of hnRNP L (Figure 6D). Knockdown of hnRNP A2/B1 under hypoxic condition and in the presence of IFN-γ and MG132 restored marked Pro hydroxylation of hnRNP L after 24 h (Figure 6E). Finally, co-IP with anti-hnRNP L antibody revealed that hypoxia induced hnRNP A2/B1 binding to hnRNP L, and completely blocked hnRNP L recognition by VHL (Figure 6F). These results indicate that the major function of hnRNP A2/B1 in the heterotrimeric switch is to protect hnRNP L from IFN-γ-triggered prolyl hydroxylation, ubiquitination, and subsequent degradation. Treatment of U937 cells with prolyl hydroxylase (PH) inhibitors L-mimosine and dimethyloxalylglycine (DMOG) blocked prolyl hydroxylation of hnRNP L and caused marked stabilization of the protein in the presence of IFN-γ under normoxia (Figure S14).
Co-IP and RNA-binding studies suggest a model in which the interaction between DRBP76 and hnRNP A2/B1 is indirect and facilitated by hnRNP L and VEGFA HSR RNA (Figure 2F). We investigated the interactions in detail by in vitro reconstitution using recombinant proteins and in vitro–transcribed RNA. DRBP76 and hnRNP A2/B1 by themselves did not bind, nor did the addition of either hnRNP L or HSR RNA restore their interaction significantly (Figure S15). However, when both hnRNP L and HSR RNA were added, then a modest interaction between hnRNP A2/B1 and DRBP76 was detected. A much stronger interaction was observed when phospho-mimetic hnRNP L (Y359D) was added together with HSR RNA, but not nonspecific RNA, thereby reconstituting the entire HILDA complex in vitro. To investigate the sufficiency of hnRNP L, hnRNP A2/B1, and DRBP76 in operating the RNA switch, we determined the regulatory activity of the purified proteins in vitro. Phospho-mimetic hnRNP L (Y359D) was used to facilitate interaction with hnRNP A2/B1. The three proteins were pre-incubated in several combinations, and their effect on in vitro translation of an FLuc reporter bearing the VEGFA HSR element (and RLuc control RNA) was determined in a wheat germ extract in the presence of 35S-Met and cytosolic extracts from IFN-γ-treated U937 cells. hnRNP L (Y359D) by itself or with either hnRNP A2/B1 or DRBP76, did not restore translation in the presence of lysates from cells treated with IFN-γ for 24 h (Figure 6G). However, the three proteins together substantially overcame the translational inhibition. Substitution of wild-type hnRNP L for the phospho-mimetic was ineffective, suggesting the posttranslational modification is not only required for maintaining a high level of cytoplasmic hnRNP L, but also is required for HILDA complex assembly. As positive controls, lysates from cells treated for 24 h with or without IFN-γ in hypoxia could rescue translation of HSR-bearing FLuc. These results support the role of the heterotrimeric HILDA complex in operating the RNA conformational switch.
The combinatorial activity of pairs of nearby elements has become an area of increasing interest, particularly with the recent recognition that microRNA binding to targets can influence protein binding to nearby target RNA elements [27]. There are few cases in which pairs of protein-binding RNA elements dictate the response. In one well-studied example, a combinatorial code in which the number and position of two elements—namely, the cytoplasmic polyadenylation element and Pumilio-binding element—determine translational activation or repression in Xenopus oocytes [28]. However, there is a dearth of studies on the mechanisms by which nearby RNA elements, and their cognate binding factors, integrate disparate environmental signals to generate a binary response and regulate gene expression. In one known case, the leader sequence of the Mg2+ transporter gene mgtA of Salmonella enterica contains a Mg2+-sensing riboswitch and an 18-codon, proline- or hyperosmotic stress-sensing ORF that integrate distinct signals to generate the cell response; however, an interaction between the disparate elements was not observed [29]. In the case of the GAIT system, we have reported that hypoxia prevents GAIT complex binding to the VEGFA 3′UTR by a switch in the conformation of RNA that masks the GAIT structural element [20] by converting the element into the ascending half of a long, double-stranded stem-loop. The switch is initiated by hypoxia-stimulated binding of hnRNP L to a 3′UTR CARE directly adjacent to the GAIT element. In this report we define the components of a heterotrimeric complex that constitutes the RNA switch, their regulation by IFN-γ and hypoxia, and their specific functions in directing the VEGFA mRNA switch in human monocytic cells.
The requirement for each of the components of the HILDA complex to drive the RNA switch was shown by knockdown experiments in cells, and their sufficiency shown by in vitro reconstitution. The HILDA complex has not been previously described, but its individual components are known to regulate distinct mRNA-related functions. DRBP76 was initially identified through its binding to double-stranded RNA and to protein kinase R (PKR) [30]. DRBP76 exhibits multiple RNA-related functions including regulation of transcription, mRNA stability [31], and translation [32]. DRBP76 also binds the VEGFA HSR in hypoxic breast cancer cells, increasing mRNA stability and translation, but the binding region within the VEGFA HSR in these experiments was not determined [33]. The double-stranded RNA-binding property of DRBP76 is most likely the critical function it performs in the context of the HILDA complex, stabilizing the conformation featuring a long, double-stranded stem loop, and disrupting the structure of the GAIT element. hnRNP A2/B1, like hnRNP L, participates in splicing of pre-mRNAs and in translational regulation [34]. hnRNP A2/B1 also serves as a molecular motor-powered transporter of select mRNAs bearing specific hnRNP A2/B1 response elements (A2RE), for example, neurogranin, Arc, and calmodulin-dependent kinase II [35]–[37]. Cytosolic complexes containing heterodimeric hnRNPs have been shown to interact with specific target mRNAs. For example, hnRNP L and I form a complex that binds murine inducible nitric oxide synthase mRNA, and regulates its translation [38]. Interestingly, the same pair of hnRNPs found in the HILDA complex, hnRNP L and A2/B1, interacts with the glucose transporter 1 (Glut1) 3′UTR, inducing translational repression and mRNA instability [18]. However, an interaction between DRBP76 and A2/B1 has not been described.
hnRNP L is a critical component of the HILDA complex because it is uniquely responsible for stimulus sensing as well as target recognition. Our results show that the steady-state level and cellular localization of hnRNP L in myeloid cells are regulated both by IFN-γ and by hypoxia. Under normoxic conditions hnRNP L is distributed between the cytoplasm and nucleus, the latter for execution of mRNA processing functions. IFN-γ induces prolyl hydroxylation of cytoplasmic hnRNP L and consequent rapid, VHL-mediated ubiquitination and proteasomal degradation (Figure 7). Near-complete cytoplasmic depletion of hnRNP L permits GAIT complex binding to the VEGFA GAIT element in the translationally silent conformer, resulting in low-level translation of VEGFA mRNA. Hypoxia induces phosphorylation of hnRNP L on Tyr359, which increases cytoplasmic localization by restricting transport into the nucleus. Hypoxia-inducible phosphorylation suggests the activity of a nonreceptor Tyr kinase such as a member of the Src, Abl, Jak, Syk, or Fak families. The sequence surrounding the Tyr359 phosphorylation site (pRRGPSR359YGPQYGHPPPPPPPP) exhibits 100% conservation in humans, rodents, rabbits, and frogs, and provides insight into the identity of the proximal kinase. “YG” is a specific Src kinase substrate motif (PhosphoMotif Finder), and the downstream polyproline motif is a binding site for SH3-containing proteins, including Src family kinases. hnRNP A2/B1 binds Tyr359-phosphorylated hnRNP L and blocks recognition by VHL-containing E3 ubiquitin ligase complex, thus permitting cytoplasmic accumulation. The precise kinetics and binding order have not been determined, but our results suggest that the phospho-hnRNP L and hnRNP A2/B1 recruit DRBP76 to form the heterotrimeric HILDA complex that binds the VEGFA CARE. The interaction is weakened by nuclease treatment, indicating that the binding of DRBP76 to other complex members is enhanced by its interaction with the long, AU-rich stem-loop within the VEGFA HSR. The HILDA complex stabilizes the translationally permissive conformer that masks the GAIT element, thus resulting in uninhibited translation of VEGFA mRNA, even in the presence of IFN-γ-induced GAIT complex.
The tumor suppressor protein VHL is an essential, target-specific component of a multifunctional E3 ubiquitin ligase complex involved in protein degradation [39]. The best-known target of VHL is hypoxia inducible factor (HIF)-1α and -2α, transcription factors that stimulate expression of multiple hypoxia-inducible transcripts, including VEGFA mRNA. In normoxia, O2-dependent prolyl hydroxylation of HIF-1α triggers recognition by VHL and consequent degradation, thereby inhibiting expression of HIF-1α targets [40]. However, prolyl hydroxylation of HIF-1α is inhibited in hypoxia, thereby stabilizing HIF-1α and increasing target mRNA transcription. Other VHL targets have been identified in renal cell carcinoma cell lines; interestingly, several are downregulated by VHL [41]–[43]. hnRNP A2/B1 has been reported to be targeted by VHL [44]. However, we find hnRNP A2/B1 binding to hnRNP L prevents targeting by VHL in human monocytic cells. Possibly, cell-type specificity of targets and directionality of regulation—i.e., up or down—are promoted by additional factors within the VHL-bearing E3-ubiquitin ligase complex. Proline hydroxylase inhibitors DMOG and L-mimosine both block hnRNP L prolyl hydroxylation and consequent degradation. Collagen prolyl-4-hydroxylase (C-P4H) is a candidate because it is induced by hypoxia [45],[46] and hydroxylates and destabilizes another RBP, Argonaute 2 (Ago2) [47]. Likewise, HIF prolyl hydroxylase (HIF-PH) is a candidate because it modifies HIF-1α for poly-ubiquitination by pVHL and proteasomal degradation [48].
Long, noncoding regions of mRNAs, because of their manifold protein- and RNA-binding elements, are potentially ideal for integration of multiple inputs into a single output—i.e., gene expression. Because of their unusually long length, the 3′UTR, which averages almost 600 nt in human mRNAs versus about 150 nt for 5′UTRs, is a particularly attractive target for signal integration [49]. A plethora of examples of posttranscriptional regulation have been described in which RBPs are activated by environmental signals that alter their binding behavior, generally by posttranslational modification and complex formation [50]. In most known cases, RBPs or complexes interact one-to-one with preformed sequence or structural elements [50],[51]. More recently, regulatory processes have been described in which signals alter the conformation of the RNA to modulate gene expression [52]. The VEGFA 3′UTR RNA switch features alternative interaction of distinct protein complexes in response to environmental signals, culminating in regulated gene expression. The CARE element is analagous to a riboswitch aptamer domain, and hnRNP L acts as a “responder/selector,” responding to environmental cues and determining HILDA complex mRNA target specificity. The AUSL element determines the expression outcome: VEGF-A expression is high when the double-stranded conformation is bound by the HILDA complex, and expression is depressed when the GAIT complex binds the GAIT element in the alternate conformation (Figure 7). To our knowledge there are not any previous reports of 3-RNA element switches. Likewise, the integration of two different signals—i.e., hypoxia and inflammatory cytokine—by the VEGFA RNA switch lacks precedent.
The principles, protein constituents, and mechanisms utilized by the VEGFA switch might be applicable to distinct mRNA switches. One possibility is that the HILDA complex recognizes other transcripts with sequence and structural elements analogous to the VEGFA switch region—i.e., CARE and GAIT elements nearby DRBP76-binding double-stranded RNA stretches. Cytoplasmic hnRNP L binds VEGF-A mRNA and other transcripts in multiple cell lines [18],[19],[38], suggesting that the HILDA complex might direct additional RNA switches. More generally, distinct RBPs may replace hnRNP L as the “specificity factor,” but likewise recruit DRBP76 to stabilize nearby stem-loop structures and drive formation of alternate regulatory conformers. High-throughput screening has identified at least two RBPs hnRNP A1 and FUS (fused in sarcoma) that bind DRBP76 and might direct alternate RNA switches [53],[54]. Alternatively, other inhibitory factors (microRNA or proteins) might replace the GAIT complex to drive the hnRNP L-directed GAIT-independent RNA switches in more general sense. We speculate that the VEGFA switch is a founding member of signal-activated, protein-directed, RNA switches that regulate posttranscriptional gene expression in vertebrates, and similar switches might be widespread RNA sensors in multicellular animals.
Phospho-safe extraction buffer was from Novagen (Madison, WI). Rabbit reticulocyte lysate, wheat germ extract, large-scale RNA production system-T7, and dual luciferase reporter assay system were from Promega (Madison, WI). Human IFN-γ was obtained from R&D Systems (Minneapolis, MN). Human monocyte nucleofactor kit was from Lonza (Switzerland). Reagents for protein purification, nuclear and cytosolic extraction, and immunoanalysis were from Pierce (Rockford, IL). Primers, dNTP mix, TRIzol LS reagent, one-step RT-PCR system, and competent cells were from Invitrogen (Carlsbad, CA). Protein A/G beads, anti-α-tubulin, anti-hnRNP A2/B1, rabbit anti-hnRNP L, and anti-GAPDH antibodies were from Santa Cruz (Santa Cruz, CA). Mouse monoclonal anti-hnRNP L antibody was from Novus (Littleton, CO). Anti-HDAC1 and anti-β-actin antibodies were from Biovision (Mountain View, CA). Anti-c-Myc, anti-HA, goat anti-rabbit/mouse IgG (Alexa Fluor® 488 Conjugate), streptavidin-HRP, and anti-ubiquitin antibodies were from Cell Signalling Technology (Danvers, MA). Anti-DRBP76 antibody was from Biorbyt (Cambridge, UK). GST monoclonal antibody was from Thermo Scientific (West Palm Beach, FL). Anti-VHL antibody was from GeneTex (San Antonio, TX). Anti-hydroxyproline antibody was from Abcam (Cambridge, MA). Anti-rabbit IgG, anti-mouse IgG, and random-primer labeling kit were from GE healthcare (UK). Translation grade [35S]methionine was from NEN-Dupont (Boston, MA), α-[32P]CTP was from PerkinElmer (Boston, MA), and [32P]orthophosphoric acid was from MP Biomedicals (Solon, OH). Actinomycin-D, DMOG, and L-Mimosine were from Sigma (St. Louis, MO). In vitro ubiquitination assay kit and ubiquitin were from Biomol (Plymouth Meeting, PA) and Boston Biochem (Cambridge, MA), respectively.
Human U937 monocytic cells (ATCC, Rockville, MD) were cultured in RPMI 1640 medium containing 10% heat-inactivated fetal bovine serum (FBS), 2 mM glutamine, and 100 U/ml of penicillin and streptomycin at 37°C and 5% CO2. PBM from healthy clinical donors were isolated by leukapheresis and countercurrent centrifugal elutriation under a Cleveland Clinic Institutional Review Board–approved protocol that adhered to American Association of Blood Bank guidelines. For preparation of cytosolic extracts, the cells were incubated for 1 h in medium containing 0.5% FBS and then with (or without) IFN-γ (500 units/ml) in presence of hypoxia (1% O2) for an additional 8 or 24 h. Cell lysates were prepared in Phosphosafe extraction buffer containing protease inhibitor cocktail. To knock down endogenous hnRNP L, DRBP76, hnRNP A2/B1, or VHL, U937 cells were transfected with appropriate concentration of (100–200 nM) gene-specific siRNA or a scrambled control siRNA using human monocyte nucleofactor kit. hnRNP L siRNAs containing 3 oligomers targeting the 3′UTR or ORF were from Origene. siRNA against DRBP76, hnRNP A2/B1, and VHL were from Santa Cruz.
The bacterial expression plasmid pRSET-hnRNP L was generated using pcDNA3-hnRNPL-c-Myc as template and cloned between BamHI and EcoRI restriction sites in the pRSET-A vector for expression and purification of His-tagged hnRNP L. HNRNPL ORF was subcloned into pGEX-4T-1 vector and the plasmid transformed into E. coli BL21(DE3) for expression and purification of GST-tagged hnRNP L. hnRNP L cDNA was subcloned into pcDNA3-c-Myc between BamHI and EcoRI restriction sites and expressed in human U937 cells as described [20]. The pcDNA3-based hnRNP L Tyr-to-Ala, -Asp, and -Phe mutants were prepared using GENEART Site-Directed Mutagenesis System (Invitrogen) according to the manufacturer's instructions. The mutation was confirmed by DNA sequencing. DRBP76 ORF was cloned into pET28-a vector between NdeI and EcoRI restriction sites. Expression of GST-tagged proteins was induced with 500 nM isopropyl-β-D-thiogalactopyranoside (IPTG) at 30°C for 6 h with 50 µg/ml ampicillin. Soluble protein was extracted and purified with B-PER GST purification kit (Thermo Fisher). His-tagged DRBP76 was generated in vitro using rabbit reticulocyte lysate in vitro translation system (Promega), and purified with MagneHis Protein Purification System (Promega). His-tagged wild-type hnRNP L and phospho-mimetic hnRNP L were expressed in E. coli BL21(DE3) with IPTG induction and in rabbit reticulocyte lysate in vitro translation system, respectively, and purified with Ni-NTA resin (Qiagen). Recombinant GST-hnRNP A2/B1 and hnRNP A2/B1 were from Novus Biologicals and Origene, respectively.
S100 extracts (4 mg) from U937 cells cultured in normoxia or hypoxia were pre-cleared by incubation for 30 min at 4°C with 2 µg 5-biotinylated, mutant antisense CARE-E RNA oligomer (5′-biotin-UCUGUGUGGGUGGGUGUAUGUAUGUAAAUA-3′), added to 200 µl of μMACs magnetic streptavidin microbeads for 10 min, and applied to μMACS separator. The cleared lysate was incubated with 2 µg of 5′-biotinylated, wild-type CARE-E RNA oligomer (5′-biotin-AGACACACCCACCCACAUACAUACAUUUAU-3′), and then with streptavidin microbeads and applied to μMACS separator as above. The column was rinsed with 100 µl protein equilibration buffer and twice with 100 µl of lysis buffer. The bound material was applied to the column and washed 4 times with 100 µl of lysis buffer to decrease nonspecific binding. 200 µl of buffer containing 300 mM NaCl was applied to the column to elute bound protein. The eluate was desalted and concentrated using Centrifugal Filter Unit (Microcon YM-3K, Millipore, Billerica, MA). Eluates were subjected to SDS-PAGE and Coomassie stain. Bands enriched only in hypoxia-treated sample were trypsinized and peptides mapped by capillary column LC-tandem MS (LTQ-linear ion trap MS system, ThermoFinnigan, San Jose, CA). The data were analyzed with Mascot using CID spectra to search the human reference sequence database. Matching spectra were verified by manual interpretation aided by additional searches using the Sequest and Blast.
Most IP experiments were done with Co-Immunoprecipitation kit (Pierce) following the manufacturer's instruction to eliminate antibody contamination of IP products. For some IP experiments, traditional method was used. Cells were lysed in Phospho-safe extraction buffer, and 500 µl of cell lysate was combined with 50 µl protein A/G agarose beads (50% bead slurry) and pre-cleared at 4°C for 60 min. The samples were centrifuged at 13,000 rpm for 10 min at 4°C and the supernatant added to 50 µl of protein A/G beads and 2 µg of antibody, and rotated for 4 h at 4°C. The beads were washed 5 times with 1 ml cold lysis buffer. Protein gel loading dye (100 µl) was added, and the samples boiled and loaded onto the gel. To avoid interference from IgG, rabbit-derived secondary antibody was used against mouse-derived primary antibody.
GST and GST-hnRNP L were generated from E. coli BL21(DE3) transformants containing pGEX-4T-1 and pGEX-4T-1-hnRNP L, respectively. Cells were sonicated and the supernatant collected after high-speed centrifugation. GST and GST-hnRNP L (1 µg of each) were incubated separately with glutathione-agarose beads for 30 min. After washing the agarose beads 4 times with 1 ml of PBS, 1 µg of recombinant DRBP76 and hnRNP A2/B1 were diluted in binding buffer (20 mM HEPES, pH 7.5, 200 mM KCl, 5 mM MgCl2, 0.2% bovine serum albumin, 10% glycerol, 0.1% Nonidet P-40, 1 mM phenylmethylsulfonyl fluoride, and complete protease inhibitor mixture), combined, and incubated at 4°C for 2 h. The agarose beads were washed 5 times with binding buffer (without bovine serum albumin and glycerol), and bound protein eluted by boiling in SDS loading buffer.
Cycloheximide (50 µg/ml) was added to 8×106 U937 cells in 4 ml RPMI1640 medium. Cells were harvested and lysed. Immunoblot was done using anti-hnRNP L antibody and the band intensity quantified and normalized by the initial value at 0-h time point.
In vitro reconstitution of hnRNP L ubiquitination was performed as described [55]. Purified His-tagged hnRNP L (0.5 µg) was preincubated with U937 cell lysate, and then incubated with a mixture of E1 and E2 enzymes, biotin-ubiquitin, and cell lysate as a source of hnRNP L E3 ligase. Recombinant hnRNP L was immunoprecipitated with anti-His tag antibody, and biotin-ubiquitin was detected by blotting with streptavidin-HRP.
The metabolic labeling assay was performed as described previously [12]. U937 cells (8×106 cells) in 4 ml RPMI 1640 medium were collected by centrifugation, re-suspended in phosphate-free medium, and metabolically labeled with a 4-h pulse of 32P-orthophosphate. The cells were collected by centrifugation and lysed with Phospho-safe extraction buffer containing protease inhibitor cocktail. hnRNP L was immunoprecipitated from lysates using mouse anti-hnRNP L antibody and protein A/G-agarose in cell lysis buffer. Proteins were resolved by 12% SDS-PAGE, and the gel was dried and applied to Phospho-screen for determination of radiolabeling.
In vitro transcribed, 32P-labeled full-length HSR or truncated HSR RNA (20 fmol) was incubated for 30 min at 4°C with purified recombinant proteins (0.2 µg) in 20 µl of buffer containing 20 mM HEPES (pH 7.5), 5 mM MgCl2, 50 mM KCl, 1 mM DTT, protease inhibitor cocktail, 0.1% Triton X-100, 0.1 mg/ml yeast total tRNA, 40 U RNasin, and 10% glycerol. The mixture was crosslinked by 15 min exposure to ultraviolet light (1,800 J/cm2) on ice in a UV crosslinker. The protein-RNA complex was incubated with 1 µl of RNase A for 20 min at 25°C. Samples were denatured in SDS-PAGE buffer under reducing conditions, and complexes analyzed by 10% SDS-PAGE and autoradiography.
The RIP assay was performed as described previously [13]. Protein A/G beads (50 µl) were incubated with 500 µl of cell lysate (4 mg protein) for 1 h at 4°C with rotation to pre-clear. The cell lysate was centrifuged and the supernatant collected. Mouse anti-hnRNP L antibody (2 µg) was added (mouse pre-immune IgG was used as negative control) and the mixture incubated at 4°C overnight with rotation. Protein A/G beads (50 µl) were added and incubated at 4°C for 4 h. The beads were washed five times with 1 ml of lysis buffer with rotation at 4°C. Total immunoprecipitated RNA was extracted with Trizol. Total RNA from the lysate was extracted and used as a positive control for RT-PCR. Immunoprecipitated RNA (3 µl) and 1 µg of total RNA were used in reverse transcriptase reaction and subsequent PCR with Taq DNA polymerase. The PCR reaction (5 out of 20 µl) was visualized by 1.5% agarose gel. The primers for semi-quantitative RT-PCR were as follows: RT_βactin-f: 5′-ATGGATGATGATATCGCCGCG-3′; RT_βactin-r: 5′-CTAGAAGCATTTGCGGTGGAC-3′; RT_VEGF-f: 5′-ACAGAACGATCGATACAGAA-3′; RT_VEGF-r: 5′-AAAGATCATGCCAGAGTCTC-3′; RT_hnRNPL-f: 5′-GAGTCCCATCTGAGCAGGAA-3′; and RT_hnRNPL-r: 5′-CAATTTTATTGAAATGTGCC-3′.
Polysome profiling was done as described [13]. CHX (100 µg/ml) was added to cells for 15 min and then collected and washed two times with CHX-containing, ice-cold PBS. 107 cells were suspended in 350 µl TMK lysis buffer and incubated on ice for 5 min. The lysates were centrifuged at 12,000 rpm for 10 min and the supernatants collected. RNase inhibitor (2 µl, 40 U/µl) and CHX (50 µl, 100 µg/µl) were added in 50 ml each of freshly prepared 10% and 50% sucrose gradient solutions just before use. Cytosolic lysates were loaded on the sucrose gradient and centrifuged at 29,000 rpm for 4 h, and 8 fractions of about 1 ml were collected and combined; light RNP, 40S, 60S, and 80S formed the translationally inactive pool, and heavy polysome fractions formed the translationally active pool. Total RNA was isolated from both combined fractions by extraction with Trizol reagent and purified by RNeasy minikit (Qiagen, Valencia, CA) following the manufacturer's procedure. The RNA was quantitated and purity determined by agarose formaldehyde gel, and used for real-time PCR analyses.
Capped, poly(A)-tailed template mRNAs was prepared using mMESSAGE mMACHINE SP6 and T7 kits (Ambion). Firefly-Luc-VEGFA GAIT element-poly(A) (200 ng) and Renilla-Luc (200 ng) reporter RNAs were incubated with U937 cytosolic lysates (500 ng of protein) from IFN-γ-treated U937 cells in the presence of 35 µl of wheat germ extract or rabbit reticulocyte lysate, and [35S]methionine. The translation reactions were performed for 90 min at 30°C and resolved by SDS-PAGE (10% polyacrylamide) and visualized by phosphorimaging. In some experiments, the FLuc and RLuc activity was measured by chemiluminescence using luminator.
U937 cells were transiently transfected with 5 µg of wild-type or mutant pCD-FLuc-VEGFA HSR using human monocyte nucleofactor kit. RLuc-expressing vector pRL-SV40 (1 µg) was co-transfected for normalizing transfection efficiency. After 12 h, transfected cells were incubated with IFN-γ under Nmx. or Hpx. for up to 24 h, lysed, and lysate luciferase activities were measured using a dual luciferase assay kit (Promega). The primers for semiquantitative RT-PCR of FLuc were as follows: RT_FLuc-f: 5′-GCCTGAAGTCTCTGATTAAGT-3′; RT_FLuc-r: 5′-ACACCTGCGTCGAAGT-3′; RT-RLuc-f: 5′-TGATTCAGAAAAACATGCAG-3′; RT-RLuc-r: 5′-ATATTTGTAATGATCAAGTA-3′.
Immunostaining of hnRNP L was as described [23]. U937 cells (106 cells/ml) in 12-well plates with glass cover slip at the bottom were incubated in hypoxia or normoxia for 24 h. Cells were centrifuged for 5 min at 2,500 rpm and washed twice with PBS and then with 4% paraformaldehyde fixing solution for 20 min. Cells were washed twice with PBS, and incubated with rabbit anti-hnRNP L polyclonal antibody (Santa Cruz, 1∶40) in blocking solution (2% BSA, 0.1% Triton X100 in PBS) at room temperature for 2 h. Cells were washed twice with PBS and centrifuged at 1,500 rpm for 5 min. Alexa Fluor 488 goat anti-rabbit secondary antibody (Invitrogen) was added (1∶50) with phalloidin (1∶50) in blocking solution for 1 h. Cells were washed with PBS three times. DAPI dye was mixed in the mounting solution and the slides imaged.
|
10.1371/journal.ppat.1005259 | How Does the VSG Coat of Bloodstream Form African Trypanosomes Interact with External Proteins? | Variations on the statement “the variant surface glycoprotein (VSG) coat that covers the external face of the mammalian bloodstream form of Trypanosoma brucei acts a physical barrier” appear regularly in research articles and reviews. The concept of the impenetrable VSG coat is an attractive one, as it provides a clear model for understanding how a trypanosome population persists; each successive VSG protects the plasma membrane and is immunologically distinct from previous VSGs. What is the evidence that the VSG coat is an impenetrable barrier, and how do antibodies and other extracellular proteins interact with it? In this review, the nature of the extracellular surface of the bloodstream form trypanosome is described, and past experiments that investigated binding of antibodies and lectins to trypanosomes are analysed using knowledge of VSG sequence and structure that was unavailable when the experiments were performed. Epitopes for some VSG monoclonal antibodies are mapped as far as possible from previous experimental data, onto models of VSG structures. The binding of lectins to some, but not to other, VSGs is revisited with more recent knowledge of the location and nature of N-linked oligosaccharides. The conclusions are: (i) Much of the variation observed in earlier experiments can be explained by the identity of the individual VSGs. (ii) Much of an individual VSG is accessible to antibodies, and the barrier that prevents access to the cell surface is probably at the base of the VSG N-terminal domain, approximately 5 nm from the plasma membrane. This second conclusion highlights a gap in our understanding of how the VSG coat works, as several plasma membrane proteins with large extracellular domains are very unlikely to be hidden from host antibodies by VSG.
| African trypanosomes have evolved two key strategies to prevent killing by the host immune response and, thus, maintain a long-term infection in a mammal. Both are based on a densely packed coat of a single protein, the variant surface glycoprotein (VSG), which covers the entire extracellular surface of the cell. The first strategy is antigenic variation, through which individual cells switch the identity of the expressed VSG at a low frequency and are selected by the host immune response. If the VSG is novel, the trypanosome proliferates, maintaining the infection; if it doesn't switch, or if the new VSG is not novel, it will be killed. In the second strategy, the VSG acts as a protective barrier, shielding the cell from innate and adaptive immune factors until there is an overwhelming titre of antibodies recognising the expressed VSG. In this review, the VSG coat is modelled, and past experiments that investigated how it protected the trypanosome are revisited using current knowledge of VSG sequence and structure. The conclusions are: (i) the identity of the individual VSGs explains early experimental variation; (ii) most of the VSG molecule is accessible to antibodies. This second conclusion highlights a gap in our understanding of how the VSG coat works, as several plasma membrane proteins with large extracellular domains are very unlikely to be hidden from host antibodies by VSG.
| VSGs are homodimers of two 50–60 kDa subunits held on the extracellular face of the plasma membrane by a glycosylphosphatidylinositol (GPI) anchor. VSGs have a large N-terminal domain of 350–400 residues and one or two small C-terminal domains of 20–40 residues each. The domains are connected to each other by flexible linkers [1–3]. The conformation of the linkers is unknown, as is their effect on the structure of the whole VSG. VSGs vary in sequence (for example, [4]), but have a conserved tertiary structure [5]. VSG molecules are free to diffuse in the plane of the membrane, and similar diffusion coefficients were obtained using the endogenous VSG coat on trypanosomes and VSG placed in the plasma membrane of mammalian cells in culture [6]. The rate of diffusion is high, similar to the rates measured for a range of other plasma membrane proteins, and equivalent to complete randomization of the VSG coat in 40 minutes [6]. The rate of diffusion provides strong evidence that there is minimal intermolecular affinity between VSG dimers, even at the high concentration present in the VSG coat. Estimates of the packing density of the VSG on the extracellular face of the plasma membrane have been derived from (i) measurements of the VSG copy number and estimates of the surface area (5.7 x 106 VSG dimers and 180 μm2 [7]), and (ii) direct measurements of the cell surface area and percentage of VSG on the extracellular face of the plasma membrane (145 μm2 and 89% [8]). Thus, the estimated area available to each VSG dimer on the cell surface is between approximately 28 nm2 (cell surface 145 μm2) and 35 nm2 (cell surface 180 μm2), using the estimated VSG copy number above. It is worth noting that the first of the values for cell surface area was measured on cells grown in rodents, whereas the second was derived from trypanosomes grown in culture, and the discrepancy between the two values may represent a real difference due to growth conditions.
The size of a VSG dimer can be derived from the structure of the N-terminal domain [5,9], and it is assumed that the long axis is perpendicular to the plasma membrane surface (Fig 1). The area taken up by each dimer can be approximated to a circle with an area of 28 nm2 [8], but note that this size estimate does not take into account any coordinated water molecules. This value is remarkably close to the estimates of the area available per VSG dimer as discussed above, strongly supporting the model that the vast majority of the plasma membrane is physically occluded by VSG. The VSG N-terminal domain has a long axis of approximately 10 nm, measured from the structure; allowing for some increase due to the C-terminal domain and GPI anchor, the thickness of the VSG coat is probably 12–15 nm. This value is in agreement with measurements from electron microscopy [10].
One conclusion that can be drawn from these estimates is that a significantly increased level of cell surface VSG can only occur if linked to an increase in cell surface area (for example, [11]). However, the estimates are not sufficiently accurate to distinguish between a model in which VSGs are always closely contacted by surrounding VSGs, forming a coat resembling a bubble raft [12], or whether there is a restricted amount of unoccupied space due to small variations in VSG-to-VSG distance.
VSG N-linked oligosaccharides have several potential functions: (i) to provide a substrate for the unfolded glycoprotein glucosyltransferase (UGGT) that catalyses the addition of glucose to a terminal mannose on an incompletely folded protein, and where export from the endoplasmic reticulum (ER) does not occur until after the glucose has been removed, reflecting a folded state for the VSG; (ii) to act as a structural element of an individual VSG; and (iii) to act as a structural element of the VSG coat. In Trypanosoma brucei, there are two oligosaccharyltransferases (OSTs) that function in the bloodstream form: OST1 and OST2. OST1 recognises an N-linked glycosylation site in a low isoelectric point (pI) context (five residues on either side of the N-X-S/T signal) and adds a paucimannose oligosaccharide that can subsequently act as a substrate for UGGT and can eventually be further modified by trimming down to three mannose residues and, sometimes further, through the addition of an N-acetyl glucosamine and galactose decorations. OST2 adds an oligomannose structure in response to an N-linked glycosylation site in a high pI context, which can be processed by trimming [13–15]. The specificities of the OSTs are overlapping, so an N-linked site in a neutral pI context could receive either oligosaccharide.
There is probably not a single fixed role for N-linked oligosaccharides in VSG function. Table 1 is an analysis of 33 distinctly expressed VSGs with A-type N-terminal domains, and shows length in residues and number, location, and nature of N-linked glycosylation sites, which are all features that will contribute to the overall dimensions of the VSG. There is an inverse correlation (R = -0.63) between the number of residues in the N-terminal domain and in the C-terminal domain, indicating that there may be an upper and lower limit on the number of residues for a functional VSG (S1A Fig). In contrast, there is only a very weak correlation (R = -0.11) between the molecular weight of the VSG and the number of N-linked sites (S1B Fig), suggesting that N-linked oligosaccharides do not normally have a role in increasing the size of VSGs, but do have a role other than structural in many cases, probably as substrates for UGGT. This view is supported by two other observations: (i) A small number of VSGs have no N-linked glycosylation sites, and so N-linked oligosaccharides can have no role in forming an effective coat. (ii) The majority of N-linked sites are in a low pI context (S2 Fig), and so will tend to have paucimannose glycans available for UGGT rather than the larger oligomannose glycans that might be more suitable for a space-filling role. If the role of the N-linked sites in most VSGs is to allow monitoring of folding, then it would follow that VSGs that fold efficiently no longer require such a site, whereas others that require reiterative folding cycles have retained one (or possibly more) sites. As such, the presence of an N-linked site could be more indicative of folding efficiency, rather than an element in forming a barrier. However, VSGs use every trick, and in some VSGs the oligosaccharide probably functions as a structural element in the barrier. The first example in which this might occur is in the minority of VSGs with multiple N-linked glycosylation sites in the N-terminal domain (Table 1), such as modelled for VSG118 [3], where the N-linked oligosaccharrides can occupy space between VSGs. However, it should be noted that in the one VSG with a known structure containing N-linked sugars, the oligosaccharide is held close to the VSG core and acts as a structural element in the molecules (Fig 1) [5,9]. A second example may be one N-linked glycosylation site location in VSG N-terminal domains, between residues 100 and 125, located at the plasma membrane distal tip of the VSG, where there appears to be selection for very high pI addition sites that would be almost exclusive addition of oligomannose (S2 Fig). A large oligosaccharide in this location could well affect access of external proteins.
The VSG coat cannot be absolutely uniform, as there are other proteins present on the extracellular face of the plasma membrane, which raises the question of how the VSG coat acts as a physical barrier to prevent access of immunoglobulins to these non-VSG proteins. The plasma membrane can be divided into three discrete areas with different non-VSG protein compositions, each separated by some form of diffusion barrier. The first is the flagellar pocket, an invagination at the base of the flagellum, where all exo- and endocytosis occurs. The second is the flagellum membrane, and the third is the cell body membrane. Various combinations of distributions have been observed for different proteins, but the mechanism of segregating a protein to one compartment but not another is not understood. Two cell surface receptors for nutrient uptake have been identified: one for transferrin [16–18] and one for haptoglobin-haemoglobin [19]. These two receptors are concentrated in the flagellar pocket, with approximately 3,000 and 300–400 copies, respectively. The density of the VSG coat in the flagellar pocket is similar to that on the rest of the plasma membrane [8], but how the VSG density is maintained in the presence of a set of receptors and many other proteins is not understood [20]. It is also not known whether the concentration of the receptors in the flagellar pocket is advantageous for nutrient uptake and/or avoidance of immunoglobulin recognition, and/or for some other unknown reason.
Many plasma membrane proteins—for example, hexose transporters—have only very small extracellular domains of less than 10 kDa, and it is likely that the VSG coat prevents access of antibodies to these domains. However, there are other proteins present on the cell body and/or flagellum plasma membrane that have extracellular domains similar in size to or even larger than the VSG. Two examples illustrate this point. First, ESAG4 and related genes encode a heterogeneous family of type 1 transmembrane proteins, some of which are localized to the plasma membrane of the flagellum. The ESAG4 family of proteins has an extracellular domain of 70–80 kDa and a cytoplasmic domain encoding an adenylate cyclase [21]. The extracellular domain is significantly larger than the VSG and can be modeled with very high confidence [22] onto a tandem di-domain of bacterial small-molecule transport proteins or substrate binding proteins (as reviewed in [23,24]). Second, the invariant surface glycoprotein (ISG) gene family also encodes type 1 transmembrane proteins localized over the whole cell surface [25–28]. ISGs have a small cytoplasmic domain and an extracellular domain that is similar size in size and structurally related to VSGs and the haptoglobin haemoglobin receptor through the use of an elaborated three-helical bundle [29,30]. ISG65 can be modeled on the haptoglobin haemoglobin receptor with a high degree of confidence, and the elongated structure has a long axis of approximately 10 nm, similar to a VSG (Fig 2) [31]. If ISGs are perpendicular to the cell surface, they would reach most or all of the way through the VSG monolayer. The copy number for individual ISGs has been estimated to between 5 x 104 and 7 x 104 [25,27]; if this level of expression is extended to the entire ISG family, it is likely that there are approximately 2 x 105 ISGs in total, roughly equivalent to one ISG for every 50 VSG molecules. These large, non-VSG proteins pose a potential threat through immunoglobulin recognition; how these proteins avoid recognition by immune effectors remains unknown.
From the description above, there is an obvious dichotomy between two possible situations. In the most simplistic explanation, the VSG coat is structured to function by simply preventing access of host immunoglobulins to molecules such as ISGs. Alternatively, the VSG coat could function by combining a limitation on access with an active system that negates any antibody binding to proteins such as ISGs. In the context of the second model, the VSG coat is not a static entity that simply expands as the cell grows through the addition of new membrane and VSG. There is rapid endocytosis and recycling of the plasma membrane and VSG [34] that processes the equivalent to the entire cell surface every 12 minutes [35]. In addition, any VSG antibody complex that forms and protrudes above the surface of the VSG layer is subject to hydrodynamic flow resulting from movement of the trypanosome that both increases the rate of diffusion relative to uncomplexed VSG and gives the diffusion directionality [36]. The effect is to selectively force the complex toward the posterior pole, effectively concentrating it near the flagellar pocket and increasing its chances of endocytosis. It is thought that these two processes allow trypanosomes to persist as the antibody titre rises in the host until a threshold concentration is reached. The hydrodynamic flow-induced increased rate of endocytosis of surface-bound immunoglobulin does not appear to have evolved in African trypanosomes as a specialized adaptation, as it also occurs in the distantly related fish pathogen Trypanoplasma borrelli [37]; both might represent a specialization of an older mechanism to harvest material from the environment.
Antigenic variation is a requirement for establishing persistent infection, as the mammalian immune system can kill trypanosomes once the immunoglobulin (Ig) titre is high enough to overwhelm the endocytosis and degradation pathway. Killing can occur through both opsonization [38] and complement-mediated mechanisms [39]. In rodent infections, near field isolates cause chronic infections lasting weeks, whereas monomorphic laboratory strains adapted for rodent growth proliferate until the rodent host dies after a few days. The difference in growth results from a loss of autoregulation of population density, leading to uncontrolled growth [40]. IgMs are important in controlling the acute infections caused by laboratory strains [41]. However, IgMs do not influence an infection when nearer field isolates are used to infect mice; the parasitaemia profile is the same in wild type and IgM-null mice [42]. This infers that the major interaction in adaptive immune system killing of trypanosomes in natural infections is probably mediated by interactions between the VSG coat and IgG. Specific VSG immunoglobulins are the mediators of clearance through the adaptive immune system, evidenced by VSG identity being the only known change in the trypanosome surface over the course of an infection. Antibodies against invariant cell surface proteins are produced during an infection but are not sufficient to produce immunity [43,44].
Binding of complement system components has also been detected. Binding of C3b and Factor B, two components of the alternative pathway C3 convertase (C3bBb), was detected after incubation in human serum [45]. Activation of the complement pathway downstream of C3 convertase did not occur, and so the trypanosomes remained viable. It is not known whether this binding is receptor-dependent and how further activation beyond C3 convertase is prevented by the trypanosome. Binding of complement C4 binding protein (C4BP), a regulatory component of both the classical and lectin pathways, has been detected, but, again, the molecular basis for the interaction has not been determined [46].
One way to determine how far extracellular proteins can penetrate toward the plasma membrane is to determine which structural features of the VSG are accessible on living trypanosomes. The proteins used have included: (i) VSG monoclonal antibodies (MoAbs), (ii) VSG monoclonal single domain antibodies (nanobodies, NAbs), (iii) polyclonal antisera recognising ISGs, (iv) lectins (in particular, Concanavalin A [Con A]), and (v) trypsin. Below, the results using each of these approaches are discussed, and some are re-evaluated in the light of more recent understanding of VSG structure to see how they illuminate the interaction between the trypanosome cell surface and molecules of the adaptive immune response.
There are several reports on the production of anti-VSG MoAbs and analysis of their binding to live trypanosomes [47–54]; some details and the results are summarized in Table 2. The fraction of the MoAbs that bind live trypanosomes in different reports ranges from two out of 20 to nine out of nine. This variation is probably a result of the different methods used in the initial screen to identify VSG MoAbs, as the majority of laboratories did not use binding to live cells. A second difficulty in interpretation is the requirement to take great care to perform live cell binding experiments at <4°C throughout to prevent localisation of the VSG antibody complex to the flagellar pocket and subsequent endocytosis [35]. This requirement may explain why one report found seven out of 30 VSG MoAbs localised to the flagellar pocket ([48] and Table 2). Pooling the experiments shown in Table 1, 43 out of 92 VSG MoAbs bind live cells. There are two observations that arise from these data: First, there are epitopes that are not accessible to antibodies—an observation consistent with dense VSG packing causing restricted access. Second, externally accessible epitopes are not a small percentage of the total number of epitopes.
It is commonly believed that MoAbs observed to bind fixed trypanosomes but not live cells result from a disruption of the surface coat during the fixation process and a concomitant exposure of epitopes inaccessible in live cells. There is a further consideration that must be made to explain the increased accessibility of some MoAbs to VSG in fixed cells or in vitro (ELISA/western blot) but not in vivo. Denaturation of the VSG will expose epitopes normally hidden by being in the huge dimerisation interface [5] and/or internal within the structure VSG. Many of the screening procedures used to select MoAbs would have resulted in complete or partial VSG denaturation, including coating plates for solid phase radioimmunoassay (RIA) or ELISA, solvent fixation, air-drying, and, possibly, formaldehyde fixation. Such MoAbs would give the appearance of recognising epitopes that were inaccessible in live cells, and no analysis of the MoAbs above was performed to determine whether they recognised epitopes only after denaturation.
A set of studies mapped the epitopes recognised by MoAbs that bound live cells onto the molecular structure of the VSG [49,50,53–55]. The first analysed nine monoclonal antibodies that were screened for VSG121 binding using solid phase RIA [49,50]. All nine MoAbs bound VSG in air-dried blood smears (no other fixation), and two bound live trypanosomes in suspension. The two MoAbs that bound live cells did not bind VSG in Western blots, whereas the other seven did. The epitopes were mapped using competition RIAs between the MoAbs and sera raised against purified cyanogen bromide peptides. The two MoAbs that recognised live cells were difficult to map but were weakly competed by anti-p19, which contains residues 1 to 111 of the mature VSG. The remaining MoAbs either recognised epitopes in p16 (residues 112 to 332) or conformational epitopes containing components from both p19 and p16. Subsequent to this report, it emerged that VSG structures are conserved [5], that p19 corresponds to the coil running the entire perpendicular length of the VSG, and that p16 contains most of the N-terminal domain (Fig 3A). Thus, it is not possible to estimate a value for the penetration of the two live cell binding MoAbs into the VSG monolayer, and the remaining seven probably recognised epitopes exposed by denaturation on drying or SDS treatment.
The second study [54] went to great lengths to isolate a VSG117 MoAb that recognised VSG on both live cells and after western blotting. Unlike the analyses above, this work was performed after the structure of a VSG had been determined and exploited the conservation of structure to map the epitope using recombinant chimaeric VSGs. The epitope was mapped to a region more than halfway down the VSG N-terminal domain (Fig 3B). This set of experiments provided very strong evidence that immunoglobulin G molecules can penetrate a minimum of 6 to 8 nm into the VSG coat, and most of the surface of the VSG N-terminal domain is accessible.
In a third study [55], a panel of seven MoAbs raised against VSG WaTat1.1 were tested for binding to a second VSG, WaTat 1.12, known to cross-react with WaTat1.1 polyclonal antisera. One of the seven MoAbs did not cross-react, and, since there are only 24 point differences in the sequences of the two VSGs, it can be assumed that one of these differences occurs in the epitope. The differences are located throughout the structure of the N-terminal domain (Fig 3C). From this comparison, it is not possible to identify the epitope recognised by the single selective MoAb. It is attractive to speculate that the difference lies between residues 73 and 82, which contain seven out of 24 differences; however, these residues are largely buried in the dimerization interface and would not be accessible in the VSG dimer.
The fourth study [53] used live trypanosomes and neutralising MoAbs that recognised VSG 78 to select mutants that escaped but still expressed a VSG recognised by a polyclonal anti-VSG 78. Several monoclonal antibodies were used to recognise different conformational epitopes. Two independent clones that escaped neutralisation with the first monoclonal antibody had changed serine 192 to arginine. The sensitivity to other monoclonal antibodies remained, showing that the overall structure was not affected by the mutation. Another mutant selected with the second monoclonal antibody expressed a VSG78 where glutamine 172 was changed to glutamic acid. The last isolated mutant selected with the third antibody had several changes in the VSG gene. There was a gene conversion in the 5′ region of the ORF and, in addition, a mutation in the codon 220 that was probably responsible for the resistance phenotype. All mutations identified are located in the loops at the membrane distal end of the VSG that would be readily accessible to antibodies on live cells.
Single domain antibodies (nanobodies, or NAbs) are derived from classes of immunogloblins that contain only two heavy chains and are unique to camelids. The variable domain is approximately 15 kDa, containing the antigen binding variable loops, and can be made as a recombinant protein. When these were produced against VSG AnTat1.1, a range of NAbs recognising different epitopes were isolated, including one that recognised the carbohydrate moiety on three different VSGs, all having N-linked oligosaccharides in the N-terminal domain [56]. The oligosaccharides on these three VSGs are located just over halfway down the N-terminal domain, and so the NAbs penetrate some distance into the VSG layer, as observed for one of the MoAbs described above [54].
While MoAb and NAb binding to the surface of the VSG N-terminal domain has been observed in multiple studies, the C-terminal domain does appear to be protected. Two polyclonal antisera to recombinant C-terminal domains both bound strongly after fixation but showed no binding to live trypanosomes [57]. This observation provides strong evidence that the VSG coat greatly reduces or eliminates penetration of immunoglobulins to the VSG C-terminal domain and, thus, the plasma membrane.
ISG65 and ISG75 are the two best-characterised invariant proteins present over the extracellular face of the plasma membrane of the entire body [25–27,43]. As detailed above, the extracellular domains of approximately 350 and approximately 440 residues, respectively, are comparable sizes to a VSG. Modeling of the structure of the domain suggests that the ISGs have a long coil of approximately 10 nm (Fig 2) [29]. Are the ISGs accessible to antibodies on live cells? The interactions with immunoglobulins were tested in two ways: first, the binding of anti-ISG immunoglobulin to fixed and live cells was compared; second, mice were immunized with recombinant protein and challenged with a trypanosome infection. There was a discrete binding of anti-ISG75, but not anti-ISG65, to live cells in one study [43]; subsequently, however, binding of anti-ISG65 has been reported with an independent antiserum [58]. The binding of anti-ISG75 was low compared to binding after fixation and was dependent on the VSG expressed, suggesting some, but not complete, limitation on accessibility. These experiments are difficult to interpret; at a simple level they could be taken to show that ISGs are accessible, but the epitopes recognised by the antisera were not characterized, and a confirmation through the use of defined MoAbs is required.
However, any ISG accessibility does not necessarily lead to immunity. Prior immunization with ISGs provided no protection against infection in mice [43]. It is also worth noting that infected people and animals produce anti-ISGs, but these do not provide protection [44]. Together, these observations allow a tentative conclusion that ISGs can be accessed by immunoglobulins, but binding is limited and tolerated by the trypanosome. The mechanism of this tolerance is probably related to the recycling of the cell surface [35]. One model for the tolerance might be that the combination of low ISG copy number and rapid recycling time does not allow the bound immunoglobulin to trigger a response. If this is the case, the VSG would shield part, but not all, of the ISG protein, ruling out a simplistic model of complete inaccessibility to non-variant surface proteins.
The Con A monomer is 29 kDa and, at pH 7.5 in physiological salt concentrations, is in approximately 1:1 dimer-to-tetramer equilibrium [59]. The dissociation constant for monomeric interaction is around 50 μM, with a dissociation rate of 4/s; consequently, any binding detected to live or fixed cells after washing must be multivalent. Con A is subject to very complex post-translational modification [60], and the properties of different batches of Con A are affected by different amounts of proteolysis of the monomeric units. The main effect of this variability is not on binding but on valency, with proteolysed subunits remaining as dimers [61]. Succinylated Con A is locked in the dimeric form and has been used to reduce variability in the reagent in some experiments [62]. Con A preferentially binds a branched mannose trisaccharide [63] and, in VSGs, will bind N-linked sites modified with oligomannose rather than paucimannose. The response to the pI context of the N-linked glycosylation site is gradual, and away from the extremes of pI values, many sites are modified with either paucimannose or oligomannose side chains. This means that predictions of whether a live trypanosome expressing a particular VSG will bind Con A (Table 1) have to be taken with a pinch of salt. The ability of Con A to bind to live trypanosomes was determined in many labs (for example, [62,64–66]). To summarize these results, Con A bound trypanosome clones expressing some VSGs but did not bind other clones expressing different VSGs. The majority fell into the second category, consistent with the predictions in Table 1. Nearly all these studies were performed before detailed sequence and structural data were available for VSGs. In light of what is known now, some of these data can be explained. The locations of the sites vary, as some VSGs have a single site in the C-terminal domain and others a single site in the N-terminal domain (Table 1). Obviously, a VSG with no N-linked sites will not bind Con A; for other VSGs, binding will depend on accessibility, which itself will be related to the location of the N-linked oligosaccharide on the tertiary structure of the VSG and on the context pI of the N-linked site. Most of the experiments to determine the nature of the binding of Con A to live trypanosomes were performed using VSGs of unknown sequence (no sequences were available at the time), but a number of the VSGs have been subsequently characterized. For example, trypanosomes expressing VSG MITat 1.6 (VSG 048) are not bound by Con A unless fixed or treated with trypsin [65], and this VSG was later shown to have a single N-linked glycosylation site in the linker between the two C-terminal domains [2]. The importance of the identity of the VSG in determining whether the N-linked oligosaccharide is accessible to Con A was clearly demonstrated in a study that used ten T. equiperdum clones expressing different VSGs; three were agglutinated, seven were not [64]. The sequences of one Con A binding VSG (BoTat 78) and one non-binding VSG (BoTat 1) have subsequently become available, and the location of N-linked glycosylation sites in these two VSGs provides an explanation for the difference: one site in VSG BoTat 1 is in the C-terminal domain; the other is at the base (plasma membrane proximal) of the N-terminal domain. In contrast, the sites in VSG BoTat 78 are located in the N-terminal domain, where they could present the oligosaccharides pointing toward the top of the VSG coat.
Trypsin is a 23 kDa protease with specificity for lysine and arginine residues. When trypsin is added to live trypanosomes, it is able to digest VSG and release fragments from the cell ([67], for example). Different VSGs are released at different rates [67]. In terms of VSG release, the most trypsin-sensitive point is the hinge between the N- and C-terminal domains. One way to explain the variations in sensitivity to trypsin is that the enzyme is near the size limit able to gain access to the hinge part of the VSG, and some VSGs block access, whereas other do not. However, this model ignores the availability of substrate, and some VSGs may simply be better substrates than others. It would be interesting to compare the trypsin sensitivity of different VSGs comparing purified protein and live cells.
Other species of African trypanosomes are much less well understood both in terms of the repertoire of functional VSGs and in non-VSG surface proteins. The genomes of T. congolense and T. vivax have been analysed for putative surface proteins [28], but there has been limited biochemical analysis to support this. In T. congolense, VSGs do not appear to have a structured C-terminal domain(s) but do have C-terminal extension of approximately 30 residues beyond the end of the structured N-terminal domain. The effect or role of this extension on the structure of the VSG coat is not known, and it may play an equivalent role to the C-terminal domain in T. brucei VSGs. T. vivax VSGs do not have any significant polypeptide extension on the C-terminus after the end of the structured VSG N-terminal domain, and there is little knowledge of experimentally identified non-VSG surface proteins.
The molecular detail of how the VSG coat negates the adaptive immune response is interesting in itself, but is also relevant to identifying therapeutic strategies. One important question to be answered is how far extracellular macromolecules can penetrate into the VSG coat. The answer to this question will provide information on the effectiveness of the VSG coat as a physical barrier and whether the cell has evolved systems to overcome immunoglobulin binding to lower copy number invariant antigens.
It is hard to draw many firm conclusions from the existing data, primarily due the absence of defined structures, ligands, and ligand binding sites. As examples: (i) The structure of ISG65 is only a model, and the epitopes recognised by the polyclonal ISG antisera have not been characterized. (ii) The structures of the N-linked oligosaccharides for some VSGs have been solved, but their location within the VSG coat is unknown (although it has been modeled [3]), and the relative affinity of Con A for the different N-linked oligosaccharides has not been determined. Another problem is that most of the data were collected before sequencing of VSGs became routine. Even now, only a subset of the experiments can be looked at with the sequence in one hand and structure-based hindsight in the other. The experiments with unambiguous data on the penetration of a macromolecule into the VSG coat provide very strong evidence that an intact immunoglobulin G could reach the lower part of the VSG N-terminal domain [54]. It is probable that the base of the VSG N-terminal domain, the region with the largest cross-sectional area perpendicular to the cell surface, represents the real physical barrier guarding the plasma membrane (Fig 4). It is possible that the C-terminal domain reinforces this barrier, as shown in Fig 4, but this location remains a model.
|
10.1371/journal.pntd.0002838 | Exploring the Potential of Flubendazole in Filariasis Control: Evaluation of the Systemic Exposure for Different Pharmaceutical Preparations | The goal of elimination of the human filariases would benefit greatly from the use of a macrofilaricidal agent. In vivo trials in humans and many experimental animal models suggest that flubendazole (FLBZ) is a highly efficacious macrofilaricide. However, since serious injection site reactions were reported in humans after parenteral FLBZ administration, the search for alternative pharmaceutical strategies to improve the systemic availability of FLBZ and its metabolites has acquired urgency in both human and veterinary medicine. The goal of the current work was to compare the systemic exposure of FLBZ formulated as either an aqueous hydroxypropyl-β-cyclodextrin (CD) or aqueous carboxymethyl cellulose (CMC) suspension or a Tween 80-based formulation (TWEEN) in rats and jirds (Meriones unguiculatus). Healthy animals of both species were allocated into four experimental groups of 44 animals each: FLBZ-CDoral and FLBZ-CDsc, treated with the FLBZ-CD formulation by the oral or subcutaneous routes, respectively; FLBZ-TWEENsc, dosed subcutaneously with the FLBZ-TWEEN formulation; and FLBZ-CMCoral, treated orally with the FLBZ suspension. The FLBZ dose was 5 mg/kg. FLBZ and its hydrolyzed (H-FLBZ) and reduced (R-FLBZ) metabolites were recovered in plasma samples collected from rats and jirds treated with the different FLBZ formulations. In both species, FLBZ parent drug was the main analyte recovered in the bloodstream. In rats, FLBZ systemic exposure (AUC0-LOQ) was significantly (P<0.05) higher after the FLBZ-CD treatments, both oral (4.8±0.9 µg.h/mL) and subcutaneous (7.3±0.6 µg.h/mL), compared to that observed after oral administration of FLBZ-CMC suspension (0.93±0.2 µg.h/mL). The same differences were observed in jirds. In both species, parenteral administration of FLBZ-TWEEN did not improve the systemic availability of FLBZ compared to FLBZ-CDoral treatment. In conclusion, formulation approaches that enhance the availability of flubendazole in the rat and jird may have therapeutic implications for a drug with poor or erratic bioavailability.
| Lymphatic filariasis and onchocerciasis are tropical parasitic diseases caused by filarial nematodes, which constitute a serious public health issue in tropical regions. Lymphatic filariasis causes debilitating lymphedema and hydrocele, resulting in temporary or permanent disability. Onchocerciasis (also known as river blindness) causes visual impairment and blindness, constituting one of the leading causes of blindness in the world. The control of human filarial infections currently depends on strategies predominantly focused at killing microfilariae (larval stage) by the use of ivermectin or diethylcarbamzine, usually in combination with albendazole. It is now generally recognized that the success of filariasis control programs in a reasonable time-frame requires the addition of a macrofilaricide (adult stage) compound. Although flubendazole has demonstrated macrofilaricidal activity in vivo, the approved formulations provide almost no oral bioavailability. The search for alternative pharmaceutical strategies to improve the systemic availability of flubendazole has acquired urgency in both human and veterinary medicine. Searching for improved flubendazole absorption, different flubendazole pharmaceutical preparations were assessed, both in rats and jirds, in the study described here. The work demonstrated that flubendazole pharmacokinetics could be markedly modified by changes in drug formulation. The resulting improved systemic exposure of flubendazole may have a significant impact on its macrofilaricidal efficacy.
| Lymphatic filariasis and onchocerciasis are tropical parasitic diseases caused by filarial nematodes in the superfamily Filarioidea, also known as “filariae”. Filariasis constitutes a serious public health issue in tropical regions. Approximately 128 million individuals suffer from lymphatic filariasis (commonly known as elephantiasis), mainly in Africa and South-East Asia. The disease causes debilitating lymphedema and hydrocele, resulting in temporary or permanent disability, impairment of physical productivity, income loss and social stigma [1]. Onchocerciasis [also known as river blindness) afflicts approximately 26 million individuals in Africa, where an estimated 746,000 are visually impaired and 265,000 are blinded by the disease, constituting one of the leading causes of blindness in the world [1]–[2]. Lymphatic filariasis is caused by Wuchereria bancrofti, Brugia malayi and Brugia timori, while Onchocerca volvulus is the cause of river blindness. Infective larvae of filariae are transmitted by blood-feeding insects, developing into fertile adults several months after infection. Chronic, long-term infections occur through suppression of host immunity [2].
Current control programs rely on three drugs, which are safe and available through donation: diethylcarbamazine (DEC), ivermectin (IVM), and albendazole (ABZ). DEC kills larval stages in the host (microfilaria) and provides long-term sterility of adults, with limited adulticidal efficacy in the regimens employed. It is contraindicated in areas where onchocerciasis is endemic, due to potentially serious and unacceptable side effects affecting the eyes and the skin of infected persons [2] as well as in pregnancy [3]. IVM is a microfilaricide and also provides long-term sterilization of adult worms, preventing re-population of the host with microfilariae for 6 months or longer, but needs to be given at least annually [4]. Like DEC, IVM has limited macrofilaricidal effects in humans or other animals, which greatly prolongs the time required for mass drug administration programs to progress to eradication [5]. Lastly, ABZ is routinely included with annual treatments of DEC or IVM in lymphatic filariasis control programs. Nevertheless, the activity of the benzimidazole component (ABZ) in this regimen is uncertain, and whether combination therapy confers benefits over DEC or IVM alone remains controversial [5]. Thus, the control of human filarial infections currently depends on strategies predominantly focused on killing microfilariae and the long-term cessation of their production [6]. It is now generally recognized that the success of filariasis control programs in a reasonable time-frame would be favored by the addition of a macrofilaricidal compound to current control strategies [4], [6], [7].
Flubendazole (FLBZ), a methylcarbamate benzimidazole (BZD), is highly active against a broad spectrum of gastrointestinal nematodes in humans and some animal species. FLBZ has also demonstrated a marked lethal effect on many filarial species in animal model hosts [6]. FLBZ has been reported to be the best macrofilaricidal molecule within the BZD group [8]. This compound is already approved for use in humans [9], which may be an advantage over other candidates for filariasis. FLBZ, like other BZDs, has limited water solubility and is commercially available for oral administration in humans as tablets or suspensions, providing low systemic bioavailability [10]. The macrofilaricidal activity of FLBZ is thought to require sustained systemic exposure, which is not achieved after administration of conventional oral formulations. The in vivo activity of FLBZ against a variety of filariid species has been reported after its parenteral administration in animal and human trials [6], [11]–[12]. FLBZ was available as a sterile suspension for intramuscular treatment [13]; however, since serious injection site reactions were reported in humans after parenteral FLBZ administration [12], the search for alternative pharmaceutical strategies to improve systemic availability of FLBZ after oral dosing has acquired urgency in both human and veterinary medicine.
Several pharmacotechnical strategies have been explored to enhance BZD systemic bioavailability. Cyclodextrins (CD), cyclic oligosaccharides used to increase drug solubility, are well-known molecular hosts capable of including water-insoluble guest molecules via non-covalent interaction within a hydrophobic cavity [14]. Enhanced aqueous solubility and bioavailability of guest molecules is a common effect observed after drug formulation with CD [15]. We have previously reported that incorporation of FLBZ into a hydroxypropyl-β-cyclodextrin (CD) formulation significantly increased its water solubility [16] and systemic exposure in mice by more than 25-fold compared to the conventional FLBZ suspension [17]. The relative bioavailability of albendazole sulphoxide (ABZSO) in mice was also increased by formulation with a CD [18]. Similar findings have been reported in humans [19].
The goal of the current study was to compare the plasma pharmacokinetic behaviour and systemic exposure of FLBZ formulated as either an aqueous CD-based solution (FLBZ-CD), aqueous carboxymethylcellulose (CMC) suspension (FLBZ-CMC) or a Tween 80-based formulation (FLBZ-TWEEN) in non-infected jirds (Meriones unguiculatus) and rats.
Pure reference standards of FLBZ, reduced-FLBZ (R-FLBZ) and hydrolyzed-FLBZ (H-FLBZ) used to develop the analytical methodology were kindly provided by Janssen Animal Health (Beerse, Belgium). Oxibendazole (OBZ), used as internal standard, was obtained from Schering Plough (Kenilworth, NJ, USA). HPLC grade acetonitrile and methanol were from Sintorgan S.A. (Buenos Aires, Argentina) and J.T. Baker (New Jersey, USA), respectively. HPBCDs were from ISP Pharmaceuticals (Cavasol, Cavitron, New Jersey, USA). Low viscosity grade sodium CMC was purchased from Anedra (Buenos Aires, Argentina). Tween 80 was purchased from Biopack (Buenos Aires, Argentina).
The FLBZ CD-based solution was prepared by dissolving FLBZ (0.1%) and CD (10%) in deionized water. The pH of the formulation was adjusted to 1.2 using hydrochloric acid (25 mM). The formulation was shaken until total dissolution of the drug and then was filtrated through a 0.45 µm filter (Whatman, NJ, USA). The final FLBZ concentration was confirmed by HPLC (n = 4). Cavitron and Cavasol were the CD used in formulations intended for oral and parenteral administration, respectively. The Tween 80-based formulation was prepared by dissolving FLBZ (0.25%) in Tween 80. The FLBZ-suspension was prepared by addition of FLBZ (0.1%) and CMC (0.1%) in deionized water (pH = 6.0) with shaking for 6 h. The FLBZ- CMC suspension was vigorously shaken immediately before intragastric administration to jirds and rats. FLBZ formulations were freshly prepared and maintained under refrigeration (3–5°C).
Chromatography was performed on a Shimadzu HPLC platform (Shimadzu Corporation, Kyoto, Japan), with two LC-10AS solvent pumps, an automatic sample injector (SIL-10A) with a 50 µL loop, an ultraviolet-visible spectrophotometric detector (UV) (SPD-10A) reading at 292 nm, a column oven (Eppendorf TC-45, Eppendorf, Madison, WI, USA) set at 30°C, and a CBM-10A integrator. Data and chromatograms were collected and analyzed using the Class LC10 software (SPD-10A, Shimadzu Corporation, Kyoto, Japan). The C18 reversed-phase column (5 µm, 250 mm×4.6 mm) was Kromasil (Kromasil®, Sweden). Elution from the stationary phase was carried out at a flow rate of 1.2 mL/min using an acetonitrile (34%)/ammonium acetate buffer (0.025 M, pH 5.3, 66%) as a mobile phase.
Plasma samples (100 or 200 µL for jirds and rats, respectively) were spiked with OBZ as internal standard. After 5 min, plasma samples were mixed with water up to 1 mL and the analytes were extracted using disposable C18 cartridges (Strata, Phenomenex, CA, USA) as previously described [17]. Identification of FLBZ and its metabolites was undertaken by comparison with the retention times of pure reference standards. Complete validation of the analytical procedures for extraction and quantification of drug and metabolites in plasma was performed before starting the analysis of experimental samples. Retention times for H-FLBZ, R-FLBZ and FLBZ were 5.7, 7.1 and 14.4 min, respectively. The calibration curves for each analyte, constructed by least squares linear regression analysis, showed good linearity with correlation coefficients ≥0.998. The limit of quantification (FLBZ and metabolites was 0.01 µg/mL), defined as the lowest measured concentration with a CV <20%, accuracy of ±20% and absolute recovery ≥70%.
The peak concentration (Cmax) and time to peak concentration (Tmax) were read from the plotted concentration–time curve for each analyte. The area under the concentration–time curve from 0 up to the limit of quantification (AUC0-LOQ) for FLBZ and metabolites was calculated by the trapezoidal rule [21], using the PKSolutionTM computer program (Summit Research Services, Ashland, OR, USA).
PK parameters are presented as arithmetic means ± SD. Non-parametric (Mann-Whitney) tests were used for statistical comparison of the pharmacokinetic data obtained from the experimental groups in each animal species. A value of P<0.05 was considered statistically significant. Statistical analysis was performed using the Instat 3.0 Software (Graph Pad Software, CA, USA).
No local tissue effects were observed in either species after sc administration of FLBZ formulated with either CD- or Tween 80. Figures 1 and 2 show mean plasma concentrations of FLBZ and metabolites after sc administration (5 mg/kg) of FLBZ in a CD-based formulation (FLBZ-CDsc) to rats and jirds, respectively. FLBZ and H-FLBZ were the main molecules detected in plasma of FLBZ-treated rats and jirds. Low R-FLBZ concentrations were detected between 15 min and 12 h post-treatment, with AUC0-LOQ values about 10% of the total drug recovered from plasma in the different groups for both species. FLBZ and H-FLBZ concentrations rapidly increased to reach peak plasma concentrations, observed as early as 0.7–3.2 h (FLBZ) and 3.0–5.2 h (H-FLBZ), according the experimental group and animal species.
The comparative plasma concentration profiles of FLBZ obtained after oral or sc administration as different formulations to rats, along with some pharmacokinetic parameters (Cmax and AUC0–LOQ), are shown in Figure 3. Table 1 summarises the plasma pharmacokinetics parameters (Cmax, Tmax and AUC0-LOQ) for FLBZ and H-FLBZ obtained after oral or sc administration of the FLBZ formulations to rats. Higher drug systemic exposure was obtained after administration of FLBZ as a CD or Tween 80-based formulation to rats compared to the CMC-based suspension, resulting in significantly higher Cmax and AUC0–LOQ values for both FLBZ H-FLBZ in the FLBZ-CDoral, FLBZ-CDsc and FLBZ-TWEEN groups (Table 1).
The sc administration of FLBZ-CDsc to rats improved its systemic exposure, resulting in significantly higher AUC0-LOQ values compared to all other experimental groups (Table 1). Additionally, FLBZ was detected in plasma for longer period (up to 16 h post-treatment) after parenteral administration (FLBZ-CDsc and FLBZ-TWEEN).
The administration of FLBZ as a Tween 80-based formulation to rats did not improve its systemic availability compared to oral administration of FLBZ-CDoral. Similar FLBZ AUC0-LOQ values were observed in the FLBZ-CDoral and FLBZ-TWEEN groups. In these groups, no differences were observed in either Cmax or AUC0-LOQ values obtained for H-FLBZ (Table 1).
Unlike rats, neither H-FLBZ nor R-FLBZ was detected in plasma at any time post-treatment of jirds with FLBZ-CMC. Only trace amounts of FLBZ were detected, and only over a period so short that it precluded pharmacokinetic analysis. However, treatment with either FLBZ-CD (oral or sc routes) or FLBZ-TWEEN solutions allowed quantification of FLBZ and its reduced and hydrolyzed metabolites in jirds. Similar to rats, FLBZ was the main analyte detected in plasma, whereas H-FLBZ concentrations represented 10–20% of the total amount of drug recovered, with even lower R-FLBZ concentrations in jirds (Figure 2). The comparative plasma concentration profiles (mean ± SD) for FLBZ after administration as CD- or Tween 80-based formulations are shown in Figure 4. Table 2 summarizes the main pharmacokinetic parameters obtained for FLBZ after oral (FLBZ-CDoral) or sc (FLBZ-CDsc and FLBZ-TWEEN) administration to jirds. The sc treatment with the Tween 80-based formulation did not improve FLBZ systemic exposure compared to FLBZ-CDoral; similar AUC0-LOQ and Cmax values were obtained in both groups (Table 2). However, FLBZ absorption in the FLBZ-TWEEN group was slower compared to sc and oral FLBZ-CD groups, since a significantly longer Tmax was observed in that group. Interestingly, FLBZ-CDsc delivered an enhanced FLBZ Cmax value (90–130%) compared to FLBZ-CDoral and FLBZ-TWEEN. A similar trend was observed in AUC0-LOQ values, but high individual variability may have obscured detection of statistically significant differences in this pharmacokinetic parameter among groups. Comparison of the relative contribution of FLBZ, H-FLBZ and R-FLBZ to the total drug plasma concentrations quantified after FLBZ treatment in different animal species, including rats, jirds, mice, pigs and sheep, is shown in Figure 5.
When pharmacological research cannot be done on humans for practical and ethical reasons, animal models constitute a practical approach to understand the parasite-active drug-host relationship. In the current work, two different animal models (rat and jird) were used to approximate what might be expected in humans. The overall plasma pharmacokinetic behaviour of BZD anthelmintics in humans is similar to other monogastric species such as mice or rats, and greatly differs from what has been reported in ruminant species (sheep, cattle). In ruminants, the rumen acts as a drug reservoir, and slowing the digesta transit time results in improved systemic availability as a consequence of greater dissolution of drug particles in the acidic pH of the abomasum (the stomach) [22]. The briefer gut transit time in monogastric species allows a shorter time for dissolution of the drug suspension compared to ruminants, limiting gastrointestinal absorption of the drug. Thus, rats or jirds could be valid animal models to obtain kinetic data extrapolatable to humans, particularly when systemic exposure of BZD anthelmintics is evaluated after the administration of different formulations. Additionally, jirds have been extensively used as animal models in drug screening studies for potential antifilarial compounds [8], [11]. Thus, the basic pharmacokinetic data reported here could be linked to efficacy trials against filarial nematodes in the same animal model.
Three main factors play important roles in activity against nematodes: i. attaining sufficient drug concentrations at the site of target parasite location to be able to therapeutically affect receptors in parasites [23]; ii. drug lipophilicity [22]; and iii. physicochemical features of the tissue/fluids surrounding the parasite [24]. Drug concentration at the site of parasite location depends on the chemical properties of the drug and the pharmaceutical preparations in which the active compound is formulated. Therapeutic failures observed in parasite control in both human and veterinary medicine may be related to exposure of parasites to sub-therapeutic drug concentrations due to poor drug dissolution and/or insufficient systemic availability of the active ingredient. Obtaining adequate drug concentrations in the compartment in which the parasite resides is a key factor that determines efficacy against systemic parasites. The physicochemical features of the parasite environment play a pivotal role in determining drug access and accumulation. Some nematode parasites living in host tissues may be protected from the deleterious effect of an anthelmintic due to low diffusion of lipophilic compounds. Furthermore, the low water solubility of BZD anthelmintics seriously limits their absorption and systemic bioavailability. Clearly, the poor oral absorption of FLBZ after administration in the conventional suspension/tablet formulations is a serious disadvantage for the treatment of systemic infections such as filariasis. Low FLBZ bioavailability has been associated with low in vivo activity against cystic echinococcosis in mice [17]. The use of pharmacotechnical strategies to overcome this limitation may markedly improve the in vivo efficacy of FLBZ against systemic parasitic nematodes.
The lack of water solubility is an important limitation for the formulation of the most potent BZD methylcarbamate anthelmintics, such as FLBZ. Irritation and post-injection precipitation are concerns in parenteral drug delivery for poorly water-soluble drugs [25]. The greater water solubility of the main active albendazole metabolite, albendazole sulphoxide (also named ricobendazole), was the starting point in the development of an injectable formulation for use in cattle currently available in some Latin American countries [26]. However, since BZD aqueous solubility is markedly higher at low pH values [27], that formulation contains ricobendazole (15% final concentration) at low pH (approx. 1–2), which produces irritation at the site of sc administration. Recently, a CD-based formulation of ricobendazole for parenteral use has demonstrated adequate tissue tolerability and bioavailability [25], but is not available in the veterinary market. Complexation with cyclodextrins has been intensively investigated as a solubilization approach for parenteral formulations. In agreement with our recent results, cyclodextrin formulations of poorly water-soluble drugs have shown little or no tendency for drug precipitation after intramuscular injection [25]. It is worth noting that most of the progress achieved to improve bioavailability of BZDs has been in formulation design [18], [28]–[29].
Prospects for an accelerated path to the elimination of onchocerciasis and lymphatic filariasis would be much enhanced if a safe and effective macrofilaricide were available [5], [6], [7]. Therefore, improvement of FLBZ systemic availability was the essential component under evaluation in the current work. We have previously reported that CDs markedly increase FLBZ water solubility [16], which was correlated with enhanced systemic drug exposure in mice [17], as demonstrated by significantly higher plasma Cmax (28 fold-higher) and AUC (27 fold-higher) values compared to a conventional suspension. Moreover, the efficacy of FLBZ against cystic echinococcosis in mice was also dramatically improved after oral administration of a CD-based formulation [17], [30]. Similar pharmacokinetic results were obtained for ABZ in mice [18], [30] and humans [19], in which a significantly higher systemic exposure for ABZ-sulphoxide was observed after ABZ administration in a CD-based solution. Consistent with previous data, the CD-based formulations FLBZ-CDoral and FLBZ-CDsc significantly increased FLBZ systemic exposure in rats compared to the FLBZ-CMC formulation. Similar FLBZ AUC0-LOQ values were observed between the FLBZ-CDoral and the Tween 80-based sc formulations. The highest FLBZ relative plasma availability was attained in the FLBZ-CDsc group, in which the AUC0-LOQ value increased by 684% compared to that observed after the oral administration of the FLBZ-CMC suspension. The FLBZ plasma detection period (up to 16 h post-treatment) was similar among the CD- and Tween 80-based formulations. CDs have the ability to complex with drugs, affording increased water solubility and improved oral bioavailability of FDA Class II compounds (poor aqueous solubility, high permeability) [15], such as the BZD anthelmintics. In the current work, the CD-based formulation induced drastic changes in FLBZ aqueous solubility, which accounted for its enhanced absorption and systemic availability in rats.
Interestingly, neither FLBZ nor its metabolites were detected in plasma after FLBZ-CMC treatment in jirds. Similar FLBZ plasma AUC0-LOQ values were observed among FLBZ-CD oral or sc treatments and the FLBZ-TWEEN groups. The H-FLBZ and R-FLBZ metabolites were recovered in plasma, although in much lower concentrations than the parent drug. In agreement with kinetic data obtained in rats, high FLBZ peak plasma concentrations were observed after sc administration as a CD-solution compared to the Tween 80 sc formulation. As previously mentioned, CD clearly improves FLBZ absorption in jirds after both oral or sc treatment. It is generally accepted that CDs enhance drug permeability by solubilizing their lipophilic components, thereby disrupting barriers to diffusion and increasing permeability. CDs may also act as permeation enhancers by carrying the drug in inclusion complexes through the aqueous barrier, from the bulk solution towards the surface of biological membranes [31].
Dominguez-Vazquez et al. [12] demonstrated that an injectable formulation of FLBZ was highly efficacious in humans against adult O. volvulus. Parenteral administration of a FLBZ Tween 80-based formulation has high efficacy against multiple filarial species in several animal hosts [6]–[8]. Since FLBZ plasma exposure obtained in the FLBZ-CDoral and FLBZ-CDsc groups was greater than obtained in the FLBZ-TWEEN group, high macrofilaricidal efficacy of the CD-based formulations may be possible. However, the potential of those formulations for treatment of humans may be limited by the high cost of the CD used in the formulation assessed in the current experimental work (hydroxyl propyl β-cyclodextrin).
Unlike other commonly used BZD anthelmintics, such as albendazole (aliphatic substitution at position -5) and fenbendazole (aromatic substitution at position -5), FLBZ contains a ketone group in that position, which has implications for its metabolism by the host. While sulphur-containing BZDs are sequentially oxidised to their sulphoxide and sulphone metabolites by both flavin-monooxygenase (FMO) and cytochrome P450 (P450) systems in the liver [32]–[33], carbonyl reductases (CBRs) are thought to be the main enzymes involved in FLBZ biotransformation [34]. The main FLBZ metabolic pathways include reduction of the ketone group to form R-FLBZ, and hydrolysis of the methylcarbamate group to form H-FLBZ. The contribution of each metabolite to the total amount of drug recovered from plasma after FLBZ treatment may vary among animal species. R-FLBZ is the main metabolite measured in plasma after FLBZ treatment in sheep [10], [16] and mice [17], [30] (Figure 5). However, while low plasma concentrations of FLBZ are detected in sheep, the parent compound was the main analyte in FLBZ treated mice. A different pattern was observed in pigs treated with FLBZ, in which H-FLBZ was the predominant molecule, representing 97% of total drug measured in the bloodstream after FLBZ treatment [35] (Figure 5). In rats, similar amounts of FLBZ and H-FLBZ were present in the bloodstream, with only trace amounts of the R-FLBZ metabolite. Although oral bioavailability of FLBZ has been estimated in humans [9], no data are available on the plasma pharmacokinetic pattern of FLBZ and metabolites. However, in vitro studies performed in our lab have shown that human microsomes biotransform FLBZ mainly to the R-FLBZ metabolite (unpublished data), which suggests similarity with the metabolic profile observed in mice and sheep (see Figure 4). Species-related differences in plasma drug exposure observed for FLBZ and metabolites may significantly influence drug efficacy. While H-FLBZ is an inactive metabolite, biological activity has been described for R-FLBZ [30], [36]–[37], which may contribute to anthelmintic efficacy observed after FLBZ treatment.
The marked improvement of FLBZ systemic availability observed after the administration of CD-based formulations to rats and jirds needs to be considered in terms of its potential usefulness as a macrofilaricide in animal models. If oral and/or parenteral administration of FLBZ-CD formulations provides satisfactory efficacy, the empirical correlation of plasma concentrations and efficacy may contribute to the development of new formulations for use in humans. The work reported here indicates that FLBZ plasma availability can be markedly improved by changes in formulation. The enhanced systemic exposure observed after treatment with the CD-based formulations has significant therapeutic implications for a drug with poor or erratic bioavailability.
|
10.1371/journal.pmed.1002326 | Modelled health benefits of a sugar-sweetened beverage tax across different socioeconomic groups in Australia: A cost-effectiveness and equity analysis | A sugar-sweetened beverage (SSB) tax in Mexico has been effective in reducing consumption of SSBs, with larger decreases for low-income households. The health and financial effects across socioeconomic groups are important considerations for policy-makers. From a societal perspective, we assessed the potential cost-effectiveness, health gains, and financial impacts by socioeconomic position (SEP) of a 20% SSB tax for Australia.
Australia-specific price elasticities were used to predict decreases in SSB consumption for each Socio-Economic Indexes for Areas (SEIFA) quintile. Changes in body mass index (BMI) were based on SSB consumption, BMI from the Australian Health Survey 2011–12, and energy balance equations. Markov cohort models were used to estimate the health impact for the Australian population, taking into account obesity-related diseases. Health-adjusted life years (HALYs) gained, healthcare costs saved, and out-of-pocket costs were estimated for each SEIFA quintile. Loss of economic welfare was calculated as the amount of deadweight loss in excess of taxation revenue. A 20% SSB tax would lead to HALY gains of 175,300 (95% CI: 68,700; 277,800) and healthcare cost savings of AU$1,733 million (m) (95% CI: $650m; $2,744m) over the lifetime of the population, with 49.5% of the total health gains accruing to the 2 lowest quintiles. We estimated the increase in annual expenditure on SSBs to be AU$35.40/capita (0.54% of expenditure on food and non-alcoholic drinks) in the lowest SEIFA quintile, a difference of AU$3.80/capita (0.32%) compared to the highest quintile. Annual tax revenue was estimated at AU$642.9m (95% CI: $348.2m; $1,117.2m). The main limitations of this study, as with all simulation models, is that the results represent only the best estimate of a potential effect in the absence of stronger direct evidence.
This study demonstrates that from a 20% tax on SSBs, the most HALYs gained and healthcare costs saved would accrue to the most disadvantaged quintiles in Australia. Whilst those in more disadvantaged areas would pay more SSB tax, the difference between areas is small. The equity of the tax could be further improved if the tax revenue were used to fund initiatives benefiting those with greater disadvantage.
| Previous real-world evaluations of a sugar-sweetened beverage (SSB) tax showed that the SSB tax led to a reduction of SSB purchases for the total population, with larger effects for lower-income households.
It was unknown what the healthcare cost savings, health gains, and financial impacts of an SSB tax would be for different income groups, in Australia or internationally.
We modelled the effect of a 20% SSB tax in Australia on life expectancy and health-adjusted life years before and after implementation of the tax, across quintiles of area-level socioeconomic deprivation.
Our model predicts that the greatest health gains would accrue to the 2 lowest quintiles (most disadvantaged), leading to the highest healthcare cost savings in these quintiles.
We estimate the increase in annual expenditure on SSBs to be AU$35 per capita in the lowest quintile, a difference of less than $5 compared to the highest quintile.
Annual tax revenue was estimated at over AU$640 million.
A 20% SSB tax in Australia is likely to decrease SSB purchase and consumption, leading to significant health gains and healthcare expenditure savings across all quintiles of socioeconomic deprivation.
The tax would generate considerable yearly revenue, which the government could use to further improve the health of the most disadvantaged.
As with all simulation models, the model results represent the best estimate of a potential effect in the absence of stronger direct evidence.
| In high-income countries, obesity is more common in the most disadvantaged groups [1]. Reducing inequalities in health between advantaged and disadvantaged groups is an important objective of public health policy [2]. The evidence of the association between sugar-sweetened beverage (SSB) intake and increased energy intake, leading to weight gain and obesity, is compelling [3,4]. Obesity is a strong risk factor for diabetes, cardiovascular disease, some cancers, osteoarthritis, and hypertension [5–7]. Individuals from lower socioeconomic groups have been found to consume more SSBs [8,9]. The prevalence of obesity-related comorbidities is also higher in lower socioeconomic groups.
A tax on SSBs is considered to be an important component of the set of recommended policy approaches to address population obesity [10–12]. Price influences SSB purchase [13], which in turn may reduce the rate of obesity [14]. There is an economic rationale for taxes when consumption results in negative externalities. In Australia, the diseases caused by obesity were estimated to cost tax payers AU$5.3 billion in healthcare costs, forgone tax, and welfare payments in 2014/2015 [15]. The economic rationale for an SSB tax essentially rests on the notion of ‘internalising the externality’ within the purchase price.
There is evidence that people with lower incomes are more sensitive to price increases [16] and are therefore more likely to change their purchasing behaviour in response to price changes. In Mexico, an evaluation of an SSB tax of approximately 10% introduced in 2014 showed a reduction in purchases of taxed beverages for the total population, with an even larger effect for lower-income households [17,18].
The financial impact of an SSB tax for different socioeconomic position (SEP) groups has been examined in terms of the predicted tax burden to individuals and households. A recent systematic review describing the financial burden of an SSB tax across different SEP groups identified 5 studies, which found that the tax would be financially regressive, but with small differences of approximately US$5 between high- and low-income households; the average additional tax paid per household as a result of the SSB tax would be less than US$30 annually across all groups [19]. Previous Australian research has predicted that an SSB tax would lead to cost savings in the health sector [20,21]. But the effect on overall healthcare cost savings and the health gains in health-adjusted life years (HALYs) across SEP groups have rarely been previously examined. The overall financial impact on individuals includes the potential healthcare costs saved by individuals, and this also has seldom been previously estimated across SEP groups. A rate of 20% is the most commonly advocated tax by public health experts [22]. South Africa and the UK have recently proposed taxes of this magnitude [23]. The main aim of this paper, therefore, is to examine the health and financial impacts of a 20% SSB sales tax for Australia across socioeconomic groups by comprehensively integrating distributional aspects into the cost-effectiveness analysis.
This study expands on previous studies in a number of ways. First, the cost-effectiveness of a 20% SSB tax for Australia by SEP subgroup was estimated, including a wide range of SSBs and substitute beverages, with a focus on the quantity and distribution of health gains in HALYs according to an area-based measure of socioeconomic disadvantage, Socio-Economic Indexes for Areas (SEIFA). Second, the distribution of financial impacts to individuals across different SEP groups was examined, in terms of out-of-pocket costs incurred from the tax and healthcare costs saved. Third, the overall economic impact of the tax was examined in terms of the balance of effects for the health sector and the general economy.
A 20% sales tax on SSBs in Australia was modelled. SSBs included soft drinks (pop, soda); flavoured water; sports, energy, and fruit drinks; and cordials (concentrates) containing added sugar. It was assumed that the full amount of the tax would be passed on to the consumer.
The model estimated the differences in life expectancy and HALYs pre- and post- implementation of the tax. These differences were based on predicted variations in 9 diseases caused by obesity. Changes to body mass index (BMI) were modelled based on projected changes in SSB consumption.
The Australian population aged 2–100 years was modelled over a lifetime, with covariates based on the Australian Health Survey (AHS) 2011–12 [24] and disease epidemiology based on a study of the US burden of diseases, injuries, and risk factors in 2010 [25]. The analysis has 2 parts: (1) a whole population analysis and (2) analyses by SEIFA quintile. Fig 1 illustrates the logic pathway of an SSB tax, identifying the steps involved in measuring the expected impact of the tax from an obesity perspective.
Table 1 outlines the general methodology of the economic evaluation. The intervention was assumed to be operating in ‘steady state’ (i.e., running at its full effectiveness potential) and was measured against current practice. Establishment costs were included in the cost of the intervention. The additional costs and the associated health benefits (HALYs) resulting from the tax were used to calculate incremental cost-effectiveness ratios (ICERs), defined as the difference in net costs of the tax compared to no tax, divided by the difference in net HALYs.
The impact of uncertainty around input values on the main outcome measures was estimated by Monte Carlo simulations (Table 2). Means and 95% confidence intervals for BMI effects on HALYs and intervention costs were reported based on 2,000 iterations using Ersatz version 1.3 software [52].
We performed several sensitivity analyses. We performed one-way sensitivity analyses to explore the effect of including flavoured milk in the SSBs. As the price elasticity for flavoured milk was not available, we assumed the same price elasticity as for soft drinks. We also tested an SSB tax rate of 30% and a 50% pass-through of the 20% tax. Another mechanism for implementing a tax—a 50¢ per litre volumetric tax was also tested. This is in line with alcoholic beverages in Australia, which are taxed per litre of alcohol. A tax of 50¢ per litre is an average 17% increase in price across all SSB categories.
A concentration index quantifies the degree of socioeconomic inequality in a specific health variable. Concentration indices were calculated for each tax scenario (sensitivity analysis) to quantify the degree to which HALYs gained are concentrated in disadvantaged groups. The index takes a negative value when HALY gains are greater amongst the most disadvantaged, and a positive value when HALY gains are greater amongst the least disadvantaged. The concentration index was calculated for each tax scenario using the following formula [54]:
C=2u∑t−1TftutRt−1
(4)
where ut is the mean number of HALYs of the tth SEIFA group, ft is its population share, and Rt is the fractional rank of SEIFA group t.
Enacting a 20% SSB sales tax in Australia was estimated to result in greater decreases in weight for the 3 most disadvantaged quintiles than for the 2 least disadvantaged quintiles for both men and women, with larger decreases in men. Quintile 5 (least disadvantaged) had the lowest predicted reductions in weight for men and women (Figs 2 and 3).
As a result of a 20% SSB tax, the Australian population was estimated to gain 175,300 HALYs (95% CI: 68,700; 277,800) and save 111,700 years of life (95% CI: 43,600; 175,800) (Table 3). The HALY gains were highest in the 2 most disadvantaged quintiles, with 49.4% of the total HALYs gained accruing to these quintiles. Quintile 5 had the lowest HALYs gained and years of life saved for men and women.
The tax was estimated to be cost saving across all intervention scenarios (sensitivity analyses) and all quintiles. Over the lifetime of the population cohort, expected healthcare cost savings were AU$1.73 billion, intervention costs were estimated to be AU$119.6m (95% CI: $91.9m; $162.1m)—approximately $4.8m (95% CI: $3.9m; $6.1m) in the first year and $3.7m (95% CI: $2.8m; $5.0m) in subsequent years, discounted at 3%. For every dollar invested in the first 10 years, the tax would result in AU$17 (95% CI: $9; $19) in healthcare cost savings. The tax revenue generated at the population level was estimated to be AU$642.9m annually (95% CI: $348.2m; $1,117.2m).
For the total population, the out-of-pocket healthcare costs saved were estimated to be AU$299.4m (95% CI: $113.8m; $476.2m). Healthcare cost savings as a percentage of household expenditure by quintile were highest in the most disadvantaged groups. Per capita, the most disadvantaged quintile was estimated to incur the most tax, at an estimated AU$35.40 (95% CI: $18.70; $62.80) per year, or 0.54% of expenditure on food and non-alcoholic drinks.
The tax revenue raised outweighed the DWL for the total Australian population, with an estimated net deadweight impact of +AU$587.9m (95% CI: +$329.2m, +$1,027.6m) per year. The DWL was more than offset by tax revenue across all quintiles, with substantial net gains in each quintile (Table 4).
The SSB tax remained cost saving when (1) the pass-through rate was 50%, (2) the rate of the tax was 30%, (3) flavoured milk was included as an SSB, and (4) a volumetric tax was applied at 50¢ per litre (see S4 Table). For each dollar invested in the first 10 years, the resulting healthcare cost savings ranged from $10 to $25 (Table 5).
All tax scenarios have a negative concentration index, indicating that the highest proportion of HALYs gained is amongst the most disadvantaged quintiles. The 50% pass-through of a 20% tax and the 50¢ per litre tax had the largest negative indices, indicating the most equitable scenarios (Table 5).
The 30% tax rate scenario resulted in the largest difference between the lowest and highest quintiles in terms of out-of-pocket costs for the tax; however, this scenario resulted in the largest health gains and healthcare costs saved across the population. Compared to a 20% tax, a 50¢ volumetric tax resulted in smaller health gains across all SEIFA quintiles, due to the level of tax translating to a lower level of price increase across all drink categories.
In our study we estimate that a 20% sales tax on SSBs in Australia would result in the largest number of obesity-related HALYs being averted in the population living in the most disadvantaged SEIFA quintiles, and it follows that the most healthcare cost savings overall would accrue to these groups. The expected out-of-pocket tax expenditure was highest in the most disadvantaged quintile; however, the difference of 0.32% points (less than 10¢) between the lowest and highest quintiles in proportion of household spending on food and non-alcoholic beverages per week was small. Our results indicate that, as a proportion of overall spending, the lowest SEIFA quintiles would have the largest out-of-pocket healthcare cost savings.
The DWLs for each SEIFA quintile, as well as for the whole population, were negative. This indicates that the loss of consumer/producer benefit would be outweighed by the amount of tax collected under our assumptions. The ‘loss in economic welfare’ is often calculated as the dollar amount of DWL in excess of dollar taxation revenue collected by the government. In our analysis there is a substantial net taxation gain suggestive of an improvement in economic welfare (this underlies the rationale for internalisation of negative externalities). But there are also behavioural responses associated with this inherently dynamic interaction that are difficult to model in these formulaic terms—desirably, industry would realign to healthier products and minimise its loss in producer surplus, consumers would realign to healthier purchases and minimise their loss in consumer surplus, and, finally, the tax revenue could be utilised for welfare-enhancing initiatives.
In the United Kingdom, it was considered reasonable to assume a pass-through rate of 100%; however, empirical evidence is mixed. The effect of manufacturers or retailers absorbing part of the tax could decrease the impact of the tax and the resulting health benefits; however, based on our predicted results for a 50% pass-through, the healthcare cost savings would nevertheless be substantial. There could also be an additional ‘halo effect’—a decrease in purchasing of SSBs from the introduction of the tax caused by increased public health awareness.
This research builds on the growing evidence that a tax on SSBs would deliver the largest health gains for the lowest socioeconomic groups. It also reinforces previous findings that the overall amount of tax per capita for a 20% value-added tax is around $30 per year, or less than 60¢ per week, and differences in tax expenditure between the lowest and highest socioeconomic groups are small [19].
The predicted body weight losses in our study are lower than those in Sharma et al.’s study [28], and this is because we took into account age and sex differences. Tax expenditures in our study are higher overall, and this is due to different price assumptions, as well as differing baseline intake of SSBs. The differences in baseline intake can be explained by the differing data collection methods. We used individual survey data recorded over 2 days from the AHS 2011–12, from which we took an average daily intake. These averages are close to the estimates from Euromonitor International of per capita purchases of SSBs in Australia [55]. Our predicted HALYs saved and expected tax revenue are slightly higher than in the previous Australian study that also modelled a 20% tax on SSBs [21], but this to be expected given that we included children in our analysis. Our predicted healthcare cost savings are higher due to a different method for calculating cancer treatment costs, based on the incidence rates rather than prevalence, and all costs have been updated to 2010.
This is possibly the first cost-effectiveness study to include the explicit health and economic outcomes by SEP and the resulting DWLs. It also expands on previous Australian research to include a wider range of SSBs and substitute beverages. We used conservative own-price elasticities that were close to half the value of other published price elasticities for soft drinks [45,56]. They take into account that SSB prices are not fixed, and households might face a quality—quantity tradeoff for each beverage and could opt for cheaper brands if they prefer quantity over quality [28].
When comparing our average changes in adult kilojoule intake per day for the population to a randomised control trial of overweight and obese adults who replaced all caloric SSBs with non-caloric beverages, the tax has approximately 20%–26% of the impact of the results from the trial (49 kcal/day decrease in our model versus 260 kcal/day and 187 kcal/day decreases for overweight and obese individuals, respectively, in the trial) [57]. This proportion of the effect is similar to the average change in own-price elasticities of consumption across all categories of SSBs of approximately 23% for all households as a result of a 20% SSB tax when compared to the trial [28].
As with all simulation models, the model results represent the best estimate of a potential effect in the absence of stronger direct evidence. We used an aggregate area-based indicator of SEP (SEIFA), as we were unable to obtain income-specific input data. We therefore assumed that price elasticities for household income groups were similar for SEIFA groups. There are also inherent limitations of survey data, such as misreporting, which may have affected the baseline intake of SSBs. Cross-price elasticities of food substitutes by SEP were not available, so these were not included [58]. Around 75% of soft drink sales are from supermarkets, and prices may be slightly different to our estimates.
In some instances, we used costing frameworks from the US and NZ in the absence of Australian estimates. NZ costs for policy advice provided by government agencies to parliament related to new laws is likely to be similar to their Australian equivalents, as the legal systems and number of new laws passed in Australia and NZ are similar. Costs to the government for compliance and administration of the introduction of plain packaging of tobacco products in Australia were similar to our estimates [46].
The model does not incorporate the effects of changes in SSB consumption on oral health or indirect costs, such as reduced productivity due to absenteeism and disability, which means that the societal savings from the intervention are likely to be substantially underestimated, especially to those in the most disadvantaged groups. The assumptions for the quality of life lost in children due to obesity are based on the best available evidence, but this evidence is from only particular age groups of school-aged children, and we have assumed the effects are similar for a wider age group.
Dedicating a portion of the substantial revenue generated from SSB taxes to efforts to reduce and prevent obesity among the most disadvantaged populations could be a way to further reduce concerns about the impact of the tax on low SEP groups. Hypothecation of taxes is also effective in generating public support [59]. There is evidence in Australia that earmarking the tax revenue for subsidising healthy food [60], tackling childhood obesity, and supporting children’s sport [61] and health promotion initiatives [62] would raise the public support for such a tax. Future studies could examine where to direct the revenue from an SSB tax for optimal equity, efficiency, and affordability.
Many countries and jurisdictions around the world have committed to an SSB tax, and this analysis shows that a 20% SSB tax is likely to result in a decrease in the purchase and consumption of sugary drinks, leading to significant health gains and healthcare expenditure savings across all quintiles of SEP. The tax would result in considerable yearly revenue that the government could use to reduce the regressive financial impacts, by funding programs to further improve the health of the most disadvantaged. Australia should consider a tax on SSBs as part of a suite of recommended policies to reduce the rates of obesity.
|
10.1371/journal.pntd.0003313 | Simple Fecal Flotation Is a Superior Alternative to Guadruple Kato Katz Smear Examination for the Detection of Hookworm Eggs in Human Stool | Microscopy-based identification of eggs in stool offers simple, reliable and economical options for assessing the prevalence and intensity of hookworm infections, and for monitoring the success of helminth control programs. This study was conducted to evaluate and compare the diagnostic parameters of the Kato-Katz (KK) and simple sodium nitrate flotation technique (SNF) in terms of detection and quantification of hookworm eggs, with PCR as an additional reference test in stool, collected as part of a baseline cross-sectional study in Cambodia.
Fecal samples collected from 205 people in Dong village, Rovieng district, Preah Vihear province, Cambodia were subjected to KK, SNF and PCR for the detection (and in case of microscopy-based methods, quantification) of hookworm eggs in stool. The prevalence of hookworm detected using a combination of three techniques (gold standard) was 61.0%. PCR displayed a highest sensitivity for hookworm detection (92.0%) followed by SNF (44.0%) and quadruple KK smears (36.0%) compared to the gold standard. The overall eggs per gram feces from SNF tended to be higher than for quadruple KK and the SNF proved superior for detecting low egg burdens.
As a reference, PCR demonstrated the higher sensitivity compared to SNF and the quadruple KK method for detection of hookworm in human stool. For microscopic-based quantification, a single SNF proved superior to the quadruple KK for the detection of hookworm eggs in stool, in particular for low egg burdens. In addition, the SNF is cost-effective and easily accessible in resource poor countries.
| Hookworm infection is widespread in resource-poor countries worldwide. Detection of hookworm eggs in human feces can be done by the Kato Katz technique (KK), sodium nitrate flotation technique (SNF) or PCR. This study was conducted to evaluate and compare the diagnostic parameters of the KK and simple SNF in terms of detection and quantification of hookworm eggs, with PCR as an additional reference test in stool, collected as part of a baseline cross-sectional study in Cambodia. PCR demonstrated the highest sensitivity for hookworm detection. By microscopy, SNF of a single stool sample proved superior for the detection of hookworm eggs in feces than quadruple Kato Katz smears. Hookworm egg counts were higher by SNF than those obtained using Kato Katz. Thus, the SNF proved superior to the quadruple Kato-Katz smears for the detection of low egg burdens and for the quantification of egg intensities. We propose the simple SNF is a superior alternative to the Kato-Katz for detection and quantification of hookworm infection in resource poor counties. The test is cost-effective and easily accessible.
| Human hookworms are estimated to infect between 576–740 million people globally and are responsible for a global burden of 3.2 million disability-adjusted life years [1], [2]. Hookworms are a leading cause of iron deficiency anemia and protein malnutrition, especially among pre- and school-aged children and untreated infections are known to result in adverse maternal-fetal outcomes in pregnant women [3]. The principal intervention strategy for hookworm infection is periodic mass drug administration of humans with the benzimidazole drugs, albendazole or mebendazole.
Diagnosis of soil-transmitted helminth (STH) infections, including hookworm has largely relied on copromicroscopy techniques based on the detection and quantification of eggs in feces. These tests aim to offer simple, reliable and economical options for assessing the prevalence and intensity of STH infections and monitoring the success of drug efficacy trials and helminth control programs. Of these, the Kato-Katz (KK) technique is currently the most widely used and accepted diagnostic technique recommended by the World Health Organization (WHO) [4]. The KK technique is relatively simple, reproducible, requires minimal equipment and the kit is mostly reusable. Hence the technique is inexpensive and commonly used as a field-based or point-of-care diagnostic test. The major disadvantage of the KK technique, however, is its lack of sensitivity for the detection of low levels and low intensities of STH infections [5]. In addition, hookworm eggs rapidly disappear in cleared slides, resulting in false negative test results if the interval between preparation and examination of the slides is too long (>30 min) [6]. For these reasons, it is necessary to increase the sensitivity of the KK technique by examining single fecal samples using multiple KK smears and/or by examining multiple fecal samples over multiple consecutive days [7], [8].
The sodium nitrate flotation (SNF) technique has been used in the veterinary field for diagnosing helminth infections for the last four decades. This method is currently the diagnostic test of choice for enteric parasites in small animals (e.g. dogs, cats) and commonly utilized with a commercial reusable stand-up fecal flotation device known as the Fecalyzer (EVSCO 014008-50, USA). Recent studies suggest that fecal flotation techniques hold promise for the diagnosis of STH infections in humans. A single fecal flotation using the FLOTAC and more recently the mini-FLOTAC device has consistently been shown superior in terms of sensitivity compared to triplicate KK and ether concentration methods for the detection of hookworm eggs in stool [5], [9], albeit at the expense of lower egg counts [10], [11], [12], [13].
A number of studies have reported the superior diagnostic parameters of molecular-based diagnostic techniques compared to those of microscopy for the detection of parasite stages in feces, including hookworms [14], [15], [16], [17].We therefore utilized a previously validated polymerase chain reaction (PCR) targeting the internal transcribed spacer (ITS)-1, 5.8S and ITS-2 region of hookworms as an additional diagnostic test for assessing the sensitivity of the coproscopy-based methods. This study was conducted to evaluate and compare the diagnostic parameters of the KK and the SNF methods in terms of detection and quantification of hookworm eggs in stool collected as part of a baseline cross-sectional study in Cambodia, with PCR as an additional reference test,.
The research was approved by the Ethics Committee of the Cantons of Basel-Stadt and Baselland (EKBB, #18/12, dated 23 February 2012), Switzerland, and by the National Ethics Committee for Health Research, Ministry of Health, Cambodia (NECHR, #192, dated 19 November 2011). Written informed consent was obtained from each participant prior to the start of the study. For participants between the ages of 2 and 18 years, written informed consent was obtained from the parents, legal guardian or appropriate literate substitute. All participants were informed of the study's purpose and procedures prior to enrolment. All parasitic infections diagnosed were treated according to the guidelines of the National Helminth Control Program of Cambodia [18].
The study was conducted in Dong village, Rovieng district, Preah Vihear province, Cambodia [9]. In brief, a total of 205 persons were randomly chosen for inclusion in this cross-sectional study. Two fecal samples were collected from each enrolled participant over two consecutive days. On the day of the first visit, informed consent was obtained from all household members and questionnaire interviews were conducted [9]. To all enrolled participants, pre-labeled stool containers were distributed. Participants were asked to defecate during the morning on the following day where stool samples were collected and a second stool container distributed. One half of the collected stool samples were transported at ambient temperature to the laboratory in the Rovieng Health Center within three hours after defecation. One part (approximately 2 g) was placed into a 15 ml centrifuge tube containing 8 ml of 10% formalin for examination using SNF and the other part (approximately 2 g) was placed into a 15 ml centrifuge tube containing 8 ml of 2.5% potassium dichromate for PCR analysis and transported to the School of Veterinary Science, University of Queensland, Gatton campus, Australia. The same collection procedure of fecal samples was carried out in the morning of the second day with samples immediately subjected to a second round of examination by the KK method.
For each stool sample two KK smears (duplicate slides) were prepared. For each person four KK smears were examined (two smears on each of two stool samples). The preparation of each slide was done following the protocol previously described [19]. Number 120-sized nylon mesh screen was used for filtering the stool and a standard plastic KK template was used to deliver 41.7 mg of stool from each sample onto each slide. The smear was examined under light microscope after allowing for clearance for 30 min. Total number of hookworm eggs observed on the slide was counted and noted. Egg counts were multiplied by 24 to obtain the number of eggs per gram feces (epg).
SNF was carried out according to a previously described protocol [20] on a single stool sample per study participant. Briefly, the formalin fixative was poured off and a fecal suspension prepared by thoroughly mixing approximately two gram of each stool sample with four times its volume of distilled water. The suspension was strained though a small funnel lined with two layers of surgical gauze directly into a 10 ml centrifuge tube and centrifuged for two min at 3,000× g. The supernatant was poured off leaving behind the fecal pellet (250 mg). Two mL of sodium nitrate solution (specific gravity 1.20) was added and the pellet mixed into a slurry using a wooden spatula. Sodium nitrate solution (specific gravity 1.20, or 315 gm/L of water) was then filled to the rim of the centrifuge tube, forming a positive meniscus and a 22 mm×22 mm cover slip was carefully placed on top. After 10 min, the cover slip was removed and placed onto a microscope slide. The entire slide was examined under light microscope at 100× magnification in a zig-zag fashion and the total number of hookworm eggs on the coverslip was counted. The observed number was multiplied by four to obtain the epg.
Genomic DNA was extracted directly from human fecal samples using the PowerSoil DNA Kit (Mo Bio, CA, USA) according to manufacturer's instructions with minor modifications and PCR carried out as previously described [17]. A positive control of each hookworm species and a negative control of distilled water were included in each run. The PCR products were visualized on 1% agarose gels in Sodium Borate (SB) buffer and stained by SYBR safe® Nucleic Acid Gel Stain (Life Technologies, Invitrogen, Eugene, USA).
The results of the fecal examinations were entered in EXCEL (Microsoft, USA) and analyzed by using STATA version 12.1 (StataCorp LP; College Station, TX). To estimate sensitivity, specificity and negative predictive value (NPV), results for the three techniques were categorized into positive and negative variables, presented in cross-tabulations, and compared for equal possibilities of being positive by using McNemar's test with 95% confidence interval (CI). The combination of the three techniques was used as diagnostic “gold standard” to estimate the sensitivity and specificity of each technique. Agreement among infection intensities of the two techniques (only KK and SNF) was estimated from their mean epg values, using paired student t-test. The “true prevalence” was calculated with the model developed by Marti and Koella, described elsewhere [21].
For a diagrammatic guide to the study design and summary of the diagnostic results refer to Fig. 1. The overall prevalence of hookworm infection in humans was 61.0% by the combined techniques, 56.1% by PCR, 26.8% by SNF and 22.0% by quadruple KK (16.6% by day 1 KK, 12.2% by day 2 KK, Table 1).
The calculated sensitivities, specificities and NPVs with 95% CI are shown in Table 1. Briefly, the sensitivity of PCR was the highest (92.0%) followed by SNF (44.0%), the quadruple KK (36.0%), day 1 KK (27.2%) and day 2 KK (20.0%) respectively. The specificity of KK, SNF and PCR was assumed 100%.
Comparison of the median intensity of hookworm infection by age group is shown in Fig. 2. The overall epg count from the quadruple KK was higher than those measured using SNF. However, the epg measured using SNF were higher than quadruple KK in two out of five age groups (21–30 years, 31–50 years). Therefore, there was no significant difference in epg between the two methods. We compared the median epg of the SNF between the samples found only positive by SNF (median epg: 160) to the samples that were analyzed by SNF and KK (median epg: 448). Using the Mann-Withney U test, there was no significant difference in epg values (P = 0.121).
The estimated “true” prevalence for hookworm infection based on the quadruple KK and gold standard were 30.3% and 70.2%, respectively which is an increase of 8.3% from our observed prevalence of 22.0% for quadruple KK and increase of 9.2% from our observed prevalence of 61.0% for gold standard (Fig. 3).
In the present study, three diagnostic techniques (KK, SNF and PCR) were assessed for the qualitative and two techniques (KK and SNF) for the quantitative detection of hookworm eggs in fecal samples from humans in Cambodia. Direct comparison of the three diagnostic techniques showed that the PCR assay had a superior sensitivity compared to the SNF, the single, duplicate and quadruplicate KK techniques. The KK when performed in duplicate with stool samples collected over two consecutive days provided a higher sensitivity (36.0%) for diagnosing hookworm infection when compared to one day KK (day 1 or day 2) alone (27.2% and 20.0%). The ten individuals that were positive by KK and negative by SNF were also found negative on PCR. Therefore it is likely that these 10 positives were false positives on the quadruplicate KK, which is yet another disadvantage of this diagnostic approach. The field of view is poor compared to the SNF and fecal artifacts can be mistaken as helminth eggs. The PCR results are likely explained by two factors: (i) false negatives - the inability to amplify these samples could be associated with failure to remove PCR inhibitors in human stool following DNA extraction [22], (ii) false-positive coproscopy results, i.e. that Trichostrongylus eggs detected in stool were misidentified as hookworm eggs. Trichostrongylus columbricformis which is present in humans in neighboring countries such as Lao People's Democratic Republic [23] and Thailand [24] produce eggs very similar to hookworms. Although there are no published reports of human infection with this species in Cambodia, T. columbricformis infection in humans cannot be disregarded because molecular identification of other strongylid nematodes was not attempted in this study.
The poorer sensitivity of the KK may be directly related to the significantly smaller amount of filtered feces examined (41.7 mg) compared to that of the SNF (∼250 mg). The addition of a washing step in the SNF procedure coupled with flotation provides a relatively ‘clearer’ and debris-free view of the hookworm eggs, thus making microscopic screening and quantification more accurate and time efficient. For a skilled parasitologist, a single SNF would take a maximum of 30 minutes to perform, including quantification of hookworm eggs. Time is thus a significant advantage for the SNF, both in terms of sampling logistics (single versus two stool samples) and preparation and examination of a single instead of duplicate slides. The sensitivity of the KK is further compromised by day-to-day and intra-specimen variation of helminth egg output [13], [25], problems related to delay from time of defecation to collection in the field and processing in the laboratories. Unlike the KK, the SNF has the advantage of indefinite formalin-based fixation at room temperature prior to examination. Rapid over-clearing of hookworm eggs by the KK may also lead to false negatives and/or an under-estimation of hookworm egg intensities [25], [26]. The SNF proved superior to the quadruple KK for the detection of low egg burdens and therefore a better method to monitor the efficacy of anthelmintic treatment programs when worm burdens are expected to be lighter.
The calculation of the “true prevalence” was done for the KK and for the gold standard. It takes into consideration the results of each examination day and estimates the prevalence if unlimited number of samples from the same individual would be examined [21]. This calculation is normally performed only for a specific diagnostic method. Yet, we also performed an estimation calculation for the gold standard, assuming the results of KK day 1, KK day 2, SNF and PCR as four results of the same diagnostic method, performed on four consecutive days.
SNF offers a number of advantages for the detection of hookworm eggs over KK methods. In addition to the superior sensitivity, the SNF did not detect a significant difference in hookworm eggs counts to the KK, a limitation of the FLOTAC technique in which egg counts are consistently reported low by comparison [27], [28]. The SNF technique is simple, quantitative and can be performed using a simple bench-top centrifuge using 10–15 ml disposable centrifuge tubes, surgical gauze, microscope slides and cover slips. Sodium nitrate can be purchased readily from chemical suppliers and if unavailable, the solution can be replaced with saturated salt of equal specific gravity (specific gravity 1.20).
In contrast with SNF, the limitations of the FLOTAC apparatus include its atypical size and the requirements of a large capacity stand-up bucket centrifuge. This requires the procedure be conducted in well-equipped laboratories only [25], usually not present in areas endemic for hookworm infections, including Cambodia. The reusable Fecalyzer device (EVSCO 014008-50, USA) is widely used and available through veterinary suppliers for less than US 1.00 each and may prove an alternative option for conducting SNF in the field or local laboratory. This stand-up fecal flotation device comes with an inbuilt filter and stirrer that in a similar fashion to the mini-FLOTAC, obviates the requirement of a bench-top centrifuge.
In conclusion, our comparison of different techniques suggests that PCR is a highly sensitive technique for the detection of hookworm infection in human parasitological surveys. It offers resource-poor communities a logistically feasible, freely available and cost-effective option to monitor the success of hookworm control programs. The SNF holds promise for the detection of human hookworm and potentially other STH infections and may become an essential tool for patient management, monitoring of helminth control programs and anthelmintic drug efficacy studies in areas with no access to the commercially produced parasitological flotation devices.
|
10.1371/journal.pcbi.0030236 | Gamma Oscillations of Spiking Neural Populations Enhance Signal Discrimination | Selective attention is an important filter for complex environments where distractions compete with signals. Attention increases both the gamma-band power of cortical local field potentials and the spike-field coherence within the receptive field of an attended object. However, the mechanisms by which gamma-band activity enhances, if at all, the encoding of input signals are not well understood. We propose that gamma oscillations induce binomial-like spike-count statistics across noisy neural populations. Using simplified models of spiking neurons, we show how the discrimination of static signals based on the population spike-count response is improved with gamma induced binomial statistics. These results give an important mechanistic link between the neural correlates of attention and the discrimination tasks where attention is known to enhance performance. Further, they show how a rhythmicity of spike responses can enhance coding schemes that are not temporally sensitive.
| Rhythmic brain activity is observed in many neural structures and is an inferred critical component of neural processing. In particular, stimulus induced oscillations in the gamma-frequency band (30–80 Hz) are common in several cortical networks. Many experimental and theoretical studies have established the neural mechanisms by which a population of neurons produce and control gamma-band activity. However, the beneficial role, if any, of gamma activity in neural processing is rarely discussed. It is increasingly apparent that gamma oscillatory power increases with subject attention to a sensory scene. Attention is associated with enhanced performance of discrimination tasks, where relevant stimuli compete with distracters. In this study we explore how gamma-band activity serves to enhance the discrimination of stimuli. We use computational models to show that the gamma rhythmicity in a population of spiking neurons drastically reduces the response variability when a preferred stimulus is present. This drop in response variability enhances stimulus discrimination and increases the overall information throughput in sensory cortex. Our results provide a much-needed link between the dynamics of neural populations and the coding tasks they perform, as well as give insight on why—rather than how—attention mediates gamma activity.
| Past work with both human and animal subjects has focused on neural correlates of attention. Attention raises the firing rate and the input–output gain of orientation-selective neurons in the visual cortex [1–3], and shifts response curves so that physiologically relevant stimuli fall in the high-gain region [4,5]. Also, when attended stimuli overlap with a recorded receptive field, gamma-band frequency components (30–80 Hz) of local field potentials and single-unit spike responses increase [6–10]. Gamma oscillations in the field potential likely reflect correlated network activity [7,9], as supported by simulations of spiking neurons with inhibitory or recurrent excitatory–inhibitory coupling [11]. Attention is thought to influence cholingergic neuromodulation [12], which presumably affects synchrony of interneuron networks involved in gamma oscillations [11,13,14]. It is well-known that correlated network discharge effectively drives postsynaptic cells [15], making gamma-band activity a signature of efficient signal propagation. This would allow attended objects to increase downstream responses, as compared to nonattended objects. In contrast, we assess the role of gamma oscillations in the signal coding of neural populations participating in gamma oscillatory dynamics.
Tasks where attention improves performance typically involve discrimination between different signals, such as visual cues with different colors, shapes, or orientations [1,6–10]. Although there are a large number of studies exploring how gamma rhythms are generated in networks of spiking neurons (for a review, see [11]), the mechanisms by which gamma oscillations modify signal discrimination are elusive in three aspects. First, the relation between gain modulation and gamma oscillations, both of which are attention-dependent, is unclear. Second, the temporal relation between a network gamma rhythm and the time course of a driving signal is often unclear. Third, gamma-induced synchronous firing may be deleterious for coding due to increased variability of population activity [16].
A popular framework for neural coding is that the number of spikes produced by a single neuron or a population of neurons carries information about a driving signal. However, in vivo spike trains often show a spike count Fano factor (ratio of the spike-count variance to the mean spike count) that is close to or even exceeds unity [16–18]. This trial-to-trial variability is deleterious to the code performance and degrades putative spike-count–based signal-discrimination schemes. In certain situations, Fano factors much less than 1 are observed in the visual cortex [19,20], the auditory cortex [21], and the salamander retina [22]. In an extreme case, if a neuron fires with high probability in response to a relevant input signal and rarely fires otherwise [21], then the signal can be estimated from the spike count with small error. In addition, spike-timing reliability, for which a neuron robustly emits just a single spike during a steep upstroke of the input and seldom fires elsewhere [23,24], is also supportive of such binary spiking.
In this study we model the essence of a gamma frequency modulation as a simple rhythmic forcing of a population of uncoupled spiking neurons. We show that gamma oscillations endow population spike counts with binomial-like statistics, which improve signal discrimination over a range of stimuli through reduced spike-count variability. In this way, we propose a connection between gamma oscillations and enhanced task performance found in behavioral experiments. Our results are both distinct and complementary to previously described influences of rhythmic network behavior in temporal coding schemes by improving spike precision [25,26] or by providing a clock for a phase-based code [27–30].
We consider signal discrimination tasks using a population of N = 100 uncoupled leaky integrate-and-fire (LIF) neurons (see Methods). The input to each neuron I(t) is the sum of the input signal s, the gamma modulation, and a fast fluctuating noise term:
For simplicity, we take the fluctuations to be broadband (e.g., white noise) with intensity σ and correlation coefficient c between neuron pairs in the population [31]. We assume that the fluctuations are correlated among neurons to comply with experimental evidence [16] and to make our discrimination task somewhat difficult, thereby allowing gamma activity to shape the results. We remark that we simply force each neuron with a sinusoidal current with amplitude A and frequency fγ = 40 Hz , rather than explicitly model the gamma oscillation as emergent from neural networks (see [11]).
We examine the statistics of the population spike count
, where Mi,T is the number of times neuron i spikes in a window of length T. In an observation window, each neuron can fire an arbitrary number of times with a maximum of T/τr, where τr is the absolute refractory period. If the firing rate approaches this upper limit, presumably by a large I(t), all neurons fire regularly with period close to τr, and M has low variance. However, 1/τr is typically hundreds of Hz, making such a saturation unreasonable for prolonged times. It is well known that the relative refractory periods enable low spike-count variability at moderate firing rates [20,22]. We explore an alternative possibility that gamma oscillations generate regular spiking at firing rates far below 1/τr when the observation window T is sufficiently large. In what follows, for simplicity we take T = 1/fγ.
To illustrate how gamma modulation influences population spike-count statistics, we switch the external signal between two static levels s = s1 and s = s2 (Figure 1). In the absence of gamma modulation (A = 0), the spike raster (Figure 1A, middle) and the spike count (Figure 1A, bottom) show a subtle but noticeable change in the statistics of M as s switches between s1 and s2. However, with finite observation time, the large trial-to-trial variability (error bars in Figure 1A, bottom) makes discrimination between s1 and s2 based on M difficult when s1 and s2 are close to one another. This difficulty is reflected by a large overlap in the spike-count probability density functions (PDFs) conditioned on s = s1 or s = s2 (Figure 1A, bottom). Further, correlated fluctuations (c > 0) bound the population spike-count variability to a nonzero value even for very large populations [16]. In contrast, with moderate A, neurons fire at most one spike per cycle because of the rhythmic nature of I(t) combined with the absolute spike refractory period (Figure 1B, middle). For larger s values, the neurons fire once every cycle with high probability, yielding a population spike count with low variability (small error bars in Figure 1B, bottom, for s = s2). The overlap of the two spike count PDFs in this case is actually smaller than that for A = 0. Consequently, discriminability between s1 and s2 is enhanced by gamma modulation (see Figure 1 caption). The remainder of the paper seeks to quantify this observation. In what follows we let s1 and s2 be constant in time; this simplification is reasonable since the observation window T is quite short compared to typical time scales of natural stimuli.
We first examine the relation between the mean spike count μ = 〈M〉 and the spike-count variance V = 〈M2〉 − 〈M〉2, where 〈·〉 is an average over gamma cycles. Figure 2A shows μ plotted against s for A = 0 (thin line) and A = 0.3 (thick line). First, the gamma modulation induces a leftward shift in the μ–s curve for s < 1. Second, a knee in the curve near μ = N (= 100) emerges when A > 0, indicating one-to-one locking of single neuron firing and the gamma cycle. The additive shift and the response saturation at moderate rates are both consistent with single-unit spike responses during attention-sensitive tasks (Figure 5A of [4]). To study how the knee region influences count variability, we plot V versus μ for A = 0 and A = 0.3 (Figure 2B). When correlated noise is both present (c = 0.12; closed symbols) and absent (c = 0; open symbols), V is smaller with gamma modulation (circles) than without (squares), conditional on s chosen so that all the neurons fire once in a window with high probability, μ ≈ N (i.e., in the knee region of the μ–s curve). When c = 0 and A = 0.3 (open circles), the relation is well-fit by that for the binomial distribution (solid line), reminiscent of binary spiking statistics for each cell in the population. When A = 0 (open squares), V does not approach low values for any μ. Nevertheless, Poisson count statistics (V = μ, dashed line), which are in rough agreement with in vivo evidence [17,18], result in a poor fit for large μ, because a large s transitions the single-cell spiking from a fluctuation driven to oscillatory regime where the large average current drives rhythmic firing (but see Figure 6). These overall trends are preserved when c > 0 (closed symbols) in spite of a larger V. Our results with A > 0 are in agreement with similar numerical studies [14] where gamma oscillations were replicated with realistic barrages of synchronous inhibitory conductances (Figure 4D in [14]).
To explore the link between gamma-induced binomial spiking and signal discrimination, we first study phenomenological models of stochastic population activity. We map the signal s to an internal parameter that characterizes the spike-count distributions. In a Poisson model we set the expected number of spikes for a single neuron to be λ = s. If neurons fire independently, then the population spike count follows a Poisson distribution: P(M = k) = e–Nλ(Nλ)k / k!, which gives μ = V = Nλ. The Poisson model represents a scenario in which reduction in spike-count variability of any kind is absent. In a binomial model, each neuron fires at most once in the window and does so with probability p (0 ≤ p ≤ 1). For each neuron, the s to p relation is a smoothed piecewise map so that for small s the map is near linear and as s → 1 the population response saturates (i.e., p → 1). If all neurons fire independently, P(M = k) = NCkpk (1 − p)N−k, where NCk is a binomial coefficient. This gives μ = Np and V = Np(1 − p).
We mimic the effect of attention in either model with an additional internal modulation sA that modifies the statistics of M. Because attention is thought to modulate spike statistics in several ways, we consider two accepted scenarios. One is an additive scenario in which s is mapped to s + sA. This is similar to attention-mediated leftward shifts of input–output curves [4,5] in the visual pathway. The other is a multiplicative scenario in which s is mapped to s(1 + sA), modeling experiments where attention multiplicatively controls the gain of orientation tuning curves in primary and middle visual areas [2,3]. These two gain manipulation schemes result in similar effects from our spike-count perspective (see below).
To quantify the discriminability of two signals, we consider the conditioned PDFs P(M|s1) and P(M|s2). Intuitively, discrimination is easier when the masses of the two PDFs are more separated. To assess discriminability, we compute the Kullback-Leibler (KL) distance [32,33] between P(M|s1) and P(M|s2) (see Methods). In short, the KL distance, which we denote by KLR (R for resistor average, see Methods), offers a method for measuring the distance between two PDFs. For Gaussian PDFs, the KL distance is equivalent to the so-called d′ discriminability [32], which is often used in psychophysical studies [34]. However, P(M|s1) and P(M|s2) are generally non-Gaussian, as is the case for binomial spike statistics, and the KL measures are more appropriate. We label KLR with a subscript P or B for statistics using the Poisson or binomial model, respectively.
Motivated by the gamma-induced additive shift in the network simulations shown in Figure 2A, we first focus on the additive model. We vary sA with s1 and s2 fixed, assuming without a loss of generality that s1 ≤ s2. For small sA, we have μB ≈ VB, and thus the binomial and Poisson models are statistically similar, yielding KLB,R ≈ KLP,R (Figure 3A). Indeed, for sA fixed at a small value, the conditional PDF for the Poisson model and those for the binomial model are nearly identical, both for s1 and s2 (Figure 3A1). As sA increases, KLB,R rises significantly, whereas KLP,R drops slightly. To understand this, we note that, in the binomial model, when s2 but not s1 saturates the population response (i.e., p2 → 1 and p1 < 1), the variance of PB(M|s2) drops significantly to reduce the overlap between PB(M|s1) and PB(M|s2) (Figure 3A2). Consequently, signal discrimination becomes easier. In the Poisson model, the population spike-count variability increases with sA, yielding an increased overlap between PP(M|s1) and PP(M|s2), which drops KLP,R. However, when sA is even larger, binomial population responses are saturated for both s1 and s2 (p1, p2 → 1), giving PB(M|s1) = PB(M|s2) ≈ δM,N , and hence KLB,R ≈ 0, whereas KLP,R > 0 (Figure 3A3).
In total, as sA is varied, KLB,R is non-monotonic, whereas KLP,R monotonically decreases over the same range of sA. Similar results are obtained for the multiplicative model except that KLP,R increases slightly with sA (Figure 3B). In Methods, we generalize these results by showing KLB,R ≥ KLP,R for any s1 and s2 pair unless both s1 and s2 saturate the binomial model response. Overall, binomial spike-count statistics can enhance signal discrimination as compared to Poisson statistics, particularly when one input signal saturates or nearly saturates the population response while the other signal is below saturation.
We next link gamma induced binomial-like spike-count statistics of a population of LIF neurons with the discrimination results obtained with the phenomenological models. In the spiking neuron population, we fix s1 and s2, as was done in Figure 3, and numerically estimate P(M|s1), P(M|s2), and the KL distance for a fixed A. Interestingly, KLR is nonmonotonic as A ranges from 0 to 0.6 (Figure 4). Specifically, when A = 0, P(M|s1) and P(M|s2) are roughly Gaussian (Figure 4A), and KLR is about 1.2. As A increases, the spike-count statistics become increasingly better described by a binomial random variable (see Figure 2), and P(M|s2) shows a reduced variance. This leads to an overall increase in KLR (Figure 4B). As A increases further, the population response is dominated by the gamma oscillation and is saturated at M ≈ N for both s1 and s2 (Figure 4C), ultimately dropping KLR significantly. This confirms the original hypothesis (Figure 1) that gamma oscillations can enhance signal discrimination of a population of spiking neurons.
A comparison between the non-monotonic trend of KLR shown in Figure 3 and that shown in Figure 4A should be done with care. In the phenomenological binomial model, the spike statistics were modulated by the attention variable sA, yet were, by design, binomial for all sA. In the network simulations, the spike-count statistics become better and better described by a binomial random variable as A increases. Although it is tempting to associate A with sA, A both shifts the population response statistics from Poisson-like to binomial-like, and at the same time modulates the spike-count statistics, similar to the variables p or λ in the phenomenological models. This is a minor point, since for moderate s1 and s2, the binomial statistics for small sA are well-approximated by a Poisson spike count (Figure 3), similar to the case of small A in the network simulations. Thus, the basic mechanism of the non-monotonic trend in Figures 3 and 4 is qualitatively the same.
To show the robustness of the increases in KLR with respect to the choice of signals, we vary s1 and s2 to cover both subthreshold (s1, s2 < 1) and suprathreshold (s1, s2 > 1) regimes. The input signal is confined to 0.85 ≤ s1, s2 ≤ 1.25, which yields moderate firing rates (8 Hz for s = 0.85 and 56 Hz for s = 1.25 without gamma modulation). For each signal pair, we determine the value of A maximizing KLR, which we label
. In Figure 5A, we plot the relative increase in discriminability
, where
is the value of KLR in the absence of the gamma. ΔKLR is large (more than 0.3 as shown in Figure 4) over a wide range, indicating that gamma-enhanced signal discrimination is a general result. The improvement is best manifested when signals are somewhat suprathreshold (1.05 ≤ s1, s2 ≤ 1.2) for which the low spike-count variability is induced by gamma oscillations. The improvement is also restricted to near the s1-s2 diagonal; far off the diagonal, signal discrimination is easy and does not require gamma oscillations.
In Figures 2A and 4, firing rates increased with A when s is not too large. Indeed, attention often increases firing rates [2–5]. However, in some cases attention raises the gamma-band power without increasing firing rates [7,9]. To show that the improved signal discriminability does not merely result from increased firing rates, we added a negative current bias to the neurons in addition to the gamma modulation so that the firing rates remain constant regardless of A. As shown in Figure 5B, ΔKLR can still be significant, although the range of signal pairs where this is apparent is reduced.
Without gamma oscillations, large static inputs place neurons in the suprathreshold regime, where the net bias drives firing. In this regime, firing is rather regular, and spike-count variability can be low (squares in Figures 2B). To examine the possibility of improved signal discrimination by excess static inputs, we set the gamma frequency fγ = 0 and shift the phase of the sinusoid by π/2 so that A corresponds to an additional bias current. To prevent very large firing rates, we assume 0 ≤ A ≤ 0.7. With the largest bias A = 0.7, the neurons fire at 81 Hz for s = 0.85 and 108 Hz for s = 1.25. As shown in Figure 5C, ΔKLR induced by a constant bias is far less impressive than that by gamma modulation (Figure 5A). Much larger firing rates would considerably increase ΔKLR, in which case the absolute refractory period of the neurons imposes periodic firing and reduction in spike-count variability, yet prolonged spiking activity at these high rates are not observed in cortical responses. This contrasts to the case with gamma modulation for which neurons fire at most fγ = 40 Hz. Overall, we conclude that gamma oscillations are an effective means of improving signal discrimination of population responses.
The population of LIF neurons used in Figures 4 and 5 produce small spike-count variances for large firing rates. This relation between the spike-count variance and the spike-count average in the absence of gamma oscillations (closed squares in Figure 2B) deviates from the Poisson relation (dashed line in Figure 2B) observed in many experiments [17,18]. To show the generality of our results, we mimicked more Poisson-like population spike-count statistics by making the input noise temporally colored, scaling the input noise intensity as the square root of the input signal, and increasing the input correlation linearly in the input signal. The first modification assumes a synaptic filter, while the last two model a presynaptic population's tendency to have both the spike-count variance and correlation grow with the mean spike count, as suggested by [17] and [31], respectively. With these modifications, the population spike-count variance in the absence of gamma oscillations is roughly equal to the spike-count average for a wide range of the firing rate (squares in Figure 6A). Also in this situation, the spike-count variance sharply drops near M = N with gamma modulation (circles in Figure 6A). Accordingly, and similar to our earlier model (Figure 4), signal discrimination improves for intermediate gamma amplitudes, as shown in Figure 6B. These final results show that gamma-enhanced signal discrimination is robust to significant changes in population response statistics.
We have shown that gamma modulation of a population of noisy spiking neurons imparts binomial-like spike-count statistics. When neurons are driven to fire at rates near gamma frequency, they phase lock with the gamma oscillation. This produces a saturation of the firing rate, reduction of spike-count variability, and importantly enhanced signal discriminability. Simple phenomenological statistical models (Figure 3) show this to be a straightforward consequence of binomial count statistics. The overall effect is robust in simulations of a population of spiking neurons (Figures 4–6).
Although we used a simple sine wave forcing as a caricature of gamma activity, experimentally measured gamma oscillations are not harmonic, and are typically broadband (30–60 Hz). Indeed, the spectral properties of the spike-train responses from our model have artificially large spike-train power at 40Hz, and a spike–spike coherence [8] value of approximately 0.5 at 40 Hz, much larger than is typically seen in vivo [8,9]. If we instead used a gamma forcing defined over a range of frequencies, then the large population rhythmicity and coherence at 40 Hz would be spread over a wider spectrum, and no single frequency would be overly dominant. We expect that such a broadband gamma modulation would not deteriorate signal discrimination because it can still elicit approximately one spike per gamma cycle, provided that the gamma band is not too broad and other sources of noise are weak, as shown in the more realistic gamma network model presented in [14]. We stress that our spiking network is only a qualitative description of gamma oscillatory neural dynamics, and not a quantitative description of cortical or hippocampal networks. The robustness of our results to changes in input s (Figure 5), changes in input statistics (Figure 6), as well as our simplified phenomenological description (Figure 3), suggests that our result may be operable in many different networks with varying response statistics.
For our theory to be operative, gamma-band activity must be exclusive to a specific subpopulation of neurons involved in a discrimination task. Our theory does not explain how such a selective gamma activity is produced. However, in support of selective modulation of gamma activity, a recent study in area LIP in the parietal cortex gives attention-related feedback projections in the gamma range to MT, which in turn feeds back to V1 [35]. A topographic overlap of feedback architecture and feedforward receptive field would therefore permit a feedback gated selective gamma response.
We dealt with population spike counts whose time resolution was quite low (T = 1/fγ = 25 ms) compared to millisecond precision on which many spike-based temporal coding schemes are based. On shorter time scales (1–5 ms), oscillatory input, for example, enhances spike-time precision by cellular resonances [24] and resets the membrane potential for improved signal discriminability [26]. Oscillatory inputs also set a rhythm for defining spike phases, which are potentially useful for coding [27–29]. These results typically assume that the downstream decoding cells are sensitive to the precise timing of input spikes. Our results are quite complementary because oscillatory activity of the same presynaptic neural populations enhances coding where decoding neurons integrate incoming spikes on much longer timescales (20–30 ms). With different kinetics of downstream neurons and synapses, both coding schemes may act in parallel.
Attention can raise firing rates [2,3], contrast gain [4,5], and gamma-band activity in both spike trains and field potentials [6,7]. In our spiking network, regardless of whether gamma activity increases firing rates or not, signal discrimination is facilitated by gamma modulation that we interpret to be generated by attention. Also in our phenomenological models, when attention is either additive or multiplicative modulation of response properties, a shift from Poisson to binomial spike statistics improves signal discrimination. This is consistent with the recent observations that attention decreases spike-count variability [36], as well as enhances the signal-to-noise ratio [37]. Thus we provide an important link between the dynamical effects of gamma oscillations and coding performance of neural populations that are attention-sensitive.
The dynamics of the i-th neuron in the population (1 ≤ i ≤ N) is described by
where vi is the membrane potential of the i-th neuron in the population, and τm = 10 ms is the membrane time constant. The correlation coefficient between the total background inputs given to two cells is denoted by c [33]. We set c = 0.12 unless otherwise stated, so that the neurons have a background correlation similar to in vivo recordings in the absence of gamma modulation [16]. The neuron fires when vi = 1 is reached from below, and then vi is instantaneously reset to the resting potential equal to 0. The absolute refractory period τr is set 2 ms. For Figures 1–5, we let the fluctuation terms ξi and ξ be uncorrelated white noise inputs with zero mean
and
). The total intensity of these inputs is σ = 0.35. In Figure 6, we replace the white noise terms ξi and ξ with an Ornstein-Uhlenbeck process (low-pass filtered white noise) with a decay time constant of 5 ms. Then we regard that the minimum input signal s is equal to 0.85 and scale the input noise intensity and the input correlation as
and
, respectively. We employ a Euler-Maruyama [38] numerical integration scheme (dt = 0.02 ms) to solve the population dynamics.
Given two conditional spike-count densities P(M|s1) and P(M|s2), we compute the Kullback-Leibler divergence [32,33] as
where (i,j) = (1,2), (2,1). Here k ranges over possible spike counts, and ΔM = 1 because the spike count is integer-valued. The KL divergence is generally asymmetric, i.e., KL12 ≠ KL21. To correct for this, we use the KL distance, or so-called resistor average [32], defined by
The KL distance approximates the optimal discrimination error better than the simple average (KL12 + KL21)/2 does [32]. A direct computation of KLij diverges if P(M = k|si) > 0 and P(M = k|sj) = 0 for some k due to numerical sampling. To accurately estimate the conditional PDFs and the KL distance, we employ the K–T estimate method [32], where 0.5 is added to all the bins in the count histogram before normalization to obtain the PDFs.
We prove that the KL divergence and the KL distance for the binomial distribution are larger than those for the Poisson distribution when at least one of the stimuli s1 and s2 does not saturate the binomial model response.
For the Poisson distributions with parameters Nλ1 and Nλ2, we obtain
For the binomial distributions with parameters p1 and p2 (0 ≤ p1, p2 ≤ 1) for a single neuron, we obtain
Although we smoothed the s-p relationship of the binomial model to produce Figure 3, the smoothing function had a very small variance. Therefore, we neglect smoothing so that p = s for 0 ≤ s ≤ 1 and p = 1 for s > 1. We equate λ1 = p1 and λ2 = p2 so that the Poisson and binomial distributions produce the same average firing rates. Using Jensen's inequality, we derive
where the equality holds if p1 = p2, or equivalently, λ1 = λ2. Finally, we obtain
These relations hold when 0 < p1, p2 < 1. If p1 or p2, but not both, is equal to 0 or 1, KLB,12 or KLB,21 goes to infinity. Even in this case, KLB,ij ≥ KLP,ij and KLB,R ≥ KLP,R hold. If p1 = p2 = 1, the two binomial distributions become delta functions so that KLB,ij = KLB,R = 0.
|
10.1371/journal.ppat.1003348 | Mutualistic Polydnaviruses Share Essential Replication Gene Functions with Pathogenic Ancestors | Viruses are usually thought to form parasitic associations with hosts, but all members of the family Polydnaviridae are obligate mutualists of insects called parasitoid wasps. Phylogenetic data founded on sequence comparisons of viral genes indicate that polydnaviruses in the genus Bracovirus (BV) are closely related to pathogenic nudiviruses and baculoviruses. However, pronounced differences in the biology of BVs and baculoviruses together with high divergence of many shared genes make it unclear whether BV homologs still retain baculovirus-like functions. Here we report that virions from Microplitis demolitor bracovirus (MdBV) contain multiple baculovirus-like and nudivirus-like conserved gene products. We further show that RNA interference effectively and specifically knocks down MdBV gene expression. Coupling RNAi knockdown methods with functional assays, we examined the activity of six genes in the MdBV conserved gene set that are known to have essential roles in transcription (lef-4, lef-9), capsid assembly (vp39, vlf-1), and envelope formation (p74, pif-1) during baculovirus replication. Our results indicated that MdBV produces a baculovirus-like RNA polymerase that transcribes virus structural genes. Our results also supported a conserved role for vp39, vlf-1, p74, and pif-1 as structural components of MdBV virions. Additional experiments suggested that vlf-1 together with the nudivirus-like gene int-1 also have novel functions in regulating excision of MdBV proviral DNAs for packaging into virions. Overall, these data provide the first experimental insights into the function of BV genes in virion formation.
| Microorganisms form symbiotic associations with animals and plants that range from parasitic (pathogens) to beneficial (mutualists). Although numerous examples of obligate, mutualistic bacteria, fungi, and protozoans exist, viruses are almost always considered to be pathogens. An exception is the family Polydnaviridae, which consists of large DNA viruses that are obligate mutualists of insects called parasitoid wasps. Prior studies show that polydnaviruses in the genus Bracovirus evolved approximately 100 million years ago from a group of viruses called nudiviruses, which are themselves closely related to a large family of insect pathogens called baculoviruses. Polydnaviruses are thus of fundamental interest for understanding the processes by which viruses can evolve into mutualists. In this study we characterized the composition of virus particles from Microplitis demolitor bracovirus (MdBV) and conducted functional experiments to assess whether BV genes share similar functions with related essential baculovirus replication genes. Our results indicate that several genes in MdBV retain ancestral functions, but select other genes have novel functions unknown from baculoviruses. Our results also provide the first experimental data on the function of polydnavirus replication genes and enhance understanding of the similarities between these viruses and their pathogenic ancestors.
| Microorganisms form associations with metazoan hosts that range from beneficial symbiosis (mutualists) to parasitic (pathogens). Mutualists serve as important sources of evolutionary innovation for hosts, while pathogens often acquire genes from hosts or other organisms that facilitate their own survival and cause disease. Although most research on obligate mutualists focuses on bacteria, several fungi and protozoans are also known to form beneficial partnerships [1]–[3]. Viruses in contrast are almost always thought to form parasitic associations [4]–[6]. A notable exception to this is the family Polydnaviridae, which consists entirely of large DNA viruses that are obligate mutualists of insects called parasitoid wasps [7], [8]. Polydnaviruses (PDVs) thus provide an opportunity for understanding the adaptations involved in the evolution of viruses into mutualists from pathogenic ancestors.
Parasitoid wasps reproduce by laying eggs into other insects (hosts) that their progeny consume [9]. The Polydnaviridae consists of two genera: the Bracovirus (BV) associated with ca. 20,000 species of wasps in the family Braconidae, and the Ichnovirus (IV) associated with ca. 18,000 species of wasps in the family Ichneumonidae [10]. Each wasp species carries a genetically unique PDV that persists in all cells as an integrated provirus. Viral replication only occurs in pupal and adult stage female wasps in a type of cell in the ovary called calyx cells. Virions from calyx cells are released via cell lysis and accumulate to high density in the lumen of the reproductive tract to form calyx fluid. Virions are also enveloped and contain multiple, circular, double-stranded DNAs of large aggregate size (190–600 kbp) that encode many virulence genes. Most PDV-carrying wasps parasitize larval stage Lepidoptera (moths) by depositing eggs containing the proviral genome plus a quantity of virions. These virions rapidly infect host cells, which is followed by expression of virulence genes that immunosuppress and alter the development of hosts in ways that are essential for survival of the wasp's progeny [11].
The origin and genomic organization of IVs is unclear [11]. In contrast, BV genomes exhibit features unlike any other known viruses [12]–[14]. As proviruses, their genomes consist of two types of DNA domains: those that contain genes with predicted roles in replication, and others that contain the virulence genes that become packaged into virions. Remarkably, these domains reside in different locations in the wasp genome [12], [13]. In addition, while genes with predicted roles in replication are transcribed in calyx cells, their transmission is entirely vertical and independent of any viral DNA replication or encapsidation [7], [11], [12]. Virulence gene-containing domains are likewise transmitted vertically. However, they also are amplified, excised from the wasp genome into circular forms and packaged into virions during replication in calyx cells [7], [15]–[20]. In all other cells of the wasp including the germ line both the replication and virulence genes of the proviral genome are silent [11]. BVs cause no apparent disease in wasps because almost no virulence genes are expressed in wasp cells and lytic replication is restricted to only calyx cells [11], [21]. In contrast, BVs cause severe disease in the hosts wasps parasitize because virions systemically infect the host insect and all of the virulence genes virions deliver are expressed [11], [21], [22]. The disease symptoms BVs cause in the host, however, are also essential for development of the wasp's offspring. Thus, BVs depend on wasps for genetic transmission, while wasps depend on BVs for parasitism of hosts.
Genes in the proviral genome of BVs with predicted roles in replication were identified because they exhibit homology with core genes from two other types of arthropod-infecting viruses: nudiviruses and their sister taxon the Baculoviridae [7], [12]. Like BVs, nudiviruses and baculoviruses replicate in cell nuclei and package large, circular ds-DNA genomes into enveloped virions. Unlike the developmentally-linked and tissue-specific replication of BVs, however, baculoviruses are virulent pathogens, which establish systemic infections in insects by undergoing lytic replication in virtually all cells of the infected host and expressing a variety of virulence genes [23]. Nudiviruses produce either systemic, lytic infections or latent infections [23], [24]. More than 60 baculoviruses have been sequenced and a survey of a subset (13) of these genomes indicates that all share 31 genes, which are collectively referred to as the baculovirus core gene set [24], [25] (Figure 1). Functional studies of model species like Autographa californica multinucleopolyhedrosis virus (AcMNPV) indicate about half of these genes have essential roles in viral replication [24]. These include genes with roles in replicating the viral genome, several genes that code for virion structural components, and four genes that code for subunits of a novel RNA polymerase, which selectively transcribes viral genes because it recognizes unique promoter sequences (Figure 1) [24]. Six nudivirus genomes have been sequenced and each contains 20 genes with homology to structural, replication and transcription components of the baculovirus core gene set [23] (Figure 1). The actual function of these genes, however, is unknown beyond inferences from baculoviruses. Data from three braconid wasps, Cotesia congregata, Chelonus inanitus, and Microplitis demolitor, indicate they lack recognizable homologs of most baculovirus core genes with roles in viral DNA replication (Figure 1), which suggests that, unlike baculoviruses, replication of BV DNAs packaged into virions is regulated by machinery from the wasp [7]. However, BVs do encode homologs of several baculovirus/nudivirus-like structural genes plus the four subunits of a baculovirus/nudivirus-like RNA polymerase [7], [12]. Each of these genes is also transcribed in ovaries when BV virions are produced. Together, these genes form a conserved gene set likely present in all BV genomes (Figure 1). However, we refrain from referring to these as “core” genes because of the small number of BV genomes currently available and their non-discrete organization in wasp genomes [7]. Other predicted members of a conserved BV gene set include a baculovirus/nudivirus-like sulfhydryl oxidase (ac92), 11 nudivirus-like genes unknown from baculoviruses, and 11 novel genes [7].
Since BV-carrying braconids are monophyletic [26], these data overall indicate that BVs evolved from an ancestral nudivirus-wasp association. Fossil calibrations estimate this association arose 100 million years ago (Mya), while the last common ancestor of BVs, nudiviruses, and baculoviruses existed approximately 312 Mya [27]. Given these timelines and the pronounced differences that exist today between BVs and baculoviruses, it is not surprising many of the genes they share have diverged to the point that homology is difficult to recognize outside of essential residues or functional domains. Indeed, algorithms like BLAST cannot detect homology between BV and baculovirus genes, while identity between predicted BV and more closely related nudivirus proteins ranges between 19–41% [7]. Such divergence, however, also begs the question of whether baculovirus-like genes in BV proviral genomes retain baculovirus-like functions. Here, we used proteomic, RNA interference (RNAi), and functional assays to characterize selected members of the conserved gene set of Microplitis demolitor bracovirus (MdBV) in the wasp M. demolitor. Our results indicate that six genes with hypothesized roles in replication exhibited conserved functions relative to baculoviruses. Our data also identified novel functions for two genes in excision of viral DNAs for packaging into virions.
BV replication in calyx cells begins with amplification of a portion of the proviral genome, which is followed by the de novo assembly and packaging of virions in nuclei [20], [28]. Calyx cells then lyse which releases virions into the lumen. In the case of MdBV, prior studies establish the timing of these events and the chronology of replication gene expression during the pupal and adult stages of M. demolitor [7]. MdBV packages multiple circular, double-stranded DNA segments into virions but each individual virion contains only a single viral DNA [14], [29]. The sequence of these DNAs as episomes and their wasp-viral boundary sequences when integrated into the genome of M. demolitor are known [14], [29], [30]. Prior studies document that these DNAs are specifically amplified in M. demolitor calyx cells, followed by excision from flanking DNA and circularization [7]. Flanking wasp DNA at the site of excision is then rejoined to form an ‘empty locus’ [7].
Given this background, we first conducted a proteomic analysis of MdBV virions to determine whether predicted conserved structural components were present. To accomplish this, we produced two independent samples of calyx fluid with the second sample further purified on a sucrose gradient that produces morphologically pure and intact virions [31]. Following separation on SDS-PAGE gels, proteins were in-gel trypsin digested and analyzed using an Orbitrap Elite mass spectrometer. Mass spectra were then compared to our previously generated M. demolitor ovary transcriptome database [7] to identify proteins present. We present our findings relative to the predicted core/conserved gene sets for baculoviruses, nudiviruses, and BVs in Figure 1 and Table S1. Four proteins (38K, VP39, VLF-1, AC92) detected in MdBV virions were homologs of baculovirus capsid or capsid/envelope components. VP39 was the most accurately detected of these proteins based on the total number of unique peptides identified, which corresponded with vp39 also being the most abundant viral gene transcript detected in ovaries during MdBV replication [7] (Table S1). We also detected several proteins related to envelope components of baculovirus occlusion-derived virus. These included variants of ODV-E66 and ODV-E56 ( = PIF-5) plus the infectivity factors P74, PIF-1 through -6 and envelope component VP91 (Figure 1, Table S1). Seven MdBV virion proteins corresponded to genes in the BV conserved gene set for which homologs are known from all or some nudiviruses but are unknown from baculoviruses (Figure 1). These included the product of the integrase-1 (int-1) gene, which is structurally related to vlf-1, plus products of several nudivirus-like genes of unknown function (HzNVORF9-1 and -2, 64, 94, 106, PMV Hypothetical Protein). We also detected products of four conserved genes or gene families unique to BVs (17A, 35A, 97A) (Figure 1, Table S1). In contrast, we did not detect any proteins in virions that corresponded to the helicase gene or the RNA polymerase subunits (lef-4, lef-8, lef-9, p47) (Figure 1).
The preceding data showed that MdBV virions contain BV conserved gene products but provided no experimental evidence for their function. We therefore selected six genes from MdBV for functional studies. Our choices included two predicted subunits of a baculovirus-like RNA polymerase (lef-4, lef-9), 2), two homologs of baculovirus capsid genes (vp39, and vlf-1), and 3) two homologs of baculovirus envelope genes (p74 and pif-1). Each of these genes is a member of the BV conserved gene set because orthologs are likely present in all other BVs studied to date (see Figure 1). Each is also a member of the baculovirus core gene set with prior studies from AcMNPV or other isolates providing experimental evidence for the function of each [24]. In addition, we selected one nudivirus-like gene (int-1), unknown from baculoviruses, for which no functional studies have been conducted. As previously noted, sequence divergence between members of the BV conserved gene set and corresponding baculovirus/nudivirus core genes is high. Identities of the above gene products from MdBV with corresponding predicted proteins from the closest known nudivirus relative, Heliothis zea nudivirus-1 (HzNV-1), were: lef-4 (25%), lef-9 (32%), vp39 (19%), vlf-1 (28%), p74 (26%), pif-1 (28%), and int-1 (30%).
The genes we selected reside in the MdBV proviral genome and each is transcribed in ovary calyx cells during replication [7]. However, conventional knock out techniques used to characterize baculovirus gene function are untenable because the DNA domains where these genes reside are not replicated and packaged into MdBV virions. We thus assessed whether RNAi could be used to knock down transcription of these genes in M. demolitor. Since MdBV replication begins in the pupal stage of the wasp, we developed methods for injecting gene-specific dsRNAs into wasp larvae after they emerged from a host caterpillar and spun a cocoon. We then compared the abundance of each targeted transcript in newly emerged adult wasps by qPCR relative to treatment with a non-specific dsRNA (ds-eGFP). Our results showed that we reduced transcript abundance on average 60–99% for each gene we targeted (Figure 2). Using an antibody we generated to MdBV LEF-9, we also confirmed that knockdown at the transcript level resulted in knockdown of the corresponding protein (Figure 2).
Before initiating any functional experiments, we further verified our approach by examining the effects of dsRNA dose, time required after treatment for target transcript degradation, and specificity. Using vlf-1 as an example, our results showed that injection of 50 ng to 5 µg of dsRNA per wasp larva yielded a similar level of knockdown (Figure S1A). Our results also indicated that injection of vlf-1 dsRNA into wasp larvae did not significantly reduce transcript abundance until day 3 of the pupal stage, which suggested that 2–3 days were required before an RNAi effect was observed (Figure S1B). We examined specificity of knockdown in two ways. Since vlf-1 and int-1 are homologous genes, we verified that int-1 dsRNA, which strongly knocked down int-1 (Fig. 2G), did not affect transcript abundance of vlf-1 via off-target effects [32] (Figure S2A). We also generated a second vlf-1 dsRNA that did not overlap the dsRNA used for the data presented in Figure 2D to verify that it had a similar knockdown effect on vlf-1 transcript abundance (Figure S2B).
Baculovirus RNA polymerases consist of four subunits (LEF-4, LEF-8, LEF-9, P47), which transcribe baculovirus genes with roles in virion formation [33]. These subunits are categorized as ‘early’ genes because they are transcribed before DNA replication and transcription of structural ‘late’ and ‘very late’ genes commences. As noted above, baculovirus RNA polymerases selectively transcribe late and very late viral genes because they recognize unique promoter elements with the consensus sequence (A/G/T)TAAG absent from host insect genes [34], [35]. The conserved gene set of BVs contains homologs of each RNA polymerase subunit. Expression data from M. demolitor and Cotesia congregata also indicate these subunits are transcribed in ovaries earlier than predicted structural genes, while sequence analysis has identified baculovirus late gene promoter elements upstream of the start codon of several predicted BV structural genes [7], [12]. Thus, if the BV RNA polymerase subunits form a functionally similar enzyme as baculoviruses, RNAi knockdown of one or more subunits should compromise transcription of MdBV structural genes but not wasp genes.
As shown above (Figure 2A, B), we knocked down lef-4, a predicted 5′ capping enzyme [36], [37] and lef-9, a predicted RNA polymerase subunit that forms part of the catalytic cleft [38], [39]. We then measured transcript abundance of two predicted MdBV structural genes (vp39 and p74), and two typical insect genes expressed in ovaries (elongation factor 1 alpha (ef1α) and DNA polymerase delta subunit (dnapolδ)). Our results showed that knockdown of lef-4 and lef-9 significantly reduced transcript abundance of vp39 and p74 while having no effect on ef1α and dnapolδ (Figure 3A–D). Given this outcome and the putative role of vp39 and p74 as structural genes we reasoned that reduced expression of vp39 and p74 could also result in production of fewer virions on average than control females. We therefore estimated viral titer by using episomal MdBV DNA segment B as a marker and a previously developed qPCR assay that includes a DNAse step to remove all non-encapsidated DNA before isolating DNA from ovary homogenates [7]. In this manner, the copy number of episomal segment B in virions, which protect the packaged DNA, could be determined. These results showed that knockdown of lef-4 and lef-9 significantly reduced the titer of DNAse-protected segment B relative to ds-eGFP-treated controls (Figure 3E).
In baculoviruses, VP39 is a major capsid protein while VLF-1 is a structural component, and is also functionally required for capsid production and very late gene expression [24], [40]–[42]. After knocking down vp39 and vlf-1 in M. demolitor (Figure 2C, D), we first assessed whether either treatment affected virion structural integrity by measuring the DNase sensitivity of packaged genomic DNAs as described above. These assays indicated that the abundance of DNase-protected segment B declined by 83% and 78% in vp39 and vlf-1 knockdown samples respectively relative to the control (Figure 4A, B). We also used the non-overlapping dsRNA, ds-vlf-1-2 in these assays, which produced the same result as ds-vlf-1 (Figure S2C).
We then examined the effects of vp39 and vlf-1 knockdown on the ability of MdBV to infect cells from the moth Chrysodeixis includens, which is a host for M. demolitor. For these assays, we used CiE1 cells, which is a continuous, hemocyte-like cell line established from C. includens that is highly permissive to MdBV infection [30], [43]. We determined by qPCR that 2–4 copies of episomal DNA segment B were present per CiE1 cell when cultures were infected at an estimated MOI of 100 with MdBV from control wasps treated with ds-eGFP (Figure 4C, D). In contrast, copy number was 85.2% and 69.6% less when cells were infected with the same amount of calyx fluid from vp39 and vlf-1 knockdown wasps. We also assessed infection using the MdBV gene product GLC1.8, which is an excellent marker because it is rapidly expressed on the surface of CiE1 cells and is easily visualized immunocytochemically [43], [44]. Normalizing the control samples, we observed that vp39 knockdown reduced the fraction of cells stained for GLC1.8 to less than 10%, while vlf-1 knockdown reduced this fraction to 26.8% (Figure 4E,F).
These results could be explained by vp39 and vlf-1 knockdown either adversely affecting virion formation, which would result in calyx fluid containing a lower titer of virus, or causing structural defects that do not reduce virion density but nonetheless compromise function. We therefore examined virion morphology in calyx fluid by transmission electron microscopy (TEM). We previously documented that MdBV virions in calyx fluid consist of a single barrel-shaped nucleocapsid surrounded by a highly elongate envelope [29], [31]. By counting the number of virions in randomly selected fields of view from treatment and control wasp sections, we determined that calyx fluid from a vlf-1 knockdown wasp contained a slightly lower concentration of virions than observed in a control wasp, whereas a vp39 knockdown wasp did not (Figure 5A–D).
We then examined MdBV morphogenesis in calyx cell nuclei. Early studies of BVs show that calyx cells exhibit a progression of development with smaller, younger cells being situated closer to the ovarioles and older, large cells being closer to the lumen of the ovary [45]. In turn, young cells show no evidence of BV replication, while old calyx cells contain an abundance of assembled virions in their nuclei [45], [46]. In control wasps, we observed that MdBV morphogenesis began with the de novo appearance of short membrane profiles in calyx cell nuclei. This was followed by the formation of nucleocapsids near virogenic stroma, which is where DNA packaging also occurs in baculoviruses. Assembled MdBV virions then formed large aggregations with a layered crystalline structure (Figure 5E). At this stage, virions were rod-shaped and of uniform length. The envelope surrounding each nucleocapsid was also not as elongated as seen for virions in calyx fluid (see Figure 5A versus 5E). Calyx cells from vlf-1 knockdown wasps in contrast exhibited an abundance of membrane profiles that appeared to be envelope progenitors (Figure 5F). Rather than elongating, these envelopes were spherical and either lacked capsids entirely or contained empty capsids (Figure 5F). A number of electron dense and empty capsids were also observed with no envelope (Figure 5F). Lastly, almost no aggregations of rod-shaped, electron dense virions were present in calyx cells from vlf-1 knockdown wasps. Calyx cells from vp39 knockdown wasps showed no distinct alterations in virion assembly, but aggregations of rod shaped, electron dense virions were consistently much smaller than those observed in control wasps (Figure 5G).
p74 and the pif genes are known as per os infectivity factors because their loss in baculoviruses such as AcMNPV disables oral infection of host insects by occlusion derived virus [47]–[49]. Each is also a component of the occlusion derived virus envelope where they form a complex with one another [50]. Unlike baculoviruses, BV virions never infect host insects orally but instead are injected into the hemocoel by wasps where they bind to host cells such as hemocytes via fusion of the envelope with the plasma membrane [51]. Nucleocapsids then travel through the cytoplasm to nuclear pores where they release their DNA into the nucleus to initiate transcription of virulence genes like glc1.8 [51], [52]. Given the differences in the known functions of per os infectivity factors in baculoviruses relative to the biology of BVs we asked whether the p74 and pif-1-like genes of MdBV still play a role in infectivity by knocking down each (Figure 2E, F) and then conducting the same assays in CiE1 cells as described above. Our results revealed no differences between knockdown and control wasps in the copy number of DNA segment B in CiE1 cells at 24 h post-infection (Figure 6A). However, the fraction of CiE1 cells expressing GLC1.8 on their surface was dramatically lower using virus from p74 and pif-1 knockdown wasps (Figure 6B).
As noted above, all baculoviruses encode a vlf-1 gene while nudiviruses and BVs also encode related integrase genes (known as vlf-1 or vlf-1a, vlf-1b-1 and -2 or HzNVORF140, and int-1 and -2 or HzNVORF144) [7], [12]. Although VLF-1 is a structural component of baculovirus virions, it along with nudivirus integrase genes are members of the tyrosine (Tyr) recombinase family, which includes several enzymes that mediate the excision and integration of genetic elements [53]. As noted above, elimination of vlf-1 from AcMNPV disables capsid formation while mutation of the conserved Tyr residue required for integrase activity in other Tyr recombinase family members produces non-infectious virus [40]. Whether baculovirus VLF-1 possesses any integrase activity, however, remains unstudied in all likelihood because baculovirus genomes persist as episomes in infected host cells and are unknown to integrate. In contrast, a key feature of BVs is their persistence in wasps as integrated proviruses that amplify, excise and package a portion of the genome when replicating in calyx cells. We therefore assessed whether vlf-1 and/or int-1 homologs from MdBV regulate proviral DNA excision.
We had previously determined that the MdBV proviral genome encodes three distinct vlf-1 genes (named vlf-1, vlf1b-1, -2) and two integrase genes (int-1, -2) that are all transcribed in ovaries during replication [7]. Phylogenetic analysis further suggested the integrase genes of BVs likely arose from duplication of vlf-1 in the nudivirus ancestor, which was then followed by duplication of each gene in M. demolitor [7]. Alignment of vlf-1 and integrase family members from MdBV, select nudiviruses, AcMNPV, and Chelonus inanitus bracovirus (CiBV) showed that MdBV vlf-1 and int-1 both retain a typical active site Tyr residue for predicted integrase activity, whereas other M. demolitor family members do not (Figure 7A). We thus knocked down vlf-1 and int-1 (Figure 2D, G), and then isolated DNA from newly emerged adult wasp ovaries to determine whether proviral DNAs had excised from the wasp genome as normally occurs. This was accomplished using MdBV segment B as a marker and qPCR assays that measured copy number of the rejoined ‘empty locus’ that only forms if proviral DNA segment B was excised from the wasp genome (Figure 7B). Copy number of the empty locus in ovaries from control wasps treated with ds-eGFP was 14.4×106, which indicated a high level of excision of DNA segment B from calyx cells. In contrast, we detected almost no copies of the empty locus in wasps treated with ds-int-1 and ds-vlf-1 (Figure 7C).
Phylogenetic data strongly support that BVs evolved from a nudivirus ca. 100 Mya, and that nudiviruses and baculoviruses shared a more ancient common viral ancestor ca. 200 Mya earlier [12], [26], [27]. Detailed studies of AcMNPV and select other species also provide important insights into the function of baculovirus core genes. In contrast, the hypothesized function of baculovirus core gene homologs in BVs (and nudiviruses) is founded on inferences from the baculovirus literature and/or expression patterns in wasp ovaries during replication. Thus, the primary goal of this investigation was to assess whether RNAi methods could be used to disrupt BV gene expression, and then to use these methods with a subset of genes to determine whether their roles in replication were consistent with or differed from baculoviruses.
Prefacing these functional studies, we conducted a proteomic analysis of purified MdBV virions to assess whether BV conserved genes that are homologs of baculovirus structural components were present. We also did this to compare MdBV virions to virions from CcBV and CiBV, which are the only other BVs for which any proteomic data are available [12], [54]. We detected all of the baculovirus-like capsid and envelope components previously identified in CcBV and CiBV as well as two additional baculovirus-like conserved genes not detected in CcBV or CiBV. These included AC92, which is associated with baculovirus nucleocapsids, and PIF-3, which is associated with baculovirus occlusion-derived virus envelopes [55]. We also detected four nudivirus-like (HzNVORF9-1 and -2, HzNVORF106, PmV) and three novel conserved gene products (17A, 35A, 97A) in MdBV virions identified in CiBV virions. In contrast, we did not detect three novel gene products (27B, 30B, 97B) reported from CiBV, yet did detect three nudivirus-like gene products (INT-1, HZNVORF64, HZNVORF94) not reported from CiBV virions.
Proteomic data must be interpreted cautiously when investigating viral structural proteins, because of the potential for non-integral proteins to become non-specifically associated with virions during assembly [24]. Nonetheless, the baculovirus literature combined with our detection of the capsid and envelope proteins in Figure 1 suggest these baculovirus-like gene products are likely structural components of MdBV virions. While no functional data from nudiviruses exist, we speculate for the same reason that products of the nudivirus-like conserved genes HzNVorf 9-1, 9-2, -64, -94, -106, PmV hypothetical protein and novel conserved genes 17a, 35a, and 97a, are also structural proteins. Our proteomic data did not identify any peptides corresponding to MdBV conserved genes with predicted roles in DNA replication (helicase) or transcription (lef-4, lef-8, lef-9, p47), which at minimum indicates our samples were not contaminated with some non-integral products transcribed in calyx cells [7]. However, it is notable that we consistently detect the products of the int-1, vlf-1b-1 and vlf-1b-2 genes, which could suggest that similar to baculovirus vlf-1 they too are capsid components or are packaged into capsids together with episomal DNA.
Because of the unique biology of BVs and limited genetic data available for their associated wasps, the options available for studying gene function are obviously constrained. RNAi is a potentially powerful method for studying BV gene function, but its efficacy in insects is also patchy with examples of successful use being more prevalent in some taxa (beetles (Coleoptera), mosquitoes (Diptera)) [56], [57] than others (moths (Lepidoptera) [58]). We thus were very careful in validating our RNAi approach for knocking down MdBV genes in M. demolitor before initiating any functional studies. Our analysis of the ovary transcriptome indicated that all genes of the siRNAi pathway are present in M. demolitor and transcribed [7]. Results presented in this study further show that ds-RNA injection into late larval stage M. demolitor effectively and specifically knocks down the genes we targeted. While we present the outcome of a number of validation experiments using vlf-1 as an example, we have conducted similar experiments with other MdBV conserved genes, which all showed the same trends.
With knockdown methods established, we used the baculovirus literature and our proteomic data to select six genes in the MdBV conserved gene set with hypothesized roles in viral transcription (lef-4, lef-9), capsid assembly (vp39, vlf-1), and envelope formation (p74, pif-1). Our rationale for selecting these genes was also driven by the strength of the functional literature for each in baculoviruses, which provided in most cases clear expectations for what a conserved phenotype should be for MdBV. Our results with lef-4 and lef-9 strongly support that MdBV produces a baculovirus-like RNA polymerase that preferentially transcribes structural genes. We also note that knockdown of lef-4 more strongly disabled structural gene expression than knockdown of lef-9. This could reflect that as a capping enzyme the effect of knocking down of lef-4 was further enhanced by degradation of cap-lacking transcripts. The transcription of reporter virus structural genes vp39 and p74 was not completely abolished for either lef-4 or lef-9 knockdowns. Despite detecting no LEF-9 protein after knockdown on immunoblots, this could reflect incomplete knockdown, and the presence of enough RNA polymerase subunit proteins to produce some functional viral RNA polymerase holoenzyme. Alternatively, viral structural genes may also be transcribed in part by wasp RNA polymerase II. Activity of replication gene transcription compared to relative silence of BV genes that are ultimately packaged into virions suggests that replication genes are transcribed from ancestral viral RNA polymerase promoters whereas virulence genes are not.
The phenotypic effects we observe in response to vlf-1 knockdown are consistent with this protein being both a structural component and a product required for virion assembly. The defects in morphogenesis of MdBV virions we observe, however, differ somewhat from the defects observed with AcMNPV where knockout of vlf-1 resulted in formation of tubular structures that appeared to be aberrant capsids that fail to package DNA [40]. The technical approaches to these studies, however, resulted in observations associated with budded virus production and precluded examination of potential defects associated with formation of occlusion-derived virus (see below). In contrast, the severe defects we observed in the assembly of MdBV virions suggest vlf-1 may be important in both DNA packaging and proper association of capsids with envelopes. Like vlf-1, knockdown of vp39 greatly increased the sensitivity of packaged DNA to DNAse treatment while also reducing infectivity. For both genes, dsRNA treatment did not completely abolish DNAse protection or infectivity of virus particles, which we presume is due to incomplete knockdown of transcript levels. Unlike vlf-1 though, knockdown of vp39 did not cause any obvious morphological defects in virion assembly with the exception that virion aggregations in calyx cells were much smaller owing possibly to a reduction in VP39 for production of capsids. Given these observations, we are unclear why knockdown of vp39 did not reduce virion density in calyx fluid.
Unlike BVs, which produce only one virion type, AcMNPV and most other baculoviruses produce two virion phenotypes named occlusion-derived virus and budded virus [24]. Occlusion-derived virus is embedded in a protein matrix called an occlusion body and is the type that initiates a midgut infection when ingested by a new host. Budded virus in contrast is non-occluded and is the form of the virus that disseminates from the midgut and other cells to systemically infect the insect. Occlusion-derived and budded virus capsids contain the same structural proteins [55] but their envelopes differ greatly with the former assembling de novo in host cell nuclei and containing products of several core genes including per os infectivity factors. Budded virus in contrast acquires an envelope when budding through host cell plasma membranes, which contains only one or two viral proteins (GP64, F) [55]. Although never occluded, the de novo assembly of BVs in calyx cell nuclei together with the envelope components detected in their virions (see above) indicate that BV particles structurally share more features with the occlusion-derived phenotype of baculoviruses. On the other hand, while per os infectivity factors are required for infectivity and binding of the occlusion-derived phenotype of AcMNPV to midgut cells, they are not required for infection of cultured cells or host larvae when injected into the hemocoel [49], [59]. We thus were unclear what effect, if any, knockdown of p74 or pif-1 might have on infectivity given that BVs infect host insects only when injected into the hemocoel by wasps. Our results reveal no defects in the copy number of MdBV DNA detected in CiE1 cells after infection with a high MOI. Yet, knockdown of each gene resulted in a large decline in the fraction of infected cells that expressed the marker gene GLC1.8. These findings are interesting because they suggest the loss of per os infectivity factors from the MdBV envelope results in improper translocation of MdBV to host cell nuclei where transcription of glc1.8 and other virulence genes occurs. No such activity has previously been associated with per os infectivity factors in baculoviruses but intriguingly GP64, the envelope fusion protein of budded virus, has been implicated in affecting baculovirus translocation to host cell nuclei [60].
In addition to targeting six baculovirus-like conserved genes, we also examined the function of nudivirus-like int-1 because this gene and vlf-1 are both tyrosine recombinase family members and BV replication requires the excision of proviral genomic DNAs from the wasp genome for packaging into virions. The near complete inhibition of empty locus formation after vlf-1 and int-1 knockdown suggests a role for both in proviral DNA excision. At this time we have little understanding of the recombination reactions VLF-1 and INT-1 potentially mediate, although detailed investigation of other tyrosine recombinases suggest they perform recombination events by establishing a synapse between their cognate binding sites and performing two consecutive strand exchanges [61], [62]. If on the same molecule recombination between two binding sites leads to excision of a circular intermediate [63]. Analysis of tyrosine-recombinase structural properties also suggest the mechanism of recombination requires binding of four enzyme monomers, which suggests the possibility for involvement of both vlf-1 and int-1 in proviral DNA excision [53]. Additionally, proviral DNA segments that become encapsidated possess direct repeats at the sequence boundaries that abut the flanking wasp DNA, which have been previously hypothesized to contain binding sites for recombinases that mediate excision [18], [19], [30]. Lastly detection of both VLF-1 and INT-1 in virions suggests these factors may also have important functions in parasitized hosts given recent findings that all DNA segments packaged into MdBV virions rapidly integrate into the genome of infected host cells [30].
Taken together, our results provide the first experimental insights into the function of a subset of MdBV genes. We fully recognize that additional experiments will be needed to more fully characterize the function of the individual genes we targeted, but by examining several key genes at once we provide evidence that: 1) RNAi can be used to knock down a number of MdBV genes, and 2) the baculovirus-like genes we targeted exhibit conserved functions despite divergence from nudiviruses more than 100 Mya. At the same time our results with vlf-1 and int-1 also identify novel functions unknown from baculoviruses but essential to the biology of BVs. With these results in hand, we are now positioned to undertake both more detailed experiments on these genes as well as studies on nudivirus-like and novel genes in the BV conserved gene set for which expectations about function are less clear.
The arms race between wasps and the hosts wasps parasitize likely underlies the high rates of speciation of BV-carrying braconids [26], [64]. The genetic mechanisms guiding host range evolution in contrast are largely unknown. One hypothesis would be that PDV virion structure has undergone rapid adaptation in response to the different lepidopteran host species each wasp species parasitizes, which could result in high variation in BV virion structure. The similarities thus far found in BV conserved genes together with the functional insights provided here, however, strongly suggest that BV gene functions will be conserved across isolates. Thus, differences in the sequence of BV genes may affect whether a given isolate can infect a given host species but we think it unlikely large differences will be found in the structure of BV virions across isolates. In contrast, the literature already indicates that the virulence genes BVs package into virions vary greatly among isolates associated with phylogenetically distant species of wasps. Thus, we would expect that differences in the virulence genes BVs deliver to hosts strongly affect host range by impacting the ability of wasp offspring to successfully develop. Finally, conservation in the function of the MdBV RNA polymerase and structural proteins suggest that key features in the evolution of BVs into mutualists do not involve radical changes in virion structural products but rather in how: 1) transcription of early factors required for DNA replication and viral transcription are regulated so that replication only occurs in calyx cells, 2) the genome is organized so that only some portions are amplified and packaged into virions, and 3) virulence genes are kept silent in wasps.
All studies were approved by the Biological Safety and Animal Care and Use Committee of the University of Georgia and were performed in compliance with relevant institutional policies, National Institutes of Health regulations, Association for the Accreditation of Laboratory Animal care guidelines, and local, state, and federal laws.
M. demolitor parasitizes several species of larval stage Lepidoptera including Chrysodeixis ( = Pseudoplusia) includens. Both species were reared at 27°C with a 16 h-light:8 h-dark photoperiod. M. demolitor has an 11 day developmental period described in detail elsewhere [7]. For this study, M. demolitor females were allowed to parasitize C. includens larvae, and wasp offspring were then allowed to develop in the host for 6 days. On day 7, wasp larvae emerge from hosts and spin a cocoon within several hours. Cocoons are slightly asymmetrical with the anterior end generally being more elevated and pointed than the posterior end that contains the wasp abdomen. Nine-12 hours after spinning their cocoon, wasps pupate and develop for four more days until emerging as an adult. Adult wasps were then maintained in constant dark at 18°C.
Virus was collected from M. demolitor ovaries. For the first replicate, 100 whole ovaries were crushed in PBS and the debris was removed by centrifugation. Then, the supernatant containing the virus was spun down at 20,000×g for 5 minutes and washed with PBS 3 times to collect virus particles. For replicate 2, calyx fluid was dissected from 100 dissected ovaries and resuspended in PBS. Virions were then isolated on a sucrose gradient as previously described [31]. Both virus samples were electrophoresed on either a 4–20% or 12.5% Tris-Glycine gel (Lonza). For each sample, the entire lane was cut into four pieces to separate proteins by size. In-gel trypsin digestion was performed for each gel slice, by overnight incubation with trypsin (20 µg/ml) in 20 mM ammonium bicarbonate. Tryptic fragments were extracted with 50% acetonitrile and 0.1% TFA and vacuum dried. Samples (4 µl) were analyzed by an Orbitrap Elite mass spectrometer, coupled to an Easy-nLC II Liquid Chromatography (LC) instrument (Thermo Fisher Scientific). Samples were desalted and pre-concentrated on a C18 Easy LC pre-column (100 um internal diameter (ID) ×2 cm, 5 µm particle packing (PP)). Peptides were eluted from a reverse-phase column (75 um ID ×10 cm, 3 µm PP) with a gradient of 10–35% B for 70 min, 35–95% B for 10 min, 95% B for 5 min (A = 0.1% formic acid in water, B = 0.1% formic acid in acetonitrile) at 300 nl/min. Nanospray ionization was performed with a spray voltage of 2 kV, with a capillary temperature of 200°C. The Orbitrap mass analyzer was used to provide resolutions of 120,000 and 30,000 for MS and MS/MS analyses, respectively. Briefly, a cycle of one full-scan mass spectrum (300–2000 m/z) was performed, followed by continuous cycles of CID and HCD MS/MS spectra acquisitions of the 2 or 5 most abundant peptide ions throughout the LC separation until the candidate ions were exhausted. Data were acquired using Xcalibur software (v2.2, Thermo Fisher Scientific).
Proteins were identified by searching against a custom database “Md” consisting of translated open reading frames (ORFs) greater than 33 amino acids in size from transcripts described in [7] using the Mascot v2.3 algorithm (Matrix Science Inc.). Transcripts can be accessed through NCBI accession numbers JO913492 through JO979916 and JR139425 through JR139430. Data were visualized with ProteomeDiscoverer v1.3 (Thermo Fisher Scientific). Peptides with scores greater than the identity score (p<0.05) were considered significant matches. Only ORFs that were matched by at least two peptide spectra were considered positive identifications.
To target individual genes for RNAi knockdown, gene-specific primers were designed with added T7 promoter adaptors to amplify a 300–400 bp template for double-stranded RNA (dsRNA) synthesis (Table S2). cDNA from adult wasp ovaries was amplified using standard PCR and the resulting products were used as template in the MegaScript RNAi Kit (Ambion). Larval stage wasp cocoons were marked within 15 h of spinning and were subsequently injected within 1–3 h. As M. demolitor is protandrous with males developing faster than females, the cocoons selected for injections were biased towards later emergence times and thus female wasps. Cocoons were affixed to double-sided tape, and oriented so that the posterior ends were facing in the same direction. Cocoons were pierced with a minuten pin in the abdomen region and wasps were injected through the cocoon with a glass needle directly into the abdomen. Approximately 0.5–1 µl of 2–4 µg/µl dsRNA was injected into each individual. All control wasps were injected with a non-specific dsRNA probe homologous to the bacterial eGFP gene. Wasps were sampled within 24 h of emerging as adults, and ovaries were removed and separated using opthalmic scissors. One ovary half was snap-frozen at −80°C for RNA extraction, and the other was used to assay RNAi phenotypes as described below.
Extraction of total RNA was performed using the QIAGEN RNeasy kit following the standard kit protocol with a 20 min on-column DNAse treatment and elution in 30 µl of RNAse-free H2O. RNA concentration was measured using a Nanodrop Spectrophotometer and cDNA was synthesized from a quantity of total RNA normalized across samples. First-strand cDNA synthesis was performed using Invitrogen reagents including the Superscript III enzyme and oligo(dT) primers as outlined by the manufacturer (Invitrogen).
We used quantitative PCR (qPCR) to detect differences in expression of target genes after dsRNA treatment. An absolute standard curve was generated via PCR amplification of the corresponding cDNA for each gene of interest using specific primers (Table S2). Each product was cloned into pSC-A-amp/kan, and after propagating and isolating each plasmid from minipreps, its identity was confirmed by sequencing. Standard curves were generated followed by determination of copy numbers from serially diluted amounts (102 to 107 copies) of each plasmid standard. qPCR was performed on a Rotor-Gene Q using the Rotor-Gene SYBR Green PCR Kit with 1 µM primers and 1 µl of undiluted cDNA per 10 µl reaction (QIAGEN). After 5 minutes of denaturation at 95°C, a two-step amplification cycle with 95°C for 5 sec denaturation and 60°C for 20 sec of annealing and extension was used for 45 cycles. Melting curve analyses were performed to ensure that amplified products were specific for the gene of interest. At least three independently acquired biological replicates were analyzed per stage for each gene, with each sample internally replicated 4 times.
A polyclonal antibody against Lef-9 was generated by PCR amplifying and cloning a portion of MdBV lef-9 into pET-30-EK-LIC (Novagen) as previously outlined [65]. Briefly, 31% of the lef-9 coding sequence was amplified and cloned to create an expression product of 22.9 kDa. This construct was confirmed by DNA sequencing, and then expressed in Escherichia coli BL21 (DE3) cells grown in 6 L Luria Broth with 50 µg/ml kanamycin at 37°C. The cultures were then induced with 0.025 mM isopropyl-β-d-thiogalactopyranoside (IPTG) for an additional 24 h at 37°C. Bacterial cells were harvested by centrifugation, lysed, and the insoluble recombinant protein was purified from the cell pellet using PerfectPro Ni-NTA (QIAGEN) agarose beads under denaturing conditions. After analysis by SDS–PAGE and immunoblotting using an anti-His monoclonal antibody, the identity of the recombinant protein was validated by mass spectrometry. A polyclonal antibody was then produced by a commercial service (Pacific Immunology) which generated antisera in rabbits by immunizing with ca. 500 µg of affinity purified truncated LEF-9. Antiserum was then purified by nitrocellulose-based immunoaffinity chromatography as previously outlined [65], [66]. The resulting antibody was then used in immunoblotting experiments with control and lef-9 knockdown wasps by explanting ovaries, separating ovary extracts on 5–20% SDS-PAGE gradient gels and transferring to PVDF (Immobilon). LEF-9 was then visualized using a goat anti-mouse secondary antibody and the ECL Advance kit as previously described [65], [67].
The titer of MdBV virions with DNAse-protected episomes of MdBV segment B was quantified by qPCR. One half of an ovary pair for each wasp individual was homogenized with a pestle in 100 µl of DNase buffer from the Roche HighPure RNA Isolation Kit, and NP40 was added to a final concentration of 1% to solubilize wasp cells and virus particle envelopes. After 20 min of gentle rocking, 1 µl of TURBO DNase from the Ambion DNA free kit was added and samples were incubated at 37°C for 40 min to digest all free wasp and episomal viral DNAs. After the addition of EDTA (10 mM) to inactivate the DNase, 250 µg of proteinase K (Roche) and 2% sarcosyl were added to samples, followed by incubation at 62°C for 1 h and by phenol∶chloroform extraction and ethanol precipitation in the presence of 0.3 M sodium acetate, pH 5.2. DNA pellets were resuspended in 30 µl of 10 mM Tris-Cl pH 8.5 and diluted to 1 ng/µl for use as template. Segment B specific primers flanking the point of circularization in the viral segment were used to amplify circularized viral DNA using qPCR as described above and previously [7] (Figure 7B, Table S2). Per-ovary copy numbers were calculated by multiplying the qPCR estimate of copy number for a half ovary by two, and by the dilution factor and elution volume. At least 3 independently acquired biological replicates were performed for each treatment with samples internally replicated 4 times. Quantification of viral segment excision was performed on genomic DNAs extracted from the entire tissue from an ovary half with the proteinase K, sarcosyl, and phenol/chloroform extraction method described above, without a DNAse step. Primers specific for the empty locus for Segment B were used in qPCR to quantify copy number (Figure 7B, Table S2). TEM was performed as in [31].
The CiE1 cell line was maintained as in [43]. The infectivity of virus preparations was measured by counting the number of Segment B viral genome copies in CiE1 cells after 24 hours of incubation [43]. Virus was collected from 24 h old wasps treated dsRNA by puncturing a half ovary in 50 µl of PBS and allowing the virus to dissolve. The entire contents of the droplet were added to a microcentrifuge tube containing 200 ul of Sf900 media with 5% fetal bovine serum and 1% antibiotics. This solution was filtered through a 0.45 µm membrane to remove any bacteria and cellular debris. Fifty µl of each virus preparation, corresponding to 0.1 wasp equivalents or an estimated MOI of 100 for control samples [29] was added to wells in a 24-well cell culture plate containing 105 CiE1 cells per well. Virus particles were incubated with cells for 2 h at room temperature, followed by removal of virus-containing media and the addition of 500 µl of fresh media. CiE1 cells were incubated for an additional 22 hours at 26°C. DNA was isolated from cells following the protocol for quantification of viral titer above without prior DNase treatment. All samples were diluted to a concentration of 50 ng/µl for qPCR amplification of circularized Segment B as described above.
Successful viral gene expression, translation and export of protein products in host cells was quantified by counting the percentage of cells displaying the MdBV protein GLC1.8 on their surface. CiE1 cells were infected with virus preparations as described above, and after 24 hours of total incubation were fixed with 3.7% paraformaldehyde and stained with a murine monoclonal antibody specific for GLC1.8 followed by goat anti-mouse Alexa-fluor 568 secondary antibody (Molecular Probes) as described [44]. One hundred CiE1 cells were counted from a randomly selected field of view using an epifluorescent, phase-contrast microscope (Leica DM IRB).
JMP v10 was used for all statistical analyses. For qPCR analyses, the number of copies of a gene or DNA product were averaged for all technical replicates within a biological replicate. For functional assays, means were calculated from experimental values derived from biological replicates. Each biological replicate represented an individual wasps' ovary. Differences between means of biological replicates were tested using a t-test assuming equal variances or ANOVA. Differences between means for experiments with more than two treatments were distinguished using Tukey's HSD test at the p<0.05 significance level.
|
10.1371/journal.ppat.0030094 | Evidence of Differential HLA Class I-Mediated Viral Evolution in Functional and Accessory/Regulatory Genes of HIV-1 | Despite the formidable mutational capacity and sequence diversity of HIV-1, evidence suggests that viral evolution in response to specific selective pressures follows generally predictable mutational pathways. Population-based analyses of clinically derived HIV sequences may be used to identify immune escape mutations in viral genes; however, prior attempts to identify such mutations have been complicated by the inability to discriminate active immune selection from virus founder effects. Furthermore, the association between mutations arising under in vivo immune selection and disease progression for highly variable pathogens such as HIV-1 remains incompletely understood. We applied a viral lineage-corrected analytical method to investigate HLA class I-associated sequence imprinting in HIV protease, reverse transcriptase (RT), Vpr, and Nef in a large cohort of chronically infected, antiretrovirally naïve individuals. A total of 478 unique HLA-associated polymorphisms were observed and organized into a series of “escape maps,” which identify known and putative cytotoxic T lymphocyte (CTL) epitopes under selection pressure in vivo. Our data indicate that pathways to immune escape are predictable based on host HLA class I profile, and that epitope anchor residues are not the preferred sites of CTL escape. Results reveal differential contributions of immune imprinting to viral gene diversity, with Nef exhibiting far greater evidence for HLA class I-mediated selection compared to other genes. Moreover, these data reveal a significant, dose-dependent inverse correlation between HLA-associated polymorphisms and HIV disease stage as estimated by CD4+ T cell count. Identification of specific sites and patterns of HLA-associated polymorphisms across HIV protease, RT, Vpr, and Nef illuminates regions of the genes encoding these products under active immune selection pressure in vivo. The high density of HLA-associated polymorphisms in Nef compared to other genes investigated indicates differential HLA class I-driven evolution in different viral genes. The relationship between HLA class I-associated polymorphisms and lower CD4+ cell count suggests that immune escape correlates with disease status, supporting an essential role of maintenance of effective CTL responses in immune control of HIV-1. The design of preventative and therapeutic CTL-based vaccine approaches could incorporate information on predictable escape pathways.
| One of the greatest challenges facing HIV-1 vaccine design today is the formidable capacity of the virus for mutation and adaptation, a characteristic that has contributed to the extensive worldwide genetic variability of HIV-1 strains observed today. On an individual basis, evolutionary selective pressures imposed by each infected person's unique immune response results in the selection and outgrowth of viral “escape” mutants capable of evading immune recognition, while on a population basis, complex evolutionary selective pressures imposed by the highly polymorphic genes of the human immune system shape HIV-1 diversity on a global level. Making sense of the seemingly infinite complexity of HIV immune escape is of paramount importance in our goal of developing a successful HIV vaccine. The current study uses cutting-edge statistical methods to identify specific sites and patterns of human leukocyte antigen (HLA) class I-restricted escape mutations in various HIV genes. Researchers summarize their findings in the form of “immune escape maps,” which highlight the differential contribution of immune imprinting to HIV genetic diversity, as well as identify specific sites in the viral genome under active immune selection pressure. Results from the present study contribute to our understanding of how human immune selective pressure contributes to variation in different HIV genes, and could help inform the development of HIV vaccines that take into consideration viral diversity.
| Genetic variation within the highly polymorphic human leukocyte antigen (HLA) class I region contributes to diversity of pathogen recognition by cytotoxic T lymphocytes (CTLs) [1], and acts as a selective force shaping viral evolution within an infected host [2–6] through selection of mutations that allow the virus to escape recognition by HLA-restricted CTLs [5,7–9]. Immune escape may also represent a significant force shaping viral evolution at the population level through an HLA “imprinting effect,” in which escape mutations selected in the context of common HLA class I alleles may become predominant in the circulating viral population if they do not revert upon transmission to new hosts [2,10,11].
One of the major challenges to HIV vaccine design is the extensive worldwide sequence diversity of this pathogen, fueled in part by the extreme mutational capacity of the virus [12]. However, despite this considerable diversity, evidence indicates that there are constraints on viral evolution [2,13,14], and that escape in response to specific immune selective pressures (similar to escape from drug selective pressures [15]) follows broadly predictable mutational patterns [13,14]. A comprehensive identification of specific sites and patterns of immune escape in clinical HIV-1 isolates will further our understanding of how immune selection contributes to viral diversity [2,16] and will also identify specific viral regions under active immune selection pressure, thus providing information relevant to the selection of candidate immunogens for an HIV-1 vaccine.
Improvements in DNA sequencing technologies and the availability of large cohorts of HIV-1 infected individuals now allow us to employ population-based genetic association approaches to identify viral amino acids (aa) under active immune selection pressure in vivo [2]; however, methodological challenges associated with identifying such mutations are now recognized [16]. Moore et al. [2] were the first to identify HLA-class I-associated polymorphisms across codons 20–227 of HIV-1 reverse transcriptase (RT) in a large clinically derived dataset using a Chi-squared association approach, thus providing evidence for HLA class I-mediated viral evolution on a population level. However, the application of standard statistical tests is inappropriate for the analysis of viral isolates with a shared phylogenetic history, since descent from a common ancestor means that viral sequences may not be treated as statistically independent entities [17]. Specifically, a cause for concern is the application of standard statistical methods to identify HLA-associated viral polymorphisms in cohorts comprising individuals of diverse genetic backgrounds (sampled from populations with differential HLA allele distributions) infected with heterogeneous viral strains. In this case, standard statistical approaches such as the Chi-squared test may identify confounding associations between strain- or lineage-specific viral polymorphisms and specific HLA alleles that are over-represented in subpopulations of individuals harboring infections with those strains. In this case, the observed “HLA-associated polymorphism” is not evidence of active HLA-mediated immune selection. Rather, the association is simply a statistical correlation between possession of a particular HLA allele observed among persons of a particular ethnic background, and a lineage-specific viral polymorphism, arising as a result of descent from a common ancestor (“founder effect”) [16]. The use of population-based, viral lineage-corrected analyses, such as those recently developed by Bhattacharya et al. [16], are therefore essential in order to accurately identify sites of active immune selection in the genomes of sequence-diverse pathogens such as HIV-1.
In addition, although there is clear evidence supporting HIV-1 adaptation to HLA class I-mediated CTL selection pressure from an evolutionary standpoint [2–5], the relevance of immune escape to clinical HIV disease progression remains unclear, due in part to the fact that many studies have focused small numbers of participants and/or escape within a limited number of HLA-restricted CTL epitopes in the viral genome [3–6,18–20]. Furthermore, no studies to date have linked HIV disease progression to HLA-associated polymorphisms corrected for lineage effects. Here we identify lineage-corrected [16] HLA class I-associated polymorphisms across select functional and accessory/regulatory HIV-1 genes in a cross-sectional analysis of a large cohort of chronically infected, treatment-naïve individuals, and investigate the relationship between these polymorphisms and clinical markers of HIV disease.
A large, well-characterized cohort of chronically HIV-1 infected, antiretroviral drug-naïve individuals from British Columbia, Canada [21], for whom HLA class I typing and HIV RNA genotyping of select functional and accessory/regulatory genes were performed, was used to identify HLA class I allele-associated viral polymorphisms across a 499 aa fragment spanning protease and most of RT p51 (n = 532 successfully genotyped), 96 aa of Vpr (n = 425), and 206 aa of Nef (n = 686). HLA class I allele-associated viral polymorphisms were identified using analytical approaches described in [16], which feature a correction for viral lineage effects by adjusting for phylogenetic relationships between sequences [16], and a correction for multiple comparisons using a q-value approach [22], which sets the false-discovery rate (20% with q < 0.2) among significant associations. The level of variation at single residues in protease, RT, Vpr, and Nef ranged from 0% to maxima of 50%, 57%, 73%, and 77% respectively (Figure 1), while the mean pairwise amino acid identity for these same genes (calculated as the percentage of codons exhibiting identical amino acids for each pairwise combination of sequences) was 92.9%, 95.3%, 89.0%, and 83.1%, respectively, indicating typical intrasubtype levels of HIV sequence diversity in this cohort of relatively homogeneous subtype distribution (97.5% HIV-1 subtype B).
It is important to note that the phylogenetically corrected methods for identification of HLA-associated viral polymorphisms developed by Bhattacharya et al. [16] do more than simply correct for confounding due to HIV intersubtype (or interclade) variation. Even among clade-homogeneous datasets, “subclade” lineage-specific effects may yield confounding associations with HLA alleles, especially if the cohort is composed of subpopulations with differential HLA allele distributions. Indeed, there was clear evidence of phylogenetic subclusters within subtype B sequences in this cohort (Figure S1).
We therefore compared phylogenetically corrected methods to a simple uncorrected test (simple Fisher), and found that even in this predominantly subtype B-infected cohort, HLA-associated polymorphisms identified using phylogenetically corrected methods had higher fractions of associations that could be independently validated by immunological data than those defined using a simple Fisher exact test (unpublished data). An example of a case in which an apparent HLA-associated polymorphism identified using a simple Fisher exact test represents an artifact of the phylogenetic tree is illustrated in Figure S1. In this analysis, therefore, we report only HLA-associated polymorphisms defined using phylogeny-based methods [16].
Application of the viral lineage-corrected method [16] yielded a total of 478 unique HLA allele-associated viral polymorphisms with q < 0.2 across the genes investigated (Table S1). These occurred at nine (9%), 28 (7%), 12 (12.5%), and 84 (41%) unique codons in protease, RT, Vpr, and Nef, respectively, highlighting a dramatic variation in HLA-associated imprinting across HIV-1 genes (Figure 1). HLA-B alleles accounted for half (241 of 478; 50.4%) of the total number of HLA-associated polymorphisms, while HLA-A and C alleles accounted for 112 (23.4%) and 125 (26.2%), respectively.
Previous studies have validated the application of such genetic association analyses of large clinically derived datasets in order to identify HLA-restricted CTL escape mutations selected in vivo [16]. Knowing that with q < 0.2, about 20% of identified HLA-associated polymorphisms will represent false-positive results, we set about classifying the 478 identified polymorphisms into putative true-positive or false-positive results based on the strength of independent biological evidence supporting each polymorphism as an escape-associated mutation. The highest level of biological support was assigned to HLA-associated polymorphisms falling within or proximal to (± 3 aa) a published CTL epitope [23] restricted by that particular HLA allele, thereby supporting these associations as in vivo-selected mutations directly or indirectly affecting MHC binding, T cell receptor recognition and/or intracellular peptide processing [24–26]. A second level of support was assigned to those associations falling within or similarly proximal to putative/novel HLA-restricted epitopes, identified by scanning the cohort consensus sequence for HLA-restricted epitope anchor residue motifs using two independent bioinformatic tools (MotifScan [Los Alamos National Laboratory], http://www.hiv.lanl.gov/content/immunology/motif_scan/motif_scan; and Epipred [Microsoft Research], http://atom.research.microsoft.com/bio/epipred.aspx).
To provide further biological support for these associations we drew upon an independent cohort of 372 HIV-1 infected individuals screened for in vitro HLA-restricted, CTL-mediated interferon-gamma (IFN-γ) responses against a set of overlapping HIV-1 subtype B consensus peptides spanning the entire viral proteome using the IFN-γ enzyme-linked immunosorbent spot assay (ELISpot) [27], in order to identify HLA class I alleles significantly associated with CTL-mediated IFN-γ production in response to stimulation with consensus HIV peptides (see Methods). HLA allele-associated polymorphisms mapping within a significantly reactive HLA allele/HIV consensus peptide pair were identified as potential escape-associated mutations to known or novel HLA-restricted CTL epitopes.
Finally, we grouped together HLA allele-specific associations clustering within these epitopes or motifs, and paired together alleles in linkage disequilibrium (Table S2) associated with the same HIV polymorphism(s), to create a series of immune escape maps capturing the minimum number of HLA-restricted epitopes and/or motifs required to explain the data (Figures 2–5). Associations that did not map within a known epitope or motif, and were not supported by ELISpot data or attributable to HLA allele linkage, were listed in a separate map (Figure 6).
After pairing together linked alleles, approximately 35% of codons in protease, RT, Vpr, and Nef exhibiting HLA-associated polymorphisms mapped inside (n = 77; 81%) or within ± 3 aa (n = 18; 19%) of a published CTL epitope specific to that HLA allele (Figures 2 and 3). Significant associations were collapsed into two categories based on the direction of the HLA selection pressure: amino acids enriched in the presence of a specific allele (positive or “escape” correlations, presumably representing the escape variant for that allele), and amino acids depleted in the presence of a specific allele (negative or “reversion” correlations, presumably representing the immunologically susceptible or “wild-type” form for that allele, and also representing the amino acid to which the residue will likely revert to upon transmission an individual lacking that allele). Overall, the majority of HLA-associated polymorphisms (58% of epitope-supported associations) represent negative (“reversion”) correlations (p = 0.002). Note that detection of a “reversion” correlation in the absence of an associated “escape” correlation may arise in the case where a specific allele selects for multiple amino acids at a given position, creating a situation where there may be sufficient statistical power to detect the “reversion” correlation but not to identify all possible escape variants.
A considerable number of codons exhibit multiple HLA associations, particularly in Nef. A total of 57 multiple associations were observed, with 2, 6, 3, and 46 occurring across protease, RT, Vpr, and Nef, respectively. In 23 of these 57 cases (for example, Nef codons 81 and 135), the same amino acid represents an escape variant for one HLA allele, but the susceptible form for another, highlighting a “tug-of-war” of differential HLA selective pressures contributing to populational HIV sequence diversity at specific codons.
There were dramatic differences in the number of HLA-associated polymorphisms across the genes investigated. Not only did Nef exhibit a much higher density of epitope-supported associations compared to protease/RT and Vpr, but the escape patterns also tended to be more complex in Nef than in other genes. A total of 53% of escaping epitopes in Nef exhibited HLA-associated polymorphisms at multiple positions within the epitope, compared with 12% and 0% in protease/RT and Vpr, respectively (p = 0.004). Similarly, epitope-proximal associations (occurring within 3 aa of a published epitope) were also observed more frequently in Nef (n = 16 [22%]) while occurring only relatively rarely in protease/RT/Vpr (total n = 2 [9%]), although this did not achieve statistical significance (p = 0.2).
Overall, HLA-associated polymorphisms were observed with relatively equal frequency across all positions within published HLA-restricted epitopes. There was no statistically significant enrichment for HLA-associated polymorphisms at anchor residues (generally defined as epitope residues 2 and C-terminal with some exceptions [28,29]) over other residues in Nef (p = 0.7) or protease/RT/Vpr (p > 0.1) suggesting that amino acid changes potentially affecting peptide binding to HLA class I molecules are not a favored mechanism of escape.
We organized a further ~50% of the identified associations into “motif-support” maps (Figures 4 and 5) that grouped HLA-associated polymorphisms within HLA-restricted epitope anchor residue motifs identified by scanning the cohort consensus sequence. Based on evidence that HLA-associated polymorphisms identified in genetic association studies predict the location of previously uncharacterized epitopes [16], we would expect that a substantial proportion of motif-supported associations represent escape mutations within novel epitopes, a hypothesis supported by the fact that many motif-supported associations (40%, 31%, 22%, and 19% in protease, RT, Vpr, and Nef, respectively) are substantiated by in vitro IFN-γ ELISpot responses to HIV-specific consensus peptides containing these motifs. Consistent with observations drawn from the epitope-support maps (Figures 2 and 3), the majority (63%) of associations in the motif-support maps represent “reversion” associations, with a much more complex pattern of escape observed in Nef compared to protease/RT/Vpr.
The remaining ~15% of HLA-associated polymorphisms did not map to known epitopes and were unlikely to lie within or proximal to novel epitopes as suggested by in vitro IFN-γ ELISpot responses or bioinformatic motif scans (Figure 6). Although these proportions are consistent with the false-discovery rate of ~20% (q < 0.2), lack of biological support cannot be used to definitively categorize these as “false-positive” associations in any particular case. In some cases, these may represent processing escape mutations occurring distant from the epitope site, compensatory mutations, unusual epitopes, or other factors. Similarly, HLA-associated polymorphisms mapping within an HLA-matched epitope or motif are likely highly enriched for mutations directly or indirectly conferring immune escape, but likely contain smaller numbers of false-positive associations as well.
Although there is clear evidence documenting the selection of escape variants over the course of HIV infection [3–5,7–9], the clinical significance of immune escape remains incompletely understood [18–20]. Moore et al. reported a significant association between HLA-associated polymorphisms and plasma viral load [2]; however, no studies to date have linked lineage-corrected HLA-associated polymorphisms with markers of disease progression on a population basis. We therefore investigated correlations between the presence of HLA-associated polymorphisms and clinical status in chronic untreated infection as measured by pretherapy CD4+ cell number and plasma viral load. In order to adopt the most conservative definition of “escape,” the primary analysis was restricted to those amino acid associations mapping inside or within ± 3 aa of a known HLA-restricted CTL epitope (Figures 2 and 3). A significant inverse dose–dependent relationship was observed between the median pretherapy CD4+ cell count and the number of epitope-associated polymorphisms observed in protease/RT (p = 0.006), Vpr (p = 0.01), and Nef (p = 0.008) (Figure 7). A trend was observed between accumulation of epitope-associated polymorphisms in protease/RT (but not other proteins) and higher pretherapy viral load (p = 0.06 [unpublished data]). The dose-dependent association between epitope-associated polymorphisms and lower CD4+ cell counts supports the ability of large genetic association studies to identify biologically relevant in vivo CTL escape-associated mutations, but more importantly, supports a clinically relevant link between immune escape and HIV disease progression.
Note that the observed association between HLA-associated polymorphisms and lower CD4+ cell count is specific to HLA-associated polymorphisms mapping within or near published epitopes, and not simply a general association between viral mutations and HIV clinical status. In a secondary analysis we investigated correlations between the presence of motif-associated (Figures 4 and 5) and unsupported (Figure 6) polymorphisms and clinical parameters. A nonsignificant trend (p = 0.07) was observed between accumulation of motif-associated polymorphisms in protease/RT (but not other proteins) and lower median CD4+ cell counts, while no significant association was observed between clinical parameters and the presence of biologically unsupported associations, consistent with a stepwise enrichment for false-positives among associations in these categories.
The present study represents to our knowledge the largest population-based investigation of HLA class I-mediated imprinting on HIV sequence to date, as well as the first characterization of HLA-associated polymorphisms in each of a functional, accessory and regulatory gene. Results identify viral polymorphisms selected in vivo in context of a wide array of class I alleles. The confirmation of the B*1501-associated polymorphism at protease codon 93 reported by Bhattacharya et al. [16] and several reported by Moore et al. [2] in RT suggest that immune escape patterns in HIV-1 subtype B are consistent across the globe. The confirmation of several functionally verified CTL escape mutations previously observed in clinically derived isolates (including escape at residues 2, 8, 2, and 5 of the HLA-B*57 restricted IW9-RT [13,30], B*51-restricted TI8-RT [31], A*24-restricted RF10-nef [32], and B*08-restricted FL8-nef [5] epitopes, respectively) confirm the utility of genetic association studies to identify escape variants commonly selected in vivo. Taken together, results provide proof of principle that population-based approaches could complement smaller functional studies by providing a whole-gene or whole-virus picture of immune escape.
Results of this large-scale, multigene analysis reveal dramatically different levels of HLA-associated polymorphisms across HIV proteins, with a previously unreported, extraordinary density and complexity of HLA-associated polymorphisms in Nef. Nef exhibits considerable sequence diversity and thus may exhibit higher levels of mutational plasticity in response to selective pressures compared to genes exhibiting structural (e.g., Gag) or functional (e.g., protease/RT) constraints; however it is important to note that protease (and to a lesser extent RT) exhibit extensive mutational capacity under antiretrovirally mediated selection pressure [15], suggesting that mutational constraints on functional genes are unlikely to fully account for the relative paucity of HLA-associated polymorphisms across these regions. Rather, results are consistent with the density of CTL epitopes across these regions, as well as the relative immunogenicity of these proteins over the course of infection [27,33]. Limited data from longitudinal studies suggest that CTL escape mutations in Nef are selected earlier in infection [33,34], and thus, in a population of chronically infected individuals, one may expect a large burden of Nef escape mutations to have already accumulated. Note that, in the current study, Nef sequences were available for a larger number of participants, thus potentially increasing power to detect significant associations.
These data are also relevant to CTL-based HIV vaccine design. First and foremost, the analysis of clinically derived datasets identifies viral epitopes under active immune selection pressure, thus identifying in vivo immunogenic viral targets. The fact that we observed such a large number of HLA-associated polymorphisms, including many instances of specific codons apparently under diametrically opposed HLA-selective pressures (an observation consistent with Iversen et al. [35]), provide some evidence against the complete disappearance of all active viral epitopes under the HLA “imprinting hypothesis” (which states that escape mutations selected in response to the most common HLA alleles may become fixed in the circulating viral population [2,10], thus resulting in a potential loss of CTL responses to these epitopes and rendering them inappropriate as candidate immunogens). Taken together with evidence supporting rapid reversion of escape mutations after transmission to a new host [36], and the fact that escape mutations in one individual may represent the susceptible form in another [16], the “HLA imprinting effect” is unlikely to result in the creation of an immunologically refractive circulating viral population by eliminating all active CTL epitopes in this population. Rather, selection pressures mediated by diverse HLA class I alleles in HIV-1 infected populations appear to be actively contributing to viral diversity thus preserving a substantial number of immunologically active epitopes in the circulating population. These active epitopes, most notably those which exhibit the “push-and-pull” of diametrically opposed HLA selection pressures, could perhaps be incorporated into a CTL-based HIV-1 vaccine strategy.
The locations of HLA-associated polymorphisms relative to known or predicted HLA-appropriate epitopes revealed no statistically significant enrichment for mutations at epitope anchor residues versus other positions. Theoretically, if the predominant mechanism of CTL escapes were abrogation of peptide-MHC binding through anchor residue mutation, a polyvalent vaccine approach may have little merit. However, these observations, combined with previous documentation of de novo T cell responses arising in response to escape variants [37], strongly support the utility of incorporating viral sequence variation into immunogen design.
Given the adaptable nature of the CTL response [37], combined with the fact that the majority of reports of CTL escape to date have focused on small numbers of individuals and/or a select few epitopes [3–6,18–20,30–32,35], it is not surprising that the clinical consequences of CTL escape remain incompletely understood. Some studies report an association between selection of escape variants and loss of viremia control [20,38] and disease progression [6,18]; however, this does not seem to equally apply to all CTL epitopes [35,39]. Here we observe a significant, dose-dependent inverse relationship between HLA-associated mutations within published epitopes in functional and accessory/regulatory genes and lower CD4+ cell counts in chronic untreated HIV infection, thus supporting a link between presence of escape mutations and HIV disease status. Although detection of escape mutations indeed preceded a loss of immune control in previous case reports [6,18], it is important to note that the cross-sectional nature of the current study precludes any inferences regarding cause and effect. Likely, a longer duration of infection (among those with lower CD4+ cell counts in this cohort) may have facilitated the accumulation of CTL escape variants, a hypothesis we were unable to investigate, because seroconversion dates were generally unknown. Other limitations of this analysis include the inherent limitations associated with the use of a single CD4+ cell measurement in a cross-sectional study design, the lack of longitudinal HIV sequence data, as well as the fact that the cohort represents a group of individuals referred for antiretroviral treatment, and thus may be biased toward more rapid progression to disease. Despite these limitations, our findings support those of Moore et al. [2] who reported that HLA-associated polymorphisms in RT predicted plasma viral load (CD4+ cell counts were not investigated). At first, results appear inconsistent with those of Iversen et al. [35] who reported higher viral loads in patients with efficient CTL selection; however, results may be reconciled by the fact that the previous study [35] investigated clinical correlates of escape to a single HLA-restricted epitope, whereas the current study evaluates HLA-associated polymorphisms across multiple genes. Ideally, however, the relationship between selection of HLA-associated escape mutations and HIV disease progression should be addressed in an unbiased, longitudinal cohort study of untreated HIV-1 seroconverters for whom infection dates, viral loads and CD4+ T cell setpoints, and rates of disease progression are known.
Although a systematic in vitro characterization of novel CTL epitopes was beyond the scope of this manuscript, the observation that a substantial number of motif-associated polymorphisms are supported by HLA-restricted, peptide-specific IFN-γ responses in an ELISpot assay suggest that they represent escape mutations within uncharacterized epitopes [16]. As the locations of published epitopes tend to be biased toward conserved regions (due to the historic use of consensus or reference strains to construct peptide libraries), the “motif maps” could complement traditional epitope mapping by identifying epitopes located in more variable regions.
After controlling for the potentially confounding effects of viral lineage [16], strong evidence for HLA class I-mediated selection is observed across functional and accessory/regulatory HIV-1 genes, with up to 40% of residues in some HIV proteins (Nef, for example) exhibiting evidence for HLA-restricted immune selection. Our results thus confirm an active and substantial contribution of human immunogenetic selection pressure on viral evolution [2] and underscore the importance of understanding how HLA class I diversity drives HIV diversity. The observed correlation between the presence of HLA-associated CTL escape mutations and lower CD4+ cell counts supports the hypothesis that maintenance of effective CTL responses plays an important role in immune control of HIV infection, although further research in additional cohorts is needed. The observation that epitope anchor residue mutation appears not to be the predominant mechanism of CTL escape supports the incorporation of HIV sequence diversity in the development of preventative and therapeutic CTL-based vaccine approaches.
In British Columbia (BC), antiretroviral drugs are distributed free of charge to HIV-infected individuals through a centralized drug treatment program (for details, see [21]). The HAART Observational Medical Evaluation and Research (HOMER) cohort is an open cohort comprising all HIV-infected, antiretroviral-naïve adults who initiated HAART since August 1996 (n > 2,200 individuals enrolled to date). A subset of HOMER, comprising all treatment-naïve individuals who initiated HAART in BC between August 1996 and September 1999 (n = 1,191) has been described in detail previously [21]. Participants in the current cross-sectional study represent a nonrandom subset (n = 765; 64%) of these 1,191 individuals at baseline (prior to initiation of HAART) included based on the availability of a peripheral blood sample for HLA typing. A comparison of pre-therapy characteristics of those included (n = 765) and excluded (n = 426) reveals no significant differences in pretherapy CD4+ cell count (280 cells/mm3); however, those included had slightly lower pretherapy plasma viral load (pVL) (median 5.07 versus 5.15 log10 copies HIV RNA/ml, p = 0.03), were on average slightly older (median 37.2 versus 36.5 y, p = 0.02), and were more likely to be male (median 88% versus 77% male, p < 0.0001) than those excluded. CD4+ cell count, plasma viral load, and HIV genotype data for each participant represent the latest pre-therapy measurement collected within 180 d prior to HAART initiation. Ethical approval for this study was granted by the Providence Health Care/University of British Columbia Research Ethics Board.
HIV RNA was extracted from a single pre-therapy (“baseline”) plasma sample using the QIAGEN (http://www.qiagen.com) viral RNA kit using a BioRobot 9600/9604 or extracted manually using guanidinium-based buffer followed by isopropanol/ethanol washes. The HIV protease (codons 1–99, HXB2 nt 2253–2549), RT (codons 1–400, or for ~25% of sequences codons 1–240 only; nt 2550–3749 or 2550–3269, respectively), Vpr (codons 1–96; nt 5559–5847), and Nef (codons 1–206; nt 8797–9414) were amplified using nested RT-PCR, and “bulk” sequenced in both the 5′ and 3′ directions on an ABI 3700 or 3100 (http://www.appliedbiosystems.com) automated DNA sequencer. HIV sequence data were analyzed using the software Sequencher (Genecodes, http://www.genecodes.com). Nucleotide mixtures were called if the height of the secondary peak exceeded 25% of the height of the dominant peak. Sequence data were aligned to HIV-1 subtype B reference strain HXB2 (Genbank accession number K03455) using a modified NAP algorithm [40]. HIV subtyping was performed by comparing HIV sequence data across HIV protease, RT, and Nef to all known subtype reference sequences in the Los Alamos HIV sequence database (http://hiv-web.lanl.gov/content/hiv-db/mainpage.html). Of total participants, 97.5% harbored subtype B infections. Consensus sequences reflecting the most common amino acid at each codon were generated from pretherapy sequences: these differed from the 2005 HIV-1 subtype B consensus at 1/99 protease, 7/400 RT, 2/96 Vpr, and 7/206 Nef codons, respectively. Genbank accession numbers of all unique HIV sequences used in this study are listed in Text S1.
Sequence-based typing (SBT) for HLA-A, B, and C was performed on DNA extracted from a PBMC-enriched frozen blood sample for each participant (n = 765). The SBT protocol is a validated “in-house” procedure based on International Histocompatibility Working Group (IHWG) protocols and involves independent, locus-specific, nested PCR amplification of exons 2 and 3 of HLA-A, B, and C followed by automated bidirectional DNA sequencing. Allele interpretation was performed by comparing SBT data against all alleles listed in the IMGT/HLA database (ftp://ftp.ebi.ac.uk/pub/databases/imgt/mhc/hla/) as of August 2005 (Release 2.10). This yields intermediate-to-high level resolution of HLA allele combinations. In order to achieve appropriately sized groups for statistical analysis, HLA alleles were summarized to two-digit resolution; note however that this approach may group together alleles which bind slightly different peptides, thus potentially reducing power to detect HLA-associated polymorphisms in some cases. Ambiguous allele combinations were resolved through incorporation of published allele frequencies and/or haplotype data. HLA-A and B typing was completed for all 765 participants, while HLA-C types were determined for 706 individuals. Although complete ethnicity data are unavailable, class I allele frequencies were consistent with those expected in a predominantly North American white population.
In order to discriminate between associations likely attributable to viral lineage effects and those that provide evidence for HLA-associated escape or reversion, we adopted the phylogenetically corrected analysis methods described in detail in [16]. Briefly, we used cohort HIV sequences to construct maximum likelihood phylogenetic trees (one for each gene). Since HLA types are available only for the infected individuals sampled, whose sequences form the tips of the tree, we used a maximum likelihood estimate of the sequence at the parental (interior) node proximate to each observation, and counted inferred escape or reversion in these last branches as independent events to be correlated with the HLA of the infected person at the terminal sequence by a Fisher exact test (method 1); alternatively, we used a likelihood ratio test to evaluate whether a model incorporating the effect of HLA association in addition to the phylogenetic structure was significantly better at explaining the data (method 2) [16]. The final list of identified associations represents the union of associations identified by both methods (Table S1).
In order to adjust for multiple comparisons, a q-value approach [22], rather than a Bonferroni correction [41], was employed: whereas a Bonferroni correction attempts to limit the probability of even a single false positive (and thus increases the rate of false-negative results), the q-statistic sets the proportion of false positives among results identified as significant (the false-discovery rate), an approach which we believe to be more appropriate for gene-wide association scans such as the present one. Associations with q < 0.2 (indicating a ~20% false-discovery rate) are presented; in this dataset this corresponded to unadjusted p-values 0.0055 > p > 3.3x10−45 for all genes. Note that the results of the lineage-corrected analysis groups associations into two broad categories based on the direction of the HLA selection pressure. Positive correlations, in which the presence of a specific HLA is associated with the presence of a particular amino acid—or, correspondingly, where the absence of the allele is associated with the absence of the amino acid—are termed “escape” associations, as they presumably reflect the escape variant for that specific HLA allele. Negative correlations, in which the presence of a specific HLA allele is associated with the absence of a particular amino acid—or, correspondingly, where a specific amino acid is enriched in the absence of a particular HLA allele—are termed “reversion” associations. In this case, the “reversion” amino acid presumably reflects the immunologically susceptible (“wild-type”) form specific for that HLA allele, as well as represents the amino acid most likely to re-emerge following transmission to an individual lacking that HLA allele.
Associations were organized into gene-specific “immune escape maps” whose goal was to capture the minimum number of epitopes (known or putative) required to explain the data. Three sets of escape maps were generated based on the strength of biological evidence supporting each association. The highest level of support was granted to those associations that fell within or proximal to (± 3 aa) a published HLA-restricted epitope (defined as all HLA class 1-restricted ≤ 15-mer epitopes listed in the Los Alamos HIV Immunology database as of December 2006 [23]). HLA-matched associations that fell within these boundaries were grouped together. Note that the ± 3 aa proximal “window” was chosen to identify putative proteasomal processing escape mutations [24–26] based on evidence indicating that the majority of such mutations occur in the three amino acids immediately flanking the epitope [42]. The secondary level of support was granted to associations which fell within or proximal to a known HLA-restricted epitope anchor residue motif (using MotifScan, http://hiv-web.lanl.gov/content/immunology/motif_scan/motif_scan) and/or a putative HLA-restricted epitope identified by an independently validated CTL epitope prediction algorithm (Epipred, http://atom.research.microsoft.com/bio/epipred.aspx [43]) based on scanning the cohort consensus sequence. Again, associations falling within the “motif ± 3 aa flanking window” were grouped together.
If specific amino acid variants were associated with additional HLA alleles in linkage disequilibrium (LD), these alleles were also grouped together within the epitope or motif. To identify HLA alleles in LD, we investigated all possible pairwise allele combinations using a simple Fisher's exact test and conservatively defined all allele pairs with p < 0.05 (q < 0.2) as linked (Table S2). In cases where LD allele pairs were associated with variation at the same codon, the allele exhibiting the strongest association (as estimated by lowest p-value) was classified as the allele driving the association.
To provide in vitro functional support to identified associations, we drew upon a partially published ELISpot dataset of 372 HIV-1 infected, non-white individuals screened for HLA-restricted, CTL-mediated IFN-γ responses against set of 410 overlapping subtype B consensus peptides (OLP) 15 to 20 amino acids in length, spanning the whole expressed HIV-1 subtype B proteome [27]. Associations between possession of individual HLA alleles and responses to specific consensus peptides in the OLP set [27] were assessed by simple Fisher exact test. HLA allele/OLP associations with p < 0.05 were considered to be “significantly reactive” and thus indicative that an HLA-restricted CTL epitope lay in the boundaries of that OLP. HLA-associated polymorphisms identified in the present study that mapped directly within an HLA-specific reactive OLP were identified and annotated as “in vitro-supported” on the immune escape maps (green; Figures 2–5). Note that the differences in ethnic composition of the current and ELISpot-characterized [27] study populations may result in an underestimation of in vitro-supported associations, due to differences in cohort HLA composition and thus power to detect significant associations. |
10.1371/journal.pcbi.1002215 | Modelling Reveals Kinetic Advantages of Co-Transcriptional Splicing | Messenger RNA splicing is an essential and complex process for the removal of intron sequences. Whereas the composition of the splicing machinery is mostly known, the kinetics of splicing, the catalytic activity of splicing factors and the interdependency of transcription, splicing and mRNA 3′ end formation are less well understood. We propose a stochastic model of splicing kinetics that explains data obtained from high-resolution kinetic analyses of transcription, splicing and 3′ end formation during induction of an intron-containing reporter gene in budding yeast. Modelling reveals co-transcriptional splicing to be the most probable and most efficient splicing pathway for the reporter transcripts, due in part to a positive feedback mechanism for co-transcriptional second step splicing. Model comparison is used to assess the alternative representations of reactions. Modelling also indicates the functional coupling of transcription and splicing, because both the rate of initiation of transcription and the probability that step one of splicing occurs co-transcriptionally are reduced, when the second step of splicing is abolished in a mutant reporter.
| The coding information for the synthesis of proteins in mammalian cells is first transcribed from DNA to messenger RNA (mRNA), before being translated from mRNA to protein. Each step is complex, and subject to regulation. Certain sequences of DNA must be skipped in order to generate a functional protein, and these sequences, known as introns, are removed from the mRNA by the process of splicing. Splicing is well understood in terms of the proteins and complexes that are involved, but the rates of reactions, and models for the splicing pathways, have not yet been established. We present a model of splicing in yeast that accounts for the possibilities that splicing may take place while the mRNA is in the process of being created, as well as the possibility that splicing takes place once mRNA transcription is complete. We assign rates to the reactions in the pathway, and show that co-transcriptional splicing is the preferred pathway. In order to reach these conclusions, we compare a number of alternative models by a quantitative computational method. Our analysis relies on the quantitative measurement of messenger RNA in live cells - this is a major challenge in itself that has only recently been addressed.
| The splicing of precursor messenger RNA (pre-mRNA) is an essential process in the expression of most eukaryotic genes. The five small nuclear ribonucleoproteins (snRNPs) and the many non-snRNP-associated proteins that constitute the splicing machinery, assemble anew on each precursor RNA to form the spliceosome complex that catalyses the two chemical reactions of splicing [1]. Both the spliceosome components and the spliceosome assembly process are largely conserved between human and yeast. The complexity of the spliceosome is indicated by the 170 proteins that are associated with it [1]. Adding to the complexity, splicing may occur partly, or entirely, concurrently with transcription. In eukaryotes, the interaction of the spliceosome with the precursor RNA can be considered to be an allosteric cascade in which early recognition steps induce conformational changes required for subsequent steps and for catalytic activation (reviewed by [2]). However, the wealth of knowledge of molecular interactions, obtained mainly through extensive biochemical and genetic analyses, has yet to be formalised as a systems model of transcription and splicing.
Spliceosome assembly is thought to occur via a series of events with many points of regulation [3]. In the first step, U1 snRNP binds to the 5′ splice site (5′SS), followed by the U2 snRNP at the branchsite. The U4, U5 and U6 snRNPs join as a tri-snRNP complex and, after the association of other, non-snRNP proteins, the spliceosome complex is activated for the first chemical step of splicing. The 5′ splice site is cleaved and, simultaneously, the 5′ end of the intron becomes covalently attached to the branchsite to form a branched, lariat structure. In the second step, the 3′ splice site (3′SS) is cleaved, which excises the intron, and the exons are joined to produce the mature mRNA. Between the two steps of splicing, a conformational change is required in the catalytic centre of the spliceosome [4], and at several stages during the cycle of spliceosome assembly, splicing and spliceosome dissociation, proofreading mechanisms are thought to operate [5]. Nascent transcripts also have to be matured at their 3′ end, by cleavage and polyadenylation. Figure 1 A illustrates spliceosome assembly and the two steps of splicing for a pre-mRNA with one intron that has already been polyadenylated and released from the DNA template.
Splicing can also occur co-transcriptionally, prior to 3′ end maturation (Figure 1 B), and there is considerable experimental evidence for functional coupling of transcription, splicing and 3′ end maturation in vivo [6]–[12]. However, little is known about the impact of coupling on kinetic rates. Splicing has been modelled, but not to the same level of detail as transcription, and models of transcription have yet to fully incorporate the splicing reaction. Quantifying the dynamics of these processes remains a challenge [13], and modelling may have an important role to play in distinguishing functional dependencies from coincidental and contemporaneous effects, and in identifying and characterising the interactions that effect coupling.
Existing models of splicing have allowed splicing efficiency to be defined [14], and have shown that transcription by RNA polymerase II (Pol II) greatly increases splicing efficiency in comparison with transcription by T7 polymerase [15]. A correlation between splicing efficiency and the pausing of Pol II on short terminal exons has been reported [11]. Splicing has been represented as a single irreversible reaction that creates the product mRNA from pre-mRNA [11], [14], and as a single irreversible reaction that creates mRNA from the pre-mRNA+spliceosome complex [15]. To-date, steps one and two of splicing have not been modelled as separate reactions, nor have the co- and post-transcriptional splicing pathways been distinguished. Further insights into splicing can be expected by more detailed modelling and analysis.
As noted above, splicing can occur during messenger RNA transcription. Transcription begins with the assembly of the pre-initiation complex at the promoter. This complex includes Pol II, which, after initiation, begins the transcript elongation process that transcribes DNA into RNA. Early in elongation, the pre-mRNA is capped at its 5′ end by the capping enzymes. Elongation involves a sequence of many hundreds of individual polymerisation reactions, and hence the time required to complete the elongation of a transcript is predicted to have less variability than a single-step process with an equivalent rate [16], [17]. The mature 3′ end of the RNA is formed by an endonucleolytic cleavage at the so-called poly A site and the newly formed 3′ end is extended by polyadenylation (reviewed by [18]). The elongation process and the 3′ end formation steps can also be accounted for when modelling transcription [16].
The recruitment of Pol II enzymes and spliceosomal proteins are important steps in transcription and splicing, but are not believed to be rate limiting under normal conditions. Kinetic studies of Pol II complexes indicate that a minority of them are actively involved in transcription at any given time. The remainder move by diffusion through the nucleus [19], as do the product mRNA molecules [20]. Three kinetically distinct populations of Pol II have been identified at the site of transcription; those bound to the promoter, those initiating transcription, and those engaged in elongation [21]. The movement of the spliceosomal proteins that catalyse the splicing reactions can be modelled as Brownian diffusion [22]: these RNPs move continuously throughout the nucleus independently of transcription and splicing.
We have developed a stochastic model that represents splicing in the context of transcript elongation and RNA 3′ end maturation, as shown diagrammatically in Figure 1 C. (All pathway models are provided as files in Dataset S1.) A stochastic formulation allows the effects of small numbers of molecules to be explored, and simulations of the model can be averaged in order to obtain the population mean over time. Experimental values for the model species (population averages in copies/cell), including fully-spliced mRNA (see Materials and Methods) and two precursor species in both 3′ uncleaved and cleaved/polyadenylated forms, have been obtained by a rapid sampling protocol that is capable of capturing transient species [23]. We first describe the structure of the pathway, then present the data, and subsequently discuss alternative representations of the steps in the RNA pathway in the light of the data. The simplest description that might be adopted for the elongation, 3′ end formation and splicing steps is a single irreversible reaction. However, we find this provides a poor fit to the available data, and consequently a number of alternative representations for these reactions are considered. The extent to which the alternative pathways fit the data is assessed by the Akaike information criterion (AIC) for optimal parameter choices.
We propose a multistep model for transcription by dividing the gene into sections to be transcribed. Each section () of the reporter DNA represents approximately 30 nucleotides, corresponding to the footprint of Pol II on the DNA [24]. As the length of the Ribo1 reporter (described below) used in the experimental studies is 1240 bases, we define 40 sections of DNA: . Each section of DNA can be occupied by at most one Pol II, and the progression of Pol II from the 5′ to the 3′ end of the gene is equated with successful extension of the transcript. The number of sections of DNA defines an upper limit on Pol II occupancy, and can limit the effective rate at which a Pol II can complete elongation. Beginning with the initiation of transcription, the reaction (see Figure 1 C) places a Pol II enzyme in the active promoter complex (APC) when the gene is active. Thereafter, this Pol II can progress along the gene at elongation rate (the number of sections of DNA transcribed per unit of time). Letting the rate of polymerisation of nucleotides be (the number of nucleotides incorporated per unit of time): . (Equivalently, the mean time for n polymerisation events: , equals the mean time for one elongation event: ). This multistep model of elongation is comparable with the kinetic model of Pol I elongation proposed in [25]. The pathway proposed here does not include a transition between active and inactive states of the promoter, as the rapid rate of mRNA production does not indicate that the promoter switches off during the period immediately after induction. However, such a transition is needed to explain the mRNA distribution in steady state [26] and can easily be included in this model.
Kinetic competition between splicing and elongation has been discussed extensively [8], [27], [28], and is modelled here as taking place at the sections of DNA after the branchsite. In these sections, the occurrence of the first step of splicing of an RNA is represented in the model by a change of state of the associated Pol II, which can make a transition to the co-transcriptional splicing path . Sections and represent the same n nucleotides of the DNA and so at most one of these sections can be occupied (by at most one Pol II). The rate for the transition between splicing pathways is 0 prior to the completion of the splicing activation process. The splicing activation process is triggered at rate when the gene switches on. When splicing is active, the transition rate is , where is a constant that determines the ratio of the competing reactions (elongation and splicing) and thereby the probability of co-transcriptional splicing. Activation of co-transcriptional splicing involves co-transcriptional spliceosome assembly as well as the first step splicing reaction (i.e. co-transcriptional spliceosome assembly alone is not sufficient). Each Pol II completes elongation either at , having completed the first step of splicing, or at having failed to do so. Subsequently, on the post-transcriptional path, 3′ end maturation () produces polyadenylated pre-mRNA, step one of splicing () produces polyadenylated lariat-exon2, and step two of splicing () produces mature mRNA and lariat, as indicated in Figure 1 C. On the co-transcriptional pathway, the second step of splicing () produces uncleaved mRNA and lariat, and 3′ end maturation (3′ cleavage, polyadenylation and release; ), produces mature mRNA. It is important to note that the species measured experimentally are pre-mRNA, lariat-exon2 (the branched lariat structure) and mRNA, and that the uncleaved and polyadenylated forms can also be distinguished (as illustrated in in Figure 1 C). The assays for these species are described below.
Initial estimates for some parameters can be obtained from the literature: the rate of initiation of transcription in yeast has been estimated as [29], [30]. Polymerisation rates in the mammalian nucleus of up to 72 nucleotides/s have been reported for polymerases that do not pause. This is a significant increase on earlier estimates of 18–42 nucleotides/s [13] that may reflect an average or effective rate. A Pol I elongation rate of 90 nucleotides/s has been reported [25]. The time for pre-mRNA 3′ cleavage in HIV-1 has been reported to be 55 s, with release taking 9 s [16]. The probability of co-transcriptional splicing is not known, and this, along with precise values for all other parameters, will be inferred from fitting the pathway to the data.
The pathway was developed to explain data from the Ribo1 reporter [23]. Ribo1 is a chimeric yeast gene that contains the single intron from the ACT1 gene and the 3′ end processing signal from PGK1, as shown in Figure 2 A (modified from [31]). The reporter gene is integrated in the genome, transcribed under the control of a doxycycline-responsive promoter in a doxycycline-inducible strain of Saccharomyces cerevisiae. By modelling splicing in this reporter, we aim to define the splicing pathways and to quantify reaction rates. The impact of splice site mutations on the coupling between splicing and transcription can also be explored.
Three replicate experiments were performed in which doxycycline was added to a culture to induce reporter gene expression, and transcript levels were measured by reverse transcription and real-time quantitative PCR (RT-qPCR; see Materials and Methods). A time series of values was obtained for accumulation of pre-mRNA, lariat-exon2, and mRNA. The RT-qPCR data were converted to copies per cell (see Materials and Methods; [23]), which allows a quantitative comparison of data obtained for the different RNA species and from different cultures. The merged time series derived from three biological replicates for Ribo1 is shown in Figure 2 B (referred to as Expt 1). In the 120 s interval 420 s–540 s after the addition of doxycycline to the cell culture, the level of Ribo1 mRNA increases from 11 to 45 copies/cell (Figure 2 B). Messenger RNA then reaches 60 copies/cell, on average, 180 s later. The high level of mRNA is notable, as is the rapid rate of transcript synthesis. The delay between the rise in pre-mRNA and the rise in mRNA may indicate a slow, or delayed, splicing reaction. In the substantive phase of transcriptional activity (after 420 s in Figure 2 B), the levels of pre-mRNA and lariat intermediate are only a fraction of the mature mRNA species which shows that the first and second steps of splicing must be rapid.
To investigate the effects of blocking the first or second step of splicing, two modified Ribo1 reporters were created with point mutations at the 5′ splice site (5′SSRibo1) or 3′ splice site (3′SSRibo1) respectively [23]. The mutant reporters were induced with doxycycline and the splicing intermediates detected using the primers shown in Figure 2 A. The merged time series for 3′SSRibo1 and 5′SSRibo1 are shown in Figure 2 C and 2 D respectively. As indicated by the error bars in Figure 2 C, the synthesis of lariat-exon2 in the mutant reporter varied between biological replicates, but technical error within each replicate remained at typical levels. The level of pre-mRNA measured in the modified reporters is greater than was observed for Ribo1. This may be attributed to changes in the rates of transcript synthesis, splicing step one or degradation, and modelling can help to resolve this question.
For practical reasons, co-transcriptional splicing is defined here as splicing that is completed before the transcript has been released from the transcription complex by 3′ end cleavage. The data shown in Figure 2 were produced using a cDNA primer that hybridises to exon2 (at position C1 in Figure 2 A), which does not distinguish between transcripts that are cleaved and polyadenylated or uncleaved at the 3′ end. Therefore, in order to differentiate between these species and to estimate the rates for 3′ end formation, a modified 3′ cleavage assay used two alternative primers for cDNA synthesis from 3′ end sequences of Ribo1; oligo (dT), anneals to cleaved and polyadenylated transcripts, whereas primer C2 is complementary to a sequence downstream of the mapped 3′ end cleavage sites (Figure 3 A; [23]). By amplifying these cDNAs with specific primers for detection of pre-mRNA, lariat-exon2 and mRNA (Figure 3 A), uncleaved and cleaved/polyadenylated pre-mRNA, lariat-exon2 and mRNA were successfully distinguished in Expt 2 (Figure 3 B and 3 C).
The 3′ cleavage assay detected a sharp, transitory peak in uncleaved pre-mRNA at 540 s, followed by a similar peak in polyadenylated pre-mRNA 30 s later (Figure 3 B). This indicates pre-mRNA that is not spliced prior to 3′ end cleavage, i.e. is not spliced co-transcriptionally. However, the rapid accumulation of uncleaved mRNA between 540 and 600 s prior to detection of polyadenylated spliced mRNA at 600 s, clearly shows that co-transcriptional splicing occurs before post-transcriptional splicing. By formally modelling the splicing pathway, we aim to quantify the extent to which mature mRNA is derived from co-transcriptional splicing, and from post-transcriptional splicing respectively.
The reactions in the model must be enabled (switched on) progressively in order to explain the data. Following the induction of transcription by doxycycline, a burst of pre-mRNA is postulated to occur. At this time, splicing is not active, and additional transcripts are not initiated. These initial pre-mRNAs are cleaved and polyadenylated, and may then splice or degrade. This process explains the accumulation of pre-mRNA in Figure 2 B, and the peak in uncleaved pre-mRNA in the 3′ cleavage data in Figure 3 B. After a delay (defined by the rate ), splicing steps 1 and 2 and the initiation of new transcripts start. This explains the drop in pre-mRNA in Figure 2 B, and the peak in polyadenylated pre-mRNA (Figure 3 B) as the activation of splicing removes these species also. Figure 1 in Text S1 illustrates the sequence of events.
The proposition that there are advantages to modelling elongation in detail can be tested. Pathways that include 40 steps of elongation are compared with simpler pathway models where competition between elongation and splicing step one is represented by two reactions and that have APC as the substrate and and as the respective products. The proportion of co-transcriptional splicing can be calculated from these reaction rates and this proportion can be compared with that predicted for the 40 step model (as defined by equation 1 in Materials and Methods). The total time allocated to elongation can also be compared in the alternative models.
On completion of elongation, the RNA transcripts undergo 3′ end formation. This involves cleavage, polyadenylation and transcript release, and requires three multi-subunit factors [32]. Polyadenylation adds up to approximately 250 nucleotides to the end of the transcript. Hence, it is uncontroversial to model 3′ end formation as a multi-step process as many steps of maturation are clearly required. When fitting the splicing pathway to the Ribo1 data, a much better qualitative and quantitative fit is obtained when 3′ end maturation is modelled as a five-step process (each of the five steps has rate ) in comparison with a single step model. The characteristics of 3′ uncleaved spliced RNA also fit the data better by modelling 3′ end formation in this way. As shown in Figure 3 C, uncleaved mRNA quickly peaks towards its steady state of 10 copies/cell rather than making a slow progression to this level. Replacing the single step model with the 5 step model of 3′ end maturation (reactions and ) significantly improves the fit to the data. It is easily shown that a process of five steps, each at the same rate, has a kinetic response that differs significantly from that of a single step. (The distribution of waiting times follows a gamma distribution rather than an exponential distribution.) We do not aim to determine the exact number of steps, rather we aim to test whether a process of multiple steps of maturation or senescence provides a better quantitative and qualitative explanation than a single reaction. Henceforth, we assume that 5 steps constitute an adequate model of a multi-step process for the purpose of testing this hypothesis. Text S2 presents an analysis of Ribo1 degradation kinetics that further illustrates the approach.
Genetic studies have identified many splicing factors, but their impact on splicing kinetics in-vivo is difficult to quantify. These factors, and the five snRNPs, are not believed to be rate-limiting and have not been included in the model: We initially consider the kinetics of the splicing intermediates alone. However, we find once more that simple unimolecular models for steps one and two of splicing do not fit the data well. Consequently, we propose two alternative characterisations of the splicing reactions prior to steady state, and quantify the extent to which these models improve the fit of the pathway to the data.
The first alternative model of splicing we propose represents these processes as a sequence of several reactions that reflect the many known steps of spliceosome assembly. The precursors and products of multi-step processes show sharp transitions in their kinetics, as observed for pre-mRNA and lariat-exon2 in the experimental data. A multi-step model of this kind has been shown to explain fluorescence recovery after photobleaching (FRAP) data obtained from a splicing reporter in mammalian cells [33].
The second alternative explanation of the rapid processing of pre-mRNA and lariat-exon2 that we propose is based on the proposition that the splicing reactions are catalysed in a manner such that the propensity of the reaction increases on successive splicing events. It is necessary for the initial propensity to be low as otherwise no accumulation of splicing precursor or lariat intermediate would be observed, and for the propensity to increase to remove the accumulation rapidly. The reduction of uncleaved lariat-exon2 over the period of time when mRNA increases rapidly (600 s-700 s in Figure 3 C) may indicate such an increase in reaction propensity: the substrate decreases while the rate of increase of product remains constant. It therefore appears that step two of splicing may not be governed by first-order kinetics when it is co-transcriptional. The observations can be modelled by positive feedback in the splicing reaction. This requires the involvement of additional molecular species in the splicing reaction - the enzyme Y - a role that can be played by factors required for step two of splicing.
The following positive feedback mechanism has the property of increasing reaction propensity: Let the enzyme Y have an initial copy number of 1, and increment the copy number on each splicing event to effectively increase the propensity. The enzyme contributes to the reaction propensity according to the formula for bimolecular reactions (). The positive feedback model is proposed for the kinetics of high rates of induction prior to steady state.
Due to the uncertainties in pathway structure, and parameter identifiability and estimation, it is of considerable value to explore multiple models of a biological system [34]. The goodness of fit to the experimental data of eight versions of the RNA processing pathway is compared in Table 1. The alternative pathways are distinguished by their representation of elongation, of co-transcriptional splicing step two, and of post-transcriptional splicing steps one and two. Elongation is modelled either as a single step or as 40 steps, as described above. The alternative models considered for the splicing reactions are: a single step, multiple steps (each at the same rate) or positive feedback. It is important to consider each pathway as a whole as the goodness of fit for each observed species is dependent on the reactions that act directly on the observed species, and those that act on its precursors and thereby shape the kinetics of the precursors. Table 1 defines each pathway and lists the AIC scores obtained using the optimal parameters (see Table 1 in Text S1 for the parameter values). Note that pathway slowromancap VIII makes the simplest assumptions about elongation and splicing steps, namely that they are single-step unimolecular reactions, and that the poor fit of this pathway to the data motivates the search for alternative models.
Pathway parameters were optimised by a simulated annealing algorithm (see Materials and Methods; [35]) that identified the best fit between each pathway and the nine data series obtained for Ribo1 (those plotted in Figure 2 B, 3 B and 3 C). The total AIC (defined in Materials and Methods) is calculated from the combined residuals from all species/experiments. All data and pathway models are provided as files in Dataset S1. An executable version of the Dizzy simulator [36] is also provided to allow the models to be executed.
The AIC scores for pre-mRNA, lariat-exon2 and mRNA are represented separately in the columns of the heat map in Figure 4. It is apparent from the A-pre-mRNA column that all pathways fit well to the pre-mRNA data in Expt 1, and fit to a comparable extent. The majority of pathways also fit the mature mRNA data well (A-mRNA and P-mRNA columns). Pathways I-IV can be optimised to the lariat-exon2 data simultaneously. In contrast, pathways V-VIII have a poor fit to one or more of the lariat-exon2 species. Pathway I has the best overall AIC as a result of fitting the nine data series most consistently.
Pathways I-IV incorporate the feedback mechanism for co-transcriptional splicing step two and this feature correlates with a good fit (low AIC) for all lariat-exon2 species. Within these pathways, a multi-step representation of post-transcriptional splicing, combined with a multi-step representation of elongation has the best overall score (pathway I). Feedback in post-transcriptional splicing, combined with a multi-step representation of elongation also explains the data well (III). Pathway VII is ranked in third place, failing to explain the polyadenylated lariat-exon2 data in Expt 2 (as indicated by the white cell in row VII, column P-lariat-exon2 in Figure 4). The predictions of the pathways for each of the nine species measured in Expt 1 and 2 are plotted in Figure 2 in Text S1. Important qualitative differences between the pathways can be seen in these graphs.
Pathways with a single elongation step require an initiation rate of 0.4, and a rate for elongation of 0.4–0.54, giving an implausible time of 2–3 s for the elongation of a 1240 nucleotide gene. As a consequence of defining a more realistic elongation time, pathways with a multi-step representation of elongation typically fit the data better, see Table 1.
In pathway I, pre-mRNA 3′ end maturation takes 35 s and uncleaved 3′ end mRNA maturation takes 49 s using the measure of the time taken for the sum of intermediate species undergoing the five steps of 3′ end processing to reduce by half (an equivalent to the half life of a single step reaction).
The completion of splicing co-transcriptionally in yeast has been a topic of debate. Genome-wide ChIP studies indicated that co-transcriptional spliceosome assembly may not have time to occur if the 3′ exon is short [28]. More recent studies provide evidence for polymerase pausing 3′ of introns, suggesting a mechanism to slow transcription, allowing more time for splicing [10], [11]. With Ribo1 we observe that the initial burst of 3′ uncleaved pre-mRNA is not spliced before it is 3′ end cleaved, as shown by the successive blue and purple peaks in Figure 3 B, and it may undergo post-transcriptional splicing. After this initial burst, the majority of transcripts splice co-transcriptionally, as seen by the accumulation of uncleaved lariat-exon2 and uncleaved mRNA prior to cleaved/polyadenylated mRNA (red, green and black, respectively in Figure 3 C). Optimisation of pathway I computes to be 11.39, and by substituting this value into equation 1 (see Materials and Methods) it follows that 12% of Ribo1 RNA transcripts splice post-transcriptionally, and 88% of transcripts splice co-transcriptionally. Values for in pathways III, V and VII imply the same proportion of co-transcriptional splicing, as do the values of and in the four remaining pathways where the proportion of co-transcriptional splicing is approximately 85%.
Pathways I-IV show a good qualitative fit to the uncleaved lariat-exon2 data (see Figure 2E in Text S1). All four pathways specify a positive feedback mechanism for with estimated rate constants in the range 0.0061–0.0068 (see Table 1 in Text S1). In pathway I , the half life of this reaction is 110 s for the first transcript to splice, and, with feedback, the half life reduces to 5.5 s at 670 s after induction. As, initially, the half life is much greater than the time to transcribe exon 2 (approximately 11 s), the decision to model the second step of splicing as a process that occurs after elongation is justified.
The feedback mechanism may be a result of the disassembly and recycling of the snRNPs of the spliceosome for subsequent rounds of splicing [37]. It has been proposed that the branchpoint binding protein (BBP) and Mud2 are recycled between two steps in pre-spliceosome assembly: BBP is released during or after the second step and efficiently recycled to promote the first [38]. The finding that snRNPs do not assemble on a nascent transcript in response to a signal, but move randomly [22], does not preclude them impacting on splicing kinetics in a transcription-dependent manner through an influence on rates of spliceosome assembly, disassembly or recycling. Maintenance of the transient Cajal body (responsible for the maturation of snRNPs) requires continuous recycling of pre-existing snRNPs after each round of spliceosome assembly [22], and may therefore be indirectly dependent on transcription when splicing is co-transcriptional. If recycling mechanisms existed for second step factors, increasing the effective second step reaction rate, this could explain the peak and dip in uncleaved lariat-exon2 in Expt 2. The allosteric effects of second step splicing factors would provide an alternative explanation.
Pathways I and III specify a multi-step representation of elongation and feedback in co-transcriptional splicing. They account for 99% of the probability mass available in the Akaike weight analysis (see Table 1). These two pathways differ on the post-transcriptional splicing mechanism: a multi-step representation is more probable (P = 0.845) but a feedback mechanism cannot be ruled out (P = 0.152). As pathway I has a better fit to the polyadenylated pre-mRNA and polyadenylated lariat-exon2 data (the precursor and products of post-transcriptional splicing), we tentatively conclude that post-transcriptional splicing has multi-step characteristics. The difficulty in modelling the post-transcriptional splicing process lies in its transient activation. The characteristic features of the feedback mechanism are not clearly revealed. For a multi-step model, the times for the sum of intermediate species undergoing splicing to reduce by half are 34 s for step one and 36 s for step two of splicing.
The 3′SSRibo1 data are explained by a variation of the transcription and splicing pathway where step one of splicing can be co-transcriptional or post-transcriptional (as in the full pathway), but where the lariat-exon2 species goes through the five-step cleavage process (at rate ) instead of step two of splicing ( = 0 and = 0). The pathway used to explain the 5′SSRibo1 data has no co-transcriptional splicing path ( cannot be reached), and no post-transcriptional splicing can occur ().
The induction of the 3′SSRibo1 reporter (Figure 2 C) shows a greater accumulation of pre-mRNA than observed for Ribo1. The lariat-exon2 product is not spliced, but accumulates and is subject to degradation. The data can be explained by pathway I using the rate inferred for Expt 1 and 2. However, to predict the pre-mRNA response is increased to 30, is reduced to 0.015, and is reduced to 0.175. The probability of step 1 occurring co-transcriptionally is therefore reduced to 56% compared with 88% in Ribo1, the time taken for splicing to become active increases two fold, and the rate for the initiation of transcription reduces to 70% of the rate in Ribo1. The prediction for lariat-exon2 is greater than observed, and this may indicate that 3′ end maturation and/or lariat-exon2 degradation pathways differ in the mutant reporter.
The induction of 5′SSRibo1 (Figure 2 D) shows that pre-mRNA accumulates and does not splice. The response can be explained by further reduction in to 0.1, that is, 40% of the rate in Ribo1.
The induction of 3′SSRibo1 was repeated using the primers of Expt 2 in order to validate the finding that the probability of co-transcriptional splicing is reduced. The new data are shown in Figure 3A in Text S1. The pathway model predicts only a slow removal of the accumulating uncleaved pre-mRNA (and consequently of polyadenylated pre-mRNA) that is consistent with the new data. In contrast, the large reduction in pre-mRNA that is predicted when the rate for is 11.39 (as inferred for Ribo1) does not fit the new data, see Figure 3B in Text S1.
The overestimation of lariat-exon2 by the model (Figure 2 C) might be explained by a significant underestimation of the degradation rate for this species. This rate has been determined in the 3′SSRibo1 ‘OFF’ strain where transcription is halted by doxycycline, see Text S2. (A second experiment using alternative primers confirms this result [23].) Alternatively, the assumption made when modelling 3SSRibo1 that uncleaved lariat-exon2 would be able to complete 3′ end maturation and contribute to the total population of polyadenylated lariat-exon2 may be incorrect. Modelling shows that polyadenylated lariat-exon2 may be the product of polyadenylated pre-mRNA alone, with no contribution from the co-transcriptional pathway.
Despite the biochemical and genetic evidence for multiple steps in the cycle of splicing events, previous in vivo studies of mRNA splicing kinetics have revealed simple first-order monomolecular reactions that exclude the action of a catalyst. The allosteric cascade is yet to be revealed at the systems level, either in terms of the existence of multiple steps, or the impact of enzyme kinetics, and we argue that this is due to the course-graining phenomena associated with stochastic processes [39] and to the lack of experimental quantification of mRNA and its precursors.
Using rapid sampling of cultures, combined with RT-qPCR assays that detect the intermediates and products of the splicing reaction in a way that permits quantitative comparisons between different RNA species and between different cultures, we are able to present kinetic data with an unprecedented level of resolution, monitoring pre-mRNA production, the two steps of splicing and 3′ end processing of a reporter transcript in yeast. Our data cannot be explained satisfactorily by single-step unimolecular splicing reactions. We conclude that a systems model of transcription and splicing must distinguish the two steps of splicing, account for their occurrence co- and post-transcriptionally, represent spliceosome assembly, and include the action of an additional partner in the splicing reactions, as we find evidence in the data for each of these processes.
While developing the model, we considered including a transition from uncleaved lariat-exon2 to polyadenylated lariat-exon2, which would permit pre-mRNAs that have already undergone the first step of splicing co-transcriptionally to undergo 3′ end maturation. However, when this transition was added to model I, the AIC was found to increase by 1.4 (after optimisation), meaning model I fits the data better without the additional transition. The proposed transition occurs very slowly, and consequently rarely, does not assist modelling the data, and, therefore, was excluded from the models we analysed further.
The model proposed here specifies that pre-mRNAs that have already undergone the first step of splicing co-transcriptionally will be fully spliced co-transcriptionally prior to 3′ end cleavage. This is in contrast with the mammalian model proposed in [40] where splicing is completed after 3′ cleavage (in HeLa nuclear extracts). Both models stipulate that partially-spliced transcripts are not released from the DNA, and both allow for a post-transcriptional splicing pathway. Our model is consistent with the recycling of splicing factors [3]. Recycling of BBP and Mud2 has been proposed for pre-spliceosome formation [38], and similar mechanisms may exist for subsequent spliceosome assembly steps. Alternatively, it has been proposed that an increase in the local concentration of splicing factors is linked to transcription via the C-terminal domain of Pol II [15]. Nuclear speckles may also have a role in keeping spliceosomal components concentrated near nascent transcripts [37]. Cooperativity in the interaction of splicing factors with the spliceosome or with the nascent pre-mRNA may also contribute to the kinetics of co-transcriptional splicing.
Addressing the interdependency between RNA processing steps, modelling indicates that mutations at the 3′ and 5′ splice sites reduce the rate of initiation of transcription, and, in the 3′SS mutant, reduce the probability of step one of splicing occurring co-transcriptionally. Quantitative analysis of the mutant data requires establishing a parameterised model for the ‘wild type’ in order to define and test the alternative explanations of the differences observed.
A half life for splicing in HeLa cell nuclear extracts of 23 min (splicing rate of 0.03/min) has been reported [15]. In vivo half-lives of 6–12 min have been reported in mammalian cells [41], as have estimates of 5–10 min for the completion of splicing after intron synthesis [42]. Half lives for splicing in the range 0.4–7.5 min have also been reported for the splicing of introns in mammalian cells [43]. The inferred rates for post-transcriptional splicing in Ribo1 equate to half lives of 0.6 min for each of steps one and two, and are at the faster end of the spectrum reported in [43]. On the co-transcriptional pathway, splicing step one is concurrent with the transcription of the 800 bases from the branchsite until the polyA site (taking approximately 11 s). Co-transcriptional step two occurs with a half life of 110 s for the first transcript, and, with feedback, the half life reduces to 5.5 s at 670 s after induction. Therefore co-transcriptional splicing is the more efficient pathway under the high induction conditions studied here.
This study proposes a mechanistic kinetic model that represents some of the complexity and flexibility of the splicing pathway that is known from biochemical and genetic studies [3]. Co-transcriptional splicing is evident in the data, and modelling shows that this pathway may be activated after a delay. Furthermore, the second step of splicing benefits from positive feedback when co-transcriptional. These could be explained by the coordination of splicing factor recruitment/recycling with transcription, possibly facilitated by polymerase pausing [10], [11] and/or dynamic chromatin modification [9], [44], [45].
To analyse the transcription, splicing, degradation and 3′ end formation of yeast pre-mRNA, the Ribo1 reporter was integrated into the yeast genome at the his3 locus. The reporter is based on a hybrid ACT1/PGK1 gene [31], modified as described in [23] by inserting two copies of the boxB sequence (57 bp each) in the ACT1 intron, enabling it to be readily distinguished by RT-qPCR from the endogenous ACT1 intron without affecting splicing.
Primer pairs were created to measure the unspliced pre-mRNA (5′ primer upstream of ATG, 3′ primer over the exon 1 - intron junction), the lariat-exon2 intermediate (5′ primer upstream of 3′end of intron, 3′ primer over exon 2; the pre-mRNA level was subtracted from this measurement) and the spliced mRNA (5′ primer upstream of ATG, 3′ primer over exon 2). Measurements of mRNA in copies per cell, averaged over a population, were obtained by carefully quantifying the efficiency of cell lysis, recovery of RNA, reverse transcription and qPCR. For full details see [23].
The first step of post-transcriptional splicing, and all transitions to the path, decrease pre-mRNA and increase lariat-exon2. The second step of splicing decreases lariat-exon2 and increases spliced mRNA, according to the pathway. All species in the pathway, with the exception of the excised intron product of step two, are measurable by RT-qPCR, provided that they extend beyond the position of the cDNA primer. Splicing events on transcripts that have not been elongated to the cDNA point are not detected until this sequence is transcribed, and the calculation of RT-qPCR signal intensity from the species in the pathway reflects this. For example, the (simulated) pre-mRNA signal is not incremented until the species is incremented, despite the PCR primers for pre-mRNA being located several hundred bases upstream.
Considering a single Pol II complex (and ignoring the effect of other Pol IIs on its movement), the probability of transitioning from states to the co-transcriptional path is simply calculated from the elongation rate and the transition rate. This choice can be made 25 times, allowing the probability of the Pol II exiting on the post-transcriptional pathway to be estimated independently of by:(1)
Unless otherwise stated, reaction rates are expressed as the probability density per unit time, per distinct combination of reactant molecules. Where there is a single reactant species, the number of distinct combinations is just the population of reactants. The half life is the time a molecular species takes to reduce by half, and is computed for unimolecular reactions by in units of seconds.
Pathway models were optimised by the simulated annealing algorithm specified in [35] (see Figure 1). Following [35], the error E is defined by equation 2 where S is a time series simulated from a pathway model, D is the observed data, n is the number of time points and d the number of dependent variables (the dimension of and is d).(2)On each iteration of the algorithm, each parameter is assigned a new value () and the error for the new set of parameters () is calculated from a simulation of the model using the updated parameter set. The new parameter value is always accepted if , otherwise it is accepted with probability , where T is the current temperature and E is the error of the current parameter set. The new parameter value is generated from the current value by adding a normally-distributed random value. We define the scale constant k in equation 3 using the error of a set of parameter values that are given as input at the start of optimisation (these must provide an approximate fit to the data), and then update each parameter value according to equation 4, where N(0,1) is a normally-distributed random value (mean 0, standard deviation 1) and and are the maximum and minimum values respectively that is allowed to take. See [35] for further details.(3)(4)
The Akaike information criterion (AIC; eqn. 5) was used to assess the fit between a time series S simulated from a pathway model of k optimised parameters, a data set D of n values [46]. Assuming normally distributed errors, AIC can be computed from the model residuals (eqn. 6) [47]. The values for total AIC incorporate the 2 k penalty for the number of parameters optimised.(5)(6)When comparing m pathway models, the Akaike weight w of model i can be defined in terms of the relative likelihood , where is the difference between the AIC for model i and the AIC of the best model [47]. Akaike weights computed by equations 7 and 8 are listed in Table 1 .(7)(8)
|
10.1371/journal.pntd.0003879 | Diagnostic Accuracy of Recombinant Immunoglobulin-like Protein A-Based IgM ELISA for the Early Diagnosis of Leptospirosis in the Philippines | Leptospirosis is an important but largely under-recognized public health problem in the tropics. Establishment of highly sensitive and specific laboratory diagnosis is essential to reveal the magnitude of problem and to improve treatment. This study aimed to evaluate the diagnostic accuracy of a recombinant LigA protein based IgM ELISA during outbreaks in the clinical-setting of a highly endemic country.
A prospective study was conducted from October 2011 to September 2013 at a national referral hospital for infectious diseases in Manila, Philippines. Patients who were hospitalized with clinically suspected leptospirosis were enrolled. Plasma and urine were collected on admission and/or at discharge and tested using the LigA-IgM ELISA and a whole cell-based IgM ELISA. Sensitivity and specificity of these tests were evaluated with cases diagnosed by microscopic agglutination test (MAT), culture and LAMP as the composite reference standard and blood bank donors as healthy controls: the mean+3 standard deviation optical density value of healthy controls was used as the cut-off limit (0.062 for the LigA-IgM ELISA and 0.691 for the whole cell-based IgM ELISA). Of 304 patients enrolled in the study, 270 (89.1%) were male and the median age was 30.5 years; 167 (54.9%) were laboratory confirmed. The sensitivity and ROC curve AUC for the LigA-IgM ELISA was significantly greater than the whole cell-based IgM ELISA (69.5% vs. 54.3%, p<0.01; 0.90 vs. 0.82, p<0.01) on admission, but not at discharge. The specificity of LigA-IgM ELISA and whole cell-based IgM ELISA were not significantly different (98% vs. 97%). Among 158 MAT negative patients, 53 and 28 were positive by LigA- and whole cell-based IgM ELISA, respectively; if the laboratory confirmation was re-defined by LigA-IgM ELISA and LAMP, the clinical findings were more characteristic of leptospirosis than the diagnosis based on MAT/culture/LAMP.
The newly developed LigA-IgM ELISA is more sensitive than the whole cell-based IgM based ELISA. Although the final diagnosis must be validated by more specific tests, LigA-IgM ELISA could be a useful diagnostic test in a real clinical-setting, where diagnosis is needed in the early phase of infection.
| Leptospirosis is an important but largely under-recognized public health problem in the tropics. A sensitive, specific and practical point-of-care laboratory test is needed to improve clinical management in resource-limited settings. We developed a recombinant LigA protein based IgM assay (IgM ELISA) and evaluated the diagnostic accuracy of the test during outbreaks in the clinical-setting of a highly endemic country, the Philippines. Sensitivity and specificity of LigA-IgM ELISA and the whole cell-based IgM ELISA, which is included in WHO guidance, were evaluated with cases diagnosed by the standard serological test, MAT, culture and loop-mediated amplification (LAMP) as the composite reference standard and blood bank donors as healthy controls. The sensitivity and the area under the receiver operating characteristic curve for the LigA-IgM ELISA was significantly greater than those of the whole cell-based IgM ELISA on admission. When we re-defined the laboratory confirmation by LigA-IgM ELISA and LAMP, the clinical findings were more characteristic of leptospirosis than the diagnosis based on MAT/culture/LAMP. Our results suggest that LigA-IgM ELISA could be a useful diagnostic test in a real clinical-setting and help timely initiation of antibiotics to prevent severe outcomes of leptospirosis.
| Leptospirosis is a ubiquitous zoonosis caused by over a hundred serovars of the pathogenic spirochetes, Leptospira spp. The disease is an important public health problem in low- and middle-income countries in the tropics, especially in Southeast Asia and Latin America, including the Philippines [1–3]. According to the World Health Organization (WHO)-International Leptospirosis Society (ILS), the estimated annual number of leptospirosis cases is 350,000–500,000 [4,5] and annual incidence is estimated from 0.1–1 per 100,000 in temperate climates to 10–100 per 100,000 in the humid tropics [6]. The importance of leptospirosis has been largely under-recognized for several reasons. The disease causes a wide range of clinical manifestations, which mimic other tropical infectious diseases such as malaria, dengue fever, typhoid fever and other hemorrhagic viral diseases [6]. The disease severity also varies from a self-limited non-specific febrile illness to fatal cases. A sensitive, specific and practical point-of-care laboratory test has not been available. The current composite reference standard for diagnosing leptospirosis is culture and microscopic agglutination test (MAT) but these tests are laborious and require a special laboratory facility and skills to interpret results [7]. Establishing a highly sensitive, specific, and practical laboratory-based diagnosis for leptospirosis is essential to reveal the true magnitude of disease burden and to improve clinical management.
Studies have shown that the detection of Leptospira DNA using conventional or real time PCR had a higher sensitivity than culture and is useful for early diagnosis [8]. More recently, Loop-mediated isothermal amplification (LAMP)-based method has been developed. This method is fast and more sensitive for testing clinical samples and cost-effective, and more suited in resource-limited settings [9–11]. Limitations of the approach are that bacteremia is transient in the early phase of infection while the spirochetes remain in urine only in a fraction of patients. In addition, in countries where antibiotics are freely available over the counter, patients are often exposed to antibiotics prior to admission, and this affects the detection rate of Leptospira DNA.
MAT is the current reference standard serological diagnostic test in leptospirosis, and most studies defined laboratory-confirmed cases based on MAT. MAT method is time-consuming and requires specialist laboratory expertise [12]. As an alternative, whole cell-based serological tests have been developed and used in endemic areas [13–16]. Most of the whole cell-based serological assays employed antigens from non-pathogenic L. biflexa serovar Patoc, with which sera from patients with leptospirosis cross-reacted, and these assays are believed to be genus-specific and detect IgM antibodies from patients, regardless of infective serovars or serogroups [17]. The disadvantage of the whole cell antigen is that it may cause cross-reaction with other infectious diseases. To overcome this limitation, several recombinant protein-based serological tests have been developed, using outer membrane proteins such as LipL32, LipL41, OmpL1, Loa22 and Lig proteins [18–20]. Leptospiral immunoglobulin-like (Lig) proteins are surface-exposed outer membrane proteins, which bind to extracellular matrix proteins and are present only among pathogenic species [21]. LigA protein might be expected to be a good candidate of target antigen for serological tests such as ELISA, and previous studies showed 62.0–92.1% sensitivity in referral human serum samples [10,11,22]. The diagnostic performance of LigA-based ELISA has not been examined in the real-clinical setting of highly endemic area.
In the Philippines leptospirosis is highly endemic. Outbreaks of leptospirosis occur during typhoon seasons particularly in urban slum areas such as Metro Manila [3,6]. In this study, we have evaluated the diagnostic accuracy of a new recombinant LigA antigen based IgM ELISA and compared it with that of the whole cell-based IgM (Patoc-IgM) ELISA in a high-endemic area in the Philippines.
A prospective observational study was conducted at a national referral hospital for infectious diseases in Manila, Philippines during three leptospirosis outbreak seasons; October to December in 2011, September to October in 2012 and August to September in 2013. Hospitalized patients were approached for enrollment if they had 1) fever plus at least 2 other signs and symptoms of leptospirosis (headache, myalgia, eye pain, nausea, vomiting, abdominal pain, diarrhea, conjunctival suffusion, jaundice, tea-colored urine, oliguria, anuria, or unusual bleeding) and 2) history of exposure to floodwater or animals [3]. Demographic and clinical data were collected from patients and medical charts using a standardized data collection sheet. Blood samples were taken both on admission (1st sample) and at discharge (2nd sample) or either one of those. Plasma was separated from blood by centrifugation and stored at -20°C. Urine samples were collected on admission and stored at -20°C. As healthy controls, 100 blood donor samples were obtained from the Philippine Blood Center of the Department of Health in Manila. No background information was available for the blood donor samples.
Two milliliters of the plasma samples were centrifuged at 16,000 × g for10 min and the pellets were subjected to DNA extraction using the DNeasy Blood & Tissue Kit (Qiagen) according to the manufacturer’s instructions. Two milliliter of the urine samples was centrifuged at 500 × g for 5 min to remove the precipitates, prior to a high-speed centrifugation (16,000 ×g, 10 min). The resulting pellets were resuspended in 20 μl of 10 mM Tris–HCl (pH 8.0) containing 1 mM EDTA (TE) and then boiled for 10 min. The supernatant of the boiled sample was used as a template for Lepto-rrs LAMP.
The C-terminal portions of LigA (amino acid position 708–1224) fused with GST or GST alone was expressed in Escherichia coli BL21 and purified as described previously [23].
The MAT was performed for detecting anti-Leptospira antibodies in patient serum samples [24] using a battery of the reference strains of serogroups Australis (serovar Australis; strain Akiyami C), Autumnalis (Autumnalis; Akiyami A), Javanica (Poi; Poi), Pomona (Pomona; Pomona), Sejroe (Sejroe; M 84), and Tarassovi (Tarassovi; Perepelitsin) and rat isolates in the Philippines of serogroups Bataviae (Losbanos; K68), Grippotyphosa (unidentified; K93) and Pyrogenes (Manilae; K72), which cover prevalent leptospiral serogroups in the Philippines [3,25]. These strains were cultivated in liquid modified Korthof's medium with 10% rabbit serum at 30°C [24].
Culture was performed using Korthof’s medium. After the sample collection, 1–2 drops of blood were put into Korthof’s medium and cultivated at 30°C up to 13 weeks. The cultures were examined weekly by dark-field microscopy. Positive cultures were identified by MAT and PCR.
Lepto-rrs LAMP was performed using previously described primers and conditions [10]. The DNA template was preheated for 5 min at 95°C followed by rapid cooling on ice before addition to LAMP reaction mix.
The microtiter plates (Immobilizer Amino, Nunc) were coated with anti-human IgM (Jackson ImmunoResearch) at the concentration of 180 ng per well in 50 μl of 100 mM sodium phosphate, pH 8.0 overnight at 4°C. Excess binding sites of the well were blocked with 200 μl per well of 20 mg/ml of BSA in 20 mM Tris, 150 mM NaCl, 0.05% Tween 20, pH 7.5 (TBST) for 1.5 h at room temperature (RT), after which the BSA solution was removed. The plate was then rinsed three times with 300 μl per well of TBST. Patient plasma samples diluted 100-fold with ELISA buffer (TBST containing 10 mg/ml of BSA) were added in a total volume of 50 μl per well and incubated for 1 h at RT. After the plasma was rinsed six times with 200 μl per well of TBST, 4 μg/ml of LigA/GST or 2 μg/ml of LigA/GST in ELISA buffer was added in a total volume of 50 μl per well and incubated for 1.5 h at RT. The antigen solution was then rinsed out as above, and rabbit anti-GST IgG solution (Santa Cruz) diluted at 200-fold with ELISA buffer was added and incubated for 1 h at RT. The antibody solution was then rinsed out as above, then replaced with 50 μl per well of peroxidase-conjugated goat anti-rabbit IgG solution (Bio-Rad) diluted 5000-fold with ELISA buffer and incubated for 1 h at RT. The goat anti-rabbit IgG solution was then rinsed out as above. Finally, 50 μl of o-phenylenediamine dihydrochloride solution (OPD tablet (Sigma) in 6 ml distilled, deionized water containing 0.02% hydrogen peroxide) was added and settled for 5 min, and the reaction was stopped by adding 50 μl per well of 1 M sulfuric acid solution. Absorbance at 492 nm of each well was quantitated in a microtiter-plate reader.
The L. biflexa serovar Patoc antigen coating plate was prepared according to the WHO guidance [5]. The plate was rinsed six times with 200 μl per well of distilled water, and then blocked with 200 μl per well of 20 mg/ml of BSA in TBST for 1.5 h at RT. The plate was then rinsed three times with 300 μl per well of TBST. Patient plasma samples diluted 400-fold with ELISA buffer were added in a total volume of 50 μl per well and incubated for 1.5 h at RT. The plasma was then rinsed six times with 200 μl per well of TBST, then replaced with 50 μl per well of peroxidase-conjugated goat anti-human IgM solution (QED Bioscience) diluted 5000-fold with ELISA buffer and then incubated for 1 h at RT. The goat anti-human IgM solution was then rinsed out as above, and the antigen-bound IgM was detected as describe above.
The sample size was calculated according to the method developed by Flahault et al [26]. For an expected sensitivity of 70% with 0.95 probability that the minimum acceptable 95% lower confidence limit of 55%, a sample size of 114 cases was required. For an expected specificity of 98% with 0.95 probability that the minimum acceptable 95% lower confidence limit of 90%, a sample size of 89 controls was required. We therefore aimed to recruit 120 laboratory confirmed cases and 100 healthy controls.
The Standards for Reporting of Diagnostic Accuracy (STARD) reporting guidelines were followed [27]. A case was defined as laboratory confirmed if 1) culture was positive, 2) specific antibodies were detected with seroconversion or at least a 4-folds increase in reciprocal MAT titer between paired samples or with a reciprocal MAT titer of > = 400 in at least 1 plasma sample, or 3) Lepto-rrs LAMP was positive for plasma or urine sample. A case was defined as severe if s/he had acute renal failure (blood urea nitrogen >50 mg/dL, creatinine >5 mg/dL, or required renal dialysis), liver dysfunction (AST >100 IU/dL, ALT >100 IU/dL, or presented with jaundice), pulmonary hemorrhage, or died during hospitalization.
Clinical characteristics were compared between groups using chi-square tests or Fisher’s exact tests for categorical variables and t-tests or Mann-Whitney U test for numerical variables. The ELISA optical density (OD) values were compared between the laboratory-confirmed cases and the blood donor controls using receiver operating characteristic (ROC) analyses. The sensitivities and specificities, with 95% confidence intervals, of each ELISA test was calculated using the mean+3 standard deviation (SD) value of blood donor controls as the cut-off limit for defining a positive result [28].
This study was approved by the Research and Ethics Review Committee of the San Lazaro Hospital and the Institutional Review Board of the Institute of Tropical Medicine, Nagasaki University, Nagasaki, Japan. Informed written consent was obtained from all participants.
In total, 349 cases hospitalized with clinically suspected leptospirosis were approached but clinical samples could not be collected from 45 patients during the peak of 2013 outbreak because of overwhelming workload (Fig 1). Consequently 304 patients with the age of 28 years (range 7–67; interquartile range 20–40) were successfully enrolled for the analysis. The demographic and clinical features of the patients are shown in the Tables 1 and S1. Men were predominant (89.1%) and the majority of patients were poor residents of slum areas and had an outside occupation. There were 167/304 (54.9%) patients with severe disease and 12/304 (4.0%) patients died; male had severe disease more frequently than female (57.8% for male and 30.3% for female, p = 0.003) but the gender difference was not observed for mortality (4.4% and 0%, p = 0.2).
The 1st blood samples were collected from 292 cases (83.7%) on admission, the 2nd blood samples were collected from 164 cases (47.0%) on discharge and paired 1st and 2nd samples were collected from 162 cases (53.3%). The median duration from onset to the 1st sample collection and the 2nd sample collection varied widely: 6.5 (range 1–39; IQR 2–19) days and 11.2 (range 3–45; IQR 4–27) days, respectively. Urine samples were collected from 195 cases (55.9%). Among the 304 cases tested for MAT and LAMP, 134 (44.1%) were positive by MAT, 58 (19.1%) were positive by LAMP in either plasma or urine or both, and 25 (8.2%) were positive by both MAT and LAMP. LAMP positive plasma samples were found only among samples collected at 9 days or earlier after the disease onset, whereas LAMP positive urine samples were seen throughout (S1 Fig). Culture was performed in 114 cases and five (4.4%) cases were positive; all culture positive cases were positive by LAMP. Taken together, 167 (54.9%) cases were confirmed as leptospirosis based on the results of culture, LAMP and MAT. The most reactive serogroup by MAT was Bataviae (n = 51, 38.1%), followed by Javanica (n = 28, 20.9%), Tarassovi (n = 26, 19.4%), Pyrogenes (n = 20, 14.9%), Sejroe (n = 18, 13.4%), and Grippotyphosa (n = 14, 10.4%). No samples were reactive with Australis, Autumnalis, and Pomona.
The sensitivity of the LigA-IgM ELISA and Patoc-IgM ELISA were calculated using the 167 laboratory confirmed cases by culture, LAMP and MAT as the reference standard. The cut-off OD value (ie. the mean+3SD value of blood donor controls) for the LigA-IgM ELISA and Patoc-IgM ELISA were 0.062 and 0.691, respectively. The sensitivity of LigA-IgM ELISA was significantly higher than that of Patoc-IgM ELISA in the 1st samples on admission: 69.5% (95% confidence interval 60.4–75.1) vs. 54.3% (95% CI 48.3–60.3), respectively; p<0.01. The sensitivities were not significantly different in the 2nd samples taken at the time of discharge: 93.8 vs. 92.0; p = 0.81. Using the blood donor samples as negative control, the specificity was calculated and found high in both assays, 98% (95%CI 96.0–100.0) and 97% (95%CI 94.6–99.5), respectively. The AUC of the ROC curve for the LigA-IgM ELISA was significantly higher than that of Patoc-IgM ELISA in the 1st samples (0.90 vs. 0.82; p<0.01), but this difference was not observed in the 2nd samples (0.97 vs. 0.99; p = 0.18) (S2 Fig).
When sensitivity of LigA-IgM ELISA and Patoc-IgM ELISA was compared according to MAT-positive serogroups (Fig 2), the sensitivity of LigA-IgM ELISA against serogroups Tarassovi, Sejroe, and Javanica were found to be lower than Patoc-IgM ELISA, though the differences were not statistically significant.
The demographic and clinical characteristics at the time of admission of the whole 304 cases clinically suspected of leptospirosis were compared, according to the laboratory results (Tables 1 and S1). When compared with clinically suspected but laboratory unconfirmed cases, the laboratory confirmed cases by culture, LAMP and MAT were significantly more likely to be male (94.6% vs. 78.1%; p<0.01). Hemoptysis, jaundice, calf pain, neutrophilia, thrombocytopenia, and renal dysfunction were significantly more common in the laboratory confirmed cases.
Of the 158 MAT-negative cases, 53 (33.5%) were positive by the LigA-IgM ELISA while 28 (17.7%) were positive by the Patoc-IgM ELISA while 12 (9.0%) of the 134 MAT-positive cases were negative by the LigA-IgM ELISA and 10 (7.4%) were negative by the Patoc-IgM ELISA. If the laboratory confirmed cases were defined as those positive by LAMP and LigA-IgM ELISA, the clinical characteristics of cases were more typical of the classic descriptions of leptospirosis when compared with those defined by culture, LAMP and MAT. Cases were clearly associated with jaundice, calf pain, neutrophilia, thrombocytopenia, and renal dysfunction, and there were further associates with myalgia and dyspnea, hypotension, a tendency to have anuria, and a coagulation disorder.
Leptospiral DNA was detected both in plasma and urine samples: 45 in plasma samples, 22 in urine samples, and 9 in both plasma and urine samples (S1 Fig). The mean (SD) duration of disease was significantly shorter among plasma LAMP positive cases than urine-only positive cases (5.9±1.7 days vs. 9.2±3.3 days; p < 0.01). The mean (SD) LigA-IgM antibody level was lower among plasma LAMP positive cases than plasma LAMP negative cases (0.023±0.511 vs. 0.589±0.587; p = 0.10). LAMP detected leptospiral DNA in 29 LigA negative cases. Intriguingly in the 146 cases that were both plasma and urine LAMP negative the level of LigA antibody was high despite the sample being taken a relatively short time after symptom onset (S3 Fig).
The sensitivities of each of the diagnostic methods were plotted in relation to the duration from disease onset (Fig 3). In the early phase of disease (0–4 days from onset), LAMP was the most sensitive diagnostic method. As the disease progressed, LAMP sensitivity decreased, whereas the ELISA sensitivities increased. The LigA-IgM ELISA showed a higher sensitivity in the early phase than the other serological tests. The MAT sensitivity was lower than the sensitivity of both ELISAs throughout.
This is the first hospital-based prospective study, evaluating the diagnostic accuracy of the LigA-IgM ELISA for leptospirosis. The study was conducted in a busy government referral hospital serving the poorest people in a highly-endemic area of Metro Manila. There were several large outbreaks during the study period. Our results demonstrated that LigA-IgM ELISA has a high clinical value because a) it had a higher sensitivity than the Patoc-IgM ELISA, especially in the first 10 days after onset, and b) leptospirosis was diagnosed in a substantial number of patients, whose diagnosis would have been missed by the conventional laboratory confirmation with MAT.
The whole cell-based serology assays are commercially available, but previous studies have demonstrated that they have a low sensitivity, especially in the early phase of leptospiral infection and low specificity [12,29]. The recombinant protein-based serology assay is thought to achieve a higher sensitivity and specificity because of the higher concentrations of immunogenic antigens and specific antigenic moieties [18]. Furthermore, antibodies detected by the whole cell-based assay persist for years after leptospiral infection and this causes a major problem of low specificity in an endemic area. Croda reported that 29% of sera from patients with previous leptospirosis remained positive by whole-Leptospira-based IgM ELISA for 4–6 years after infection [30].To reduce this problem, in our study, we determined the cut-off OD for the whole cell-based serology assay by using the local blood donors as negative controls. If the cut-off OD had been determined by using non-leptospirosis Japanese febrile patients as negative controls (n = 15), the cut-off OD was lower, resulting in a higher sensitivity (73.2% in the 1st samples) but a lower specificity (78.0% in the 1st samples). The cut-off OD for the LigA-IgM ELISA did not significantly differ depending on the negative control samples used.
Several studies have evaluated the performance of LigA-IgM ELISA. The sensitivity has varied depending on sample population studied, the duration from the disease onset and also the definition of the composite reference standard. For an appropriate evaluation, we need to use several diagnostic assays to provide an adequate composite reference standard and a sufficiently large sample size. Very few studies fulfilled such criteria. Srimanote et al, reported a sensitivity of LigA-IgM ELISA was over 94% in both acute and convalescent phase when the cases were defined by a positive MAT result, but their sample size was very small (n = 46) and detection of pathogen was attempted only by culture [28]. Chen et al., reported the sensitivity of recombinant protein-based ELISA using LipL32, LipL41 and LigA antigens to be as low from 62 to 65%, although it could be increased to 90% by combing more than 2 antigens [20]. Very recently Kanagavel et al., compared the LigA-IgM ELISA and whole cell-based IgM ELISA in a study with a sufficient sample size (n = 140) and demonstrated that the sensitivity and the specificity of the LigA-IgM ELISA was higher than that of the whole cell-based ELISA [31]. These studies only evaluated the sensitivity using MAT positive samples referred for reference laboratory testing of unknown clinical background. There have been no previous studies evaluating the performance of the assay in a real clinical setting.
Our results suggested slight differences in sensitivity of the LigA-IgM ELISA depending on the infecting serogroups. Although the differences were not statistically significant, the sensitivity of LigA-IgM ELISA tended to be lower against L. borgpetersenii (serogroups Sejroe, Tarassovi and Javanica) but higher against L. interrogans (serogroups Bataviae, Pyrogenes and Grippotyphosa) when compared with the whole cell-based ELISA. A recent study suggested that L. interrogans expresses genes of both LigA and LigB but L. borgpetersenii has only LigB [32,33]. A lack of LigA in L. borgpetersenii may explain the lower sensitivity. It is plausible that, although the LigB molecule has the same repeated immunoglobulin-like domains where antibodies bind to, the C-terminal 80 kDa domain of LigB might mask the antibody epitope regions [34].
Culture and MAT is still regarded as the composite reference standard for leptospirosis referral laboratory confirmation. This definition has a number of limitations in the clinical setting. Both tests demand well-established laboratory where specific culture media is available for leptospirosis and the MAT assay needs an experienced laboratory technician with several Leptospira serovar strains alive. Culture requires up to 13 weeks to have results and can be rendered negative by antibiotic pre-exposure. To overcome these limitations, in this study, we used a LAMP assay that does not demand sophisticated laboratory equipment and that gives a result within 2 hours with almost equivalent sensitivity to real-time PCR. When we applied a new definition by LAMP and LigA-IgM ELISA and compared with culture/LAMP/MAT, we identified a patient population whose clinical symptoms and signs were more compatible with leptospirosis than culture/LAMP/MAT. We propose that the LAMP/LigA ELISA combination is practical and reasonable for laboratory confirmation in the clinical setting.
There are several limitations in this study. The study was conducted in a busy government hospital, many patients were discharged within seven days after admission and did not return to the clinic, thus convalescent samples were often not available at the optimum time of at least two weeks apart. This is a reflection of a real clinical setting. The majority of our patients were males reflecting the high incidence of leptospirosis among males in our setting [3]. However, this imbalance did not affect our findings; the sensitivity and specificity of LigA-IgM ELISA and Patoc-IgM ELISA did not differ between the sexes. Most laboratory tests were conducted in Japan, and some LAMP results might have been affected by improper condition of sample transportation. Finally, the lack of a satisfactory laboratory composite reference standard is a limitation in defining the diagnostic accuracy of new diagnostic tests in leptospirosis.
In summary, the findings of this study indicate that LigA-IgM ELISA can be a good diagnostic tool with high sensitivity especially in early phase of illness. We propose that a combination of molecular and serology assays can improve diagnosis and help timely initiation of antibiotics to prevent severe outcomes of leptospirosis in an endemic area.
|
10.1371/journal.ppat.1005105 | Inhibiting the Recruitment of PLCγ1 to Kaposi’s Sarcoma Herpesvirus K15 Protein Reduces the Invasiveness and Angiogenesis of Infected Endothelial Cells | Kaposi’s sarcoma (KS), caused by Kaposi’s sarcoma herpesvirus (KSHV), is a highly vascularised tumour of endothelial origin. KSHV infected endothelial cells show increased invasiveness and angiogenesis. Here, we report that the KSHV K15 protein, which we showed previously to contribute to KSHV-induced angiogenesis, is also involved in KSHV-mediated invasiveness in a PLCγ1-dependent manner. We identified βPIX, GIT1 and cdc42, downstream effectors of PLCγ1 in cell migration, as K15 interacting partners and as contributors to KSHV-triggered invasiveness. We mapped the interaction between PLCγ1, PLCγ2 and their individual domains with two K15 alleles, P and M. We found that the PLCγ2 cSH2 domain, by binding to K15P, can be used as dominant negative inhibitor of the K15P-PLCγ1 interaction, K15P-dependent PLCγ1 phosphorylation, NFAT-dependent promoter activation and the increased invasiveness and angiogenic properties of KSHV infected endothelial cells. We increased the binding of the PLCγ2 cSH2 domain for K15P by substituting two amino acids, thereby creating an improved dominant negative inhibitor of the K15P-dependent PLCγ1 activation. Taken together, these results demonstrate a necessary role of K15 in the increased invasiveness and angiogenesis of KSHV infected endothelial cells and suggest the K15-PLCγ1 interaction as a possible new target for inhibiting the angiogenic and invasive properties of KSHV.
| Kaposi’s Sarcoma (KS), etiologically linked to Kaposi’s sarcoma herpesvirus (KSHV), is a tumour of endothelial origin characterised by angiogenesis and invasiveness. In vitro, KSHV infected endothelial cells display an increased invasiveness and high angiogenicity. Here we report that the KSHV protein K15, which increases the angiogenicity of endothelial cells, contributes to KSHV-mediated invasiveness by the recruitment and activation of the cellular protein PLCγ1 and its downstream effectors βPIX, GIT1 and cdc42. We explored the functional consequences of disrupting the K15-PLCγ1 interaction by using an isolated PLCγ2 cSH2 domain as a dominant negative inhibitor. This protein fragment, by interacting with K15, reduces K15-driven recruitment and activation of PLCγ1 in a dose-dependent manner. Moreover, the PCLγ2 cSH2 domain, when overexpressed in KSHV infected endothelial cells, reduces the angiogenesis and invasiveness induced by the virus. These findings highlight the role of the K15-PLCγ1 interaction in KSHV-mediated invasiveness and identify it as a possible therapeutic target.
| Kaposi’s Sarcoma Herpesvirus (KSHV), also known as human herpesvirus 8 (HHV-8), was first identified in Kaposi’s sarcoma (KS) tissues by representational difference analysis [1,2]. Found to be the etiological agent of KS, the virus was successively linked also to primary effusion lymphoma (PEL) and multicentric Castleman’s disease (MCD) [3,4], two rare lymphoproliferative disorders.
KS, a highly vascularized tumour, is one of the most frequent AIDS-related cancers and a major health problem in sub-Saharan Africa [5,6]. Histologically KS is characterised by the presence of an inflammatory infiltrate of neutrophils, B and plasma cells as well as aberrant angiogenesis [7,8]. In advanced KS lesions, the main proliferative elements are the KSHV infected endothelial cells, which lose their typical morphology, become spindle-shaped and acquire invasive properties [9]. Although the majority of KSHV infected endothelial cells in KS tissue harbour the virus in a latent phase, a small population undergoes lytic (productive) reactivation [10]. Moreover, lytically infected cells secrete viral and cellular factors able to promote the pathological angiogenesis and invasiveness of latently infected cells in a paracrine manner [11–13]. In particular, upon lytic reactivation, virus infected endothelial cells show an increased secretion of VEGF, Ang2, ephrin B2, MMPs, IL6 and 8 [12,14,15]. Among the KSHV-encoded proteins that possess angiogenic and invasive properties are the latency associated nuclear antigen 1 (LANA1), the viral homolog of G-protein coupled receptor (vGPCR, a constitutively active IL8 receptor), the viral homolog of interleukin 6 (vIL6), two viral chemokine homologs, vCCL1 and 2, as well as the K1 protein [12,14–16]. In this context, we recently reported that the KSHV K15 protein also contributes to KSHV-mediated angiogenesis [17].
The K15 gene, located at the “right end” of the long unique region (LUR) of the viral genome, consists of eight exons that are alternatively spliced [18–20]. Two main K15 alleles, termed P (predominant) and M (minor) have been identified with the M allele most likely being the result of a homologous recombination with an unknown γ2 herpesvirus [21]. For both K15P and M a mRNA comprising all eight exons is the most strongly expressed transcript and encodes a protein of 12 predicted transmembrane domains and a cytoplasmic tail involved in signalling. Although K15P and M share as little as 33% of amino acid sequence homology, the cytoplasmic tail of both K15 alleles contains a putative src homology -3 (SH3) (PLPP) and two SH2 (YASIL, YEEVL) binding sites (Fig 1A, upper panel) [18–20]. Previous studies reported that both K15 alleles activate NF-κB and the Ras/MAPK signalling [22–25]. Microarray studies revealed that K15 upregulates the expression of genes involved in angiogenesis and cell migration [22,24,25], including, among others, COX2 and an NFAT-dependent upregulation of DSCR1 [17,24,25]. We previously showed that K15 contributes to the increased angiogenic properties of KSHV infected primary endothelial cells in a matrigel-based capillary tube formation assay [17]. This process involves the K15P-dependent recruitment of phospholipase C γ1 (PLCγ1) and its phosphorylation on tyrosine 783, the successive activation of the calcineurin/NFAT pathway and the downstream upregulation of DSCR1 [17].
Apart from its role in angiogenenesis, PLCγ1 is also required for cell migration and cytoskeleton remodelling in both cancer and normal endothelial cells [26–30]. Upon integrin engagement, PLCγ1 associates with the βPIX (β-PAK interacting/exchange factor)-GIT1 (G-protein-coupled receptor kinase-interacting protein 1) complex leading to the activation of the small GTPase cdc42 and subsequent cell spreading and migration [27]. Furthermore, in cancer patients, PLCγ1 overexpression and increased phosphorylation levels are correlated with poor prognosis and metastasis formation [31,32]. Activating somatic mutations in PLCγ1 have recently been observed in several cases of angiosarcoma [33–35]. In addition to the ubiquitously expressed PLCγ1, the PLCγ protein family also includes PLCγ2 which is found mainly in hematopoietic cells. These two isoforms share the same domain organization and a high degree of sequence homology (Fig 1A, bottom panel). Both are characterized by the presence of the γ-specific array, a protein region of approximately 500 amino acids inserted between the two modules of the catalytic domain (Fig 1A, bottom panel). The γ-specific array consists of a split pleckstrin homology domain (spPH), two SH2 as well as a SH3 domain, and is required for the regulation of PLCγ enzymatic activity.
In this study, we report that the recruitment and activation of PLCγ1 by K15 contributes to the invasiveness of KSHV infected endothelial cells in a GIT1- βPIX- and cdc42-dependent manner. We also show that the overexpression of an isolated SH2 domain, derived from PLCγ2, can disrupt the K15P-PLCγ1 interaction and can inhibit K15-dependent downstream signalling and thereby the increased invasiveness and angiogenic properties of KSHV infected endothelial cells.
Previously reported microarray experiments have shown that K15 increases the expression of genes involved in inflammation and cell migration such as chemokines and MMPs [17,25]. Moreover, we and others [17,36,37] had suggested that K15 expression increased the invasiveness of endothelial cells. This is illustrated in Fig 1B and S1A Fig, which compare the invasiveness of K15-transduced immortalized endothelial cells (HuAR2T) to that of cells transduced with the control vector. In this experimental system, K15 overexpression significantly increases the invasiveness of endothelial cells. Since we previously reported that K15 recruits and activates PLCγ1 [17], which has been found to trigger cellular invasiveness in many cellular contexts [28,32], we tested whether the recruitment of PLCγ1 by K15 was also important for K15-driven invasiveness. We performed invasion assays using K15-transduced endothelial cells (HuAR2T) transfected with either control siRNA or siRNA targeting PLCγ1 (Fig 1C and S1B Fig). Silencing PLCγ1 was sufficient to decrease K15-mediated invasiveness to the basal level, indicating that K15 increases the invasiveness of endothelial cells in a PLCγ1-dependent manner. We and others have previously shown that, upon lytic reactivation, KSHV infected endothelial cells display an increased ability to invade the extracellular matrix [12,14,16]. To elucidate the contribution of K15 and PLCγ1 to the KSHV-mediated invasiveness induring the induction of the lytic cycle, KSHV infected immortalized endothelial cells (HuAR2T rKSHV.219) were transfected with siRNA targeting either K15 or PLCγ1 and the invasiveness of the cells was evaluated in a matrigel-based invasion assay (Fig 1D and S1C Fig). Upon induction of the lytic cycle, the cells showed increased invasiveness (Fig 1D, compare lane 3 to lane 6). Depletion of either K15 or PLCγ1 by siRNA significantly reduced the number of invading cells (Fig 1D, compare lanes 1 and 2 to lane 3). Furthermore, reduced invasiveness was also observed in HEK 293 stably harbouring a recombinant KSHV genome lacking K15 (KSHV BAC36 ΔK15) as compared to cells harbouring KSHV wt genome (KSHV BAC36) (S2 Fig). These results point to K15 and PLCγ1 as contributors to KSHV-mediated invasiveness (Fig 1D and S1C and S2 Figs).
Since it has been reported that βPIX, GIT1 and cdc42 act downstream of PLCγ1 and contribute to integrin-mediated invasiveness [27,28], we silenced these proteins (using siRNA) and tested the invasiveness of KSHV infected endothelial cells, following the reactivation of the lytic cycle. Although the siRNA pools used in this experiment did not completely suppress the expression of βPIX, GIT1 and cdc42, they significantly reduced the invasiveness of KSHV infected endothelial cells, compared to the control siRNA treated cells (Fig 1E and S1D Fig). Therefore, we conclude that K15 contributes to KSHV-induced invasiveness and that this process is mediated, at least in part, by PLCγ1 and its downstream effectors GIT1, βPIX and cdc42.
We next investigated the cellular localization of K15, PLCγ1, βPIX and GIT1 in KSHV infected cells by immunofluorescence assay. K15 staining has previously been performed in HEK293T or HeLa cells in the context of transient transfection [22,25]. Under these experimental conditions, K15 was localised at the plasma membrane showing a punctate pattern or in large patches with a perinuclear distribution. A similar distribution has also been reported in HEK 293 cells stably transfected with KSHV BAC 36 [25]. Using a newly generated rat monoclonal antibody to K15 (see Materials and Methods), we were able to detect endogenously expressed K15 in KSHV infected endothelial cells (Fig 2). In the cultures shown here, about 15–20% of KSHV infected cells were expressing K15 to a detectable level. The majority of K15-expressing cells showed a perinuclear as well as a plasma membrane localization with a punctate pattern, with bigger dots localizing in the perinuclear region (Fig 2A–2C and Fig 3A), in line with a previous report from our group in which a polyclonal antibody to K15 was used [25]. Interestingly, in these dot-like structures, K15 protein co-localized strongly with total PLCγ1 and partially with phosphorylated PLCγ1 (pY783 PLCγ1), as well as with βPIX and GIT1 (Fig 2A–2C).
In order to address the question if K15 co-localises with activated (phosphorylated) PLCγ1 in infected cultures undergoing the early stages of lytic replication, we co-stained KSHV infected HuAR2T cells, in which the lytic replication cycle had been induced, with antibodies to K15 and PLCγ1 phosphorylated on Y783. We confirmed the induction of the lytic cycle by staining the early KSHV protein encoded by orf 59 (polymerase processivity factor) (Fig 3A). Similarly to what we observed in latently infected HuAR2T (Fig 2A–2C), in lytically induced cultures, we also observed partial co-localisation of phosphorylated PLCγ1 and K15 (Fig 3B). In these cultures, this co-localisation was mainly observed in orf59-negative cells (Fig 3B) and it was less evident in orf 59 positive cells (Fig 3C). Owing to the lack of antibodies to βPIX and GIT1 suitable for co-localisation studies with K15 in cultures stained for orf59, we could not assess whether, as in the case of phosphorylated PLCγ1, βPIX and GIT1 co-localize with K15 in HuAR2T-KSHV cell cultures after the activation of the lytic cycle.
We also investigated the interaction between K15, PLCγ1, cdc42 and GIT1 by co-immunoprecipitation. As shown in Fig 3D and 3E, we co-immunoprecipitated PLCγ1 from KSHV infected and uninfected endothelial cells (HuAR2TrKSHV.219 and HuAR2T) and detected its interaction with K15 in the former, but not the latter cell line. We also observed the previously reported interaction of PLCγ1 with GIT1 in infected and uninfected cells (Fig 3E). Moreover, we observed that, in KSHV infected but not in uninfected cells, PLCγ1 associates not only with K15 and GIT1, but also with cdc42 (Fig 3E). Furthermore, when we immunoprecipitated PLCγ1 from KSHV infected endothelial cells following reactivation of the lytic cycle (Fig 3E), we also observed a complex composed of K15, PLCγ1, GIT and cdc 42 (Fig 3E).
Thus, K15 associates with PLCγ1, GIT1 and cdc42 in cells of both latently and lytically induced cultures. Based on the immunofluorescence experiment shown in Fig 3A–3C, we conclude that the association of K15 with PLCγ1, GIT1 and cdc42 occurs mainly in latent cells (i.e. cells lacking orf59 expression). Unfortunately, low levels of βPIX expression in combination with the lack of an antibody suitable for western blot did not allow us to assess the presence of βPIX in this complex by co-immunoprecipitation.
Since it has been previously shown that GIT1 forms a stable complex with βPIX and that these proteins serve as a scaffold for the assembly of pro-migratory cytoplasmic protein complexes induced by activated PLCγ1 [38], K15 might recruit these proteins to trigger virus-induced cellular invasiveness.
Previously, we showed that the C-terminal SH2 binding motif (YEEV) and SH3 binding motif (PPLP) of K15P (Fig 1A) both contribute to the interaction with PLCγ1 [17]. In this study we wanted to identify the PLCγ1 region involved in this interaction. Since the γ-specific array region of PLCγ1contains two SH2 domains (nSH2 and cSH2) (Fig 1A, bottom panel, Fig 4A and [39–41]), we focused on this portion of PLCγ1. The SH2 domain is a protein module of approximately 100 amino acids with an evolutionarily conserved phospho-tyrosine binding site. In particular, an arginine within the binding pocket is highly conserved in human SH2 domains and provides nearly half of the ligand binding energy [42,43]. To investigate the potential role of the two PLCγ1 SH2 domains in the interaction with K15, we mutated the conserved arginine residues within both the n and c SH2 domains (R586L and R694L, respectively; see Figs 1A and 4A) and tested these PLCγ1mutants for binding to both the K15P and M variants, after co-transfection with K15 expressing vectors into HEK293T cells. While mutation of the nSH2 domain (R586L) moderately decreased the binding to K15P, mutation of the cSH2 domain (R694L) had a stronger effect on this interaction (Fig 4B). In the case of K15M, only the mutant of the nSH2 domain of PLCγ1 (R586L) showed a reduced binding (Fig 4C). These results indicate that between K15P and K15M there is a subtle difference with regard to how they interact with PLCγ1. While the N-terminal SH2 domain of PLCγ1 seems to be involved in the interaction with both K15P and K15M, the cSH2 domain is only important for the binding to K15P.
In metazoans, there are two isoforms of PLCγ, the ubiquitously expressed PLCγ1, and PLCγ2, whose expression is restricted mainly to hematopoietic cells. Although they are functionally distinct, the two isoforms share some binding partners. We therefore explored the possibility that K15 might bind to PLCγ2 as well. To this end, we performed a GST pulldown assay with a fusion protein of GST and the K15P cytoplasmic domain (aa 347–489) bound to glutathione beads and a lysate of HEK293T cells that had been transiently transfected with plasmid expressing PLCγ2 or individual PLCγ2 domains. Although we could not observe an interaction between the K15P cytoplasmic domain and the full length PLCγ2 protein (Fig 5A, left panel), the PLCγ2 cSH2 domain, in isolation, did bind to K15P (Fig 5A, right panel).
Since the interaction between K15 and PLCγ1 is required for both KSHV mediated angiogenesis [17] and invasiveness (Fig 1D), we wondered if a targeted inhibition of this interaction could have therapeutic potential. PLCγ1 is ubiquitously expressed and depletion of the PLCγ1 gene in mice leads to embryonic lethality [44], hence its inhibition would be expected to be associated with side effects in uninfected cells. However, an inhibitor that would specifically target the interaction between K15 and PLCγ1, without interfering with the physiological functions of PLCγ1 might counteract only KSHV-dependent signalling processes. Since Decker et al [45] successfully impaired pathological osteolysis by uncoupling the adaptor and the catalytical functions of PLCγ2 by ectopic expression of tandem PLCγ2 SH2 domains, we used a similar approach and tested the PLCγ2 cSH2 domain as a potential inhibitor of the K15-PLCγ1 interaction. We first characterized the interaction between K15P and the isolated PLCγ2 cSH2 domain in more detail. To confirm that the interaction between K15 and the PLCγ2 cSH2 domain takes place through the YEEV SH2 binding site of K15, which has been shown to participate in the interaction with PLCγ1 [17], we mutated the conserved tyrosine in this motif to phenylalanine (K15 Y481F, from now on called K15 YF). Fig 5B shows that the K15P YF mutant does not associate with the PLCγ2 cSH2 domain. We also mutated the conserved arginine (R672L) (Fig 1A, bottom panel) within the PLCγ2 cSH2 domain. Fig 5C shows that this mutation impairs binding to K15P. Therefore, the interaction between the PLCγ2 cSH2 and K15P involves the Y481 of K15 and R672 of the PLCγ2 cSH2 domain.
It has been previously reported [46] that the phospho-tyrosine binding pocket of the SH2 domain of the cellular kinase Fyn can be optimized in order to obtain stronger binding to a phosphorylated binding site, without changing the binding specificity of this SH2 domain. The authors obtained this superbinder (SB) SH2 domain by mutating specific amino acids (T181V; S186A; K204L) within the Fyn SH2 domain. As both sequence and structure of SH2 domains are highly conserved, we identified and mutated the corresponding amino acids in the isolated PLCγ2 cSH2 domain (Fig 5D) in order to obtain a SB mutant. Of the residues altered to obtain the Fyn SB SH2 domains (Fig 5D), one, the A681, was already an alanine in the PLCγ2 cSH2 domain. We therefore mutated only two residues, S677 and R695 to valine and leucine, respectively. Subsequently, we tested the ability of K15P to associate with the PLCγ2 cSH2 S677V/R695L mutant (from now on called superbinder, SB) by co-immunoprecipitation (Fig 5E). K15 was precipitated more efficiently with the SB mutant than with the wt cSH2 domain (Fig 5E). This result was confirmed by GST pulldown from lysates of cells transfected with expression vectors for the PLCγ2 cSH2 wt or SB domain (Fig 5F). The K15P cytoplasmic tail fused to GST bound more efficiently to the SB than to either of the single mutants (S677V and R695L) of the cSH2 domain (Fig 5F). This result is in line with the observations reported by Kaneko et al. [46] that the properties of the SB mutant are the result of a synergistic effect obtained by mutating multiple residues in the SH2 domain. To confirm that the cytoplasmic domain of K15 binds directly to the PLCγ2 cSH2 domain or its SB mutant, we used surface plasmon resonance (SPR) to measure the interaction of purified, prokaryotically expressed GST-K15P and the PLCγ2 cSH2 wt and SB domains, as described in Materials and Methods. We obtained equilibrium dissociation constants (KD) of 5.8 10−9 for PLCγ2 cSH2 wt, and 3.8 10−9 for the SB mutant This result shows that the interaction of K15 with the PLCγ2 cSH2 domain is of high affinity, in the nanomolar range, even though in this experiment K15, having been expressed in bacteria, was not phosphorylated on Y489. This lack of phosphorylation most likely explains the similar affinity observed for the interaction of K15 with the PLCγ2 cSH2 wt and SB domains.
Since the cSH2 domain of PLCγ1 has been shown to contact the phosphorylated Y783 in PLCγ1 [47,48], thereby mediating an intramolecular interaction required for the activation of this protein, we also investigated if the isolated PLCγ2 cSH2 would interact with PLCγ1. Purified His-tag fused PLCγ2 cSH2 domain bound to Ni-beads was incubated with cleared cell lysate and was probed via western blot for its binding to endogenous PLCγ1. The PLCγ1 cSH2 domain was included in the experiment as a positive control. Endogenous PLCγ1 was pulled down by the isolated PLCγ1 cSH2 domain, but not by the PLCγ2 cSH2 domain (Fig 5G). Thus, the PLCγ2 cSH2 isolated domain interacts with K15 (via its YEEV motif, Fig 5B), but not with PLCγ1.
It was previously reported that the overexpression of the PLCγ1 γ-specific array in cancer cell lines has a dominant negative effect on PLCγ-meditated cell migration [29,30]. Moreover, as shown above, K15 co-localizes with PLCγ1, GIT1, βPIX and these proteins contribute to KSHV-mediated invasiveness (Figs 1, 2 and 3). Since we also observed that the PLCγ2 cSH2 domain interacts with K15 via its YEEV motif, which is required for downstream signalling (Fig 5B), we decided to test if it has a dominant negative effect on the interaction of K15 with PLCγ1 and the subsequent signalling events. In an interaction assay, increasing amounts of the PLCγ2 cSH2 domain gradually reduced the binding of PLCγ1, cdc42 and GIT1 to the GST-K15 cytoplasmic tail (Fig 6A). In this assay, higher concentrations of the isolated PLCγ2 cSH2 domain were needed to disrupt the binding of PLCγ1 and GIT1 to K15 than to reduce the binding of cdc42. In addition, the presence of the isolated PLCγ2 cSH2 domain reduced in a dose-dependent manner K15-mediated PLCγ1 phosphorylation levels in HeLa cells, as well as in primary HUVECs (Fig 6B).
We have previously shown that K15 induces angiogenesis by binding and activating PLCγ1 thereby triggering NFAT-mediated signalling [17]. Therefore, we also investigated the effect of the PLCγ2 cSH2 domain on K15-mediated activation of the NFAT-dependent promoter in a luciferase based reporter assay (Fig 6C). Increasing amounts of the PLCγ2 cSH2 domain gradually decreased the ability of K15 to activate an NFAT-dependent promoter, further confirming that the PLCγ2 cSH2 domain could be used as a dominant negative inhibitor of K15-mediated signalling. In contrast, the nSH2 domain of PLCγ2, which does not bind to K15 (Fig 5A), also did not inhibit the K15-dependent NFAT activation.
K15 mediates the activation of several cellular signalling cascades, including the NF-κB pathway [22,23,25]. We previously showed that the activation of NF-κB by K15 occurs via a region in the K15 cytoplasmic tail located close to the last transmembrane domain [23]. To explore the specificity of the dominant negative effect observed with the PLCγ2 cSH2 domain, we tested whether it would also affect the K15-mediated activation of NF-κB (Fig 6D). Neither the isolated PLCγ2 cSH2 domain, nor the isolated PLCγ2 nSH2 and SH3 domains (which do not bind K15, Fig 5A) compromised the ability of K15 to activate NF-κB-dependent transcription. Furthermore, we tested the effect of the SB (S677V/R695L) and the R672L mutants of the PLCγ2 cSH2 domain on the K15-mediated NFAT- as well as NF-κB activation (Fig 6E and 6F). While the SB showed a stronger effect than the PLCγ2 cSH2 wt, the R672L mutant had no effect on the ability of K15 to activate NFAT-dependent transcription (Fig 6E). In contrast, neither the SB nor the R672L mutant had any effect on the ability of K15 to activate an NF-κB-dependent promoter (Fig 6F). Therefore, we conclude that the PLCγ2 cSH2 domain specifically impairs K15-mediated NFAT activation, but has no effect on NF-κB-dependent transcription. In addition, increasing the affinity of the PLCγ2 cSH2 domain for the K15 YEEV motif, by introducing two substitutions (Fig 5D), enhances its dominant negative effect on K15-dependent NFAT activation, but has no influence on NF-κB signalling.
As described above, the other K15 allele, K15M, also binds to PLCγ1 via the PLCγ1 nSH2 domain but differs subtly from K15P, whose binding to PLCγ1 appears to involve additionally the PLCγ1 cSH2 domain (Fig 4C). Similar to K15P, K15M can also activate NFAT-dependent promoters [24]. Therefore we investigated the effect of the PLCγ2 cSH2 domain on K15M-mediated NFAT-dependent transcriptional activation (Fig 7A). Irrespective of the amount of transfected PLCγ2 cSH2 plasmid, we did not detect any significant difference in K15M-mediated NFAT-dependent activation, although in the same assay a dose dependent inhibitory effect was observed for K15P. Consequently, we decided to test the ability of K15M to bind the isolated PLCγ2 cSH2 domain but could not detect any association between these two proteins (Fig 7B). Furthermore, we tested whether the expression of the PLCγ2 cSH2 SB could affect K15M-triggered activation of an NFAT-dependent promoter. To this end, we compared the ability of both K15M and P to activate NFAT signalling in the presence of the wt or SB PLCγ2 cSH2 domain. While the presence of both wt and SB PLCγ2 cSH2 domain reduced K15P-dependent NFAT activation, neither of these two proteins affected the ability of K15M to activate the NFAT-dependent promoter (Fig 7C). To explore whether the PLCγ2 cSH2 domain SB would interact with K15M, we also tested the binding of the wt and the SB cSH2 domain to K15M in a co-immunoprecipitation assay (Fig 7D). While the PLCγ2 cSH2 SB domain, as expected, showed increased binding to K15P, it failed, as did the wt cSH2 domain, to interact with K15M (Fig 7D). Therefore, while the PLCγ2 cSH2 domain binds K15P and thereby blocks K15P-dependent NFAT signalling, it does not bind K15M and, consequently, has no influence on the K15M-mediated activation of NFAT-dependent transcription. Mutations that increase the ability of the PLCγ2 cSH2 domain to interact with K15P do not confer binding to K15M.
So far, our results showed that the isolated PLCγ2 cSH2 domain inhibits K15-mediated NFAT activation by blocking the recruitment of PLCγ1. We further wanted to test the effect of the PLCγ2 cSH2 domain on the role of K15- dependent invasiveness (see above) and angiogenesis of KSHV infected endothelial cells [17]. As a first step, we performed an invasion assay using immortalized endothelial cells (HuAR2T) that had been transduced with a retroviral vector for K15P and a lentiviral vector for the PLCγ2 cSH2 domain. We observed that the presence of the PLCγ2 cSH2 domain significantly reduced K15-mediated invasiveness (Fig 8A and S3A Fig). We were able to confirm this result in the context of the HuAR2T cell line that had been stably infected with a recombinant KSHV carrying the K15P allele (HuAR2TrKSHV.219). In these cells, following the reactivation of the lytic cycle, the number of invading cells was significantly reduced in PLCγ2 cSH2-expressing cells (Fig 8B, compare lanes 1 and 3, and S3B Fig).
We have previously shown that K15 triggers angiogenesis via PLCγ1 and subsequent NFAT activation [17] and that the PLCγ2 cSH2 domain can reduce the level of NFAT activation in the case of the K15P allele (Fig 6B, 6C and 6E and Fig 7A). As a next step, we tested whether we could measure a reduction in K15-dependent angiogenic tube formation in primary endothelial cells (HUVEC) after the transduction with a lentivirus expressing the PLCγ2 cSH2 domain and a retroviral vector for K15P (Fig 8C). Indeed, in the presence of the PLCγ2 cSH2 domain, the increased formation of angiogenic tubes, which we previously showed to be induced by K15P [17], was significantly reduced (Fig 8C compare lanes 1 and 2, and S3C Fig). Remarkably, despite its inhibitory effect on K15-mediated angiogenesis, the PLCγ2 cSH2 domain did not decrease the number of capillary tubes in VEGF-stimulated HUVECs (compare lanes 5 and 6 of Fig 8C, and top row S3C Fig), thus indicating that the VEGF-dependent angiogenic signalling is not impaired by the PLCγ2 cSH2 domain. We further investigated the inhibitory effect of the PLCγ2 cSH2 domain on the increased angiogenesis displayed by KSHV infected primary endothelial cells, following the reactivation of the lytic cycle. To this end, we infected primary HUVECs with a recombinant KSHV (rKSHV.219) and, following lytic reactivation and transduction with the PLCγ2 cSH2 domain, we performed an in vitro capillary tube formation assay. As shown in Fig 8D and S3D Fig the angiogenic index was significantly increased in KSHV infected HUVECs, upon lytic reactivation (lane1 and 5), as compared to the uninduced samples (lanes 2 and 6) and to the uninfected controls (lanes 7–12). Transduction with the PLCγ2 cSH2 domain (lane 3), but not control vector (lane1), reduced the number of capillary tubes.
Therefore, we conclude that the isolated PLCγ2 cSH2 domain impaired both the increased invasiveness and angiogenesis observed in endothelial cells infected with a KSHV strain carrying the K15P allele.
Aberrant angiogenesis and cell migration represent important features in the pathogenesis of KS [49]. Our group previously showed that, in an in vitro assay (angiogenic tube formation on matrigel), performed in primary endothelial cells, KSHV triggers angiogenesis via activation of PLCγ1 by K15 [17]. In the present study, we demonstrate that K15 also triggers migration and invasiveness of KSHV infected endothelial cells in a PLCγ1-GIT1-βPIX-cdc42-dependent manner (Fig 1D and 1E). Confocal images showed that K15 co-localizes with PLCγ1 in KSHV infected endothelial cells (Fig 2A). K15 was found to partially co-localize with phosphorylated PLCγ1 (pTyr 783) (Fig 2B and 2C and Fig 3A–3C) and to co-localize strongly with βPIX (Fig 2B) and GIT1 (Fig 2C). In endothelial cell cultures, in which the lytic replication cycle had been activated, this co-localisation was mostly seen in orf59-negative cells, which are either latent or show a restricted viral gene expression pattern (Fig 3A–3C). Moreover, co-immunoprecipitation of PLCγ1 in virus infected endothelial cells showed the presence of a complex formed by K15, PLCγ1, GIT1, and cdc42 (Fig 3D and 3E). Similar results were obtained in the GST-pulldown assay performed with a GST-fused K15 cytoplasmic tail (Fig 6A). Since several studies reported previously that βPIX and GIT1 form a stable complex in endothelial cells [50,51], we hypothesize that K15 recruits a complex formed by PLCγ1, βPIX, GIT1 and cdc42, which is then responsible for the increased invasiveness of KSHV infected endothelial cells that are either latent or show a restricted viral gene expression pattern. In our invasion assay, the majority of migrating cells do not express RFP, a marker of early lytic gene expression in the recombinant KSHV.219 used in this study [52], suggesting that the majority of migrating cells have not entered the lytic cycle. On the other hand, we did have to activate the lytic cycle in order to observe an increased invasiveness of the KSHV infected HuAR2T cells. We therefore believe that additional viral or cellular factors, possibly acting in a paracrine manner, have to be released from the small percentage of lytic cells (see Fig 3A) in order to promote the invasiveness of the majority of KSHV infected cells that have not entered the lytic replication cycle. In this scenario K15 would therefore be a necessary (since its silencing abrogates invasiveness) but not sufficient factor for the increased invasiveness of KSHV infected endothelial cells.
PLCγ1 contains two SH2 domains (Fig 1A), the nSH2 domain, which binds to receptor tyrosine kinases (RTKs), and the cSH2 domain, which participates in an intra-molecular interaction necessary for PLCγ1 enzymatic autoregulation [47,48,53]. Similar to RTKs, both the M and P variants of K15 associate with the nSH2 domain of PLCγ1, since mutation of the conserved R586 in PLCγ1 responsible for contacting the phosphorylated tyrosine in the SH2 binding site of the interacting protein, decreases the affinity of both K15 variants for PLCγ1 (Fig 4). However, in the case of K15P, but not of K15M, mutation of a similar arginine (R694) in the PLCγ1 cSH2 domain reduced this interaction more significantly. Since the PLCγ1 cSH2 domain is involved in an intramolecular interaction with the phosphorylated Y783 (Fig 4A, [47,48,53]), one interpretation of our observation could be that a change in the PLCγ1 conformation, resulting from a loss of the intramolecular interaction between Y783 and the cSH2 domain, could dramatically affect the affinity for K15P. However, we also found that the cSH2 domain of PLCγ2 directly binds to the YEEV motif of K15P (but not K15M) (Figs 5A–5C, 7B and 7D). By analogy, we postulate that the PLCγ1 cSH2 domain can also directly interact with K15P. In view of the importance of the cSH2 domain for the enzymatic activation of PLCγ1, we speculate that the recruitment of this domain by K15P could explain the constitutive activation of PLCγ1 induced by K15P [17]. We previously reported that a mutation of either the tyrosine residue in the K15P YEEV motif or the proline residues in the K15P SH3 binding motif (PPLP), reduced the interaction with PLCγ1 and that both mutations together abolished it completely [17]. We therefore conclude that, most likely, the PLCγ1 SH3 domain also contributes to the recruitment of PLCγ1 by K15, in addition to either one or both the two SH2 domains.
Since our data suggest that the interaction between K15 and PLCγ1 contributes to both angiogenesis and invasiveness in KSHV infected endothelial cells, we wanted to explore the functional effects of disrupting this interaction. We have previously shown that the commercially available PLCγ1 inhibitor, U73122, blocks K15-dependent NFAT activation [17]. However, PLCγ inhibitors have so far not been adopted for clinical use. Therefore, to elucidate the contribution of the association between K15 and PLCγ1 to virus mediated invasiveness and angiogenesis, we wanted to target more specifically the K15-dependent activation of PLCγ1. Others had already shown, as evidence of an involvement of PLCγ1 in cell motility, that a dominant negative fragment of PLCγ1 can impair the migration and invasiveness of cancer cells in vivo [54]. Based on the observation that the isolated cSH2 domain of PLCγ2 interacts with the YEEV motif of K15P (Fig 5B) and that this interaction involves R672 of the PLCγ2 cSH2 domain (Fig 5C), we explored whether the PLCγ2 cSH2 domain could act as a dominant negative inhibitor of the K15P-PLCγ1 interaction. We showed that the PLCγ2 cSH2 domain can compete with PLCγ1, GIT1 and cdc42 for the binding to K15 (Fig 6A), releasing these factors from the complex and thereby decreasing the levels of PLCγ1 phosphorylation (Fig 6B) as well as the NFAT-dependent promoter activation (Fig 6C) induced by K15. Although the binding of K15 to PLCγ1 was only moderately impaired in the presence of the highest PLCγ2 cSH2 concentration (2 μg of transfected DNA) (Fig 6A, lane 4), a significant impairment in NFAT activation was observed already with lower amounts of the transfected PLCγ2 cSH2 domain (0.5 μg of transfected DNA) (Fig 6C). This result suggests that even a partial inhibition of the K15-PLCγ1 interaction, which as discussed above, likely involves also the K15 PPLP motif and the PLCγ1 SH3 domain, is sufficient to significantly decrease the K15-dependent intracellular signalling. In line with these data, the K15-dependent phosphorylation of PLCγ1 on Y783 in endothelial cells could be efficiently inhibited by overexpressing the PLCγ2 cSH2 domain (Fig 6B).
Activation of NFAT signalling by K15M, which binds PLCγ1 through its nSH2 domain (Fig 4B), and does not associate with the PLCγ2 cSH2 domain (Fig 7B and 7D), could not be significantly inhibited by overexpressing the PLCγ2 cSH2 domain (Fig 7A and 7C), thus suggesting that the inhibitory effect of the PLCγ2 cSH2 domain on K15P is due to its ability to interact with the latter, and not due to an indirect effect on other cellular proteins. Apart from binding to the phosphate moiety, which provides half of the ligand binding energy, SH2 domains also contact their cognate binding sites through additional residues in their binding pocket, which confer increased affinity and specificity [43]. We therefore hypothesize that different residues in the K15P and M SH2 binding sites, involved in contacting the phosphotyrosine-binding pocket allow K15P, but not M, to associate with the PLCγ2 cSH2 domain.
A recent report [46] showed that the affinity of SH2 domains for their ligands can be increased by altering amino acids which flank the binding pocket for the residues surrounding the phosphorylated tyrosine in the corresponding SH2 binding site. We therefore explored if the dominant negative effect of the PLCγ2 cSH2 domain on the K15P-PLCγ1 complex could be enhanced by strengthening its binding to K15P. We found (Figs 5D–5F, 6E and 6F) that altering two residues, S677V and R695L, achieves this objective and that the resulting PLCγ2 SB mutant (S677V/R695L) shows an increased ability to inhibit the K15-dependent activation of NFAT, but not of NF-κB, thus demonstrating the specificity for the K15-dependent recruitment and activation of NFAT. Using surface plasmon resonance we observed a similar affinity of the purified PLCγ2 cSH2 wt and SB domains for a purified GST-fusion protein containing the K15 cytoplasmic domain. On one hand, this experiment confirms that the PLCγ2 cSH2 wt and SB interact directly with the K15 cytoplasmic domain with significant affinity (KD of 4–6 x 10−9). On the other hand, we only observed a moderate increase in affinity of the PLCγ2 cSH2 SB domain over the wt counterpart, unlike Kaneko et Al [46] who found a 300 fold increase in the affinity of the Fyn SH2 domain SB to the phosphorylated EGFR peptide. Since tyrosine phosphorylation (on the SH2 binding site) increases the affinity for the SH2 domain on the order of 100 fold [42,43], a possible interpretation of our results is that the K15 fusion protein used in our SPR assay was not phosphorylated and we therefore failed to observe an increased affinity of the SB domain when using recombinant proteins, in contrast to the results obtained when we analysed the binding of PLCγ2 cSH2 wt and SB domains in lysates of transfected cells (Figs 5E, 5F, 7B and 7D).
We could also show that the PLCγ2 cSH2 domain was able to impair the K15- and KSHV-mediated angiogenic effect (angiogenic tube formation) as well as the increased invasiveness of primary and immortalised endothelial cells (Fig 8 and S3 Fig). Together, the reported results demonstrate an involvement of the KSHV K15 protein in the increased migration, invasiveness and angiogenesis of KSHV infected endothelial cells. This function of K15 relies on the recruitment and activation of PLCγ1 and downstream signalling complex involving βPIX, GIT1 and cdc42. Therefore, at the molecular level, K15 exploits cellular mechanisms of integrin-dependent cell motility.
Our results, though, do not exclude the previously described involvement of other viral proteins [12,15,16,49,55] in KSHV-induced invasiveness and angiogenesis. In fact, the observation that K15 overexpression alone can induce invasiveness (Fig 1B and 1C) whereas in KSHV infected endothelial cells, the contribution of K15 the increased invasiveness only becomes measurable if there are lytic cells present in the culture (Fig 1D and 1E), suggests that, when expressed in its physiological context from the viral genome, K15 is necessary, but not sufficient, for KSHV-induced invasiveness and angiogenesis. We believe that the difference in invasiveness between K15 overexpression and its expression in the context of KSHV genome is due to a much broader range of cellular genes being induced by the former [17]. It is therefore plausible that, in KSHV infected cells, other viral proteins may complement the angiogenic effect of K15. This may include viral proteins typically expressed during the lytic replication cycle, which would explain why we observe increased invasiveness/angiogenesis (and a role for K15) primarily in cell cultures in which the lytic (productive) replication cycle has been induced (Figs 1D, 1E; and 8D). Taken together, our results indicate that the necessary role of K15 in KSHV-dependent angiogenesis/invasiveness could be a potential therapeutic target. We identified the isolated cSH2 domain of PLCγ2, as well as its SB derivative, as dominant negative inhibitors of K15P-dependent NFAT signalling and the resulting increased invasiveness and angiogenesis of endothelial cells infected with a KSHV strain carrying the K15P allele. This observation could provide the basis for developing small molecule inhibitors that would target the ability of KSHV to increase the invasiveness and angiogenesis of infected endothelial cells. Our findings therefore provide a proof of principle that the association between K15 and PLCγ1 represents a potential therapeutic target for the development of a protein-protein interaction (PPI) inhibitor able to counteract the increased invasiveness and angiogenesis of virus-infected cells. Although we have so far only been able to antagonize the recruitment of PLCγ1 by K15P, and not by its other allele, K15M, our results may provide the foundation for a search for dominant negative protein fragment-based inhibitors or small molecules which will also target K15M and help to mitigate some of the pathogenic features of KSHV in infected endothelial cells.
The use of human umbilical cords was approved by the Hannover Medical School Ethics Committee and experiments were performed in agreement with the Declaration of Helsinki. Written informed consent was obtained from parents of umbilical cord donors.
GST-K15 contains a synthetic K15P cytoplasmic tail (aa 347–489)(sK15P), with a nucleotide sequence optimized for prokaryotic expression (GENEART GmbH). The sK15P fragment was inserted into pGEX-6P-1 using EcoRI and BamHI sites.
The K15-P and M wt constructs were previously described [22,25,56]. The NF-κB reporter p3EnhκB conA–Luc construct was kindly provided by A. Eliopoulus (University of Crete Medical School, Heraklion, Greece). The reporter vector pNFAT-TA-Luc, which contains three NFAT binding sites cloned from IL2 promoter upstream of the firefly luciferase gene, and the corresponding control vector pTA-Luc were purchased from Clontech. Vectors expressing PLCγ1wt and PLCγ2 specific array domains (cSH2, nSH2, SH3, spPH) were kindly provided by M. Katan (University College London, London, UK); all of these were based on the pTrieX-4 vector [53]. PLCγ1 R586L and PLCγ1 R694L were generated by site-directed mutagenesis of PLCγ1 wt using the following primers: F 5′-CTTCCTCGTGCTAGAGAGTGA-3′; R 5′-CTCACTCTCTAGCACGAGGAA-3′; F 5′-CCTTCCTGGTGCTGAAGCGGAATGAACCC-3′; R5′-GGGTTCCTTCCGCTTCAGCACCAGGAAGG-3′, respectively. PLCγ2 cSH2 R672L was generated by site-directed mutagenesis of PLCγ2 cSH2 using the primers F 5′-CTTCCTGATCCTGAAGCGAGAGGG-3′ and R 5′- CCCTCTCGCTTCAGGATCAGGAAG-3′. PLCγ2 cSH2 S677V was generated by site directed mutagenesis of PLCγ2 cSH2 wt using the primers:F 5`-CATAGGAGTCGACCCCCTCTC-3`and R5`-GACAGGGGGTCGACTCCTATG-3`. PLCγ2 cSH2 R695L was generated by site directed mutagenesis of PLCγ2 cSH2 wt using the primers:F5`-AAAGCATTGTCTCATCAACCG-3`; and R5`-CGGTTGATGAGACAATGCTTT-3`. To generate the PLCγ2 cSH2 S677V/R695L (SB) mutant, the S677V mutant was amplified using the R695L primers. In order to generate the pRRL.PPT.SF.PLCγ2 cSH2 lentiviral construct, the DNA fragment containing PLCγ2cSH2 fused with the pTriEx-4 containing tags (S-tag, and His-tag) was amplified using the primers F5′-CCTGCTAGCTCGGACCGAAATTAATACG3′and R5′-GGTACATGTTTACGTTGAGGAGAAGCCCGG-3′. The amplified segment was then inserted in the lentiviral pRRL.PPT.SF.GFP vector (kindly provided by A. Schambach, Hannover Medical School) using the NheI and BsrgI sites.
All cells were cultured at standard conditions: 37°C in a 5% CO2 incubator. HEK293T (ACC 305 from the German Collection of Microorganisms and Cell Cultures-DMSZ) HEK293 (ATCC CRL-1573) and HeLa CNX (ACC 57 from the German Collection of Microorganisms and Cell Cultures-DMSZ) cells were cultured in Dulbecco Modified Eagle′s Medium (DMEM) (Cytogen) supplemented with 10% foetal calf serum (Hyclone). HEK293 stably infected with KSHV BAC36 wt or ΔK15 have been previously described [17]. Human umbilical vein endothelial cells (HUVECs) were freshly isolated from umbilical cords by collagenase digestion as described previously [57] and grown in EGM-2MV medium (Lonza). HuAR2T, an endothelial cell line obtained from HUVECs conditionally immortalized with a doxycycline inducible human telomerase reverse transcriptase (hTERT) and simian virus 40 (SV40) large T antigen transgene expression [58], (a kind gift of Dagmar Wirth, HZI, Braunschweig) was grown in EGM-2MV medium containing 200 ng/ml of doxycycline. The HuAR2T cell line stably harbouring recombinant KSHV.219 (r.KSHV.219) was obtained as previously reported [59]. For transfection of HeLa CNX and HEK293T, either 1-2x 105cells/ml were plated in six well plates, twenty-four hours later FuGENE transfection reagent (Promega) was used at a ratio 3μl:1μg of DNA. 1μg of each construct was transfected, unless otherwise stated. For transfection of small interfering RNA (siRNA) into HUVECs and HuAR2T, 100pmole of siRNA were transfected into 105 cells using the Neon transfection system (Invitrogen) according to the manufacturer′s instructions. The following siRNAs (siGenome SMART pool) were purchased from Dharmacon, Thermo Scientific: control (non-targeting siRNA pool 2 D-001206-14-20), siPLCγ1 (M-003559-01), siβPIX (M-009616-00), siGIT1 (M-020565-02) and siRNA against KSHV K15 protein targeting exon 8 (CAACCACCUUGGCAAUAAU) [17,22,25]. For retrovirus production, HEK293T were transfected using the calcium phosphate method with either pSF91-K15-IRES or pSF91-IRES vector [17], together with the packaging plasmids pM57DAW (gag/pol) and pRD114 envelope protein. Retroviruses containing pSF91-IRES or pSF91-IRES-K15 were produced as described previously [17]. For lentivirus production, HEK293T were transiently co-transfected using the calcium phosphate method with the helper plasmids (pMDLGg/p, pRSV-REV and pMD.G) and either pRRL.PPT.SF.GFP or the pRRL.PPT.SF.PLCγ2 cSH2 plasmid; lentiviral stocks were produced as described previously [59]. For transduction, cells were transduced with the indicated lenti- or retrovirus in the presence of 5 μg/ml of polybrene and centrifuged for 30 min at 450g. Medium was changed 8 hours post transduction. Experiments were performed with 60–70% of transduced cells, 48 hours post transduction. Sf9 insect cells (ACC 125 from the German Collection of Microorganisms and Cell Cultures-DMSZ) were grown in Grace′s medium (Gibco) supplemented with 10% fetal bovine serum, 100 U/ml of penicillin and 50 ug/ml of streptomycin. Cloning and production of RTA expressing baculovirus were previously described [52].
rKSHV.219 was produced from BJAB (ACC 757 from the German Collection of Microorganisms and Cell Cultures-DMSZ) infected with rKSHV.219 as described in [60]. Briefly, 6x105 BJAB rKSHV.219 were grown for seventy-two hours in spinner flasks (agitation at 60rpm) in RPMI1640 medium supplemented with 10% FCS and in the presence of 2.5 μg/ml of anti-IgM antibody. Subsequently, the culture was centrifuged and the supernatant was first filtered (0.45 μm pore size filter) and then centrifuged 5 hours at 15000 rpm. The pellet was resuspended in EBM2 medium.
A codon optimized version for bacterial usage of the cytoplasmic tail of K15P (aa 347–489) was purchased from GENEART and cloned in pGEX-6P using BamHI and EcoRI restriction sites. The plasmid was transformed in Rosetta E. coli and grown at 37°C in LB medium with ampicillin. Once an optical density at 600nm (OD600) of 0.8 was reached, bacterial cultures were induced for 4 hrs with 1mM isopropyl-γ-D-thiogalactopyranoside (IPTG) at 30°C and subsequently pelleted by centrifugation. Bacteria were lysed in 1x PBS plus protease inhibitors and sonicated 3 times for 30 seconds. Subsequently, 0.1% Triton-X 100 was added and the lysate was cleared by centrifugation at 14000 rpm for 10 minutes. Cleared lysate was incubated with glutathione sepharose beads (Amersham Bioscience) overnight at 4°C. The PLCγ1 and 2 cSH2 domains were produced and purified as previously described [53]. Briefly, plasmids harbouring the PLCγ1 or 2 cSH2 domains were transformed into Rosetta E.coli and grown to an OD600 of 0.4. Protein expression was induced with 100 μM IPTG for approximately 10 hours at 25°C. Bacteria were pelleted and stored at -20°C until further processing. Pellets were resuspended in 10 ml of chilled lysis buffer (25mM tris HCl, 250mM NaCl, 40mM imidazole, 10mM benzamidine, 1mM MgCl2 and 100μM CaCl2) per 1L of culture. Lysis was further accomplished by sonication (3x30 sec) and by the subsequent addition of 10% of Triton X-100. Bacterial lysate was cleared by two rounds of 15 min centrifugation at 14000g at 4°C. Supernatant was then incubated with 2 ml washed Ni beads (Quiagen), for 4 hours rolling at 4°C. Subsequently beads were washed 3 times with lysis buffer and stored at -80°C.
50 μg of the purified GST-K15 (aa 347–489) were injected intraperitoneally (i.p.) and subcutaneously (s.c.) into LOU/C rats using incomplete Freund's adjuvant supplemented with 5 nmol CpG 2006 (TIB MOLBIOL, Berlin, Germany). After six weeks interval a final boost with 50μg K15 and CpG 2006 was given i.p. and s.c. three days before fusion. Fusion of the myeloma cell line P3X63-Ag8.653 with the rat immune spleen cells was performed according to standard procedures. Hybridoma supernatants were tested in a solid-phase immunoassay with K15-GST or GST bound to ELISA plates via mouse anti-GST antibody. Antibodies from tissue culture supernatant recognizing K15 were detected with HRP conjugated mouse mAbs against the rat IgG isotypes (TIB173 IgG2a, TIB174 IgG2b, TIB170 IgG1 all from ATCC, R-2c IgG2c homemade), thus avoiding mAbs of IgM class. HRP was visualized with ready to use TMB (1-Step Ultra TMB-ELISA, Thermo). Monoclonal antibodies (mAbs) that reacted specifically with the K15 were further analysed by western blot and IFA. In experiments, antibodies from clone number 10A6 (IgG G1) and 18E5 (IgG 2b) were used for western blot and immunofluorescence, respectively.
In order to assay the expression levels of specific proteins, cells were lysed in 1x SDS sample buffer (62.5 mM tris-HCl pH 6.8, 2% w/v) SDS, 10% glycerol, 50 mM DTT, 0.01% (v/v) β-mercaptoethanol and 0.01% (w/v) bromophenol blue). For the detection of K15 protein, samples were not boiled prior to SDS-PAGE. Proteins were separated by SDS-PAGE and transferred to nitrocellulose membrane (Amersham). Membranes were blocked 1 hour in 5% (w/v) milk in PBS-T. Specific proteins were identified using the following primary antibodies: rat monoclonal anti-K15 antibody (see above), rabbit anti-PLCγ1(#2822), rabbit anti-phospho Y783 PLCγ1(#,2821), rabbit anti S tag (#8476), rabbit anti-cdc42 (#2462) were purchased from Cell Signaling Technology; mouse anti-KSHV ORF 45 (sc-53883), mouse anti-KSHV Kb-Zip (sc-69797), mouse anti-βPIX (sc-136035) were obtained from Santa Cruz Biotechnolocy Inc.; rat anti KSHV ORF73 (LNA-1)(13-210-100) and mouse anti-β-actin (A5441) were purchased from Advanced Biotechnologies, and Sigma Aldrich, respectively; mouse anti GIT (611388) was obtained from BD Transduction Laboratories; mouse anti KSHV ORF 59 was purchased from Advanced Biotechnology Inc. (13-211-100). A polyclonal rabbit antibody to the FLAG epitope was purchased from Sigma (F7425). The HRP-labelled polyclonal secondary antibodies: goat anti rabbit (P0448), rabbit anti mouse (P0260) and rabbit anti rat (P0450) were purchased from DAKO.
Matrigel-coated invasion chambers (354483, BD Biosciences) were used. HuAR2T or HuAR2Tr.KSHV were transfected with the indicated siRNA, twelve hours later cells were transduced with the indicated lenti- or retroviral vectors. Twenty-four hours after siRNA transfection, the lytic cycle was induced with 10% or 15% RTA (v/v) supernatant and 1mM sodium butyrate; twenty-four hours later, cells were starved in EBM2 supplemented with 2% FBS for twelve hours and subsequently cells were seeded in quadruplicate in Matrigel-coated invasion chambers under starvation conditions and were allowed to invade for twenty-four hours.
Alternatively, uninfected HEK293 cells, or HEK293 cells stably harbouring KSHV BAC36 wt or a ΔK15 mutant, were plated on a 6-well plate at a density of 6X105 cells per well and the lytic cycle was induced twenty-four hours later. Cells were starved overnight in DMEM and thirty-six hours after induction, 5x104 cells were plated in 0.5ml of DMEM with 0.1% BSA in quadruplicate in the inner insert of an invasion chamber. In the outer insert 0.75ml of DMEM supplemented with 5% FCS was added and the cells were incubated for twenty-four hours. Before plating of cells on the inner chamber, the Matrigel inserts were rehydrated with 0.5 ml of DMEM for two hours. Subsequently, cells were washed in 1xPBS, fixed in 4% PFA, permeabilized with 0.2% Triton X-100, stained with DAPI (Sigma-Aldrich) and the bottom of the chambers was cut out and mounted on coverslip,. Invading cells were counted using a fluorescence microscope. Cells were counted in four random fields per chamber using Cell Profiler 2.0 software [61]. Each experiment was performed three times. To determine whether there was a statistically significant difference between the different conditions, a Kruskall-Wallis test with a Dunn’s post- test was performed.
2.5x104 HuAR2T r.KSHV.219 cells were plated on glass coverslips and thirty-six hours later washed with 1xPBS and fixed with 100% ice cold methanol for 20 min at -20°C. Cells were then thoroughly washed with 1xPBS and incubated for one hour in 10% FCS in PBS at 37°C. Coverslips were then incubated with a rat monoclonal antibody to K15 (clone 18E5) for one hour at 37°C, washed 3 times in 1x PBS and incubated for one hour with Cy3-labelled anti-rat secondary antibody (712-165-153 Jackson Immuno Reasearch) for one hour at 37°C. Coverslips were washed, fixed with 4% PFA 20 minutes, PFA was quenched with 1xTBS 10 min at room temperature. Coverslips were then stained as indicated. The following fluorescently labelled secondary antibodies were used: FITC conjugated donkey anti-rabbit (711-095-152 Jackson Immuno Research), Cy5 conjugated goat anti-mouse (115-175-164 Jackson Immuno Research). Images were taken with ZEISS LSM 510 Meta scan head connected to an inverted microscope Axiovert 200M. To analyse the co-localisation the JACOP tool was used and the Pearson’s coefficient (PC) was calculated and shown in Figs 2 and 3A[52].
Thirty hours after transient transfection with the indicated plasmid, HEK293T were lysed in IP buffer (150mM NaCl, 25mM tris HCl pH7,6, 1mM EDTA, 1% glycerol, 0.2% NP-40). 200 μl of precleared lysate was incubated overnight with gentle shaking at 4°C with appropriate beads, as indicated in the figure legends. For the S-tag constructs S protein agarose beads (Novagen) were used. For PLCγ1 immunoprecipitation, protein A sepharose beads (GE Healthcare) were incubated with pre-cleared HuAR2T r.KSHV.219 lysate and anti-PLCγ1 antibody according to the manufacturer′s instructions. The His-tag pull-down assay was performed by incubating overnight, with gentle shaking at 4°C, equal amounts of Ni beads (Quiagen) bound with the indicated fusion proteins together with pre-cleared lysate of HEK293T cells. GST pulldown assays were performed by incubating equal amounts of GST-fusion protein and control GST coated beads with pre-cleared cellular lysate of HEK293T cells transfected with the indicated constructs overnight with gentle shaking at 4°C. Beads were washed three times in ice cold lysis buffer and the bound proteins complexes were analysed by SDS-PAGE and western blot.
HEK293T or HeLa cells were transiently transfected in six well plates with 50 ng of reporter plasmid and the indicated expression constructs. Forty hours after transfection cells were harvested in 300 μl/well of 1x reporter lysis buffer (Promega). Luciferase activity was measured in cleared lysates with a luciferase buffer according to the manufacturer’s instructions (Promega). Each experiment was performed at least three times in duplicate; the error bars in the graphs represent the standard deviation across all three experiments.
Surface plasmon resonance (SPR) was performed using a Biacore X100 (GE Healthcare) instrument. Purified recombinant GST-K15P cytoplasmic tail, dissolved in acetate buffer at pH 5.5, was immobilized in the Flow cell 2 of a CM4 chip by amine-coupling until reaching 207 relative units (Rmax approx. 100 RU). For binding assays, purified recombinant His-tagged wild type and mutant PLCγ2 cSH2 domains were injected at different concentrations in HBS-EP buffer (10 mM Hepes, 150 mM, NaCl, 3 mM EDTA, 0.005% (vol/vol) surfactant P20, pH 7.4) at a flow rate of 10 μl/min. To determine the association and dissociation constants single cycle kinetics analysis was used. Briefly, 1.5, 3, 6, 12, 24 nM of wild type and SB PLCγ2 cSH2 domain were injected at a flow rate of 30 μl/min in HBS-EP buffer during 120 sec followed by 900 sec of dissociation. The chip surface was regenerated by injecting 2M MgCl2. The sensorgrams obtained were analysed with the Biacore X100 Evaluation 2.0.1 software. Bulk refractive index changes were removed by subtracting the reference flow cell (Flow cell 1) responses, and the average response of a blank injection was subtracted from all analyte sensorgrams to remove systematic artefacts.
HUVECs were either transduced with the indicated retroviral vectors or infected (MOI 20) with rKSHV.219. In the case of transduced cells, they were starved in EBM2 medium supplemented with 2% FCS for thirty hours after transduction. Forty-eight hours after transduction the assay was performed. Three days after infection with recombinant KSHV, HUVECs were transduced, with the lentiviral vector carrying either the PLCγ2 cSH2 domain or the control vector. Four days after infection, infected HUVECs were treated with 1mM of sodium butyrate and 10–15% RTA to induce the lytic cycle. The capillary tube formation assay was performed thirty-six hours after reactivation. For this assay, 8x103 cells were plated in wells precoated with growth factor reduced matrigel (BD Biosciences) and incubated in a 37°C, 5% CO2 incubator according to the manufacturer′s instructions and as previously reported [17]. For each well four different fields were photographed with a NIKON T200 fluorescence microscope. The angiogenic index was calculated as the number of branching points in a visual field after 4–6 hours of incubation. The number of branching points was averaged and standard deviation ±95% CI was calculated. Each experiment was performed independently three times in quadruplicates.
To determine whether there was a statistically significant difference between the different conditions, a Kruskall-Wallis test with a Dunn’s post- test was performed.
|
10.1371/journal.pntd.0004018 | Differential Gene Expression and Infection Profiles of Cutaneous and Mucosal Leishmania braziliensis Isolates from the Same Patient | Leishmaniasis is a complex disease in which clinical outcome depends on factors such as parasite species, host genetics and immunity and vector species. In Brazil, Leishmania (Viannia) braziliensis is a major etiological agent of cutaneous (CL) and mucosal leishmaniasis (MCL), a disfiguring form of the disease, which occurs in ~10% of L. braziliensis-infected patients. Thus, clinical isolates from patients with CL and MCL may be a relevant source of information to uncover parasite factors contributing to pathogenesis. In this study, we investigated two pairs of L. (V.) braziliensis isolates from mucosal (LbrM) and cutaneous (LbrC) sites of the same patient to identify factors distinguishing parasites that migrate from those that remain at the primary site of infection.
We observed no major genomic divergences among the clinical isolates by molecular karyotype and genomic sequencing. RT-PCR revealed that the isolates lacked Leishmania RNA virus (LRV). However, the isolates exhibited distinct in vivo pathogenesis in BALB/c mice; the LbrC isolates were more virulent than the LbrM isolates. Metabolomic analysis revealed significantly increased levels of 14 metabolites in LbrC parasites and 31 metabolites in LbrM parasites that were mainly related to inflammation and chemotaxis. A proteome comparative analysis revealed the overexpression of LbrPGF2S (prostaglandin f2-alpha synthase) and HSP70 in both LbrC isolates. Overexpression of LbrPGF2S in LbrC and LbrM promastigotes led to an increase in infected macrophages and the number of amastigotes per cell at 24–48 h post-infection (p.i.).
Despite sharing high similarity at the genome structure and ploidy levels, the parasites exhibited divergent expressed genomes. The proteome and metabolome results indicated differential profiles between the cutaneous and mucosal isolates, primarily related to inflammation and chemotaxis. BALB/c infection revealed that the cutaneous isolates were more virulent than the mucosal parasites. Furthermore, our data suggest that the LbrPGF2S protein is a candidate to contribute to parasite virulence profiles in the mammalian host.
| Leishmaniasis is a critical public health problem worldwide. The clinical outcome of leishmaniasis depends on the infecting parasite species, host genetics and immune response and insect species. Leishmania braziliensis is a major etiological agent of cutaneous and mucosal leishmaniasis in Brazil. Fewer than 10% of L. braziliensis-infected patients with CL develop the mucosal form (a severe clinical manifestation). The small number of parasites in the mucosae increases the difficulty of obtaining clinical isolates, and parasite samples are frequently derived from individuals with different genetic backgrounds. Therefore, clinical isolates from cutaneous and mucosal sites from the same patient represent unique tools to understand parasite factors that contribute to disease outcome and pathogenesis. In this study, we investigated parasite factors involved in disease progression using two pairs of L. (V.) braziliensis isolates from mucosal (LbrM) and cutaneous (LbrC) sites of the same patient. In conclusion, the murine infection and proteome and metabolome data suggest that the differences between the cutaneous and mucosal isolates are mainly related to inflammation and chemotaxis. Our data also suggest that the LbrPGF2S protein plays a role in parasite virulence in the mammalian host.
| Leishmaniases, which are endemic in 98 countries (predominately in tropical and subtropical regions worldwide), represent a critical public health problem [1]. These diseases develop distinct clinical manifestations depending on the infecting Leishmania species, the composition of the sandfly vector saliva and the mammalian host’s genetic and immunological profile [2–4]. The output of infection varies widely; the symptomatic forms may be subdivided into tegumentary and visceral diseases. The tegumentary diseases may develop from mild (localized cutaneous leishmaniasis—LCL) to severe forms that include diffuse cutaneous leishmaniasis (DCL) and mucocutaneous disease (MCL) [5].
Approximately 20 species of Leishmania cause human infection, and tegumentary diseases may be caused by several species in different endemic countries. Each clinical form has been linked to one or a few species. For example, the Leishmania species from the Viannia subgenus (Leishmania (V.) braziliensis, Leishmania (V.) guyanensis and Leishmania (V.) panamensis), which are widely distributed in the Americas, are associated with not only CL but also MCL, which emerges in 5–10% of L. braziliensis infections [6–8]. In MCL, the oral and nasopharyngeal areas of the face are the most commonly affected and display tissue destruction characterized by intense inflammation and a low parasite load [7, 9]. The parasite factors determining the disease have been widely explored (but remain poorly understood) and it has been suggested that the L. braziliensis genotypes may be associated with specific disease manifestations [10].
The mechanisms that trigger the migration of L. braziliensis from the primary cutaneous site of infection to the facial mucosae are not understood. Various studies have suggested that migration of macrophages after infection may play a role in the dissemination of the parasite to other body regions, contributing to the development of MCL [11, 12]. However, the role of the parasite in the divergent behavior of host cells is unknown.
Although research using parasites collected from different patients are relevant to improving our understanding of mucosal disease, each patient has a distinct immunogenetic background and may respond to parasite infection differently. Therefore, clinical isolates from cutaneous and mucosal sites on the same patient represent unique tools that can be used to understand parasite factors that contribute to disease outcomes and pathogenesis.
Typically, mucosal lesions are diagnosed months or years after the primary cutaneous lesion. One study in an endemic area in Brazil evaluated 200 patients with CL by otorhinolaryngological examination and detected parasites in the nasal mucosae of six patients despite the absence of mucosal lesions [13]. There may be genetic differences between parasite populations isolated from cutaneous and mucosal sites. The genomes of the three Leishmania species L. (L.) major, L. (L.) infantum and L. (V.) braziliensis are highly conserved at the level of protein-coding gene content; however, the ample genetic plasticity of Leishmania is clear, suggesting that the key to understanding the diverse clinical manifestations, virulence and tropism of the parasite may be differential genome expression [10, 14, 15].
Here, we describe comparative expressed genome analysis and infectivity studies of two pairs of clinical isolates obtained from the primary site of infection and the mucosae of the same patient [13].
Leishmania (Viannia) braziliensis were rescued from the biopsies of cutaneous lesions and mucosae of two patients from the endemic area of Jequié (Bahia/Brazil) with mucocutaneous leishmaniasis [13]. The isolates from cutaneous lesions were denoted LbrC1 (MHOM/BR/00/BA778) and LbrC2 (MHOM/BR/00/BA776), whereas parasites recovered from the mucosae were denoted LbrM1 (MHOM/BR/00/BA779) and LbrM2 (MHOM/BR/00/BA777); the numbers 1 and 2 refer to the respective patients. We have employed this simplified notation for the isolates throughout the text to facilitate comprehension. Wild type and transfectant promastigotes were maintained at 26°C in M199 medium supplemented as previously described [16]. Transfectants were maintained in liquid medium containing 6 x LD50 G418.
PCR was performed using GoTaq Flexi DNA Polymerase (Promega, Madison, USA) according to the manufacturer’s instructions. To confirm that the isolates belonged to the Viannia subgenus, genomic DNA (gDNA) extracted from the parasites was subjected to PCR using the primers 5’-CGGATCGCCCATGTACTC-3’ and 5’-GCATCGCAATAGTCCCACAT-3’ to amplify LbrM.23.0390 (RNAse III), which is specific to this subgenus [17].
All of the isolates were analyzed for the presence of the LRV virus (Leishmania RNA virus) by RT-PCR. cDNA was generated using reverse transcriptase (Invitrogen) and 1 μg of the extracted RNA according to the manufacturer’s protocol. PCR was performed using the primers LRV-for (5’-CGGTAGAGCATTAAGGGCTAGC-3’) and LRV-rev (5’-CGGCAGTAACCTGGATACAACC-3’), which amplify a genomic region conserved among all virus subtypes (kindly shared by MVG da Silva).
Low-melting agarose plugs containing gDNA were prepared according to the protocol of Beverley et al. (1988) and Cruz et al. (1991) [18, 19]. The gDNA was fractionated in a 1% agarose gel by PFGE using two different programs. Program 1 was used to separate large chromosomes (4.5 V/cm, 144 h, 16°C, initial pulse: 360 s and final pulse: 800 s, 1x TBE buffer), whereas program 2 was used for smaller chromosomes (4.5 V/cm, 44 h, initial pulse: 50 s and final pulse: 120 s, 0.5x TBE buffer). The gels were stained with ethidium bromide for 90 min.
The use of mice and hamsters was approved by the Ethical Commission of Ethics in Animal Research (CETEA) at the Ribeirão Preto Medical School, University of São Paulo. They certified that Protocol n° 159/2011 (“Investigation of host-parasite interaction: exploring models of study of virulence and tropism”) is consistent with the ETHICAL PRINCIPLES IN ANIMAL RESEARCH adopted by Brazilian College of Animal Experimentation (COBEA) in 8/27/2012.
Hamsters were obtained from the facilities of the ANILAB Animais de Laboratório (Street Servidão Quatro, 292, Paulínia, SP, Brazil), and BALB/c mice were obtained from the institutional animal facility (Ribeirão Preto Medical School, USP, Ribeirão Preto, SP, Brazil). Mouse infection was conducted according to the guidelines of the Ethics Committee on Animal Experimentation from Ribeirão Preto Medical School, USP.
BALB/c mice and hamsters were infected in vivo with stationary-phase promastigotes (105 viable parasites/10 μL PBS) by intradermal injection into the right ear (5 animals per experiment). Lesion progression was recorded once each week for six weeks by measuring ear swelling with a digital Vernier caliper using the non-infected contralateral ear as a control [20]. The parasite load was determined in the ear 4 weeks p.i. as described previously [21].
Peritoneal macrophages from BALB/c mice were maintained in RPMI medium supplemented with 10% (v/v) FBS. The cells (5x105) were infected with late stationary promastigotes at a ratio of 10 parasites per macrophage for 4 h at 37°C. The infected cells were washed 3 times with incomplete RPMI 1640 (Life Technologies, Carlsbad, CA, USA) to remove non-internalized promastigotes and incubated at 5% CO2, 37°C for 0, 24 and 48 h. At the end of the assay, the infected macrophages were stained using the Diff Quick kit (LABORCLIN, Pinhais, Paraná, Brazil), and intracellular parasites were counted using a Leica DM500 microscope with a 100x objective. The parasite burden was verified by counting the number of infected macrophages in 300 cells (technical triplicates).
The production of IL-4 and IFN-γ cytokines was quantified using ELISA (Mouse IL-4 ELISA kit and Mouse IFN-γ ELISA kit- BD OptEIA) according to the manufacturer’s protocol (BD Biosciences, San Diego, CA, USA). Cell suspensions were prepared from the lymph nodes and spleens of BALB/c mice infected for 4 weeks with LbrC1, LbrM1, LbrC2 and LbrM2 promastigotes. Overall, 5x106 cells/mL were plated per well in 24-well tissue culture plates and stimulated with 40 μg/mL L. braziliensis particulate antigens (SLA) as previously described [22].
The protein extracts of logarithmic phase promastigotes (5x108 parasites) were obtained by precipitation with trichloroacetic acid, and the samples were subjected to 2D gel electrophoresis using Immobiline DryStrip gels (13 cm/4-7 pI) (GE HealthCare, Piscataway, New Jersey, USA). Protein extraction and 2D electrophoresis were performed as described previously [23, 24]. A comparative analysis of the digitized proteome maps of the LbrC and LbrM isolates was performed using ImageMaster platinum v6.0 software (GE Healthcare). Genes were considered differentially expressed when the spot intensity was increased 1.5-fold between isolate pairs. All of the analyses were performed in biological triplicate, and we used a two-sample t-test to compare the differentially expressed spots. The peptides were identified by MALDI-TOF/TOF mass spectrometry at the Center for Protein Chemistry (University of São Paulo, Ribeirão Preto) and analyzed using the MASCOT program (Version 2.2.04) and the GeneDB genome of L. braziliensis.
The LbrPGF2S CDS flanked by 800 bp was amplified from the gDNA of L. (V.) braziliensis strain MHOM/BR/75/M2904 using the primers 5’FLR-LbrPGF2S_NheI-for (5’-GCTAGCAGGTGTGCTACAGGTAAGGAAGC-3’) and 3’FLR-LbrPGF2S_HindIII_XbaI-rev (5’-TCTAGAAAGCTTGCATGAAGAAGAGGGTCCAG-3’) and cloned into the pGEM-T easy vector (Promega) according to the manufacturer's instructions. The 2.3-kb fragment resulting from digestion with the NheI and HindIII enzymes containing LbrPGF2S CDS and its 3’ and 5’ UTRs [25] was subcloned into the pXNEO vector [26] for overexpression in Leishmania.
Promastigotes were transfected with the pX63NEO and pX63NEO-LbrPGF2S plasmids by electroporation. Transfectants were selected in semi-solid M199-agar medium in the presence of the G418 antibiotic (Sigma, St. Louis, MO). The G418 LD50 was determined for each isolate, and four- or six-fold LD50 was used. Promastigotes were maintained at 26°C in M199 medium supplemented as previously described [16]. For in vitro infection, the mutant promastigotes were cultivated in a liquid medium containing six-fold G418 LD50.
Total protein extracts were obtained from 1x107 promastigotes by precipitation with trichloroacetic acid (a similar extraction was used for the 2D analysis). The proteins were fractionated by 12% SDS-PAGE and blotted onto Hybond ECL membranes in a TE22 mini transfer unit (both from GE Healthcare) [23]. The membranes were blocked in 3% BSA blocking buffer for 1 h, incubated with chicken anti-LbrPGF2S (1:10000) (produced by our group) for 1 h, washed, and incubated with peroxidase-conjugated anti-chicken IgY (1:800,000) (Sigma) for 1 h at room temperature. Antigen-antibody interactions were detected using an ECL kit (GE Healthcare); chemiluminescence was visualized using an ImageQuant LAS 4000 (GE Healthcare).
Genomic DNA from each isolate (LbrC1, LbrC2, LbrM1 and LbrM2) was sequenced from paired-end DNA libraries constructed with the Nextera XT DNA Sample Preparation kit. The Illumina MiSeq platform (FMRP-USP) was used for library sequencing (150-nt long reads). Read quality, length and number were verified using the FastQC tool (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/); reads with an average quality value of less than 30 were removed. The Illumina adapter was removed using cutadapt software (version 1.4.1) (http://journal.embnet.org/index.php/embnetjournal/article/view/200).
The reads were aligned against the LbrM2903 version 8.0 reference genome (S1 Table) available at TriTrypDB [27] using BWA [28] (Version 0.7.10-r789) to generate alignments in the sam format. The BWA-MEM algorithm with default values was applied for 150-bp Illumina reads. SAMtools [29] was used to convert the sam files into binary format and sort, index and count the reads from each chromosome in bam files. These bam files were visualized with Artemis [30]. The chromosome somy in each library was calculated independently according to the method of Zhang et al. [31].
For each compared group, LbrC and LbrM, seven biological replicates were collected in the late-log phase of promastigote culture. Before metabolite extraction, metabolism was quenched by shaking the 10-mL culture bottle in an ethanol/dry-ice bath for 40 sec. For HPLC-MS and CE-MS, 4x107 promastigotes were centrifuged at 2,000 x g for 10 min at 4°C, washed 3 times in cold (4°C) PBS (137 mM NaCl, 8 mM Na2HPO4, 2.7 mM KCl, and 1.5 mM KH2PO4; pH 7.0) and lysed in 450 μL of cold (4°C) CH3OH/H2O (4:1, v/v). The cells were mechanically disrupted by 3 freeze/thaw cycles in liquid N2, followed by lysis for 10 min at 50 Hz in a TissueLyser LT (Qiagen) with glass beads (50 mg, 425–600 μm, Sigma). The cellular debris was removed by centrifugation at 15,700 x g at 4°C for 10 min. For HPLC-MS, 200 μL of clarified supernatant was transferred into a glass vial and submitted to HPLC-MS analysis. For CE-MS, 200 μL of clarified supernatant was transferred into a new tube, dried and re-suspended in 200 μL of milli-Q water.
For GC-MS, metabolites were extracted by the same process in 350 μL of CH3OH/CHCl3/H2O (3:1:1, v/v/v) at 4°C. The supernatant (200 μL) was clarified by centrifugation and evaporated to dryness in a SpeedVac at 30°C. Next, 10 μL of O-methoxyamine hydrochloride (15 mg/mL in pyridine) was added to each GC vial, mixed vigorously for 5 min using a vortex FB 15024 (Fisher Scientific, Madrid, Spain), and incubated in darkness at room temperature for 16 h for methoximation. Then, 10 μL of BSTFA (N,O-bis(trimethylsilyl)trifluoroacetamide) with 1% TMCS (v/v) (trimethylchlorosilane) was added, and the vials were vortexed for 5 min and incubated for 1 h at 70°C for the silylation reaction. Finally, 100 μL of heptane containing 10 mg/mL C18:0 methyl ester (internal standard) was added, and the samples were vortexed. Two blank samples were prepared following the same extraction and derivatization procedures.
Quality controls (QCs) were independently prepared for each technique by pooling equal volumes of each sample. The controls were analyzed at the start of each analysis to reach system equilibration and throughout the run to provide a measurement of the system’s stability and the reproducibility of the sample treatment procedure.
Considering the large chemical diversity of metabolites, the samples were analyzed by HPLC-MS, CE-MS and GC-MS to ensure wide coverage encompassing hydrophobic, hydrophilic, acidic, basic and neutral molecules. The HPLC-MS, CE-MS and GC-MS instrumentation and settings for metabolomic analysis were as previously described by Canuto et al. [32].
Background noise and unrelated ions were removed from the resulting data files (HPLC-MS and CE-MS) using the Molecular Feature Extraction (MFE) tool in Mass Hunter Qualitative Analysis software (B.05.00, Agilent). Primary data treatment (filtering and alignment) was performed using Mass Profiler Professional software (B.02.01, Agilent). Data treatment for GC-MS analysis was conducted through compound identification using the Fiehn retention time locked (RTL) library and the National Institute of Standards and Technology mass spectra library with MSD ChemStation software (G1701EA E.02.00.493, Agilent) and a correct assignment based on the coincidence of the retention time and the spectrum profile [33]. For all of the analytical platforms, features that did not appear in at least 50% of the QCs with a coefficient of variation less than 30% were excluded from the analysis. The metabolic profiles were analyzed by principal component analysis (PCA) (S1 Fig). We considered a metabolite to have a differential profile between LbrC and LbrM only in the following situations: (i) when there was a statistically significant differential abundance in the samples from the two phenotypes (Student’s t-test, p value < 0.05); and (ii) when the metabolite was consistently detected in 100% of the biological replicates per group. The accurate masses representing statistically significant differences were searched in MassTrix [34] and CEU Mass Mediator (http://ceumass.eps.uspceu.es/mediator/). The heatmap was designed using MetaboAnalyst (v. 2.0) [35].
All sequence information was deposited in GenBank under bioproject ID PRJNA292004. DNA sequencing data can be accessed from the SRA database using accession no. SRP062173.
Paired clinical isolates were recovered from two patients in a leishmaniasis endemic area in Brazil where Leishmania (Leishmania) amazonensis and Leishmania (Viannia) braziliensis species are responsible for CL, according to the Manual for Surveillance of American Integumentary Leishmaniasis [36]. All of the isolates were subjected to PCR using primers to amplify a fragment of a Viannia-specific gene (LbrM.23.0390, RNase III domain gene). Reactions using parasite genomic DNA from all four isolates amplified the control DNA (SSU 18S) and the RNase III domain gene, confirming the isolation of the Leishmania (V.) braziliensis species (Fig 1A). We used a Leishmania (Leishmania) major strain as a negative control for the Viannia-specific fragment, from which no amplification was obtained (LV39 lane).
Based on a recent study suggesting a positive correlation between the presence of LRV virus in Leishmania (Viannia) spp. and metastatic behavior [37], we searched for the presence of LRV RNA in all four isolates. We could not detect the presence of the virus by RT-PCR in the LbrC and LbrM isolates; L. (V.) guyanensis (M4147 strain) was used as a positive control (Fig 1B). This result suggests that the metastatic behavior of the LbrM isolates is not associated with LRV.
To investigate possible genomic differences between the LbrC and LbrM isolates recovered from the same patient, agarose-embedded genomic DNA was fractionated by pulsed field gel electrophoresis (PFGE). Using two different pulse conditions to fractionate small/medium or large chromosomes, we observed a similar karyotype, with the exception of one extra band representing a large chromosome that was exclusively present in both LbrM isolates (Fig 1C). Comparative analysis of the signal intensity of this ‘novel’ band in the LbrM2 karyotype with the corresponding region of the LbrC2 (marked with *) is suggestive of one allele size increment. Overall, the karyotype analysis revealed high similarity among the four isolates, suggesting that all belong to the same strain.
To confirm the similarity of genomic content and investigate possible chromosome somy changes, we subjected both of the LbrC/LbrM pairs to NGS genomic sequencing. As shown in Fig 2 and S2 Table, no consistent differences between paired isolates were detected, confirming the lack of major genomic changes.
We investigated whether the parasites rescued from the primary site of infection versus those rescued from the mucosae of the same patient had a different infection profile. In vivo infection in hamsters and BALB/c mice revealed that both LbrC isolates displayed a more severe clinical manifestation. Hamsters infected with the LbrC isolates produced larger lesions than hamsters infected with the LbrM isolates (Fig 3A). BALB/c mice were used to evaluate the virulence of the isolates. The parasite burden was increased at the inoculation site (ear) of LbrC-infected animals compared to LbrM-infected mice at 4 weeks p.i. (Fig 3B). Therefore, both parasite load and lesion size measurement revealed a different murine infection profile between the LbrC and LbrM isolates. Also, significantly increased IL-4 levels were detected in cell cultures derived from the lymph nodes and spleens of LbrM2-infected mice compared with cultures from mice infected with LbrC2 (Fig 3C). In contrast, LbrC2-infected cells released more IFN-γ than cells infected with LbrM2. In both cases, the cells were stimulated with the same antigens (SLA).
To obtain an overview of the global physiological differences between LbrC and LbrM, we subjected one of the pairs (LbrC1 and LbrM1) to a comparative metabolomic analysis. The tight cluster of QCs (quality controls) in the unsupervised PCA model scatter plots for HPLC-MS, CE-MS and GC-MS confirmed technical reproducibility (S1 Fig).
S3 Table summarizes the metabolomic data obtained from the different techniques, including the monoisotopic mass, retention time, percentage change, p value and biological role.
A heat map was constructed to visualize differences in the LbrC and LbrM metabolomes (Fig 4). The heatmap revealed important differences in the intracellular concentration of metabolites. The hierarchically clustered heat map revealed 31 metabolites whose levels were significantly decreased in the LbrC parasites and 14 metabolites that were decreased in the LbrM parasites. The metabolome differences indicated a clear metabolomic dichotomy in the parasite population from the cutaneous site versus the mucosae of the same individual. Most of the metabolites that were present at different levels in the compared samples were related to inflammatory processes.
Several phospholipids and related metabolites, such as choline, saturated fatty acids (myristic and palmitic acids) and ketocholesterol, were identified among the up-regulated metabolites in the LbrM samples. Others metabolites, such as phosphatidic acid (PA) and phosphatidylglycerol (PG), were detected at lower levels in LbrM (Fig 4).
The analysis also revealed differences in “amino acids and derivatives” between the compared lines. Purine metabolism differed between the two lines, with some metabolites up-regulated or down-regulated in the metastatic line (Fig 4). Chalcone levels were higher in LbrM, whereas trypanothione disulfide (a reduced form of trypanothione) levels were lower (Fig 4). Other metabolites from different chemical classes that participate in pathways related to purine and polyamine metabolism and/or redox routes were expressed at different levels in the LbrC and LbrM samples.
As part of the global comparative analysis of the LbrC and LbrM isolates, we investigated possible modifications of gene expression between parasites from mucosal and cutaneous sites through a comparative analysis of proteome profiles. The proteomes were evaluated by protein fractionation through two-dimensional gel electrophoresis, followed by identification of differentially expressed spots by mass spectrometry.
We considered a differential expression positive if the differences in spot signal intensities were greater than 1.5-fold between the compared proteomes. In each replica of the fractionated proteome, we detected 477, 488 and 525 spots for LbrC1, 358, 351 and 434 spots for LbrM1, 448, 440 and 345 spots for LbrC2 and 406, 346 and 387 spots for the LbrM2 samples. The percentage of corresponding (matched) spots among the LbrC1/LbrM1 and LbrC2/LbrM2 pairs was satisfactory (61.10% and 70.49%, respectively).
Differentially expressed spots were excised from the gels and subsequently analyzed by mass spectrometry. Twenty-four polypeptides were identified under these conditions in the comparative analysis of the LbrC1 and LbrM1 protein extracts, and 23 polypeptides were identified in the analysis of the protein extracts of LbrC2 and LbrM2 (S4 Table). Among these, only LbrM.31.2410 (prostaglandin f2-alpha synthase- LbrPGF2S) and LbrM.28.2990 (HSP70, putative) were consistently over-represented in both cutaneous isolates (LbrC1 and LbrC2) and less abundant (or undetectable) in both mucosal isolates (LbrM1 and LbrM2).
Both the proteome and metabolome data suggested that PGF2S was an important target warranting further investigation. Thus, we generated LbrC1 and LbrM1 parasites overexpressing LbrPGF2S to confirm the correlation between LbrPGF2S levels and virulence. The gene was inserted into pX63NEO and expressed ectopically. LbrPGF2S overexpression was confirmed by Western blotting using a polyclonal anti-LbrPGF2S antibody (S2 Fig).
Infection of peritoneal macrophages from BALB/c mice with the wild type gene and transfectants resulted in average macrophage infection rates of 84%, 95%, 78%, 90% and 96% (t0 post-infection) for LbrC1 wild type, LbrC1 [pX63NEO], LbrC1 [pX63NEO-PGF2S], LbrM1 wild type and LbrM1 [pX63NEO-PGF2S], respectively, indicating that the percentage of internalization during early infection did not vary significantly. Nevertheless, after 24 h, the infection index decreased to 2.8% and 4.5% for LbrC1 wild type and LbrM1 wild type, respectively, and to 21.6% and 34.6% for the LbrPGF2S-overexpressing transfectants. Thus, the LbrPGF2S-overexpressing parasites exhibited a significantly increased infection percentage 24 h p.i. This difference persisted until 48 hours p.i. for LbrC1 [pX63NEO-PGF2S] (Fig 5). Immediately before the in vitro infection experiments, the overexpression of LbrPGF2S was confirmed by Western blotting (S2 Fig).
The number of amastigotes within macrophages decreased at 24 h and 48 h p.i. for all groups. Nevertheless, the differences between the wild type and LbrPGF2S-overexpressing infections persisted at these time points. The average numbers of intracellular amastigotes were 7.3, 5.8, 59.3 and 69.8 per macrophage (300 cells counted) for infections with LbrC1 wild type, LbrM1 wild type, LbrC1 [pX63NEO-PGF2S] and LbrM1 [pX63NEO-PGF2S], respectively. Both of the overexpressor transfectants produced a significant increase in amastigotes inside macrophages compared with the wild type; this increase was most apparent at 24 h p.i.
Here, we used a powerful tool to explore parasite factors involved in the pathogenesis of leishmaniasis. We demonstrated that significant differences in the proteome, metabolome and parasite virulence may emerge from two subpopulations of the same L. braziliensis strain collected from different tissues in the same human host. We propose that parasite factors other than the presence of the Leishmania RNA virus (LRV) are involved in specific manifestations in tegumentary leishmaniasis when the immunogenetic background is the same. The data presented here are the first to suggest that the enzyme LbrPGF2S may participate in the virulence profile in the host. Our findings were reproducible for two “same-host” pairs isolated from two different patients.
We demonstrated that these L. (V.) braziliensis isolates consistently exhibited differences in the expressed genome, metabolome and pathogenesis in vivo. We determined that the four isolates possessed similar molecular karyotypes and therefore likely originated from a single circulating strain. In addition, complete genome sequencing did not reveal significant somy differences between the LbrC and LbrM isolates. The inclusion of these samples in the Viannia subgenus was confirmed by PCR based on the presence of a domain from the Dicer-like gene that is an integrant of the RNA interference pathway exclusive to the Viannia subgenus [17].
The lack of LRV in the LbrC and LbrM isolates is relevant in light of the reported correlation between the presence of LRV and the severity of clinical manifestations of tegumentary leishmaniasis. The parasite virus triggers host macrophage recognition, promoting inflammation and modifying the immune response during infection, thereby conferring parasite fitness advantages within the host cell [37]. Therefore, it is possible to speculate that differences in the expressed genomes of the parasites found at the infection site and those isolated from the nasal mucosae are partially responsible for the diverse pathogenesis of LbrC and LbrM isolates.
In vivo infection experiments revealed that the analyzed parameters—i.e., parasite load and lesion progression—differed between the LbrC and LbrM isolates. Hamsters and BALB/c mice infected with the LbrM isolates exhibited smaller lesions and lower parasite burdens at the site of primary infection and the draining lymph nodes compared to animals infected with LbrC. These findings are in accordance with those of Jara et al. [9], who previously have demonstrated a lower parasite burden in the mucosae compared to that at cutaneous sites. However, we must emphasize that the parasite populations analyzed in this study were recovered from patients during the acute phase of infection, whereas the mucosal isolates from the study performed by Jara et al. were recovered from a chronic infection.
The differences in IL-4 and IFN-γ levels suggest that the LbrC and LbrM isolates stimulated different host cell responses in BALB/c mice. Also, higher levels of IFN-γ in the supernatants of cells infected with LbrC2 are consistent with the increased lesions observed in hamsters infected with LbrC2. Nevertheless, to pursue a robust characterization of the immune response associated to mucosal or cutaneous isolates, it will be necessary to quantify other Th1 or Th2 pattern-specific cytokines.
We investigated the features of the phenotypic expression that could be associated with the pathogenesis of the parasite at the mucosae. The comparison of the proteomes and metabolomes of the paired strains revealed interesting features.
The observed metabolome differences indicate a clear dichotomy between the parasite populations resting at the cutaneous site versus the population localized to the mucosae of the same individual. Many of the metabolites that significantly differed between isolates may affect several cellular processes, and an investigation of the relevant metabolic pathways is needed to understand the role of these metabolites in pathogenesis. However, the metabolome profiles of LbrC and LbrM indicate that the differences in pathogenesis involve the differential production of metabolites related to inflammation and chemotaxis.
The complex interaction of molecules that determine the migration of L. braziliensis-infected host cells from the primary lesion site to the mucosal regions remains undetermined. However, previous studies have demonstrated that Leishmania promastigotes release chemotactic factors that regulate cell migration and the activation of the innate immune system at the primary site [38, 39]. In this study, we used proteomic and metabolomic analyses to obtain a more global understanding of the physiological differences between the LbrC and LbrM isolates and to identify chemotactic networks that guide or contribute to the differential pathogenesis of L. braziliensis.
Some of the fatty acids that were elevated in LbrM (i.e., myristic and palmitic acids) may affect the inflammatory reaction, playing a key role in parasite tropism or exert pro-inflammatory activities. Others may affect protein anchoring to membranes (which is critical for the recognition and attachment of parasites to host cells) or signal transduction pathways [40–43].
Resolvins were markedly elevated in LbrM. Resolvins are lipid mediators that stimulate pro-resolving mechanisms during sepsis [44]. Resolvins may decrease the migration of inflammatory and dendritic cells [44, 45], suppress NF-κB activation [46], enhance phagocytosis and anti-inflammatory cytokine production and stimulate host-protective actions in inflammatory responses [47].
Several metabolites (i.e., phosphatidylcholines, phosphatidylethanolamines and their derivatives) involved in phospholipid synthesis were up-regulated in the LbrM samples. Parasites rely on a complex system of uptake and synthesis mechanisms to obtain lipids at different life stages [48], and lipid metabolism is crucial for the production of factors related to pathogenesis. The increased levels of phospholipids suggest that the Kennedy pathway [49] could be more active in the LbrM strain. However, a targeted metabolomics approach is needed to more closely evaluate the levels of Kennedy metabolites. The platelet-activating factor (PAF) was also increased in LbrM; PAF is involved in a variety of inflammatory processes such as vascular permeability, oxidative burst, chemotaxis and activation of leukocytes, stimulation of arachidonic acid metabolism [50] and cellular differentiation and infectivity in Trypanosoma cruzi [51].
Cyclohexanecarbonylpentadecylamine and phthalic acid mono-2-exthylhexyl ester, metabolites related to arachidonic acid metabolism, were decreased in LbrM [52]. The cellular starting material for prostaglandin biosynthesis is usually membrane phosphatidylinositol, whereas prostaglandins are predominantly synthesized from arachidonic acid, which is also a prostaglandin synthase substrate in protozoan parasites that release prostaglandins [53].
Chalcone, a metabolite that can affect local inflammatory responses and the prostaglandin synthesis pathway, was up-regulated in LbrM. Chalcones may suppress the mitogen-activated protein kinase (MAPK) pathway, thereby inhibiting pro-inflammatory mediators such as nitric oxide (NO), prostaglandin E(2), tumor necrosis factor-alpha (TNF-alpha) and the production of reactive oxygen species [54, 55]. The combination of the up-modulation of prostaglandin biosynthesis metabolites in LbrC and the up-modulation of inhibitory metabolites in the prostaglandin pathway (4-metoxychalcone) in LbrM suggests that prostaglandin metabolism could play a role in L. braziliensis pathogenesis.
Interestingly, the proteome comparative analysis reinforced the potential relevance of the prostaglandin pathway to the pathogenesis of L. braziliensis infection. The proteome analysis enabled the detection of two proteins with matching differential expression patterns between different host isolates from cutaneous site in the two patients: prostaglandin F2 alpha synthase (LbrM.31.2410) and putative HSP70 (LbrM.28.2990). These proteins were uniquely expressed or over-expressed in the LbrC1 and LbrC2 samples, respectively. Remarkably, survival of L. infantum promastigotes of nitric oxide donor exposure [56] has been associated with HSP70 and PGF2S overexpression and increased in vitro infectivity. These results support the hypothesis that the HSP70 and PGF2S proteins are involved in parasite infectivity or infection patterns in the vertebrate host.
Prostaglandin synthesis has been reported to occur in metazoan and protozoan parasites in addition to mammals. However, the molecular mechanisms of prostaglandin production and their biological role in parasites have not been well elucidated [53]. High levels of prostaglandin F2-alpha (PGF2α) and PGF2α synthase have been reported in T. brucei [57].
Interestingly, the overexpression of PGF2α synthase increased the infectivity of the LbrC and LbrM isolates in vitro by improving parasite survival within host cells. Further investigation is needed to identify the mechanisms involved in the effect of PGF2α synthase on parasite virulence.
It has been shown that PGF2S protein is present in the secretome of L. (V.) braziliensis [58] and the exosome of L. (L.) donovani [59]. According to the TDR Targets Database (tdrtargets.org), 13 putative antigenic epitopes in the PGF2S protein of L. major are responsible for 77.8% of its antigenicity, making it one of the most antigenic proteins produced by L. major. Using the same database, this protein was identified as having high potential as a drug target, with a 0.8 druggability index (range: 0.0 to 1.0). Therefore, LbrPGF2S may represent a relevant target for studies of parasite-host interactions.
In conclusion, this study identified parasite-derived factors that contributed to the pathogenesis pattern of L. braziliensis. The genetic, proteomic and metabolomic results indicate that the inoculated population of parasites may contain a subpopulation of cells with a divergently expressed genome, leading to physiological differences that alter the modulatory response of the host.
|
10.1371/journal.ppat.1004906 | Latent Membrane Protein LMP2A Impairs Recognition of EBV-Infected Cells by CD8+ T Cells | The common pathogen Epstein-Barr virus (EBV) transforms normal human B cells and can cause cancer. Latent membrane protein 2A (LMP2A) of EBV supports activation and proliferation of infected B cells and is expressed in many types of EBV-associated cancer. It is not clear how latent EBV infection and cancer escape elimination by host immunity, and it is unknown whether LMP2A can influence the interaction of EBV-infected cells with the immune system. We infected primary B cells with EBV deleted for LMP2A, and established lymphoblastoid cell lines (LCLs). We found that CD8+ T cell clones showed higher reactivity against LMP2A-deficient LCLs compared to LCLs infected with complete EBV. We identified several potential mediators of this immunomodulatory effect. In the absence of LMP2A, expression of some EBV latent antigens was elevated, and cell surface expression of MHC class I was marginally increased. LMP2A-deficient LCLs produced lower amounts of IL-10, although this did not directly affect CD8+ T cell recognition. Deletion of LMP2A led to several changes in the cell surface immunophenotype of LCLs. Specifically, the agonistic NKG2D ligands MICA and ULBP4 were increased. Blocking experiments showed that NKG2D activation contributed to LCL recognition by CD8+ T cell clones. Our results demonstrate that LMP2A reduces the reactivity of CD8+ T cells against EBV-infected cells, and we identify several relevant mechanisms.
| Epstein-Barr virus (EBV) is carried by most humans. It can cause several types of cancer. In healthy infected people, EBV persists for life in a "latent" state in white blood cells called B cells. For infected persons to remain healthy, it is crucial that they harbor CD8-positive "killer" T cells that recognize and destroy precancerous EBV-infected cells. However, this protection is imperfect, because the virus is not eliminated from the body, and the danger of EBV-associated cancer remains. How does the virus counteract CD8+ T cell control? Here we study the effects of latent membrane protein 2A (LMP2A), which is an important viral molecule because it is present in several types of EBV-associated cancers, and in latently infected cells in healthy people. We show that LMP2A counteracts the recognition of EBV-infected B cells by antiviral killer cells. We found a number of mechanisms that are relevant to this effect. Notably, LMP2A disturbs expression of molecules on B cells that interact with NKG2D, a molecule on the surface of CD8+ T cells that aids their activation. In this way, LMP2A weakens important immune responses against EBV. Similar mechanisms may operate in different types of LMP2A-expressing cancers caused by EBV.
| Epstein-Barr virus (EBV), which belongs to the human herpesvirus family, is a persistent virus carried by more than 90% of the adult population worldwide. EBV has a preferential B cell tropism, and latently infected B cells constitute the viral reservoir in healthy carriers [1]. Acute infection can lead to infectious mononucleosis (IM), a self-limiting lymphoproliferative disease characterized by expansion of EBV-infected B cells and virus-specific CD8+ T cells [2]. EBV is an oncovirus, and can contribute to the development of various cancers, such as Burkitt lymphoma, nasopharyngeal carcinoma and Hodgkin lymphoma [3,4]. In healthy carriers, EBV infection is under control of a diverse repertoire of antigen-specific T cells, and an important role is played by CD8+ T cells that recognize viral protein-derived peptides presented by MHC class I molecules [2]. In contrast, immunosuppressed patients who lack EBV-specific T cell responses, such as patients after transplantation, are prone to developing EBV-associated lymphoproliferative disease. This condition can be treated or prevented by transfer of EBV-specific T cells [5–7].
In immunocompetent EBV carriers, a majority of EBV-infected B cells in peripheral blood carry EBV without expressing any viral protein, a state that is called "true latency" or "latency 0" [4,8]. Thus, such latently infected B cells are invisible to EBV-specific T cells.
In contrast, during lytic EBV replication many viral proteins are expressed [9,10]. In this situation, the virus would be particularly vulnerable to immune control. Thus, EBV has evolved a number of proteins expressed in the lytic cycle that interfere with the display of viral antigens to CD8+ T cells. These proteins include BNLF2a, which inhibits the transporter of antigen processing [11], BILF1, which induces MHC class I internalization and degradation [12], and BGLF5, which inhibits cellular protein biosynthesis [13].
In proliferating infected B cells, EBV installs another program of gene expression, the "growth" or "latency III" program. This type of latency is found in in vitro EBV-induced lymphoblastoid cell lines (LCLs), in post-transplant lymphoproliferative diseases [14], as well as in EBV-infected B cells in lymphoid organs during primary and persistent EBV infection, where this program is thought to lead to amplification of EBV load through proliferation of infected cells [4,8]. Several immunogenic EBV antigens, the latent membrane proteins (LMP1, LMP2A, LMP2B) and the Epstein-Barr nuclear antigens (EBNA1, -2, -3A, -3B, -3C, -LP), are expressed in latency III EBV-infected B cells [9,10]. How do B cells expressing the EBV "growth program" manage to escape from recognition and elimination by virus-specific T cells? Previous studies on immunoevasion in EBV latency have focused on the nuclear protein EBNA1 or the latent membrane protein LMP1. EBNA1 interferes with its own presentation to CD8+ T cells through its glycine-alanine repeat domain [15,16], which reduces processing by the proteasome [17] and interferes in cis with EBNA1 translation [18–20]. As a result, presentation of EBNA1 epitopes on MHC class I to T cells is reduced. Likewise, LMP1 interferes in cis with presentation of its own epitopes [21]. Although several other viral proteins are expressed in the EBV growth program, it has remained unknown whether presentation to T cells of epitopes from these proteins may be suppressed by viral mechanisms.
The EBV latent protein LMP2A is a regular constituent of the EBV growth program, and is also expressed in a variety of EBV-associated cancers [9,10]. LMP2A has various functions in infected cells. Reminiscent of the accessory subunits of the B-cell receptor, the N-terminal cytoplasmic domain of LMP2A activates protein tyrosine kinases and induces downstream pathways of B cell activation [22,23]. Accordingly, LMP2A can stand in for deficient B-cell receptor signaling in mouse or human models, ensuring B cell survival [24,25]. In EBV-infected B cells, however, LMP2A counteracts lytic EBV reactivation triggered by cross-linking of the B-cell receptor [26–28]. No consensus has been reached yet on the importance of LMP2A in B cell proliferation and transformation [25,29–35].
Given these complexities, we hypothesized that LMP2A may have other functions that are not cell-intrinsic or directly related to virus replication, but related to immune control. This possibility was already suggested by the observation that LMP2A modulates signalling of type I/II interferon receptors in epithelial cells [36], that the presence of LMP2A alters the expression of several immune-related genes [37], and that LMP2A increases expression of the cytokine interleukin-10 (IL-10) [38], which may exert immunomodulatory functions. In this study, we investigated the influence of LMP2A in recognition of infected cells by immune effector cells. We show that LMP2A reduces recognition of infected B cells by EBV-specific CD8+ T cells, and we describe several mechanisms that may contribute to this effect.
We established EBV-transformed B cell lines (lymphoblastoid cell lines, LCLs) with an EBV deleted for LMP2A [25]. This virus is deleted for the promoter and the first exon of LMP2A on a background of EBV strain B95.8. Expression of LMP2B is still possible in this mutant. In line with previous findings [25,29], we found that the LMP2A-deficient virus (ΔLMP2A) had reduced efficiency of B cell transformation. To facilitate the establishment of LMP2A-deficient LCLs, we infected primary B cells with mutant EBV on a layer of murine fibroblasts overexpressing human CD40 ligand (CD40L). Infection with recombinant EBV 2089 [39] that contains the complete B95.8 EBV genome (here denoted "wild-type", WT), which is parental to the ΔLMP2A construct, was carried out in parallel under the same conditions. Outgrowing B cell cultures were expanded and maintained in the absence of CD40L stimulators. Under these conditions, WT and ΔLMP2A LCLs could be established with similar efficiency, and expanded in parallel using the same procedures. A closer analysis of established WT and ΔLMP2A LCLs showed that the rate of apoptosis was the same, but proliferation was somewhat slower in ΔLMP2A LCLs (S1 Fig). Thus, LMP2A increased the efficiency of EBV transformation in vitro, but was not essential for the proliferation of established LCLs.
We analyzed the reactivity of EBV-specific CD8+ T cells to ΔLMP2A and WT LCLs (Fig 1). We found that CD8+ T cell clones specific for epitopes from all latent antigens tested (EBNA1, EBNA3A, LMP2) showed a higher IFN-γ release in response to ΔLMP2A LCLs than to WT LCLs (Fig 1A and 1C). CD8+ T cells specific for the LMP2 epitope CLG recognized ΔLMP2A LCLs, because the CLG peptide is derived from a transmembrane region that is shared between LMP2A and LMP2B. CD8+ T cell clones specific for lytic-cycle antigens (BRLF1, BZLF1) showed weak recognition of both types of LCLs, and therefore differences in recognition could not be detected (Fig 1B). Thus, LMP2A interferes with CD8+ T cell recognition of EBV latent antigens.
To confirm that these differences in T cell recognition were caused by LMP2A and not some other unrecognized deviations between the two EBV constructs, we tested the effect of LMP2A on CD8+ T cell recognition in isolation, in the absence of an EBV genome (Fig 1D). Co-transfection of LMP2A reduced CD8+ T cell recognition of 293T kidney cells transfected with the HCMV antigen pp65. This experiment showed that the effect of LMP2A on T cell recognition was not limited to the context of the EBV genome.
In the early stages of infection, there are differences in EBV gene transcription in B cells carrying LMP2A-negative EBV as opposed to LMP2A-positive EBV [35]. Thus, we investigated whether the observed differences in T cell recognition of ΔLMP2A and WT LCLs were related to differential expression of EBV antigens. Average transcript levels of several EBV latent antigens (EBNA1, EBNA3A, LMP2) appeared to be increased in ΔLMP2A LCLs (Fig 2A). However, this difference reached p < 0.05 only for EBNA1. No difference between WT and ΔLMP2A LCLs was seen for median expression of the lytic-cycle genes BZLF1 and gp350. Thus, LMP2A may downmodulate the expression of some latent antigens in EBV-infected B cells, in particular EBNA1. This may contribute to the reduced presentation of these antigens to CD8+ T cells by WT LCLs.
LMP1 is an EBV protein that may alter CD8+ T cell recognition of infected cells, in particular by inducing MHC I pathway components through NF-κB, but also by inducing immunomodulatory genes [21,40]. We found that expression of LMP1 at the protein level was somewhat reduced in ΔLMP2A LCLs (Fig 2B). This argued against a possible role of LMP1 in contributing to increased recognition of ΔLMP2A LCLs by upregulating MHC I presentation.
Next, we investigated whether LMP2A modulated the reactivity of CD8+ T cells to EBV-infected B cells by mechanisms other than altering the availability of EBV antigens. We loaded WT and ΔLMP2A LCLs exogenously with peptides CRV and VLE, derived from the human cytomegalovirus (HCMV) protein IE-1, and we analyzed LCL recognition by HCMV-specific CD8+ T cell clones (Fig 3). Peptide-loaded ΔLMP2A LCLs were more strongly recognized by these CD8+ T cells than peptide-loaded WT LCLs, resulting in higher IFN-γ release. We also investigated direct killing by cytotoxic CD8+ T cells, but did not observe differences in killing of WT and ΔLMP2A LCLs loaded with HCMV peptides (S2 Fig). The reasons for differential regulation of IFN-γ secretion and direct cytotoxicity in this setting remain to be elucidated.
Because the intracellular antigen processing machinery was bypassed in these peptide-loading experiments, LMP2A appears to act on CD8+ T cells through mechanisms other than regulation of EBV antigens or of intracellular processing pathways. Therefore, we studied the effect of LMP2A on cell surface-residing or secreted factors relevant for CD8+ T cell recognition.
It was recently shown that LMP2A increases IL-10 production in infected B cells [38]. The possibility of a similar effect in our system was intriguing, because cellular IL-10 and its viral homolog reduce the antiviral activity of different types of immune effector cells [41–43]. In accordance with Incrocci and colleagues [38], we found that WT LCLs released higher amounts of IL-10 than LCLs lacking LMP2A (Fig 4A). These levels of secreted IL-10 were not mirrored by transcription levels for human IL-10 (Fig 4B), which suggested an effect of LMP2A on post-transcriptional regulation of IL-10 [44]. In contrast to cellular IL-10, transcription of viral IL-10 was very low in each type of LCL (Fig 4B), in accordance with its description as a lytic-cycle gene [45]. To determine whether differences in IL-10 release could directly influence T cell reactivity to LCLs, we used specific antibodies to block IL-10 receptor on CD8+ T cells (Fig 4C and 4E), or to neutralize IL-10 in the supernatant (Fig 4D and 4F). In each case, recognition of WT or ΔLMP2A LCLs was not altered. Thus, modulation of IL-10 secretion by LMP2A did not directly affect the ability of CD8+ T cells to recognize infected B cells. This experiment did not rule out indirect effects of secreted IL-10, which may act back on the LCLs over time in culture and modulate their immunogenicity.
We continued by analyzing ΔLMP2A and WT LCLs for cell surface molecules involved in the interaction between CD8+ T cells and LCLs. First, we determined the levels of total MHC-I and individual MHC-I allotypes (Fig 5). MHC-I was marginally increased in LCLs deleted for LMP2A as compared with WT LCLs (p = 0.0046). A similar tendency was observed for some of the individual MHC-I allotypes, but did not reach p < 0.05.
Next, we examined whether expression of selected costimulatory and immunomodulatory molecules on the surface of LCLs was altered in the absence of LMP2A (Fig 6). We found strong differences in expression for some of these molecules. The coinhibitory B7 family molecule PD-L1 (B7-H1) was (somewhat unexpectedly) induced in ΔLMP2A LCLs, whereas the costimulatory B7 molecule CD86 was equally expressed on ΔLMP2A and WT LCLs. CD11a, the α chain of the integrin LFA-1 that plays important roles in the immunological synapse, was strongly downregulated in the absence of LMP2A, whereas ICAM-1 (CD54), its counterpart, was expressed equally in the presence or absence of LMP2A. So far, these alterations were not obviously connected with the increased susceptibility of ΔLMP2A cells to CD8+ T cell recognition.
Recent reports suggested that EBV infection induces ligands of the coactivatory receptor NKG2D, a molecule expressed on T and NK cells [46–49]. However, a comprehensive analysis of NKG2D ligands on LCLs has not previously been performed. Our analyses by flow cytometry showed that EBV infection induced the expression of three NKG2D ligands (MICA, MICB and ULBP4) on LCLs (Fig 6). These molecules were not expressed on primary B cells. Markedly higher levels of MICA and ULBP4 were detected on ΔLMP2A LCLs as compared to WT LCLs, whereas MICB levels did not differ (Fig 6). We could not detect expression of the other five NKG2D ligands (ULBP1, 2, 3, 5, 6) on the surface of WT or ΔLMP2A LCLs with available monoclonal antibodies, but this does not rule out that these molecules may as well be modulated by LMP2A. Our results suggested a possible contribution of NKG2D ligands to differential recognition of LCLs by CD8+ T cells.
We tested the functional relevance of differential NKG2D ligand expression for CD8+ T cell recognition. An analysis of NKG2D levels on several CD8+ T cell clones showed that all were positive for NKG2D (Fig 7A). Differences in NKG2D expression levels were not correlated with antigen specificity. When we blocked NKG2D on EBV-specific CD8+ T cells with a specific antibody, IFN-γ release after contact with LCLs was reduced (Fig 7B–7D). A reduction in the reactivity of CD8+ T cells to both WT and mutant LCLs was observed after blocking, but reduction was even slightly stronger for ΔLMP2A LCLs than for WT LCLs (Fig 7C and 7D). Likewise, blocking NKG2D on HCMV-specific CD8+ T cell clones led to reduced recognition of peptide-loaded LCLs (Fig 7E). Thus, NKG2D ligands on LCLs contribute to their recognition by CD8+ T cells irrespective of antigen specificity. LMP2A reduces CD8+ T cell recognition of EBV-infected B cells by reducing the expression of NKG2D ligands.
Since expression of PD-L1, a ligand of the immunomodulatory receptor PD-1 on T cells, was increased on ΔLMP2A LCLs (Fig 6), the question emerged whether PD-L1 may counteract T cell recognition of ΔLMP2A LCLs. In this case, even greater differences in T cell recognition of ΔLMP2A LCLs as opposed to WT LCLs might be revealed by masking the effects of PD-L1. Since it was reported that stimulation of PD-L1 on LCLs induces their apoptosis in a T-cell-independent manner [50], we used a PD-1-blocking antibody, EH12.2H7, that was described to interfere with T-cell-inhibitory interactions of PD-L1 and PD-1 [51]. Thus, we tested blocking antibodies to NKG2D and PD-1 in T cell recognition assays. Interestingly, blockade of PD-1 did not increase T cell recognition of ΔLMP2A LCLs, but reduced it, although less so than blockade of NKG2D (Fig 8). Addition of PD-1 antibody to NKG2D antibody did not further modify recognition of ΔLMP2A LCLs, although recognition of WT LCLs was additively reduced by the two antibodies. We conclude that the increased amounts of PD-L1 on ΔLMP2A LCLs did not counteract T cell recognition and resulting IFN-γ production.
In this report, we show that LMP2A interferes with CD8+ T cell recognition of latently infected B cells, and identify several mechanisms that may contribute to this interference. First, we found that LMP2A decreased mRNA expression levels of EBV latent antigens targeted by CD8+ T cells, in particular EBNA1. Second, LMP2A downregulated MHC class I, although to a limited extent. Third, two ligands of the coactivatory receptor NKG2D were strongly upregulated in LMP2A-deficient LCLs, and blocking of NKG2D reduced T cell recognition of infected cells. We conclude that LMP2A hampers CD8+ T cell recognition of infected cells through different mechanisms including regulation of NKG2D ligands.
A basis for the present work was the efficient generation of ΔLMP2A LCLs. The importance of LMP2A for human B cell transformation by EBV has been controversial: some studies did not identify a role of LMP2A [30–34], but others reported that LMP2A increases B cell proliferation and transformation [25,29,35]. In our experience, LMP2A is important for establishment of EBV latent infection in vitro, as we found it difficult to establish ΔLMP2A LCLs under standard conditions. However, when we supplemented a strong CD40 stimulus for the first days after infection [52,53], ΔLMP2A LCLs and WT LCLs could be generated with similar yield, and ΔLMP2A LCLs could then be maintained autonomously. This finding confirmed that LMP2A is not essential for maintenance and proliferation of established LCLs, as long as the B cell receptor is expressed [25].
Another important function of LMP2A is its role in "stabilizing latency", i.e. prevention of lytic-cycle induction. A complex picture has emerged, in which LMP2A interferes with lytic induction after exogenous B-cell receptor stimulation [26–28,30,54], but induces basal levels of lytic gene expression in the absence of such a stimulus [55]. In accordance with earlier results [35], we found that baseline lytic gene expression was low both in WT and ΔLMP2A, and therefore was without consequence for recognition by T cells [2,56]. Thus, the "latency-stabilizing" function of LMP2A does not appear to be relevant for T cell recognition of B cells that express the growth program in the absence of exogenous triggers. It remains to be investigated whether the immunomodulatory functions of LMP2A extend to cells in lytic cycle.
Two EBV latent proteins, EBNA1 and LMP1, were previously shown to modulate antigen presentation to CD8+ T cells. EBNA1 does not affect presentation of other antigens, but specifically interferes in cis with its own presentation by blocking its own proteasomal processing [17] and modulating its translation [19,20], both through its glycine-alanine-rich domain. LMP1 is a strong inducer of MHC I presentation through activation of the NF-κB pathway [21,57], but contains a structural element that acts in cis to protect epitopes derived from its own sequence from efficient presentation [21]. LMP1 is also unusual in that peptides derived from secreted LMP1 were shown to interfere with T cell activation [58]. However, this requires amounts of LMP1 that are much higher than those secreted by EBV-infected cells [58]. To our knowledge, it has remained untested whether EBV latent antigens more generally affect recognition by and activation of CD8+ T cells.
Suggestions regarding a role of LMP2A in T immune modulation emerged from comprehensive microarray-based analyses of LMP2A-mediated changes to the transcriptome of mouse and human B cells [37]. Interestingly, transcription of genes in the inflammation/immunity category, including interferon-regulating factors, was repressed by LMP2A in human LCLs, whereas no such genes were induced [37]. Among genes of direct relevance for the B-cell—T-cell interface, CD86 was induced and LFA-1 was repressed by LMP2A in BJAB cells, but no differential expression of these genes was found in LCLs with or without LMP2A [37]. These findings highlight that the effects of LMP2A depend on the cellular context, and that T-cell-modulatory functions of LMP2A in more restricted modes of EBV latency may hypothetically be even stronger than in the LCL system studied here. For example, the ability of LMP2A to interfere with signaling through interferon receptors [36] may further contribute to LMP2A-mediated evasion from T cell recognition.
Our data demonstrated that LMP2A markedly reduced the reactivity of EBV-specific CD8+ T cells against LCLs. This was true for all latent EBV antigens investigated (LMP2, EBNA1, EBNA3A). The epitopes we analyzed are processed by different pathways for their presentation: CLG and FLY are TAP-independent epitopes, with FLY being immunoproteasome-dependent [59,60], while RPP and HPV are TAP-dependent [61]. A reduction of CD8+ T cell reactivity was also observed on LCLs loaded with exogenous peptides from a different pathogen, which makes it clear that LMP2A does not exclusively affect intracellular mechanisms of antigen provision and presentation. Reduced CD8+ T cell reactivity in the presence of LMP2A was observed in the context of all HLA allotypes that were studied: HLA A*0201, B*0702, and B*3501 for intracellularly processed EBV epitopes, HLA A*0201 and C*0702 for exogenously loaded HCMV epitopes. Thus, our data indicate that LMP2A affects CD8+ T cell reactivity regardless of the identity of the peptide presented, the mechanism of processing, or the presenting HLA molecule.
However, our results also suggested an antigen-specific aspect to the immunomodulatory effects of LMP2A, because we found a trend toward elevated expression of several latent genes in the absence of LMP2A. This is in line with the idea that LMP2A may mediate global B-cell transcription factor regulation to reduce expression of EBV latency proteins [62,63]. This was not true for LMP1, however, whose protein expression in the absence of LMP2A was reduced. Our findings are in line with similar tendencies in EBV latent gene expression in the first seven days after B cell infection with EBV ΔLMP2A [35]. It is intriguing that EBNA1 was the EBV latent antigen whose mRNA expression was most clearly reduced by LMP2A, since both antigens are part of the restricted EBV gene expression spectrum in latency II EBV malignancies such as nasopharyngeal carcinoma and Hodgkin lymphoma [64,65]. If LMP2A interferes with presentation of EBNA1-derived and other peptides also in latency II type cancers, this will have important implications for their immune surveillance.
Among immunomodulatory cytokines, IL-10 was a particularly interesting candidate in our context, because it is constitutively produced at high levels by EBV-transformed B cells [66,67], and a recent report showed that LMP2A increased IL-10 production in Burkitt lymphoma cell lines [38]. Moreover, EBV encodes a viral homologue of human IL-10 [68]. Both human and viral IL-10 were suggested early on to interfere with cellular immune responses to EBV [41,69], but it may be difficult to distinguish an immunomodulatory function of cellular or viral IL-10 from their function as growth factors for EBV-infected B cells [66,70,71]. vIL-10 contributes to downregulation of the transporter of antigen processing 1 (TAP1) and MHC-I in the early phase of B cell infection [43], but recognition of early-infected B cells by EBV-specific CD8+ T cells was not increased in the absence of vIL-10 [42]. Our data showed that LCLs lacking LMP2A released lower amounts of IL-10 compared to WT LCLs, but reactivity of CD8+ T cell clones was not altered by neutralization of IL-10 or blocking of the IL-10 receptor. However, a more indirect role of IL-10 remains possible. Therefore, LCL-secreted IL-10 may act back on the LCLs over time, and thus downregulate MHC-I or other relevant molecules [43,72] in WT LCLs more strongly than in ΔLMP2A LCLs.
PD-1 is an immunomodulatory receptor found on activated T cells, on exhausted virus-specific T cells in chronic infection, but also on functional EBV-specific effector memory T cells in latent infection [73]. Blocking of the interaction between PD-1 and its ligand, PD-L1, may restore antiviral T cell function [74]. Somewhat counter-intuitively, we found PD-L1 to be downregulated in LCLs in the presence of LMP2A. When we blocked PD-1 on CD8+ T cells, we did not observe increased reactivity to LCLs, but rather a reduction in reactivity. The possibility remains that PD-L1 on EBV-infected B cells exerts a more long-term influence on shaping specific CD8+ T cell repertoires and functions that was not tested in our experiments.
NKG2D is an agonistic receptor on T and NK cells and recognizes a number of ligands that are upregulated on target cells in conditions such as malignant transformation, viral infection or heat shock [75]. Increased expression of some NKG2D ligands after EBV infection was described [46–48,76], but a comprehensive analysis of NKG2D ligands on LCLs has been lacking. Our analysis of the eight known NKG2D ligands showed that EBV infection induced the expression of three of them (MICA, MICB, and ULBP4), and that induction of MICA and ULBP4 was even more increased in the absence of LMP2A. In addition, we demonstrated that blocking of NKG2D on CD8+ T cells distinctly affected the recognition of LCLs by these effector cells. A recent study has shown that in patients with genetic deficiencies in the magnesium transporter MAGT1, who are particularly susceptible to EBV infection and EBV+ lymphomas, NKG2D plays an important role in the control of EBV infection by NK and CD8+ T cells [46]. A role for NKG2D in control of EBV-associated cancer has been further illustrated in a mouse model of LMP1-induced cancer that could be therapeutically targeted through NKG2D [76]. Targeting of the NKG2D ligand MICB by an EBV-encoded miRNA may decrease susceptibility of EBV-infected B cells to lysis by NK cells [77]. Thus, NKG2D ligands represent important coagonists for EBV-specific adaptive and innate immunity, and it appears an efficient strategy for the virus to blunt cellular immune responses by decreasing surface expression of NKG2D ligands through the action of LMP2A.
Taken together, we describe here a functional immunomodulatory effect for the EBV protein LMP2A, and show that LMP2A mediates partial escape of infected B cells from recognition by CD8+ T cells. Several immunoevasive mechanisms are driven by LMP2A in EBV-infected B cells. Thus, it will be urgent to determine the role played by LMP2A in evasion from T and NK cell recognition in other modes of EBV infection, and in different types of EBV-associated lymphoproliferative and malignant diseases.
Mononuclear cells were isolated from standard blood donations by anonymous healthy adult donors purchased in the form of buffy coats from the Institute for Transfusion Medicine, University of Ulm, Germany, or from voluntary healthy adult blood donors providing written informed consent. The institutional review board (Ethikkommission, Klinikum der Universität München, Grosshadern, Munich, Germany) approved this procedure. All work was conducted according to the principles expressed in the Helsinki Declaration.
Virus-producing cell lines for recombinant EBV 2089 (EBV WT), derived from EBV strain B95.8 [39], and its ΔLMP2A-deleted derivative EBV 2525 [25] were provided by Wolfgang Hammerschmidt [25,39]. In 293HEK-derived EBV producer cell lines, which stably carry the EBV genome in an episomal form, the viral lytic cycle was induced by transient transfection of expression plasmids coding for transcription factor BZLF1 and glycoprotein gp110/BALF4 [78]. EBV-containing supernatant was harvested three days later, centrifuged to reduce cellular debris, filtered (0.8 μM), and stored at 4°C. Titer of infectious virus was determined by infecting Raji cells for three days and quantifying GFP-positive cells as described [78]. Infection of B cells was performed at 0.1 virus units per cell.
Standard cell culture medium was RPMI 1640 with 10% foetal bovine serum, 100 U/ml penicillin, 100 μg/ml streptomycin (all from Invitrogen), and 100nM sodium selenite (ICN). Stimulator cell line LL8 was generated by stable transfection of L929 mouse fibroblasts with an expression plasmid for human CD40 ligand carrying a G418-selectable marker, followed by two rounds of single-cell cloning under selection. We found this stimulator cell line to be functionally analogous to the one described earlier [53,79]. Lymphoblastoid cell lines were established from primary B cells purified from freshly isolated PBMCs. Untouched B cells were negatively isolated using Human B Cell Isolation Kit II (Miltenyi Biotec, Bergisch Gladbach, Germany). Enrichment of B cells was verified by flow cytometry (anti-CD19 clone HIB19; anti-CD3 clone HIT3a, Biolegend), and was in the range of 95–98%. B cells were plated at 1×105 cells/well in 24-well plates on an adherent cell layer of irradiated (180 Gy) CD40 ligand-expressing LL8 cells in standard medium supplemented with 1 mg/mL cyclosporine A (Novartis). B cells were infected with 0.1 virus units per cell. Half of the culture supernatant was exchanged at day 1 post infection. Outgrowing cultures were transferred after 1–2 weeks to a fresh plate, and further cultivated autonomously in the absence of LL8 cells. Presence of mutant EBV and absence of endogenous EBV wild type was confirmed by PCR every few weeks with primers L2BRC (5'-GCTTCCTCGTGCTTTACGGTATC-3') and L2BRD (5'-AAGAACTTTGACCTGTTGTCCCTG-3') for amplification of a product bridging the LMP2A deletion, primers L2INA (5'-CATTGCGGGTGGATAGCCTC-3') and L2BRD for amplification of the deleted sequence. Proliferating EBV-infected LCLs were analyzed and used in T cell assays between 1.5 and 4 months after infection.
DNA transfection experiments were performed with 293T human embryonic kidney cells with a plasmid expressing HCMV pp65 fused to a destabilized green fluorescent protein (GFP) under the HCMV immediate-early promoter (provided by Manuel Albanese and Wolfgang Hammerschmidt) (pp65-GFP). A plasmid expressing full-length LMP2A under the control of the SV40 early promoter (pSV-LMP2A) and the corresponding control (pSV) were used for co-transfection. The pp65-GFP plasmid and pSV plasmids were used at a 1:10 ratio. One day before transfection, 293T cells were plated in 24-well plates (1×105 cells/well). Transfection was performed with a mix of 1.17 μl of Metafectene Pro (Biontex) and 390 ng of plasmid DNA in a volume of 100 μl of OptiMEM (Gibco) for each well, according to the manufacturer's protocol. Two days after transfection, cells were used for T cell assays and assessment of transfection efficiency by FACS. About 40–50% of cells were GFP-positive.
To identify EBV-positive donors among anonymously purchased buffy coats, serum-containing cell-free supernatant was tested for IgG specific for EBV EBNA1 and VCA by a rapid immunofiltration assay (RDT EBV IgG Assay, Bio-Rad). EBV-specific T cells were directly isolated from PBMCs of EBV-seropositive, HLA-typed donors after stimulation with a matched peptide and IFN-γ secretion assay (Miltenyi Biotec). For single T cell cloning, isolated IFN-γ-secreting cells were seeded into round-bottom 96-well plates at 0.7 or 2.5 cells per well in 200 μl of medium supplemented with 1000 U/mL rIL-2, 1×105/mL irradiated (50 Gy) HLA-matched LCLs, and 1.5×106/mL of a mixture of irradiated (50 Gy) allogeneic PBMCs from at least three different donors. Outgrowing T cell clones were expanded in round-bottom 96-well plates by restimulating every 2 weeks under the same conditions. The specificity of the T cell clones was determined by IFN-γ ELISA with individual antigenic peptides (see below), and by staining with HLA-peptide pentamers (Proimmune, Oxford, UK). The T cell clones were specific for the following epitope peptides, abbreviated by their first three amino acids in one-letter code: CLG (CLGGLLTMV, LMP2, A*0201) [80], YVL (YVLDHLIVV, BRLF1 A*0201) [81], FLY (FLYALALLL, LMP2, A*0201) [60,82], RPP (RPPIFIRRL, EBNA3A, B*0702) [83], RAK (RAKFKQLL, BZLF1, B*0801) [84], HPV (HPVGEADYFEY, EBNA1, B*3501) [85]. HCMV-specific CD8+ T cells clones specific for NLV (NLVPMVATV, pp65, A*0201) [86], CRV (CRVLCCYVL, IE-1, C*0702) [87] and VLE (VLEETSVML, IE-1, A*0201) [88] were obtained as described [87].
Flow cytometric analysis was performed with a BD FACS Calibur or a BD LSR Fortessa machine. Analysis of WT and mutant LCLs lacking LMP2A established from the same donor was always conducted in parallel and for at least one WT line and one ΔLMP2A line. Generally, 1–1.5×105 cells were stained in a V-bottom 96-well plate at 4°C for 20 minutes in a volume of 20 μl, washed twice in 200 μl of buffer (PBS + 2% FCS), resuspended in buffer, and analyzed immediately. When unlabeled antibodies were included in the staining that required counterstaining with labeled anti-Ig antibodies, two to three rounds of staining and washing were performed as necessary. Antibodies anti-MICA (clone 159227, unlabeled), anti-MICB (clone 236511, unlabeled), and anti-ULBP1 (clone 170818, unlabeled) were from R&D Systems and were counterstained with anti-IgG-APC (clone Poly4053) from Biolegend. Anti-ULBP2/-5/-6 (clone 165903, PE-labeled) and anti-ULBP3 (clone 166510, PE-labeled), were from R&D Systems; anti-ULBP4 (6E6, unlabeled) was from Santa Cruz; anti-CD11a (clone G43-25B, PE-labeled) was from BD Bioscience; anti-CD48 (clone 5F4, PE-labeled), anti-PD-L1 (clone 29E.2A3, APC-labeled), anti-CD86 (clone IT2.2, APC-labeled), anti-ICAM-1 (clone HCD54, APC-labeled), anti-HLA-ABC (clone W6/32, APC-labeled), and anti-HLA-A2 (clone BB7.2, PE-labeled) were from Biolegend; and anti-HLA-B7 (clone BB7.1) was from Millipore. Antibody anti-HLA-A3 (clone 4i87, IgM, USB) was counterstained with anti-IgM-PE (clone RMM-1; BioLegend). A hybridoma producing the HLA-C/HLA-E-specific antibody DT9 (IgG2b) was kindly provided by Véronique Braud, Nice, France [89], and counterstained with anti-mouse IgG-APC (clone Poly4053) from Biolegend. HLA-Bw6 was stained with a PE-labeled human antibody (REA143, 130-099-843) from Miltenyi Biotec. Isotype controls were IgG2A (clone 133304) and IgG2B (clone 133303) from R&D Systems; IgG1 (clone MOPC-21), IgG2A (clone MOPC-173), and IgG2B (clone MG2b-57) from Biolegend.
T cells were stained with antibodies anti-NKG2D (1D11, APC-labelled), anti-CD8 (RPA-T8, Pacific Blue- or FITC-labelled), anti-CD3 (HIT3a, PE-Cy5-labelled) from BioLegend.
Combined analysis of proliferation and apoptosis of LCLs was performed using CellTrace Violet (Life Technologies), AnnexinV-Cy5 conjugate (ApoScreen, Southern Biotech), and propidium iodide (PI, Life Technologies). One million cells was stained with 1 μl of CellTrace Violet in 1 ml PBS, washed, cultivated in 3 ml of full medium in a 12 well plate at 1x106 cells/well, and incubated for 4 days. Cells were harvested, counted, and 2.5×105 cells were stained in 200 μl buffer with 2 μl of AnnexinV-Cy5 and propidium iodide at 1 μg/ml, before proceeding to flow cytometric analysis.
For IFN-γ ELISA, effector cells (2.5×104 cells/well) and target cells (5×104 cells/well) were co-cultivated in 200 μl/well in a V-bottom 96-well plate at 37°C and 5% CO2. For IL-10 ELISA, LCLs were plated at 5×105 cells/ml in a 12-well or V-bottom 96-well plate and incubated at 37°C and 5% CO2. Supernatants were harvested after 16–18 hours. ELISA was performed according to the manufacturer's instructions (Mabtech, Nacka, Sweden). Blocking by specific purified antibodies was performed where indicated. Antibody was added to the effector cells (anti-NKG2D, anti-IL10R) or to the target cells (anti-IL10) at a pre-established concentration and incubated for 1 hour at 37°C prior to the addition of the target or effector cells, respectively. We used antibodies to IL-10 and IL-10R that were previously shown to neutralize activity at the same or lower concentrations [90–92]. Antibodies used for blocking, and matched isotype controls, were all low-endotoxin, azide-free (LEAF) and purchased from Biolegend: anti-NKG2D (clone 1D11, used at 50 μg/ml) with isotype (mouse IgG1, clone MOPC-21), anti-IL10R (clone 3F9, used at 20 μg/ml) with isotype (rat IgG2a, clone RTK2758), anti-IL10 (clone JES3-9D7, used at 20 μg/ml) with isotype (rat IgG1, clone RTK2071), anti-PD-1 (clone EH12.2H7, used at 10 μg/ml) with isotype (mouse IgG1, clone MOPC-21). Investigation of the recognition by CD8+ T cell clones of WT and ΔLMP2A LCLs established from the same donor was always performed in parallel and for at least one WT line and one ΔLMP2A line. Statistical analysis was performed with GraphPad Prism software.
The cytotoxic reactivity of CD8+ T cell clones against target cells was measured by calcein-release assay. Target cells (4×105) were labeled with 1 μg/ml in 500 μl medium. After incubation for 1 hour at 37°C cells were washed 3 times with sterile PBS, and 5×103 target cells/well were plated in a V-bottom 96-well plate (final volume 200 μl/well). For each target cell type, spontaneous release (no effector cells, 0% lysis) and maximal release (addition of 0.5% of triton-X 100, 100% lysis) was determined. Effector cells were co-incubated with target cells for 3 hours at a 2:1 ratio. Afterwards, 100 μl of supernatant were collected and transferred to a fresh flat-bottom 96-well plate and fluorescence intensity at 485/535 nm was measured in an Infinite F200 PRO fluorometer (Tecan). RPMI without phenol red was used to reduce background fluorescence.
Total RNA was extracted from LCLs with the RNeasy Mini Kit, and cDNA synthesis was performed with the QuantiTect kit, both from Qiagen, Hilden, Germany. Quantitative PCR was performed on a LightCycler 480 (Roche, Basel, Switzerland) using the SYBR Green LC480 Mix. Primers were as follows: human IL-10 (forward: 5'-GCAGGTGAAGAATGCCTTTA-3', reverse: 5'-CCCTGATGTCTCAGTTTCGT-3'), BZLF1 unspliced (forward: 5'-GCACATCTGCTTCAACAGGA-3', reverse: 5'-CCAAACATAAATGCCCCATC–3'), EBNA1 (forward: 5'-CGCAAGGAATATCAGGGATG-3', reverse: 5'-TCTCTCCTAGGCCATTTCCA-3'), gp350 (forward: 5'- TTGTGAAATTTCGCCATCCT-3', reverse: 5'-CAAAACCCCGTGTACCTG-3'). Primers specific for BCRF1 (vIL-10), EBNA3A, LMP2AB, and GUSB were described before [42]. Specific mRNA expression was standardized to the housekeeping gene β-glucuronidase (GUSB) [91,93]. WT and ΔLMP2A LCLs simultaneously established from the same donor were always analyzed in parallel.
Cells were incubated for 15 min on ice with lysis buffer (50 mM Tris/HCl pH 7.4, 150 mM NaCl, 1% NP40, 0.5% DOC, 0.1% SDS) together with protease inhibitor (completeMini, Roche). Protein concentration was determined with the Bio-Rad Protein Assay. Proteins were separated on an 8% SDS-PAGE gel and transferred to a nitrocellulose membrane by semi-dry blotting. Blots were probed with antibodies specific for LMP1 (1G6, provided by Elisabeth Kremmer, 1:25 dilution) [94] and GAPDH (1A7, 1:10 dilution). Blots were further probed with secondary antibodies conjugated to horseradish peroxidase, and immunoreactive proteins were detected by incubation with chemoluminescence substrate (0.1M Tris/HCl, pH 8.8, 200 mM p-Coumaric Acid in DMSO, 1.25 mM Luminol in DMSO) and exposure of CEA RP NEW films (Agfa HealthCare, Belgium).
|
10.1371/journal.pbio.3000100 | A transient helix in the disordered region of dynein light intermediate chain links the motor to structurally diverse adaptors for cargo transport | All animal cells use the motor cytoplasmic dynein 1 (dynein) to transport diverse cargo toward microtubule minus ends and to organize and position microtubule arrays such as the mitotic spindle. Cargo-specific adaptors engage with dynein to recruit and activate the motor, but the molecular mechanisms remain incompletely understood. Here, we use structural and dynamic nuclear magnetic resonance (NMR) analysis to demonstrate that the C-terminal region of human dynein light intermediate chain 1 (LIC1) is intrinsically disordered and contains two short conserved segments with helical propensity. NMR titration experiments reveal that the first helical segment (helix 1) constitutes the main interaction site for the adaptors Spindly (SPDL1), bicaudal D homolog 2 (BICD2), and Hook homolog 3 (HOOK3). In vitro binding assays show that helix 1, but not helix 2, is essential in both LIC1 and LIC2 for binding to SPDL1, BICD2, HOOK3, RAB-interacting lysosomal protein (RILP), RAB11 family-interacting protein 3 (RAB11FIP3), ninein (NIN), and trafficking kinesin-binding protein 1 (TRAK1). Helix 1 is sufficient to bind RILP, whereas other adaptors require additional segments preceding helix 1 for efficient binding. Point mutations in the C-terminal helix 1 of Caenorhabditis elegans LIC, introduced by genome editing, severely affect development, locomotion, and life span of the animal and disrupt the distribution and transport kinetics of membrane cargo in axons of mechanosensory neurons, identical to what is observed when the entire LIC C-terminal region is deleted. Deletion of the C-terminal helix 2 delays dynein-dependent spindle positioning in the one-cell embryo but overall does not significantly perturb dynein function. We conclude that helix 1 in the intrinsically disordered region of LIC provides a conserved link between dynein and structurally diverse cargo adaptor families that is critical for dynein function in vivo.
| The large size and complex organization of animal cells make the correct and efficient distribution of intracellular content a challenge. The solution is to use motor proteins, which harness energy from ATP hydrolysis to walk along actin filaments or microtubules, for directional transport of cargo. The multi-subunit motor cytoplasmic dynein 1 (dynein) is responsible for transport directed toward the minus ends of microtubules. An important question is how dynein is recruited to its diverse cargo, which includes organelles such as endosomes and mitochondria, proteins, and mRNA. In this study, we use nuclear magnetic resonance spectroscopy to show that the light intermediate chain (LIC) subunit of human dynein uses a short helix in its disordered C-terminal region to bind structurally distinct adaptor proteins that connect the motor to specific cargo. We then use genome editing in the animal model C. elegans to demonstrate the functional relevance of the C-terminal LIC helix for dynein-dependent cargo transport in neurons. Thus, dynein recruitment to cargo involves a highly conserved interaction between LIC and adaptor proteins.
| Microtubule-based cargo transport and force production are critical for a wide range of cellular and developmental processes. In animal cells, the 1.4-MDa complex cytoplasmic dynein 1 (dynein) is the major molecular motor with motility directed toward microtubule minus ends. Dynein-driven cargo transport is particularly crucial in highly polarized cells such as neurons. In axons, whose microtubule plus ends are uniformly oriented toward the axonal tip, dynein is responsible for the retrograde transport of diverse vesicle and organelle cargo toward the cell body. Mutations in dynein that alter axonal transport kinetics have been linked to a variety of nervous system disorders, including spinal muscular atrophy, motor neuron disease, Perry syndrome, and Charcot-Marie-Tooth 2 disease [1,2]. In addition to transporting cargo, dynein can exert pulling forces on microtubules when the motor is stably anchored at subcellular sites such as the cell cortex. A striking example of dynein-dependent force production occurs during mitosis, when cortically localized dynein pulls on astral microtubules to position and orient the bipolar spindle, which in turn defines the axis along which the cell will divide. Dynein's functional versatility implies tight regulation of localization and motor activity, the molecular basis of which has only recently begun to be understood.
Dynein is a 12-subunit complex consisting of a dimerized heavy chain (HC) with a C-terminal motor domain and two copies each of five accessory chains that bind along the HC N-terminal tail: dynein intermediate chain (IC), light intermediate chain (LIC), and three types of light chain (LC). In vivo, dynein function requires the cofactor dynactin, which is itself a 1.1-MDa complex built around a short filament of actin-related protein 1 (ARP1) [3]. In recent years, a number of coiled-coil proteins, referred to as activating adaptors [4], have been shown to recruit dynein to cargo while simultaneously promoting the association of dynein with dynactin [5,6]. Dynein, dynactin, and the N-terminal coiled-coil region of activating adaptors form a stable three-way assembly capable of highly processive motility in vitro [7,8], whereas the C-terminal region of adaptors links to cargo [4]. Cryo–electron microscopy (EM) studies with the adaptors bicaudal D homolog 2 (BICD2), BICD-related protein 1 (BICDR1), and Hook homolog 3 (HOOK3) have revealed the molecular arrangement within the dynein-dynactin-adaptor assembly [3,9–11]: the adaptor coiled-coil region binds along the length of the dynactin ARP1 filament with the adaptor N terminus located at the filament's barbed end, and the N-terminal tail of dynein HC makes contact with both the ARP1 filament and the coiled-coil region of the adaptor.
An additional contact between dynein and adaptors involves the C-terminal region of the LIC subunit (LIC-C) [12,13]. Whereas the highly conserved N-terminal GTPase-like domain of LIC binds tightly to the HC [12,14], the LIC-C sequence is more divergent and predicted to be disordered. Vertebrates possess two genes for LIC (LIC1 and LIC2), which may specify distinct dynein populations. Adaptors known to bind LIC1-C include BICD2 and Spindly (SPDL1), which are likely related [15,16], as well as the structurally distinct adaptors HOOK3, RAB11 family-interacting protein 3 (RAB11FIP3; hereafter referred to as FIP3), and RAB-interacting lysosomal protein (RILP) [12,17]. Whether LIC2-C also interacts with these adaptors has not been examined. SPDL1/BICD2 and HOOK3 bind to LIC1-C through a motif in their first coiled-coil segment (the CC1 box) and the N-terminal Hook domain, respectively, and point mutations in these adaptors that abrogate binding to LIC1-C compromise the formation of the dynein-dynactin-adaptor assembly [16–18]. Consequently, HOOK3 mutants that fail to bind LIC1-C do not support processive dynein runs in vitro [17]. A mutation in the CC1 box of Drosophila melanogaster BicD causes a hypomorphic loss-of-function phenotype [19], indicating that the BicD-LIC interaction is functionally relevant in vivo. Numerous other loss-of-function studies have implicated LIC in many dynein-dependent processes, including mitosis and retrograde cargo transport in axons. However, the extent to which LIC loss-of-function phenotypes reflect an important role for LIC-C is less clear, as the dynein complex becomes destabilized when LIC is absent in D. melanogaster and Aspergillus nidulans [20,21], as well as in LIC1-deficient mice [22]. Indeed, biochemical analysis indicates that the N-terminal LIC domain plays an important structural role within the dynein complex [14,23]. Thus, although LIC is clearly essential for dynein function in vivo, the specific contributions of LIC-C remain to be determined.
Here, we dissect the interaction between LIC-C and dynein adaptors in vitro and in the animal model C. elegans. Nuclear magnetic resonance (NMR) analysis shows that human LIC1-C is intrinsically disordered and possesses two short segments with helical propensity. In agreement with a recent report [13], we show that helix 1 of LIC1-C is essential for binding to BICD2, SPDL1, and HOOK3, and we extend this finding to the adaptors FIP3, RILP, ninein (NIN), and trafficking kinesin-binding protein 1 (TRAK1), as well as to LIC2-C. Finally, we show that LIC-C mutants generated by genome editing in C. elegans have major defects in postembryonic cell division and retrograde axonal cargo transport, demonstrating the crucial importance of LIC-C helix 1 for dynein function in vivo.
To gain mechanistic insight into how LIC-C interacts with dynein adaptors that are diverse in structure and function (Fig 1A and 1B, S1 Fig), we first characterized LIC1-C (residues 388–523) by NMR spectroscopy. The 15N-1H heteronuclear single quantum coherence (HSQC) spectrum of LIC1-C at 25°C showed a narrow 7.8 to 8.6 parts per million (ppm) amide-proton chemical shift range, indicating a predominance of structural disorder (Fig 1C). Near-complete backbone resonance assignment was achieved (104 out of the 117 nonproline residues; Fig 1C, S2 Fig). An overlay of 15N-1H HSQC spectra of two smaller constructs, consisting of residues 388–471 and 472–523, reproduced the spectrum obtained from the entire LIC1-C (S2A Fig). This finding indicates that there are no long-range interactions between N- and C-terminal segments of LIC1-C.
To assess residual secondary structure in LIC1-C, we computed the secondary structure propensity (SSP) score developed by Forman-Kay and colleagues [24]. The method combines different backbone chemical shifts into a single residue–specific SSP score. The results closely matched those obtained from the GOR4 secondary structure prediction program [25] (Fig 1D, S1 Data). An SSP score of +1 and −1 indicates a fully formed α helix and β sheet, respectively. The SSP scores for LIC1-C residues with predicted α-helical propensity (residues 444–454 and 497–504) were positive values up to 0.35. However, assuming a linear relationship between the score and the population, helical structure was only transiently sampled at about 20%. These two segments of LIC1-C, which are highly conserved (S1A Fig), will be referred to as helix 1 and helix 2, respectively. The remaining segments scored mostly slightly negative SSP values and GOR4 values around zero, indicating complete structural disorder (note that β sheet and random coil have both extended secondary structure with similar scores in practice).
To characterize the dynamic properties of LIC1-C, we performed 15N NMR relaxation measurements (Fig 1E, S3 Fig). The relaxation rates are sensitive probes of the amplitudes and timescales of residue-specific structural fluctuations. The longitudinal relaxation rates R1, which report on motion faster than nanoseconds, were 1.2 s−1 on average. Although such a value is not compatible with a globular protein, it is indicative of an intrinsically disordered protein that undergoes extensive local reorientation. Residues 441–456 and 492–508 had larger R1 values, which are suggestive of increased structural order of the residues with α-helical propensity. The residues between helices 1 and 2 had consistently lower-than-average R1 values, indicating a highly flexible interhelical linker.
The fast dynamics inferred from R1 measurements reduce the transverse relaxation rates R2 typically observed for folded proteins, but in addition, potential motion slower than nanoseconds might increase them. Whereas the measured R2 average of 4.7 s−1 is again only reconcilable with high structural disorder, the helical regions showed larger values in line with partial helix formation (Fig 1E). Since the R1 values suggest less flexibility in the helices on the fast timescale, it is likely that the increased R2 values originate also exclusively from reduced fast-motion amplitudes. In support of this, relaxation dispersion experiments did not indicate any slow micro-millisecond dynamics. The segment with the smallest values was again the interhelical linker. Since slow motion is absent throughout the protein, the ratio R2/R1 allowed extraction of an effective residue-specific reorientational tumbling time that is independent of the motional amplitude [26]. The average tumbling time was 4.5 ns, which is much smaller than a typical value expected for a folded protein consisting of approximately 140 residues.
The 1H-15N nuclear Overhauser enhancement (NOE) is another probe of the amplitudes of motions taking place on the sub-nanosecond timescale (Fig 1F). The values of nearly all residues fall below 0.5, which is again indicative of structural disorder with large amplitudes. In support of the findings described above, the largest values were typically found for the helix 1 and 2 segments, which suggests higher structural order than the other regions.
In conclusion, the relaxation data show that LIC1-C is intrinsically disordered with large amplitudes of motions faster than nanoseconds. In helix 1 and 2, the amplitudes are smaller and the local reorientation slower but still on the fast timescale. These results support the highly transient character of the helices derived from the SSP score.
For independent confirmation of the residual helical propensity in LIC1-C, we collected circular dichroism (CD) spectra of LIC1-C (S4 Fig). Whereas fully formed α-helical proteins show prominent negative bands at 222 and 208 nm and a positive band at 193 nm, disordered proteins have very low ellipticity above 210 nm and negative bands near 195 nm [27]. The spectrum of LIC1-C corresponded largely to structural disorder. However, there were negative values at 208 nm and a negative shoulder at 222 nm. These features are indicative of residual helical structure, in agreement with our findings from NMR spectroscopy. The helical content was maintained over a large temperature range (25, 37, 50, 60, and 70°C), similar to previous reports on other intrinsically disordered proteins [28].
To identify adaptor binding sites in LIC1-C, we recorded 15N-1H HSQC spectra of LIC1-C bound to SPDL1(2–359), BICD2(2–422), and HOOK3(2–239) using molar ratios of 1:0, 1:0.5, and 1:1 (Fig 2A, 2B, 2C and 2D). The spectra obtained from each titration series revealed similar modes of interaction for the three adaptors. Rather than chemical shifts being perturbed, many peaks were significantly attenuated because of binding. This indicates decrease in the local motion of the bound residues; increase of the hydrodynamic radius of the entire complex, resulting in increase in the effective tumbling time; and/or contributions from conformational and chemical exchange [29]. Inspection of the peak height ratios between bound and free LIC1-C showed that the residues affected were mainly located in helix 1 and to a lesser extent in helix 2 (Fig 2B, 2C and 2D). In addition, we observed intensity reduction for the peaks of residues 418–421, which also tend to be conserved (S1A Fig). The quenching of segment 418–421 and helix 2 was more pronounced for SPDL1(2–359) than for BICD2(2–422) or HOOK3(2–239) (Fig 2B, 2C and 2D).
To assess binding affinities between LIC1-C and adaptors, we conducted surface plasmon resonance (SPR) experiments in which LIC1-C::6xHis was immobilized on a nitrilotriacetic acid (NTA) chip. We obtained a KD of 5.7 ± 1.9 μM for the interaction between LIC1(388–523) and SPDL1(2–359) (Fig 2E). LIC1(388–471), which contains helix 1 but not helix 2, bound to SPDL1(2–359) with the same affinity as LIC1(388–523) (5.40 ± 0.25 μM). Similarly, LIC1(388–523) and LIC1(388–471) bound to BICD2(2–422) with comparable affinity (KD of 1.07 ± 0.05 μM and 0.7 ± 0.01 μM, respectively) (Fig 2F). We also measured binding to fluorescently labeled LIC1(388–523) in microscale thermophoresis (MST) experiments. We obtained KD values of 13.1 ± 0.4 μM and 6.0 ± 1.3 μM for the interaction with SPDL1(2–359) and BICD2(2–422), respectively, which is in reasonable agreement with the SPR analysis (S5A and S5B Fig). From SPR and MST experiments with LIC1(388–523) and HOOK3(2–239), we obtained KD values of 1.9 ± 0.4 μM and 4.0 ± 0.8 μM, respectively (Fig 2G, S5C Fig), which are similar to those obtained with BICD2(2–422). We conclude that SPDL1, BICD2, and HOOK3 bind to LIC1-C with an affinity in the single-digit micromolar range and that an LIC1-C fragment comprising residues 388–471 is sufficient for binding. These results are in agreement with the NMR titration experiments, which showed the most pronounced intensity quenching for helix 1 residues.
NMR spectroscopy and SPR analysis strongly suggested that helix 1 in LIC1-C is the major binding site for BICD2, SPDL1, and HOOK3. To directly test whether helix 1 is important for adaptor binding, we performed in vitro pull-down experiments using purified glutathione S-transferase (GST)-tagged versions of LIC1-C (Fig 3A and 3B) and LIC2-C (Fig 4A and 4B) as bait. In addition to BICD2(2–422), SPDL1(2–359), and HOOK3(2–552), we purified full-length versions of RILP (residues 1–401) and FIP3 (residues 2–756), as well as an N-terminal fragment of NIN (residues 1–693) (Fig 3C, S1B Fig). Using Coomassie Blue staining and immunoblotting for the Strep-tag II at the adaptor C terminus, all adaptors were readily detected in GST pull-downs with LIC1(388–523) and LIC1(388–471) (Fig 3D). By contrast, no significant binding was observed with LIC1(472–523). The same result was obtained with corresponding constructs of LIC2 (residues 375–492, 375–450, and 451–492) (Fig 4C). Deleting helix 1 in LIC1-C (Δ440–455), or mutating either the two phenylalanines or the two leucines in helix 1 to alanine (F447A/F448A and L451A/L452A) abrogated binding between LIC1-C and all six adaptors (Fig 3D). Mutating the two leucines in helix 1 of LIC2-C to alanine (L436A/L437A) had the same effect (Fig 4C). We conclude that helix 1 in LIC1-C and LIC2-C is essential for binding to cargo adaptors that are diverse in structure and function.
Next, we asked whether helix 1 of LIC1-C (residues 440–455) was sufficient to bind adaptors. In GST pull-down experiments, RILP bound robustly to LIC1(388–523) and LIC1(440–455) (Fig 3A, 3B and 3E). SPR experiments in which LIC1-C was immobilized through its N-terminal GST tag confirmed that RILP binds LIC1(440–455) with similar affinity as LIC1(388–523) (KD of 1.67 ± 0.21 and 1.18 ± 0.14 μM, respectively; S6 Fig). We conclude that helix 1 of LIC1-C is sufficient for RILP binding. By contrast, none of the other five adaptors were detected in GST pull-downs with LIC1-C helix 1, suggesting that efficient binding requires additional segments in LIC1-C (Fig 3E).
NMR spectroscopy suggested that, in addition to helix 1, residues N-terminal to helix 1, as well as residues in helix 2, also make contact with adaptors. To investigate the contribution of these sites to adaptor binding, we first truncated LIC1(388–471) from its N terminus to generate progressively smaller fragments (residues 402–471, 414–471, 424–471, and 440–471), all of which retained helix 1 (Fig 3A). SPDL1(2–359), BICD2(2–422), HOOK3(2–552), FIP3(2–756), and NIN(1–693) showed reduced binding to LIC1-C with decreasing fragment size, whereas RILP bound equally well to all LIC1-C fragments (Fig 3E). Thus, for most adaptors, efficient binding to LIC1-C requires residues located N-terminally of helix 1. To assess the contribution of helix 2, we generated a fragment from the beginning of helix 1 until the end of LIC1-C (residues 440–523). Although SPDL1(2–359), BICD2(2–422), and HOOK3(2–552) were not detected in pull-downs using LIC1(440–455), they bound weakly to LIC1(440–523), potentially reflecting a modest contribution of helix 2 to the interaction (Fig 3E). Overall, our binding experiments with purified components demonstrate that helix 1 of LIC-C is essential for the interaction with adaptors but suggest that most adaptors also need to make additional contacts with the flexible LIC-C scaffold for efficient binding. Specifically, residues preceding helix 1 appear to make a significant contribution. The exception is RILP, for which LIC-C helix 1 on its own is sufficient for binding.
To determine whether helix 1 of LIC-C contributes to dynein function in vivo, we turned to the animal model C. elegans, which expresses a single dynein LIC, dli-1 [30]. Just as in other LIC orthologs, the C-terminal helix 1 is highly conserved in DLI-1 (Fig 5A). We used clustered regularly interspaced short palindromic repeat/CRISPR-associated 9 (CRISPR/Cas9)–mediated genome editing to mutate either the two phenylalanines or the two leucines in helix 1 to alanine (F392A/F393A and L396A/L397A), in analogy to the human LIC-C mutations we characterized in vitro (Fig 5B). For comparison, we also deleted the entire DLI-1 C-terminal region (Δ369–443). For all three dli-1 mutants, homozygous offspring from heterozygous mothers developed to adulthood but were sterile, suggesting that the C-terminal region of DLI-1, and helix 1 in particular, is essential for dynein function in vivo (Fig 5B). Because RNA interference (RNAi)-mediated depletion of DLI-1 showed that DLI-1 is required for the stability of dynein HC 1 (DHC-1; S7 Fig), we addressed the possibility that the phenotype of our dli-1 mutants was a consequence of reduced DHC-1 levels. Immunoblotting of homozygous adults with an antibody against DHC-1 demonstrated that this was not the case. On the contrary, DHC-1 levels appeared slightly increased in all three dli-1 mutants (Fig 5C and S7B Fig). This indicates that the C-terminal DLI-1 mutants do not interfere with the binding of the N-terminal GTPase-like domain to DHC-1.
Differential interference contrast imaging revealed severe morphological defects in day 1 adults homozygous for the dli-1 mutations (Fig 5D, 5E and 5F). All three dli-1 mutants had a slightly dumpy body with a protruding vulva and a highly disorganized and underdeveloped gonad (Fig 5D and 5E), which is indicative of problems with postembryonic cell divisions. Furthermore, the prevalence of vacuoles suggested widespread necrotic-like cell death (Fig 5F). Consistent with this, dli-1 mutants exhibited severe defects in locomotion (Fig 5G), which was assessed by determining the body bending frequency in liquid medium (wild type, 1.50 ± 0.02 Hz; Δ369–443, 0.50 ± 0.03 Hz; F392A/F393A, 0.46 ± 0.04 Hz; L396A/L397A, 0.50 ± 0.03 Hz). Dli-1 mutants had a significantly shorter life span compared to wild-type animals (wild type, 18.3 ± 0.7 d; Δ369–443, 13.7 ± 0.4 d; F392A/F393A, 12.3 ± 0.5 d; L396A/L397A, 13.0 ± 0.5 d) (Fig 5H). None of the dli-1 mutant animals lived beyond 24 d, whereas 27% of wild-type animals did. We conclude that deletion of the DLI-1 C-terminal disordered region and point mutations in the conserved helix 1 cause similar defects in development, locomotion, and life span. These phenotypes are reminiscent of those reported for other dynein mutants, including null mutants of dli-1 [30,31].
Next, we sought to assess the impact of the C-terminal dli-1 mutations on dynein-dependent cargo transport. We chose to examine the distribution and transport kinetics of membrane cargo in axons of touch receptor neurons, which are mechanosensory neurons whose processes extend along the length of the animal just underneath its cuticle (Fig 6A). Axonal microtubules in touch receptor neurons are polarized with their plus ends oriented toward the axonal tip, and dynein is therefore responsible for retrograde transport of cargo toward the cell body [31,32]. Using Mos transposase-mediated single-copy insertion [33], we constructed transgenic animals expressing mKate2::RAB-5 from the mec-7 promoter and crossed them with animals expressing soluble green fluorescent protein (GFP) from the mec-4 promoter (allele zdIs5). This allowed simultaneous visualization of early endosomes (mKate2), a cargo of dynein, and neuronal morphology (GFP) (Fig 6A). In contrast to control day 1 adults, all three dli-1 mutants exhibited a pronounced misaccumulation of mKate2::RAB-5 at the axonal tips and nerve ring synapses of anterior lateral mechanosensory (ALM) and anterior ventral mechanosensory (AVM) neurons (Fig 6B, 6C, 6D and 6E). This defect resembled the previously described misaccumulation of synaptic vesicle proteins in touch receptor neurons of dynein mutants [31].
To determine how dli-1 mutants affect the kinetics of axonal transport, we imaged mKate2::RAB-5-labeled early endosomes at high temporal resolution in an axonal segment close to the cell body of ALM neurons at the larval L4 stage (Fig 6A and 6F and S1 Movie). Quantitative analysis of kymographs revealed that the three dli-1 mutants had near-identical defects in axonal transport. The fraction of particles moving in the retrograde direction was reduced significantly, with a corresponding increase in the fraction of stationary particles (Fig 6G). Analysis of individual runs showed that particles spent less time in retrograde motion and paused more frequently and for longer periods of time (S8A Fig). Particles also had a decreased run length and moved at reduced speed, especially in the retrograde direction (Fig 6H and 6I). Overall, this analysis revealed that all three dli-1 mutants strongly impaired retrograde axonal transport of early endosomes, consistent with compromised dynein function.
We also examined the axonal distribution of two additional types of dynein cargo, each labeled with a marker expressed from a single-copy transgene: synaptic vesicles (SNB-1 [synaptobrevin 1]::mKate2) and mitochondria (TOMM-20 [translocase of outer mitochondrial membrane 20][1–54]::mKate2). Because the three dli-1 mutants had identical defects in the distribution and transport kinetics of mKate2::RAB-5, we restricted our analysis to the dli-1(L396A/L397A) mutant. We found that dli-1(L396A/L397A) animals misaccumulated SNB-1::mKate2 at axonal tips (S8B, S8C and S8D Fig), similar to the results of a prior study that examined the distribution of SNB-1::GFP in the null mutant dli-1(js351) [31].
When examining the distribution of TOMM-20(1–54)::mKate2 particles (Fig 7A), we found that axons of ALM neurons contained an average of 21 ± 1 mitochondria in control day 1 adults, corresponding to a density of 5.6 mitochondria per 100 μm, and that mitochondria were evenly distributed along axons (Fig 7B, 7C, 7D and 7E). These findings are consistent with prior reports [34,35]. Mitochondria remained evenly distributed in ALM axons in dli-1(L396A/L397A) animals (Fig 7E). However, the average number of axonal mitochondria increased to 33 ± 1, corresponding to a density of 9 mitochondria per 100 μm (Fig 7C and 7D). Close inspection of mitochondrial morphology suggested that the increase was primarily due to an excess of small TOMM-20(1–54)::mKate2 puncta, which may represent fragmented mitochondria (Fig 7B). Thus, the dli-1(L396A/L397A) mutation does not result in distribution bias of mitochondria along the axon but appears to promote mitochondrial fission. Similar results were obtained for the dli-1(Δ369–443) mutation (Fig 7C, 7D and 7E). Analysis of mitochondrial transport kinetics in dli-1(L396A/L397A) animals revealed a reduction in the speed of transport, with a more pronounced effect in the retrograde direction (Fig 7F and 7G, and S2 Movie). However, none of the other transport parameters were significantly altered (S8E and S8F Fig).
Given the modest impact of the dli-1(L396A/L397A) mutation on mitochondrial transport, we asked whether the mitochondrial adaptor TRAK1, which contains a CC1 box similar to SPDL1 and BICD2, binds LIC1. GST pull-downs with purified human proteins showed that maltose-binding protein (MBP)::TRAK1(103–187), which corresponds to the first coiled-coil segment, was able to bind GST::LIC1(388–523), albeit weakly and only when artificially dimerized using a general control nondepressible 4 (GCN4) sequence (Fig 7H and 7I). The interaction was abolished when the two leucines in LIC1-C helix 1 were mutated, suggesting that LIC1 binds TRAK1 through a similar mechanism as the other adaptors.
Just like helix 1, helix 2 in LIC-C shows high sequence conservation, including in C. elegans (residues 420–428; Fig 8A and S1A Fig). To assess the role of helix 2 in vivo, we used genome editing to generate animals that express DLI-1 without its C-terminal 30 residues (Δ414–443). This truncated version of DLI-1 retains helix 1 (residues 388–400) (Fig 8A). In contrast to the helix 1 point mutants, homozygous dli-1(Δ414–443) animals were fully viable and fertile (embryonic viability 98.4% ± 0.6% and 97.3% ± 0.7% for wild type [>300 progeny, 12 mothers] and mutant [>300 progeny, 14 mothers], respectively). Furthermore, we observed no increase in mKate2::RAB-5 signal at axonal tips of touch receptor neurons, indicating that dynein-dependent transport of early endosomes was not significantly affected in this mutant (Fig 8B, 8C and 8D). Because homozygous dli-1(Δ414–443) animals produced progeny, we were able to examine the first mitotic division of embryos coexpressing GFP::β-tubulin and mCherry::histone H2B, which mark microtubules and chromosomes, respectively (Fig 8E). During mitosis, dynein-dependent pulling on centrosome-nucleated microtubules is essential for the separation and positioning of centrosomes, which form the spindle poles after nuclear envelope breakdown (NEBD). At the time of NEBD in control animals, centrosomes were fully separated and the centrosome–centrosome axis was oriented approximately parallel to the anterior–posterior axis (Fig 8E and 8F). In dli-1(Δ414–443) embryos, centrosomes separated normally, but the centrosome–centrosome axis was frequently severely tilted relative to the anterior–posterior axis (angle > 45° in 7 out of 21 mutant embryos versus 0 out of 13 controls), resulting in misoriented spindles. Despite initial misorientation, spindles ultimately correctly aligned with the anterior–posterior axis, and chromosome segregation completed without defects. The delay in spindle orientation indicates a subtle impairment of dynein-dependent pulling forces [36,37]. To further probe the functional significance of helix 2, we combined the dli-1(Δ414–443) mutation with a null allele of the dynein cofactor nud-2, nud-2(ok949), which represents a sensitized dynein background that also exhibits a spindle orientation delay and that we had previously exploited for an enhancer screen [36,38]. We observed neither an increase in lethality nor enhanced defects in spindle orientation in the double mutant relative to the single mutants (embryonic viability 85.9% ± 1.6% and 89.8% ± 1.4% for ok949 [>300 progeny, 15 mothers] and ok949; dli-1[Δ414–443] [>300 progeny, 19 mothers], respectively), consistent with the idea that dynein function in dli-1(Δ414–443) animals is relatively unperturbed.
We conclude that helix 2 of the DLI-1 C-terminal region makes a modest contribution to dynein-dependent pulling forces during mitosis but overall plays a minor role in vivo compared to helix 1, which is critical for dynein function. These results are consistent with our biochemical characterization of human LIC-C in vitro, which shows that helix 1, but not helix 2, is essential for the interaction between LIC-C and cargo adaptors.
How dynein achieves specificity for cargo is a key question underlying the functional versatility of the motor. Studies over the past few years have revealed that the N-terminal region of cargo adaptor proteins helps bring together the motor with its essential cofactor dynactin and that multiple distinct protein–protein interactions participate in the assembly of a processive dynein-dynactin-adaptor transport machine. Here, we dissected the structural determinants and the functional relevance of the interaction between the C-terminal region of the dynein LIC subunit and cargo adaptors. We present high-resolution molecular evidence that LIC-C is disordered and that a conserved segment with helical propensity, helix 1, is the main binding site for seven adaptors that differ in structure and cargo specificity. Our results agree with recent work that identified LIC1-C helix 1 as the critical structural element for the interaction with HOOK1/3, SPDL1, and BICD2 [13], and we extend these findings to RILP, NIN, FIP3, and TRAK1. The highly conserved phenylalanines and leucines of the amphipathic helix 1 are essential for the interaction, which occurs with single-digit micromolar affinity for SPDL1, BICD2, HOOK3, and RILP, suggesting a similar binding mechanism. NMR spectroscopy analysis shows that helix 1 is transient. The presence of transiently sampled secondary elements in intrinsically disordered proteins is common and may facilitate the transition to more rigid states upon binding [39–41]. The observed signal quenching upon binding to adaptors is unlikely to be caused entirely by an increased local tumbling time. Instead, micro-millisecond conformational exchange between various states of the helical elements or binding kinetics contribute significantly to the quenching. Both mechanisms suggest residual structural disorder. Indeed, various examples of disordered proteins that maintain partial disorder during interaction with other proteins have recently been reported [40,42–44]. A prior study used X-ray crystallography to show that LIC1-C helix 1 inserts into a hydrophobic cleft on HOOK3(1–160) [13]. An analogous hydrophobic cleft is likely formed by the CC1 box, the LIC1-C binding region in SPDL1 and BICD2 [13,16]. CC1 boxes are also present in other adaptors whose N-terminal regions are structurally similar to SPDL1 and BICD2, such as HAP1 and TRAK [45]. Consistent with this, we show that the first coiled-coil segment of TRAK1 interacts with LIC1-C in a helix 1-dependent manner. RILP, NIN, and FIP3 contain neither a Hook domain nor a CC1 box. The structural element that accommodates the LIC-C helix 1 in these adaptors remains to be identified.
LIC1-C helix 1 is essential for adaptor binding, but we find that helix 1 on its own—i.e., LIC1(440–455)—is not sufficient for efficient binding to 5 out of 6 adaptors tested, RILP being the exception. Pull-down assays with LIC1-C truncations show that segments N-terminal to helix 1 contribute to adaptor binding. LIC1-C binding to SPDL1, FIP3, and NIN appears to be particularly sensitive to N-terminal LIC1-C truncations, as we find that LIC1(424–471) has markedly reduced affinity relative to LIC1(388–471). This is in agreement with NMR analysis, which indicates that residues 418–421 participate in the interaction with SPDL1. NMR analysis also indicates that LIC1-C helix 2 participates in the interaction with adaptors, but in vitro binding experiments show that the contribution of helix 2 to the overall binding affinity is relatively minor.
Vertebrates possess two LIC paralogs that differ significantly in their C-terminal region. Prior studies in cultured cells suggest that there is functional redundancy among LIC1 and LIC2 but also hint at functional specialization in dynein-dependent cell division processes and membrane trafficking [46–53]. One attractive possibility is that the C-terminal regions of LIC1 and LIC2 discriminate between adaptors. Our results show that LIC1 and LIC2 use the same helix 1–based mechanism for adaptor binding, and we did not detect any striking differences in the ability of the two C-terminal regions to interact with our set of six adaptors. Nevertheless, it is plausible that LIC1 and LIC2 exhibit subtle preferences for specific adaptors. For example, we find that NIN binds slightly better to LIC2 than LIC1. Posttranslational modifications could possibly further modulate the affinity for adaptors in an isoform-specific manner.
Previous work in vitro showed that the dynein-dynactin-adaptor assembly fails to form in HOOK3 and SPDL1 mutants that cannot bind to LIC1 [16,17] and that addition of excess LIC1-C helix 1 inhibits processive movement mediated by HOOK3 or BICD2 fragments in motility assays [13]. Here, we present direct evidence that LIC-C helix 1 is important for dynein function in vivo. Like other invertebrates, the nematode C. elegans expresses only one LIC isoform, which facilitates the identification of functionally critical regions. In all assays, the outcome of mutating the two conserved phenylalanines or leucines in the C-terminal helix 1 of DLI-1 is identical to that of deleting the entire C-terminal region. Unlike DLI-1 depletions, the phenotype of the C-terminal DLI-1 mutations cannot be attributed to destabilization of DHC-1 and therefore likely reflects defective adaptor binding. On the animal level, sterility and other developmental defects suggest failure of postembryonic cell division, similar to what was described for dli-1 null mutants [30,31]. We show on the cellular level that the C-terminal DLI-1 mutations disrupt the distribution of early endosomes and presynaptic vesicles in axons of touch receptor neurons. Since we find that axonal length is normal in day 1 adults of dli-1(L396A/L397A) and dli-1(Δ369–443) mutants, this is unlikely to be an indirect consequence of a developmental defect. Instead, the misaccumulation at the axonal tip is suggestive of impaired retrograde transport by dynein, as previously described for other dynein mutants, including the dli-1 null mutant js351 [31]. Consistent with this, we show that the frequency, velocity, and run length of early endosome movement is significantly decreased in our dli-1 mutants, with a predominant effect on retrograde movement. Neurodegeneration (i.e., axon beading) starts to become evident in dli-1 mutant day 1 adults and likely contributes to the animals' severe locomotory deficit and shortened life span.
The residual retrograde movement of early endosomes in axons of dli-1 mutants could indicate that dynein retains a limited ability to form processive dynein-dynactin-adaptor complexes without the interaction between DLI-1 and adaptors. This may also explain the modest effect of dli-1 mutants on mitochondrial transport. However, it is difficult to rule out that a small fraction of wild-type maternal DLI-1, passed on from the heterozygous mother to homozygous mutant progeny, could persist in touch receptor neurons at the L4 stage and promote residual dynein activity. Nevertheless, the observation that dli-1 mutants have a more pronounced effect on early endosomes and synaptic vesicles than on mitochondria indicates that DLI-1 binding to adaptors may not be equally important for all cargo transport.
In contrast to the point mutants in the DLI-1 C-terminal helix 1, there are no obvious developmental defects in animals expressing the DLI-1(Δ414–443) mutant that lacks the C-terminal helix 2, and mKate2::RAB-5 is not misaccumulated at the axonal tip of touch receptor neurons in this mutant. We have not examined neuronal cargo transport kinetics, but given that dli-1(Δ414–443) animals are healthy and propagate normally, any defects are unlikely to be substantial. We do, however, observe a delay in mitotic spindle positioning in one-cell dli-1(ΔΔ414–443) embryos. The mild phenotype contrasts with the failure of spindle assembly observed in one-cell embryos after DLI-1 depletion [30]. Furthermore, the mitotic defects of dli-1(Δ414–443) embryos are not enhanced by a null allele of the dynein cofactor nud-2, which partially compromises dynein [36,38]. Thus, analysis in vivo and in vitro indicates that LIC-C helix 2, despite its high sequence conservation, makes a relatively modest contribution to dynein function compared to helix 1.
Together with the work of Lee and colleagues [13], our study establishes the molecular mechanism used by LIC to interact with structurally diverse cargo adaptors. An interesting open question is whether LIC-C could have additional binding partners besides adaptors. Two recent cryo-EM studies revealed that two dyneins can be recruited by a single dynactin-adaptor complex [10,11]. In one of the structures, an extra density, most likely corresponding to LIC-C of the first dynein, packs against the N-terminal coiled-coil of BICDR1 while simultaneously contacting one of the HCs of the second dynein [10]. It is tempting to speculate that this interaction between LIC-C and HC facilitates the incorporation of a second dynein into dynein-dynactin-adaptor assemblies.
The cDNAs for expression of human DYNC1LI1 (UniProt ID: Q9Y6G9; residues 388–523, 388–471, 402–471, 414–471, 424–471, 440–471, 440–455, 440–523, and 472–523) and human DYNC1LI2 (UniProt ID: O43237; residues 375–492, 375–450, and 451–492) were cloned into vector pGEX-6P-1 with a single N-terminal tryptophan and a C-terminal linker (GSGSG) followed by 6xHis. The cDNAs for human BICD2 (UniProt ID: Q8TD16; residues 2–422), RAB11FIP3 (UniProt ID: O75154; residues 2–756), HOOK3 (UniProt ID: Q86VS8; residues 2–239 and 2–552), NIN (UniProt ID: Q8N4C6; residues 1–693), SPDL1 (UniProt ID: Q96EA4; residues 2–359), and TRAK1 (UniProt ID: Q9UPV9; residues 103–167 and 103–187, with and without a C-terminal fusion to the GCN4 dimeric coiled-coil sequence VKQLEDKVEELLSKNAHLENEVARLKKLV [GCN4CC]) were cloned into a 2CT-derived vector containing an N-terminal 6xHis::MBP fusion followed by a linker with a Tobacco Etch Virus nuclear-inclusion-a endopeptidase (TEV protease) cleavage site and containing a C-terminal linker (GSGSGR) followed by the Strep-tag II. The cDNA of RILP (UniProt ID: Q96NA2; residues 1–401) was cloned into the pACEBac1 vector with a C-terminal linker (GSGSGR) followed by the Strep-tag II.
All bacterial expression vectors were transformed into the Escherichia coli strain BL21, except for the NIN and HOOK3 constructs, which were transformed into the E. coli strain Rosetta. Expression was induced with 0.1 mM IPTG at 18°C overnight at an OD600 of 0.9, and cells were harvested by centrifugation for 20 min at 4,000g.
For GST::LIC::6xHis constructs used in pull-down experiments, bacterial pellets were resuspended in lysis buffer A (50 mM HEPES, 250 mM NaCl, 0.1% [v/v] Tween 20, 10 mM EDTA, 10 mM EGTA, 5 mM DTT, 1 mM phenylmethanesulfonyl fluoride [PMSF], 2 mM benzamidine-HCl, 1 mg/mL lysozyme [pH 8.0]), disrupted by sonication, and cleared by centrifugation at 34,000g for 45 min. GST::LIC::6xHis was purified by tandem affinity chromatography using glutathione agarose resin (Thermo Fisher Scientific) followed by HisPur Ni-NTA resin (Thermo Fisher Scientific). Glutathione agarose resin was incubated in batch with the cleared lysate and then washed with wash buffer A (25 mM HEPES, 250 mM NaCl, 0.1% Tween 20, 1 mM DTT, 2 mM benzamidine-HCl [pH 8.0]), and proteins were eluted on a gravity column with elution buffer A (50 mM HEPES, 150 mM NaCl, 10 mM reduced L-glutathione, 1 mM DTT, 2 mM benzamidine-HCl [pH 8.0]). Fractions containing the recombinant proteins were pooled, incubated in batch with Ni-NTA resin, and washed with wash buffer B (25 mM HEPES, 250 mM NaCl, 25 mM imidazole, 0.1% Tween 20, 1 mM DTT, 2 mM benzamidine-HCl [pH 8.0]). Proteins were eluted on a gravity column with elution buffer B (50 mM HEPES, 150 mM NaCl, 250 mM imidazole, 1 mM DTT, 2 mM benzamidine-HCl [pH 8.0]). Fractions containing the proteins were pooled and dialyzed against storage buffer (25 mM HEPES, 150 mM NaCl [pH 7.5]) or further purified by size-exclusion chromatography using a Superose 6 10/300 column (GE Healthcare) equilibrated with storage buffer. Glycerol and DTT were added to final concentrations of 10% (v/v) and 1 mM, respectively, and aliquots were flash-frozen in liquid nitrogen and stored at −80°C.
Purification of LIC1::6xHis (residues 388–523, 388–471, and 472–523) for NMR spectroscopy, SPR, and MST experiments was carried out as described above with the following modifications: GST::LIC1::6xHis was captured using a GSTrap FF column (GE Healthcare) and eluted with elution buffer A. The GST moiety was cleaved off in solution with PreScission Protease, glutathione was removed by dialysis (50 mM HEPES, 150 mM NaCl [pH 8.0]), and the sample was applied again to a GSTrap FF column to remove GST and GST-tagged Prescission Protease. The flow-through containing LIC1::6His was subjected to size-exclusion chromatography using a Superdex 75 increase 10/300 GL column (GE Healthcare) in NMR buffer (50 mM sodium phosphate, 150 mM NaCl [pH 6.5]).
For purification of cargo adaptors, bacterial pellets were resuspended in lysis buffer B (50 mM HEPES, 250 mM NaCl, 10 mM imidazole, 0.1% Tween 20, 1 mM DTT, 1 mM PMSF, 2 mM benzamidine-HCl, 1 mg/mL lysozyme [pH 8.0]), disrupted by sonication, and cleared by centrifugation at 34,000g for 45 min. The 6xHis::MBP::adaptor::Strep-tag II proteins were purified by tandem affinity chromatography using HisPur Ni-NTA resin followed by Strep-Tactin Sepharose resin (IBA). HisPur Ni-NTA resin was incubated in batch with the cleared lysate and then washed with wash buffer B, and proteins were eluted on a gravity column with elution buffer B. Fractions containing the recombinant proteins were pooled, incubated overnight with TEV protease to cleave off the 6xHis::MBP moiety (except for TRAK1 fragments, which were not cleaved), incubated in batch with Strep-Tactin Sepharose resin, and washed with wash buffer A. Proteins were eluted on a gravity column with elution buffer E (100 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 2.5 mM desthiobiotin [IBA] [pH 8.0]). Fractions containing the proteins were pooled and dialyzed against storage buffer or further purified by size-exclusion chromatography using a Superose 6 10/300 column equilibrated with storage buffer. Glycerol and DTT were added to final concentrations of 10% and 1 mM, respectively, and aliquots were flash-frozen in liquid nitrogen and stored at −80°C.
Bacmid recombination and virus production were performed as described previously [54]. A 500-mL culture (SFM4 medium; Hyclone) of Sf21 cells (0.8 × 106 cells/mL) was infected with RILP::Strep-tag II-encoding virus. Cells were harvested by centrifugation at 800g for 5 min. Pellets were resuspended in lysis buffer C (50 mM HEPES, 250 mM NaCl, 1 mM DTT [pH 8.0]) supplemented with EDTA-free cOmplete Protease Inhibitor Cocktail (Roche), sonicated, and cleared by centrifugation at 34,000g for 45 min. RILP::Strep-tag II was purified by batch affinity chromatography using Strep-Tactin Sepharose. The resin was washed with wash buffer C (25 mM HEPES, 250 mM NaCl, 0.1% [v/v] Tween 20, 1 mM DTT [pH 8.0]), and the protein was eluted on a gravity column with elution buffer E. Fractions containing RILP::Strep-tag II were pooled and dialyzed against storage buffer. Glycerol and DTT were added to final concentrations of 10% and 1 mM, respectively, and aliquots were flash-frozen in liquid nitrogen and stored at −80°C.
Purified GST::LIC::6xHis constructs (50 pmol) were incubated with 250 pmol SPDL1(2–359)::Strep-tag II, 50 pmol BICD2(2–422)::Strep-tag II, 250 pmol NIN(1–693)::Strep-tag II, 50 pmol HOOK3(2–552)::Strep-tag II, 50 pmol RAB11FIP3(2–756)::Strep-tag II, 50 pmol RILP(1–401)::Strep-tag II, 250 pmol 6xHis::MBP::TRAK1(103–187)::Strep-tag II, 250 pmol 6xHis::MBP::TRAK1(103–167)::GCN4CC::Strep-tag II, or 250 pmol 6xHis::MBP::TRAK1(103–187)::GCN4CC::Strep-tag II for 1 h at 4°C in 150 μL pull-down buffer (50 mM HEPES, 100 mM NaCl, 5 mM DTT [pH 7.5]) containing 0.1% Tween 20 and supplemented with 15 μL of glutathione agarose resin. After washing the resin with 3 × 500 μL of the same buffer, proteins were eluted with pull-down buffer containing 15 mM reduced L-glutathione.
For immunoblots of purified proteins, samples were resolved by 10% SDS-PAGE and transferred to 0.2-μm nitrocellulose membranes (GE Healthcare). Membranes were blocked in PBS (4 mM KH2PO4, 16 mM Na2HPO4, 115 mM NaCl [pH 7.4]) containing 3% (w/v) BSA and 0.5% (v/v) Tween 20 and probed at 4°C overnight with mouse StrepMAB-Classic antibody (IBA) at 1 μg/mL in PBS containing 0.2% BSA and 0.1% Tween 20. Membranes were washed three times with PBS/0.1% Tween 20 (PBST), incubated with goat anti-mouse antibody coupled to HRP (Jackson ImmunoResearch, 1:10,000) for 1 h at room temperature, and washed again three times with PBST. Proteins were visualized by chemiluminescence using Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific) and X-ray film (Amersham, GE Healthcare).
For immunoblots of C. elegans, 100 adult hermaphrodites were collected into M9 buffer and processed for immunoblotting as described [37]. Samples were resolved on a gradient gel (4%–20%) and transferred to 0.2-μm nitrocellulose membranes. Membranes were blocked with 5% (w/v) nonfat dry milk in TBST (20 mM Tris, 140 mM NaCl, 0.1% Tween [pH 7.6]) and probed at 4°C overnight with rabbit anti-DHC-1 antibody GC4 (1:1,400, made in-house), mouse anti-FLAG M2 antibody (1:1,000, Sigma-Aldrich), or mouse anti-α-tubulin B512 antibody (1:5,000, Sigma-Aldrich). Membranes were sequentially rinsed 3× with TBST, 1× with 5% nonfat dry milk in TBST, and 3× with TBST. Membranes were incubated with goat secondary antibody coupled to HRP (Jackson ImmunoResearch, 1:10,000) for 1 h at room temperature and washed again 3× with TBST, 1× with 5% nonfat dry milk in TBST, and 3× with TBST. Proteins were visualized as described above.
For backbone resonance assignment of LIC1(388–523)::6xHis, 15N-1H HSQC, HNCACB, CBCA(co)NH, HNCO, HN(ca)CO, and HNN spectra [55] were recorded on a triple-resonance Varian 900 NMR cryoprobe spectrometer at 25°C using a 13C/15N-labeled sample in 50 mM sodium phosphate and 150 mM NaCl at pH 6.5 and 375 μM protein concentration in a standard 5-mm Shigemi tube. The 3D spectra were acquired with a nonuniform sampling (NUS) scheme generated by the NUS@HMS scheme generator software [56] with 1,024 complex data points in the direct dimension and 30% sampling of the original 96 and 80 points in the indirect 13C and 15N dimension, respectively. The spectral widths were 14,045 Hz (1H), 3,200 Hz (15N), 3,770 Hz (C = O), and 15,835 Hz (Cα/Cβ); the interscan delay was 1.7 s; and the number of scans was 16 for all experiments. The NUS-acquired data were reconstructed using the hmsIST software [56]. Zero-filling was achieved by addition of 256 points in both indirect dimensions. A solvent subtraction function was applied in the direct dimension. Further data processing and visualization were performed using NMRPipe/NMRDraw [57] and NMRFAM Sparky [58]. Resonance assignment was performed using CCPNmr Analysis software [59].
Because of high sequence redundancy and extensive peak overlap, we used a “divide-and-conquer” approach for chemical shift assignment. We measured and overlaid 15N-1H HSQC spectra for two smaller LIC1 constructs, 388–471 and 472–523, with 388–523 to facilitate the verification of assignment. The 15N-1H HSQC spectra of the small constructs were measured with 128 scans and 128 complex points in the indirect dimension.
We assessed residual secondary structure using the SSP score program developed by Forman-Kay and colleagues with the re-referencing algorithm for 13CA and 13CB shifts [24]. The method combines different chemical shifts into a single residue–specific SSP score. Our input shifts were those of 1HN, 15N, 13CO, 13CA, and 13CB.
The 15N R1 and R1ρ relaxation rate constants and 15N-1H heteronuclear NOEs of LIC1(388–523)::6xHis were measured on a Bruker 700 MHz spectrometer equipped with a triple-resonance cryoprobe at 25°C using a 15N-labeled sample at 400 μM protein concentration in 50 mM sodium phosphate buffer, 50 mM NaCl, 0.05% (w/v) NaN3, and 5 mM DTT at pH 6.5. For the R1 and R1ρ experiments, the sampling time points were 40, 88, 136, 192, 288, 392, 592, 688, 792, and 992 ms and 30, 60, 120, 150, 180, and 210 ms, respectively. During the R1ρ relaxation time, a 15N spin-lock field of 1,433-Hz strength was applied. R2 was calculated from R1 and R1ρ using the following equation: R2 = R1ρ + (R1ρ–R1)*tg2(θ), where θ = tan-1(2πΔν/γNB1), Δν is the resonance offset, |γNB1|/2π is the strength of the spin-lock field B1, and γN the gyromagnetic ratio of the 15N spin. The 15N-1H heteronuclear NOEs were determined from two spectra recorded in presence and in the absence of 1H saturation in an interleaved manner.
For NMR titration analysis, 15N-1H HSQC spectra with 128 complex points in the indirect dimension and 128 scans were recorded on a triple-resonance Varian 900 NMR cold-probe spectrometer at 25°C of 40 μM samples of 15N-labeled LIC1(388–523)::6xHis alone and in the presence of 0.5 and 1 equivalent of unlabeled SPDL1(2–359)::Strep-tag II, BICD2(2–422)::Strep-tag II, or HOOK3(2–239)::Strep-tag II in 50 mM sodium phosphate and 150 mM NaCl at pH 6.5. We used freshly prepared samples for each titration step to limit confusion with potential degradation peaks. Data processing and visualization were performed using NMRPipe/NMRDraw [57] and NMRFAM Sparky [58].
The 1H, 13C, and 15N chemical shift assignments have been deposited in the BioMagResBank database (http://www.bmrb.wisc.edu) under the accession number 27401.
SPR analysis with SPDL1, BICD2, and HOOK3 fragments was conducted with a Biacore 3000 system. His-tagged LIC1(388–523) and LIC1(388–471) constructs (ligand) were immobilized on three different flow cells (FC1–3) on an NTA sensor chip at densities of 200–300 RU, whereas the fourth flow cell (FC4) was spared for blank sensogram measurement. Concentrated stocks of Strep-tagged SPDL1(2–359), BICD2(2–422), and HOOK3(2–239) (analyte) were dialyzed exhaustively against HBS-P flow buffer (10 mM HEPES, 150 mM NaCl, 0.005% [v/v] surfactant P20 [pH 7.4]). The background-blank sensogram was subtracted from sensograms measured with immobilized ligands. Injections for each analyte concentration were performed in triplicate. Data processing was done on Biacore evaluation software.
SPR analysis with RILP::Strep-tag II was performed using a Biacore X100 system equipped with a CM5 sensor chip (GE Healthcare). Anti-GST antibody was immobilized using amine-coupling chemistry using the Amine Coupling Kit (GE Healthcare) and the GST Capture Kit (GE Healthcare) according to manufacturer's instructions. The surfaces of flow cells 1 and 2 were activated for 7 min with a 1:1 mixture of 0.1 M N-hydroxysuccinimide and 0.4 M 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide at a flow rate of 5 μL/min. Anti-GST antibody at a concentration of 30 μg/mL in 10 mM sodium acetate (pH 5.0) was immobilized at a density of 7,500 RU on flow cells 1 and 2. Surfaces were blocked with a 7-min injection of 1 M ethanolamine (pH 8.0). Anti-GST antibody high-affinity sites were blocked with 3 cycles of a 3-min injection of recombinant GST at 5 μg/mL in running buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.05% surfactant P20 [pH 7.4]) followed by a 2-min injection of regeneration solution (10 mM glycine-HCl [pH 2.1]). To collect kinetic binding data, at the beginning of each cycle, GST::LIC1(388–523)::6xHis or GST::LIC1(440–455)::6His in running buffer (ligand) was injected over flow cell 2 at a density of 720–530 and 550–460 RU, respectively. Flow cell 1 was injected with GST::6xHis at a density of 575–430 RU to serve as a reference surface. RILP::Strep-tag II (analyte) was injected in running buffer over the two flow cells at concentrations of 50, 11, 4.5, 1.8, and 0.3 μM at a flow rate of 30 μL/min and a temperature of 25°C. The complex was allowed to associate and dissociate for 120 and 600 s, respectively. Surfaces were regenerated with two 2-min injections of regeneration solution. Three independent runs were performed for each condition. The data were fit to a 1:1 interaction model using the evaluation module of the Biacore X100 software, version 2.0.1 (GE Healthcare).
Measurements were carried out on a NanoTemper Monolith NT.115 pico instrument (NanoTemper Technologies) at 25°C using medium power and 20% excitation power (auto-detect-pico-red). LIC1(388–523)::6xHis was fluorescently labeled by interacting 100 μL protein solution (200 nM concentration) with 100 μL Red-Tris NTA dye (100 nM concentration). The reaction mixture was incubated at room temperature for 30 min followed by centrifugation for 10 min at 4°C with 15,000g. Then, 10 nM of labeled LIC1(388–523)::6xHis and 16 two-fold dilution series of the adaptor were loaded into 16 standard capillaries (NanoTemper Technologies) (SPDL1[2–359]::Strep-tag II highest concentration 42.5 μM; BICD2[2–422]::Strep-tag II highest concentration 36.5 μM; HOOK3[2–239]::Strep-tag II highest concentration 70 μM). We observed sigmoidal behavior of the fluorescence level over time, which allowed characterization of the interactions. Raw data were analyzed using the NanoTemper software (MO affinity analysis v 2.2.7). The signal-to-noise ratios for SPDL1(2–359)::Strep-tag II, BICD2(2–422)::Strep-tag II, and HOOK3(2–239)::Strep-tag II were 18.8, 22.1, and 7.1, respectively.
CD spectra of LIC1(388–523)::6xHis were collected on a J-815 CD spectrometer (Jasco) with a wavelength range from 190 to 250 nm, data pitch 1 nm, standard sensitivity, 1 nm bandwidth and 20 nm/min scanning speed, at temperatures of 25, 37, 50, 60, and 70°C. The baseline of the spectrum was obtained from measurement of the buffer (50 mM sodium phosphate, 150 mM NaCl [pH 6.5]) and subtracted from the spectra of the samples to remove artificial CD signals that might originate from the optical system. Data measurement and analysis were performed using Spectra Manager version 2 (Jasco). Molar ellipticity was calculated as m°M/(10LC), where m° is the degree value (mdeg), M is the molecular weight in g/mole, L is the path length in cm, and C is the concentration in g/L.
Worm strains (S1 Table) were maintained at 20°C on standard nematode growth media (NGM) plates seeded with OP50 bacteria. A Mos1 transposon-based strategy (MosSCI) was used to generate strains stably expressing mKate2::RAB-5, SNB-1::mKate2, and TOMM-20(1–54)::mKate2 in touch receptor neurons [33]. Transgenes were cloned into pCFJ151 for insertion on Chromosome 2 (ttTi5605 locus), and transgene integration was confirmed by PCR. C-terminal mutants of dli-1 (Δ369–443, F392A/F393A, L396A/L397A, and Δ414–443) and 3xflag::dli-1 were generated by CRISPR/Cas9-mediated genome editing, as described previously [60,61]. Genomic sequences targeted by sgRNAs are listed in S2 Table. Modifications in genomic DNA sequence were confirmed by sequencing, and strains were outcrossed 6 times against the wild-type N2 strain to remove potential background mutations. Other fluorescent markers were subsequently introduced by mating. The dli-1 mutants Δ369–443, F392A/F393A, and L396A/L397A were maintained as heterozygotes using the GFP-marked genetic balancer nT1 [qIs51]. Homozygous F1 progeny from balanced heterozygous mothers were identified by the absence of GFP fluorescence.
Animals were collected at the L4 stage (day 0) and transferred every 2 d to a new NGM plate with bacteria. Animals were scored as alive or dead every 1–3 d. Animals were considered dead if they did not respond when touched with a platinum wire and if there was no evidence of pharyngeal pumping. Animals that were found dead on the edge of the plate, escaped, or died because of internal hatching of progeny were excluded from the assay.
Image analysis was performed using Fiji software (Image J version 1.52d).
Statistical analysis was performed with GraphPad Prism 7.0 software. Values in figures and text are reported as mean ± SEM. The type of statistical analysis performed is described in each figure legend. Differences were considered significant at P < 0.05.
|
10.1371/journal.pgen.1005291 | mTOR Directs Breast Morphogenesis through the PKC-alpha-Rac1 Signaling Axis | Akt phosphorylation is a major driver of cell survival, motility, and proliferation in development and disease, causing increased interest in upstream regulators of Akt like mTOR complex 2 (mTORC2). We used genetic disruption of Rictor to impair mTORC2 activity in mouse mammary epithelia, which decreased Akt phosphorylation, ductal length, secondary branching, cell motility, and cell survival. These effects were recapitulated with a pharmacological dual inhibitor of mTORC1/mTORC2, but not upon genetic disruption of mTORC1 function via Raptor deletion. Surprisingly, Akt re-activation was not sufficient to rescue cell survival or invasion, and modestly increased branching of mTORC2-impaired mammary epithelial cells (MECs) in culture and in vivo. However, another mTORC2 substrate, protein kinase C (PKC)-alpha, fully rescued mTORC2-impaired MEC branching, invasion, and survival, as well as branching morphogenesis in vivo. PKC-alpha-mediated signaling through the small GTPase Rac1 was necessary for mTORC2-dependent mammary epithelial development during puberty, revealing a novel role for Rictor/mTORC2 in MEC survival and motility during branching morphogenesis through a PKC-alpha/Rac1-dependent mechanism.
| The protein kinase mTOR is frequently activated in breast cancers, where it enhances cancer cell growth, survival, and metastastic spread to distant organs. Thus, mTOR is an attractive, clinically relevant molecular target for drugs designed to treat metastatic breast cancers. However, mTOR exists in two distinct complexes, mTORC1 and mTORC2, and the relative roles of each complex have not been elucidated. Moreover, as pathways that regulate normal tissue growth and development are often highjacked to promote cancer, understanding mTOR function in normal mammary epithelial development will likely provide insight into its role in tumor progression. In this study, we assessed the role of mTORC1 and mTORC2 complexes in normal mammary epithelial cell branching, survival, and invasion. Interestingly, while mTORC1 was not required for branching, survival and invasion of mammary epithelial cells, mTORC2 was necessary for these processes in both mouse and human models. Furthermore, we found that mTORC2 exerts its effects primarily through downstream activation of a PKC-alpha-Rac1 signaling axis rather than the more well-studied Akt signaling pathway. Our studies identify a novel role for the mTORC2 complex in mammary morphogenesis, including cell survival and motility, which are relevant to breast cancer progression.
| Post-natal mammary epithelial morphogenesis is a complex process during which an extensively branched ductal network develops from a rudimentary epithelial bud [1]. Branching morphogenesis is most active during puberty and is regulated by endocrine hormones and local paracrine interactions with mesenchymal stroma [2]. In response to hormonal and growth factor cues, mammary epithelial cells (MECs) within the terminal end buds (TEBs), the club-shapes structures at the distal epithelial tips [1, 2], proliferate and collectively invade surrounding stroma. Differentiation of epithelial progenitors in the TEB populates the ducts with mature luminal MECs, and apoptosis canalizes the lumen. TEB bifurcation results from mechanical restraints at the TEB midline, forming new primary ducts. Side-branches sprout laterally from the trailing ducts as proliferative out-pouchings. Primary and side branching reiterates, filling the entire mammary fat pad [1, 2]. The dynamic processes that occur during puberty in the mammary epithelium are carefully coordinated by many molecular signaling pathways.
The intracellular serine/threonine kinase mammalian target of rapamycin (mTOR) regulates cellular metabolism, protein and lipid synthesis, cell survival, and cytoskeletal organization, processes that are required for proper mammary morphogenesis. mTOR regulates these processes through phosphorylation of its target substrates, including translation initiation factor 4E (eIF4E)-binding protein 1 (4E-BP1), p70S6 kinase (S6K), Akt, SGK1, and protein kinase C-alpha (PKC-alpha) [3]. A complex of associated protein co-factors regulates mTOR substrate specificity. As such, mTOR functions in two distinct complexes, each defined by the specific co-factors in complex with mTOR kinase and by their relative sensitivity to rapamycin. The rapamycin-sensitive mTOR complex (mTORC)-1 requires the co-factor regulatory-associated protein of mammalian target of rapamycin (Raptor), whereas mTORC2 requires the co-factor rapamycin-insensitive companion of mammalian target of rapamycin (Rictor). Although mTORC2 is relatively insensitive to acute rapamycin treatment, more recent studies determined that prolonged rapamycin treatment can inhibit mTORC2 complex assembly [4–7]. The intracellular serine/threonine kinase Akt is phosphorylated at S473 directly by mTORC2 and is key effector for many of the biological effects initiated by mTORC2. Akt is also linked to activation of mTORC1 downstream of PI3-kinase, making Akt a point of intersection between mTORC1, mTORC2, and their associated effectors [3].
Though mTOR regulates MEC growth in cell lines [8, 9] and milk protein expression [8, 10–12], mTOR-mediated regulation of mammary ductal morphogenesis remains under-investigated. The signaling complexity of mTOR, its pleiotropic functions, and a lack of mTORC2-specific inhibitors present a challenge to dissecting the relative roles of mTORC1 and mTORC2 in mammary development. Given the importance of mTOR in breast cancer progression and treatment, an understanding of mTORC1 and mTORC2 in untransformed MECs is needed. We assessed the impact of tissue-specific Rictor and Raptor ablation on mammary morphogenesis. Rictor loss impaired mTORC2 activity, reduced ductal lengthening and secondary branching, and reduced MEC proliferation and survival in vivo and ex vivo. Surprisingly, genetic disruption of mTORC1 via Raptor ablation resulted in distinct and milder effects on the developing mammary ductal epithelium, revealing non-overlapping roles for mTORC1 and mTORC2 during mammary morphogenesis. Interestingly, we found that mTORC2 controls mammary morphogenesis through downstream effectors PKC-alpha and Rac1, but not Akt.
To assess the role of Rictor/mTORC2 during mammary morphogenesis in the context of the native mammary microenvironment, we bred MMTV-Cre mice [13] to RictorFL/FL mice [14], allowing mammary-specific Cre recombinase to disrupt Rictor expression at floxed (FL) Rictor alleles. Immunohistochemistry (IHC) for Rictor revealed expression in luminal and myoepithelial MECs in Rictor+/+MMTV-Cre (RictorWT) mice (Fig 1A–upper panel). Rictor expression was not seen in RictorFL/FLMMTV-Cre (RictorMGKO) luminal MECs, and was slightly reduced in the myoepithelium, consistent with luminal but not myoepithelial Cre expression in MMTV-Cre mice. Akt phosphorylation at S473, the mTORC2 phosphorylation site, was decreased in MECs of RictorMGKO mice versus RictorWT, confirming decreased mTORC2 signaling upon Rictor ablation (Fig 1A–lower panel). Immunofluorescent (IF) staining for cytokeratin (CK)-8 and CK14, molecular markers of luminal and myoepithelial MECs, respectively, confirmed that Rictor loss did not affect the relative spatial organization of luminal and myoepithelial MECs (Fig 1B–upper panel), but revealed the presence of apically mis-localized nuclei in RictorMGKO MECs (yellow arrows), versus basally located nuclei and an organized, smooth apical border in RictorWT samples (white arrows). IF for the tight junction (TJ) protein Zona Occludens-1 (ZO-1) revealed apical ZO-1 localization in RictorWT samples. However, ZO-1 was aberrantly localized along baso-lateral membranes in RictorMGKO MECs (Fig 1B–lower panel). In contrast, the baso-lateral localization of the adherens junction (AJ) protein p120 was relatively unaltered by Rictor loss. These results suggest that Rictor loss disrupts the proper apical distribution of ZO-1 in MECs. The apically mis-localized nuclei apparent in histological mammary sections from 6-week old RictorMGKO female mice contributed to an irregular apical border (Fig 1C, black arrows). Additional structural alterations were seen in TEBs, including sloughing of body cells (the multi-layered TEB population comprised of mature and progenitor luminal MECs; Fig 1C–lower panel, arrow) within TEB lumens, and stromal thickening at the neck between maturing ducts and TEBs (Fig 1C–lower panel, *). Morphological alterations were seen throughout whole mounted, hematoxylin-stained RictorMGKO mammary glands (Figs 1D, arrows, and S1A). Because mammary ducts lengthen distally at a predictable rate during puberty, we measured ductal length in mammary glands from 6 week- (mid-puberty) and 10 week-old (late puberty) mice. Ductal length was significantly reduced in RictorMGKO mammary glands at both time points (Fig 1E–left panel, and S1B Fig). Primary (Y-shaped) and side (T-shaped) branches were counted in each mammary gland, revealing a significant reduction in T-shaped side branches at 6 and 10 weeks of age in RictorFL/FLMMTV-Cre samples as compared to RictorWT (Fig 1E–right panel).
IHC analysis of Ki67 in both ducts and TEBs was used as a relative measure of cellular proliferation in the mammary epithelium (Figs 1F—upper panel, and S1C–upper panel), revealing decreased Ki67+ nuclei in RictorMGKO samples as compared to RictorWT at 6 weeks of age but not at 10 weeks (Fig 1G—left panel). Cell death in ductal MECs or TEBs, measured using TUNEL analysis (Figs 1F—lower panel, and S1C–lower panel), demonstrated a remarkable increase in TUNEL+ MECs in RictorMGKO samples at 6 and 10 weeks of age (Fig 1G—right panel). These results demonstrate that Rictor loss impairs mTORC2 activity, P-Akt, MEC growth, and MEC survival during mammary morphogenesis.
Western analysis of whole mammary lysates harvested from 10-week old female mice confirmed decreased P-Akt S473 in RictorMGKO mammary glands, and revealed increased phosphorylation of the mTORC1 effector ribosomal protein S6 ([15]; Fig 2A) confirming that Rictor loss decreases mTORC2 activity, but not mTORC1. To dissect more precisely how Rictor signaling affects mammary morphogenesis, we used primary mammary epithelial cells (PMECs) and primary mammary organoids (PMO’s) harvested from RictorFL/FL mice. Adenoviral infection of RictorFL/FL PMECs with Ad.Cre significantly reduced Rictor and P-Akt S473 levels relative to cells infected with control Ad.LacZ, and increased P-S6 levels (Fig 2B), similar to the impact of Rictor ablation in vivo. Consistent with structural alterations were seen in our RictorMGKO model in vivo (e.g. sloughing of body cells in TEBs, irregular ductal tracts, multiple cell layers), confocal analysis of Rictor-deficient PMOs stained for E-cadherin revealed multiple cell layers in acinar structures and poor lumen formation relative to control PMOs infected with Ad.LacZ, which formed a well-defined lumen surrounded by a single layer of epithelial cells (S1D Fig). Rictor loss did not significantly impact PMEC proliferation, as measured by bromodeoxyuridine (BrdU) incorporation into genomic DNA (Fig 2C–left panel). However, the percentage of TUNEL+ PMEC nuclei was increased >2-fold following Ad.Cre infection (Fig 2C–right panel), consistent with increased cell death in Rictor-null MECs in vivo. Similar results were seen using MCF10A immortalized human MECs, in which Rictor gene targeting with Rictor-specific zinc finger nucleases (ZFNs) genetically impaired Rictor expression and decreased P-Akt S473 (Fig 2D), thus validating our findings in a human MEC model. Increased cell death was also seen in MCF10A-RictorZFN cells as compared to parental MCF10A cells, as shown by AnnexinV-FITC binding (Fig 2E). Therefore, Rictor is necessary for mTORC2 signaling and cell survival in human and mouse MECs.
We cultured adenovirus transduced RictorFL/FL mammary organoids in three-dimensional (3D) Matrigel to assess collective epithelial morphogenesis (Fig 2F). Mammary organoids accurately model epithelial autonomous molecular events of mammary morphogenesis in a stroma-free environment that preserves the native relationship between luminal and myoepithelial MECs and permits cell-cell and cell-matrix interactions in three dimensions [16]. GFP fluorescence in organoids infected with Ad.GFP or Ad.Cre-IRES-GFP confirmed efficient infection in basal and luminal cells of organoids (S2A and S2B Fig). IF staining for pan-cytokeratin confirmed that organoids were epithelial-derived (S2C Fig). Ad.Cre infection of RictorFL/FL PMECs substantially reduced organoid size and branching (Fig 2F and 2G) and reduced Rictor expression levels (Fig 2H). In contrast, Ad.Cre infection of RictorFL/+ PMECs only modestly reduced Rictor expression levels (Fig 2H) and did not significantly affect organoid size or the number of branches formed in RictorFL/+ organoids (Fig 2F and 2G). These data suggest that Rictor is necessary for multicellular morphogenesis of the mammary epithelium, faithfully recapitulating ex vivo the consequences of Rictor ablation that are seen in vivo and demonstrating the utility of this model to examine branching mammary gland morphogenesis.
Previous studies demonstrated that Rictor knock-down reduces migration of breast cancer cell lines [17–19]. We therefore assessed PMEC invasion and motility through Matrigel-coated transwell filters upon Rictor ablation ex vivo. Fewer RictorFL/FL PMECs invaded through Matrigel when infected with Ad.Cre, as compared to RictorFL/FL PMECs infected with control Ad.LacZ (Fig 2I). Similarly, invasion through Matrigel-coated transwells was profoundly reduced in MCF10A-RictorZFN cells as compared to parental MCF10A cells (Fig 2J). Under these conditions, there were a similar number of viable cells remaining in the upper transwell chamber after 24 hours of culture of both MCF10A and MCF10A-RictorZFN cells (S2D Fig), suggesting that cell death may not be the primary reason underlying the reduced ability of MCF10A-RictorZFN cells to migrate/invade in these transwell assays, but rather that cell invasion, per se, is decreased in the absence of Rictor. Collectively, these data demonstrate that Rictor promotes MEC invasion and migration, two processes necessary for mammary ductal lengthening and branching.
Because Rictor loss reduced P-Akt S473, we tested the hypothesis that Akt phosphorylation by Rictor-regulated mTOR complex 2 is necessary for survival and morphogenesis of MECs. Adenoviral expression of myristoylated Akt1 (Ad.Aktmyr) was used to express a membrane-localized (and thus, constitutively active) variant of Akt1. Indeed, expression of this Akt variant in mammary epithelium delays involution and the onset of apoptosis in vivo [20]. Additionally, we repeated experiments using an alternative adenoviral, constitutively active Akt construct, Ad.AktDD. Ad.Aktmyr or Ad.AktDD restored P-Akt S473 in Ad.Cre-infected RictorFL/FL PMECs (Figs 3A and S3A). Surprisingly, RictorFL/FL organoids infected with Ad.Cre + Ad.Aktmyr or Ad.AktDD were morphologically similar to and harbored little to no statistically significant difference in the numbers of branches compared to those infected with Ad.Cre alone (Figs 3B and S3B). Further, size of Rictor-deficient organoids was not fully rescued by expression of Ad.Aktmyr or Ad.AktDD (Figs 3C and S3C). We found that blockade of Akt using the allosteric Akt inhibitor 5J8 blocked Akt phosphorylation at S473 (Fig 3D), reduced the number of branches per organoids, and reduced organoid size by nearly 50% (Fig 3E). These data suggest that while Akt is necessary for mammary branching and growth, restoring Akt function is not sufficient to completely rescue defects caused by loss of Rictor/mTORC2 function. Indeed, expression of Ad.Aktmyr did not reduce the number of Rictor-null PMECs undergoing cell death (Fig 3F), nor did it increase the number Rictor-null PMECs invading through Matrigel-coated transwells (Fig 3G). Taken together, these observations suggest Rictor is necessary for Akt phosphorylation in MECs, but that Akt is not the primary effector of mTORC2 that regulates MEC survival, invasion, and side branching. Thus, while Akt is necessary for proper mammary epithelial morphogenesis, it is not sufficient to compensate for loss of Rictor/mTORC2 function.
Previous studies showed that mTORC2 phosphorylates PKC-alpha [21] Consistent with these findings, Rictor loss reduced P-PKC-alpha in PMECs, as well as total PKC-alpha (Fig 4A). We also observed decreased P-PKC-alpha by IF in mammary gland sections from 6 week old RictorMGKO mice, as compared to RictorWT controls (S4A Fig). Adenoviral PKCα expression rescued P-PKC-alpha in Rictor-null PMECs (Fig 4B), rescued branching morphogenesis in Rictor-null organoids (Fig 4C) and increased Rictor-null organoid size (Fig 4D). Similar to what was seen in mouse PMECs, P-PKC-alpha and total PKC-alpha were diminished in MCF10A-RictorZFN cells relative to parental MCF10A (Fig 4E). Restoration of PKC-alpha by adenoviral transduction increased P-PKC-alpha in both parental MCF10A and MCF10A-RictorZFN cells (Fig 4F).
Rac1, a small GTPase involved in actin cytoskeletal dynamics, is necessary for migration of many breast cancer cell lines, regulates apical polarity in MECs, and is a downstream effector of mTORC2 signaling. Importantly, Rac1 is also a known effector of PKC-alpha in MECs [22–24], but the linear relationship between Rictor, PKC-apha, and Rac1 in MECs is currently unknown. We examined Rac1 activation in MCF10A cells using agarose beads conjugated to recombinant p21-activated kinase binding domain (PBD), which specifically binds to active GTP-bound Rac. Western analysis to detect Rac1 in PBD pull-downs revealed decreased Rac-GTP in MCF10A-RictorZFN cells as compared to parental MCF10A (Fig 4G). However, Ad.PKC-alpha increased Rac-GTP in Rictor-null cells, confirming that PKC-alpha activates Rac downstream of Rictor. Additionally, Ad.PKC-alpha increased invasion of MCF10A-RictorZFN cells through Matrigel-coated transwells (Fig 4H), and significantly reduced apoptosis in MCF10A-RictorZFN, as measured by Annexin V-FITC staining (Fig 4I). A pharmacological PKC-alpha inhibitor profoundly decreased invasion of parental MCF10A cells through Matrigel-coated transwells (Fig 4J), providing validation that PKC-alpha is necessary for MEC motility. These data suggest Rictor-mediated PKC-alpha signaling in MECs controls Rac1 activation, branching morphogenesis, cell survival and motility.
To confirm the role of Rictor in Rac1 activation in vivo, we examined mammary epithelium in situ for GTP-bound Rac1 using a glutathione-S-transferase (GST)-PBD fusion protein as a probe for Rac-GTP. IF detection of GST-PBD binding was decreased in RictorMGKO mammary glands compared to RictorWT (Figs 5A, 5B, and S4B). Importantly, IF detection of GST-PBD binding in WT PMECs was abolished by a pharmacological Rac1 inhibitor (Figs 5C and S4C), confirming the specificity of the assay for detection of Rac1-GTP. In contrast to the abundant Rac1-GTP detected in WT PMECs, RictorFL/FL PMECs infected with Ad.Cre displayed a 10-fold decrease in GST-PBD binding to Rac-GTP relative to Ad.LacZ infected controls (Figs 5C and S4C). Phalloidin staining revealed cortical actin overlapping with GST-PBD binding in Ad.LacZ-infected RictorFL/FL PMECs (S4D Fig). However, Ad.Cre-infected RictorFL/FL PMECs showed increased formation of actin stress fibers, bearing no overlap with GST-PBD. Constitutively active Rac1 (Ad.caRac1) expression (Fig 5D) restored GST-PBD binding in Rictor-null PMECs (Fig 5E), suggesting that Rictor is necessary for Rac1-GTP in PMECs.
Ad.caRac1 was used to determine if restoration of Rac1-GTP could rescue invasion in Rictor-null MECs. Ad.caRac1 increased invasion 2.5-fold over Ad.LacZ in Rictor-null PMECs (Fig 5F). P-Akt S473 was unaffected by caRac1 (Fig 5D), suggesting that while Akt and Rac1 are both effectors of Rictor-dependent signaling, they exist in two separable pathways in MECs. Despite having no impact on P-Akt, Ad.caRac1 decreased cell death in Rictor-null PMECs (Figs 5G and S4E), suggesting that Rictor-dependent Rac1-GTP is necessary for PMEC survival. Ad.caRac1 also rescued branching morphogenesis of Rictor-deficient organoids (Fig 5H). Conversely, Rac inhibition using a pharmacologic Rac1 inhibitor decreased organoid size and branching in WT organoids (Fig 5I), confirming that Rac1 is necessary for mammary epithelial branching morphogenesis and appears to function downstream of mTORC2. Thus, Rictor is required for Rac1-GTP signaling, and restoration of Rac1 activity rescued branching morphogenesis and survival of Rictor-deficient PMECs. To determine if Rictor/mTORC2-mediated branching morphogenesis and ductal outgrowth are dependent on PKC-alpha/Rac versus Akt in the context of the native mammary gland environment in vivo, we transduced PMEC from RictorFL/FL mice with control Ad.GFP versus Ad.Cre in the presence or absence of Ad.PKC-alpha, Ad.caRac, or Ad.Aktmyr and transplanted them into the cleared inguinal mammary fat pads of 4 week old recipient female mice. We harvested mammary glands from these animals 6 weeks post-transplantation and assessed epithelial architecture and branching morphogenesis in whole-mount preparations. Consistent with data from MMTV-Cre/RictorFL/FL mice, transplanted Rictor-deficient MEC produced structures characterized by shortened ductal outgrowths with fewer branches relative to GFP controls (Fig 5J and 5K). Restored PKC-alpha or Rac activity (Fig 5J and 5K) rescued these defects and produced epithelial outgrowth that resembled endogenous epithelium in contralateral controls (S4F Fig). Consistent with our ex vivo organoid culture analyses, restored Akt activity was unable to fully rescue defects produced by loss of Rictor (Fig 5J and 5K). These data suggest that Rictor/mTORC2-dependent mammary epithelial morphogenesis relies primarily upon downstream activation of PKC-alpha and Rac-GTPase.
Rapamycin is a pharmacologic inhibitor of mTOR originally thought to preferentially inhibit mTORC1 over mTORC2. However, sustained rapamycin treatment impairs both mTORC1 and mTORC2 in a cell type-dependent manner [25–28]. Consistent with this idea, acute rapamycin treatment for 1 hour (1 h) decreased P-S6 (an mTORC1 effector) but not P-Akt (an mTORC2 effector), whereas sustained rapamycin treatment (24 h) decreased both P-S6 and P-Akt S473 (Fig 6A). Rapamycin treatment for 10 days significantly decreased branching morphogenesis and organoid size in WT organoids (Fig 6B). Although PMEC survival was not affected by acute rapamycin treatment, cell death increased after 24 h with rapamycin treatment (Fig 6C). Proliferation of WT PMECs, as measured by BrdU incorporation, was unaffected by acute (30 min) or sustained (24 h) pre-treatment with rapamycin (Fig 6D). The effects of sustained rapamycin treatment, including reduced MEC survival, branching morphogenesis formation, and diminished organoid size were similar to the effects achieved by Rictor ablation in MECs. Also similar to what was seen with Rictor-deficient MECs, the phenotypic effects of rapamycin treatment were rescued by Ad.PKC-alpha (Fig 6E) and Ad.caRac1 (Fig 6F), including rescue of branching morphogenesis and colony size. Ad.caRac1 also rescued rapamycin-mediated inhibition of cell motility (Figs 6G and S5). Because the mTOR inhibitor rapamycin impairs mTORC1 and mTORC2, and recapitulates the morphological and molecular effects of Rictor ablation in MECs, these results suggest that Rictor is acting in complex with mTOR to regulate MEC survival, motility, and branching morphogenesis, supporting a role for mTORC2 in the developing mammary gland. However, these findings do not rule out the contribution of mTORC1 to mTOR-mediated mammary morphogenesis.
To understand how mTORC1 participates in mammary morphogenesis, we infected PMECs harvested from female RaptorFL/FL mice [29] with Ad.Cre. Western analysis confirmed loss of Raptor and decreased P-S6 in serum-deprived cells (Fig 7A). However, P-Akt S473 was unaffected by Raptor ablation, confirming that genetic ablation of Raptor causes selective inhibition of mTORC1, while Rictor ablation inhibits mTORC2. RaptorFL/FL mammary organoids infected with Ad.LacZ formed multi-branched colonies, as expected (Fig 7B). Surprisingly, infection with Ad.Cre did not affect branching morphogenesis in RaptorFL/FL organoids, the number of branches per organoid, or colony size (Fig 7B). Additionally, Raptor ablation had no significant impact on PMEC migration in wound healing assays (Fig 7C).
RaptorFL/FLMMTV-Cre (RaptorMGKO) mice were used to assess the impact of Raptor ablation on mammary morphogenesis in vivo. IHC detected Raptor and the mTORC1 effector P-S6 in RaptorWT mammary glands at 10 weeks of age but did not detect P-S6 in age-matched RaptorMGKO mice (Fig 7D). Western analysis of whole mammary lysates from 10 week-old mice confirmed loss of Raptor (S6A Fig). Immunofluorescent (IF) staining for cytokeratin (CK)-8 and CK14, molecular markers of luminal and myoepithelial MECs, respectively, confirmed that Raptor loss did not affect the relative spatial organization of luminal and myoepithelial MECs (S6B Fig–upper panel). Additionally, no alterations in localization or staining pattern of ZO-1 were observed (S6B Fig–lower panel). Proliferation, as measured by IHC for Ki67, was significantly decreased in RaptorMGKO ducts in 6-week old mice, but not in TEBs (S6C Fig). By 10 weeks, however, proliferation in ducts had recovered to levels seen in RaptorWT (Fig 7D). TUNEL analysis revealed similar ratios of TUNEL+ MECs in RaptorMGKO and RaptorWT samples harvested from 6 and 10 week old animals (Fig 7D). Consistent with these observations, only mild defects in side branching and ductal length were found in mammary glands from 6-week old RaptorMGKO mice (Fig 7E), and these were resolved by 10 weeks of age.
Taken together, these results demonstrate that mTOR uses Rictor to activate PKCα/Rac1-dependent survival, motility, and branching morphogenesis in the mammary epithelium and that Rictor does not rely fully on Akt signaling to promote ductal morphogenesis in the breast.
Postnatal mammary epithelial morphogenesis requires precise coordination of cell proliferation, apoptosis, differentiation, and motility in order to turn rudimentary epithelial buds into an organized, branched ductal network permeating the entire mammary fat pad by the end of puberty [1, 2]. mTOR is a central regulator of proliferation, apoptosis, differentiation, and motility, integrating numerous upstream signals to generate the desired biological outcome. Therefore, we assessed how mTOR signaling contributes to mammary morphogenesis. We found that pharmacologic mTOR inhibition reduced the size and branching complexity of mammary organoids in culture, phenotypes recapitulated by mTORC2 loss of function via Rictor ablation, but not upon mTORC1 inhibition through Raptor ablation. We also observed a disorganized epithelial architecture and stromal thickening around TEB upon tissue-specific Rictor ablation. The MMTV-Cre model has been reported to be leaky, leading to expression in tissues other than luminal mammary epithelium [30] thus it is possible that some of these defects may be due to loss of Rictor in stromal components. Alternatively, changes in basal epithelium may be a secondary effect of luminal cell misolocalization in the absence of Rictor, or Rictor expression in the luminal compartment may regulate expression and function of mTOR signaling intermediates in the basal cell layer through an indirect, juxtacrine signaling mechanism. We are actively investigating the role of Rictor/mTORC2 in luminal versus basal epithelium in our ongoing research. As our epithelial branching and survival phenotypes were recapitulated in the ex vivo stroma-free organoid culture model, however, it is likely that the effects on stroma are, at least in part, secondary to the loss of Rictor in epithelium.
Genetic inhibition of mTORC2 also reduced ductal branching and lengthening in vivo, diminished P-Akt and P-PKC-alpha, and impaired activation of the GTPase Rac1. Akt restoration only modestly enhanced branching morphogenesis in Rictor-deficient mammary organoids and was not sufficient to rescue cell survival or PMEC invasion through Matrigel. However, Akt inhibition did decrease organoid branching and colony size suggesting that Akt provides a critical signal in growth control, but is not sufficient to drive branching morphogenesis in the absence of Rictor. This is consistent with the data from our analysis of transplanted Rictor-deficient/AktMyr expressing MEC in vivo and with the phenotype of Akt1 deletion, which did not affect mammary epithelial cell differentiation but did impair lactation [31]. Deletion of Akt1 and one allele of Akt2 enhanced this defect [32, 33]. Moreover, Akt activation did not completely inhibit luminal apoptosis during MCF10A acinar morphogenesis in culture [34], suggesting that other factors also regulate cell survival during normal mammary epithelial development. In contrast to Akt, restoration of PKC-alpha signaling to Rac1, or Rac1 activation independently of upstream signals, fully rescued all phenotypes resulting from Rictor loss in culture and in transplanted Rictor-deficient MEC in vivo, suggesting that Rictor-dependent mTORC2 is essential for PKC-alpha-Rac1 signaling to drive mammary morphogenesis. While not directly tested here, at least one additional study has elucidated mechanisms downstream of Rac1 that can control cell survival. One report using lymphoma cells demonstrated direct inhibition of apoptosis through Rac1-stimulated phosphorylation of the Bcl-2 family member, Bad, which occurred in an Akt-independent manner [35]. We observed a modest decrease in cell viability upon prolonged treatment with the Rac1 inhibitor in organoid culture coupled with the decreased branch extension, consistent with previous studies that also reported regulation of branching initiation and extension via PI3K-mediated Akt and Rac1, respectively [36].
Interestingly, levels of mTORC1 target P-S6 are elevated in MEC upon Rictor loss relative to controls. This could reflect shift of mTOR kinase to complex 1 in the absence of a stable mTORC2 complex. It will be of great interest to track mTOR kinase association with the two complexes over the course of mammary epithelial development to better understand its functions. Activation of the Akt signaling pathway upon mTOR inhibition via a negative feedback loop has been observed in many cell types, including breast cancer cell lines (Reviewed in [37]). In our study, rapamycin preferentially inhibited mTORC1 upon acute treatment (e.g. reduction in P-S6 without affecting P-Akt-S473 levels) and as prolonged treatment inhibited both complexes (e.g. reduction in both P-S6 and P-Akt-S473). These data are consistent with the observation that rapamycin is an effective inhibitor for activity of both complexes in many cell types [5], including MECs. The differences in response to rapamycin between normal MECs and breast cancer cell lines could be due to differences in insulin-like growth factor receptors (IGFRs), which are expressed at higher levels in cancer cells and mediate feedback to Akt upon mTOR inhibition (Reviewed in [37–39]).
Given the known roles of mTORC1 in cell growth, metabolism, and protein and lipid synthesis [3], it was surprising that Raptor loss produced only a transient delay in ductal lengthening. It is possible that other signaling pathways may compensate for loss of mTORC1 function in Raptor-deficient mammary epithelium, such as RSK-mediated activation of S6 [40]; [41]. However, we observed similar decreases in cellular proliferation in the absence of Raptor and Rictor expression at 6 weeks that recovered by 10 weeks, suggesting that MEC proliferation may rely on both mTORC1 and mTORC2. Decreased MEC proliferation upon genetic mTORC1 ablation is consistent with other reports of rapamycin-mediated cell growth inhibition in lactating mouse mammary explants, in lactating mice, and in milk-producing HC11 cells. Based on these previous studies, it will be important to determine the effects of Raptor and Rictor ablation on growth, differentiation, and milk production in alveolar mammary epithelium during pregnancy and lactation in vivo.
The PI3-kinase (PI3K)/mTOR pathway is aberrantly activated up to 60% of clinical breast cancers, facilitating tumor cell growth, survival, metabolism, and invasion [42, 43]. Moreover, increased PI3K activity in MMTV-Cre/PTENFL/FL mice increases mammary epithelial branching and decreases apoptosis during pubertal development [44], suggesting that PI3K signaling is important in branching and survival in the breast. This idea is consistent with the phenotype produced by MMTV-Cre-driven Rictor loss, in which loss of a PI3K pathway mediator produces decreased branching and survival. While inhibitors of mTORC1 show limited clinical efficacy as single agents, anti-PI3K agents combined with dual mTORC1/2 inhibitors appear to be more effective [45–48], underscoring the clinical relevance of mTORC2 in breast cancer. Importantly, these recent clinical observations parallel the data shown here demonstrating that mTORC2 inhibition due to either sustained rapamycin treatment or to Rictor deletion profoundly affected the complex series of events driving mammary morphogenesis, and these mTORC2-dependent processes occur in a manner unique and separable from mTORC1. Interestingly, preferential targeting of mTORC2 versus mTORC1 reduced breast cancer cell motility and survival in culture and in vivo [18, 49], and Rictor knockdown suppressed anchorage-independent growth of MCF7 breast tumor cells [50]. Although at least one report suggests elevated Rictor levels correlate with higher overall and recurrence-free survival [51], Rictor overexpression was observed in clinical invasive breast cancer specimens relative to normal breast tissue, as well as in lymph node metastases [18], supporting the clinical relevance of mTORC2 in invasive breast cancer. Given our findings that Rictor/mTORC2 is required in the normal mammary epithelium for PKC-alpha-Rac1 activation which drives MEC survival, motility, and invasion, it will be interesting to determine if the mTORC2-PKC-alpha-Rac signaling axis is used by breast cancer cells to drive metastasis. If so, mTORC2-specific targeting or PKC-alpha inhibition could represent potential therapeutic strategies to limit metastatic spread of breast tumors, and to limit survival of disseminated tumor cells.
Although data shown herein are the first demonstration of mTORC2-mediated regulation of normal MEC migration and invasion, several lines of evidence suggest that cancer cells exploit Rictor-dependent signaling pathways to facilitate invasion and metastasis. For example, siRNA-mediated Rictor knockdown inhibited MCF7 and MDA-MB-231 breast cancer cell migration [18, 49]. Rictor knockdown inhibited transforming growth factor beta (TGF-beta)-mediated epithelial-to-mesenchymal transition (EMT) in breast cancer lines [52]. In contrast to our findings that untransformed MECs use Rictor to activate PKC-alpha and Rac1-mediated invasion, breast cancer cells used Rictor to drive motility through protein kinase C-zeta (PKC-zeta; [18]), integrin-linked kinase (ILK; [52]) and Akt [49]. Although Akt phosphorylation at S473 required Rictor/mTORC2 in primary MECs, restoring Akt function was not sufficient to rescue survival, motility, or branching morphogenesis in the absence of Rictor. Restoration of Rac1 activity, an essential regulator of mammary epithelial branching morphogenesis [16, 53] and a downstream effector of mTORC2 and PKC-alpha, rescued survival and migration defects induced by genetic mTORC2 inhibition. While not specifically linked to Rictor in breast cancer cells, Rac1-mediated invasion and metastasis of breast cancer cells has been reported previously [54–56]. Together, these data suggest that Rictor/mTORC2-dependent Rac signaling could promote breast cancer invasion, paralleling its function normal MEC branching morphogenesis. It is possible that breast cancer cells can engage multiple pathways (PKC-zeta, ILK, Akt, Rac, and others) to regulate tumor cell metastasis, and it is interesting to speculate that Rictor may lie at the intersection of each of these pathways.
In summary, our data demonstrate distinct, non-overlapping functions of mTORC1and mTORC2 in post-natal mammary morphogenesis. Whereas Raptor-dependent mTORC1 signaling regulates proliferation, Rictor-dependent mTORC2 is essential for cell survival, cell junctions, motility, and branching morphogenesis. These findings underscore the importance of understanding the distinct roles for mTORC1 and mTORC2 in normal physiology of the breast and in breast cancer in order to intelligently develop and administer mTOR-directed therapies.
All animals were housed under pathogen-free conditions, and experiments were performed in accordance with AAALAC guidelines and with Vanderbilt University Institutional Animal Care and Use Committee approval. RictorFL/FL mice (C57BL/6) were kindly provided by Dr. Mark Magnuson (Vanderbilt University) and have been previously described [14]. RaptorFL/FL mice ([29], C57BL/6) were purchased from the Jackson Laboratories (Bar Harbor, ME). MMTV-Cre mice ([13] FVB) were purchased from the Jackson Laboratories. All analyses were performed on age-matched siblings resulting from F1 (1:1, FVB:C57BL/6) intercrosses.
Primary mammary organoids were generated from freshly collected, partially disaggregated mouse mammary glands using a modification of previously described methods [16]. Primary mouse mammary epithelial cells (PMECs) were harvested as described previously [57]. Organoids were immediately embedded in growth factor reduced Matrigel (BD Bioscience) at 50 organoids/100 microliters. Once polymerized, Matrigel-embedded cultures were overlain with Growth Media [DMEM:F12 supplemented with 5 micrograms/ml porcine insulin (Sigma-Aldrich), 10 picograms/ml each estrogen and progesterone (Sigma-Aldrich), 5 nanograms/ml human epidermal growth factor (R&D Systems), 100 I.U./ml penicillin-streptomycin (Life Technologies)]. PMEC were maintained in Growth Media. For some experiments, cells were maintained for 24 hour in Starvation Media [Growth Media supplemented with penicillin-streptomycin only] or treated with Fibroblast-Conditioned Media (DMEM:F12 supplemented 100 I.U./ml penicillin-streptomycin cultured with mouse mammary fibroblasts for 48 hours and passed through a 0.2 micron filter) for wound closure migration studies. Rapamycin (Sigma-Aldrich, 20 nanomolar), InSolution Rac1 inhibitor (Calbiochem/Millipore, 20 micromolar), and adenoviral particles (Ad.Cre, Ad.LacZ, Ad.caRac1, Ad.Aktmyr, and Ad.PKC-alpha, Vector Biolabs) were purchased. Freshly collected organoids were incubated with adenoviral particles (5 X 108 particle forming units/ml) with constant rocking for 3–5 hours at 37°C, washed, and embedded in Matrigel. We analyzed 20–30 independent organoids isolated from 5–6 independent mice in 2–3 experiments for each condition.
Morphogenesis in organoids was scored by counting the number of branches/organoid in 10 or more organoids/culture condition. For structures that appeared more spherical and less branched (e.g. cultures treated with Ad.Cre or inhibitors), we counted bifrucations and/or small protrusions from ball-shaped structures as branches in order to be as rigorous and conservative in our quantifications as possible. Organoid size was scored using NIH Image J software to quantify pixel area in 10 or more organoids/culture condition.
MCF10A and MCF10A RictorZFN were purchased from Sigma-Aldrich and cultured in Growth Medium [DMEM:F12 supplemented with 5% Horse Serum (Life Technologies), 10 μg/ml porcine insulin (Sigma-Aldrich), 20 nanograms/ml human epidermal growth factor (R&D Systems), 10 nanograms/ml cholera toxin (Sigma-Aldrich), 100 micrograms/ml hydrocortisone (Sigma-Aldrich), 100 I.U./ml penicillin-streptomycin (Life Technologies)]. For some experiments, cells were maintained for 24 hour in Starvation Media [Growth Media without serum or EGF] prior to stimulation and/or analysis. PKC-alpha inhibitor GO6976 (Sigma-Aldrich, 2 nm) and adenoviral particles (Ad.RFP and Ad.PKC-alpha, Vector Biolabs) were purchased. Cells were incubated with adenoviral particles (5 X 108 particle forming units/ml) for 3–5 hours at 37°C and cells were allowed to recover for 48 hours prior to experimental analysis.
Matrigel-emdedded organoids cultured on coverslips were fixed 8 minutes in 1:1 methanol:acetone at -20°C, permeabilized in 0.5% Triton-X 100/PBS for 10 min, blocked [130 millimolar NaCl, 7 millimolar Na2HPO4, 3.5 millimolar NaH2PO4, 7.7 millimolar NaN3, 0.1% bovine serum albumin, 0.2% Triton-X 100, 0.05% Tween-20] and stained with rabbit anti-pan-cytokeratin (Santa Cruz Biotechnology, 1:100) and AF621-goat anti-rabbit (1:100), counterstained with TO-PRO-3 Iodide (Invitrogen), and imaged using the Vanderbilt Cell Imaging Shared Resource Zeiss LSM 510 confocal microscope and LSM Image Browser software. For E-cadherin staining, the primary antibody used was anti-E-cadherin (BD Transduction Laboratories) and visualized with anti-mouse Alexa 594 secondary antibody (Invitrogen, Molecular Probes). Confocal images of 3D structures were visualized using an LSM 510 META inverted confocal microscope with a 20X/0.75 plan apochromat objective.
Cells and tissues were homogenized in ice-cold lysis buffer [50 millimolar Tris pH 7.4, 100 millimolar NaF, 120 millimolar NaCl, 0.5% NP-40, 100 micromolar Na3VO4, 1X protease inhibitor cocktail (Roche)], sonicated 10 seconds, and cleared by centrifugation at 4°C, 13,000 x g for 5 min. Protein concentration was determined using BCA (Pierce). Proteins were separated by SDS-PAGE, transferred to nitrocellulose membranes, blocked in 3% gelatin in TBS-T [Tris-buffered saline, 0.1% Tween-20), incubated in primary antibody overnight and in HRP-conjugated anti-rabbit or anti-mouse for 1 hour, and developed using ECL substrate (Pierce). Antibodies used: alpha-actin (Sigma-Aldrich; 1:10,000); AKT and S473 P-Akt (Cell Signaling; 1:2,000 and 1:500, respectively); S6 and P-S6 (Cell Signaling; 1:1,000); Rictor (Santa Cruz; 1:250); Raptor (Cell Signaling; 1:500); Rab11 (Cell Signaling; 1:1,000); PKC-alpha and T638/641 P-PKC-alpha (Cell Signaling; 1:2,000); Rac (BD Transduction; 1:200). GST-Pak-PBD effector pulldown assays were performed using reagents from Millipore as per manufacturer’s protocol.
Mammary glands were whole-mounted on slides, cleared of adipose, and stained with hematoxylin as described previously [57]. Sections (5 micron) were stained with hematoxylin and eosin. In situ TUNEL analysis was performed on paraffin-embedded sections using the ApopTag kit (Calbiochem). IHC on paraffin-embedded sections was performed as described previously [58] using: Ki67 (Santa Cruz Biotechnologies), P-S6 (Cell Signaling Technologies); P-Akt S473 (Cell Signaling Technologies); Rictor (Santa Cruz), E-cadherin (Transduction Labs). Immunodetection was performed using the Vectastain kit (Vector Laboratories), AF488-conjugated anti-rabbit, or AF621-conjugated anti-mouse (Life Technologies), according to the manufacturer’s directions.
Methanol-fixed PMECs were probed 1 hour with GST-PBD (Millipore) diluted 1:50 in PBS. GST (lacking PBD) was used as a negative control. Samples were washed then probed with AF488-conjugated anti-GST (1:100), stained with DAPI or AF621-phalloidin, and mounted.
MECs (105) were added to upper chambers of Matrigel-coated transwells in starvation medium and incubated 5 hours to score migration in response to 10% serum-containing medium in the lower chamber. Filters were swabbed and stained with 0.1% crystal violet, [59] and cells on the lower surface were counted. For wound closure, 50,000 MECs were plated on Matrigel-coated 24 well plates, grown to confluence, serum-starved for 24 hours, and wounded with a P200 pipette tip. Migration in response to mammary fibroblast conditioned medium [60] was scored by measuring the [width of the wound area at 24 hours] ÷ [width of the wound area at 0 hours] as described previously [61].
Mammary gland whole-mounts and transwell filters were imaged with Olympus SZX12 Inverted Microscope. Slides were imaged with Olympus BX60 Stereo Microscope. Organoids, annexin V-FITC-staining, and wound closure assays were imaged with Olympus IX71 Inverted Microscope. All images were acquired by Olympus DP 72 Digital Camera and CellSens software at ambient temperature.
All animals were housed under pathogen-free conditions, and experiments were performed in accordance with AAALAC guidelines and with Vanderbilt University Institutional Animal Care and Use Committee approval. The laboratory animal care program of Vanderbilt University (PHS Assurance #A3227-01) has been accredited by AAALAC International since 1967 (File #000020). The AAALAC Council on Accreditation's most recent review of VU's program was done in 2011 and resulted in "Continued Full Accreditation.” Isofluorane was used for anesthesia, as well as euthanasia. For human euthanasia, cervical dislocation was used following isofluorane overdose.
|
10.1371/journal.ppat.1000875 | Electron Tomography Reveals the Steps in Filovirus Budding | The filoviruses, Marburg and Ebola, are non-segmented negative-strand RNA viruses causing severe hemorrhagic fever with high mortality rates in humans and nonhuman primates. The sequence of events that leads to release of filovirus particles from cells is poorly understood. Two contrasting mechanisms have been proposed, one proceeding via a “submarine-like” budding with the helical nucleocapsid emerging parallel to the plasma membrane, and the other via perpendicular “rocket-like” protrusion. Here we have infected cells with Marburg virus under BSL-4 containment conditions, and reconstructed the sequence of steps in the budding process in three dimensions using electron tomography of plastic-embedded cells. We find that highly infectious filamentous particles are released at early stages in infection. Budding proceeds via lateral association of intracellular nucleocapsid along its whole length with the plasma membrane, followed by rapid envelopment initiated at one end of the nucleocapsid, leading to a protruding intermediate. Scission results in local membrane instability at the rear of the virus. After prolonged infection, increased vesiculation of the plasma membrane correlates with changes in shape and infectivity of released viruses. Our observations demonstrate a cellular determinant of virus shape. They reconcile the contrasting models of filovirus budding and allow us to describe the sequence of events taking place during budding and release of Marburg virus. We propose that this represents a general sequence of events also followed by other filamentous and rod-shaped viruses.
| The filoviruses, Marburg and Ebola, cause lethal hemorrhagic fever and are highest-priority bioterrorism agents. Filovirus particles contain a rod-like nucleocapsid and are normally filamentous, though other shapes are seen. It is poorly understood how such large filamentous particles are assembled and released from infected cells. Here we have studied Marburg virus production in infected cells using electron tomography. This technique allows virus particles to be visualized in three dimensions at different stages during assembly. We find that in early stages of virus production, highly infectious filamentous viruses are produced, whereas after prolonged infection poorly infectious spherical viruses are released. We also define the sequence of steps in filamentous virus release. The intracellular nucleocapsid first travels to the plasma membrane of the cell, where it binds laterally along its whole length. One end is then wrapped by the plasma membrane and wrapping proceeds rapidly until the virus protrudes vertically from the cell surface. The rear end of the virus particle then pinches off from the cell. We propose that other important filamentous and rod-shaped viruses also follow this series of steps of assembly and budding.
| Marburg virus (MARV) and Ebola virus, the two genera in the family Filoviridae, cause fulminant hemorrhagic disease in humans and nonhuman primates, resulting in high mortality rates [1], [2], [3]. Outbreaks of MARV disease in sub-Saharan Africa underline the emerging potential of this virus, which is classified as a highest-priority bioterrorism agent by the Centre for Disease Control [4], [5], [6], [7], [8].
The filoviruses are members of the order Mononegavirales and contain a single-stranded negative-sense RNA genome, which is encapsidated by the nucleoprotein (NP). The MARV genome encodes seven structural proteins [9], [10]: the polymerase (L), VP35 and VP30 associate with NP to generate the helical nucleocapsid (NC) [11], [12], [13]. The viral glycoprotein (GP), which is inserted in the viral envelope, mediates cell entry [14], [15]. The major matrix protein VP40 plays a key role in virus assembly, and VP24, the second matrix protein, is suggested to support the template function of the NC [12], [16], [17], [18].
MARV infected cells develop viral inclusions in the perinuclear region [19], [20], [21]. These contain NC proteins and are most likely centres of NC assembly [22]. MARV particles bud from the plasma membrane (PM) of long filamentous cellular protrusions that contain parallel actin bundles and other markers of filopodia [23]. The released virus particle has a membrane envelope and contains an NC that is surrounded by the viral matrix protein VP40. It is unclear how NCs are transported from viral inclusions to the PM, whether they adopt their virion conformation before, during, or after, transport, or where NCs associate with VP40 that is not co-transported with NCs but is necessary for budding. Released MARV particles appear filamentous, hooked, six-shaped or round by electron microscopy (EM) [24] but their three-dimensional (3D) morphology is unclear. It is also unknown whether production of differently shaped viruses depends on different budding mechanisms and whether they differ in infectivity. Experiments to address these issues are complicated by the need to perform all infection experiments under BSL-4 containment conditions.
The processes of assembly, budding and release of spherical viruses have been extensively studied [25], [26], [27] and it is well established that spherical enveloped viruses are produced by budding away from the cytoplasm in a process that is related and topologically equivalent to the formation of small vesicles in multivesicular bodies [28], [29]. In contrast, the basic steps of assembly and release for large, filamentous, enveloped particles such as the filoviruses are poorly understood. EM studies have shown MARV and Ebola virus particles protruding perpendicularly from the cell [23], [30]. These observations, together with similar findings of protruding intermediates of other important filamentous or rod-shaped viruses such as rabies virus [31], influenza virus [32], [33], [34], or vesicular stomatitis virus (VSV) [35] have led to the suggestion of a vertical “rocket-like” mode of budding. Ebola virus NCs have been seen associated parallel to the PM [36], leading to the suggestion that a second, horizontal, “submarine-like” mechanism is the major mode of budding. This unusual mechanism has not been described for any other virus.
In this study we use electron tomography (ET) to describe and analyse in 3D the structure of MARV budding intermediates and released viruses at different stages of infection. In contrast to conventional EM of ultrathin sections, ET allows complete MARV virions and budding structures to be studied in 3D. This permits unambiguous determination of virus morphology, dimensions, stage of budding and position relative to the infected cell. The samples are prepared by high-pressure freezing followed by embedding in resin and staining with heavy metals. Unlike cryo-ET of vitreous samples, this method is not appropriate for the study of high-resolution protein structure. Nevertheless, it gives excellent preservation of the features studied here such as membranes and protein assemblies including viral NCs [37], [38]. It also has, for this particular purpose, substantial advantages over cryo-ET. The features of interest are imaged at much higher contrast, including at high tilt angles. The samples are also easier to handle and stable under the electron beam, allowing efficient screening, which facilitates the collection of larger datasets.
Our observations suggest that interplay between the virus and the infection state of the cell determines virus morphology. Furthermore, the 3D data allow us to describe the sequence of steps that take place in infected cells during assembly and budding of filamentous virions, and to reconcile the horizontal and vertical models for filovirus budding as representing different snapshots of a single budding process.
To determine the time course of production of infectious MARV during a prolonged infection period, supernatants of HUH-7 cells, infected with MARV under BSL-4 conditions, were collected from day one to four post infection (p.i.) and tested for viral protein content, specific infectivity and virion morphology. The amounts of viral NP and VP40 released from the cells were measured by quantitative immunoblotting. This showed that viral protein release peaked between day one and two p.i. (Figure 1A). The TCID50 of each supernatant was determined and normalized to the amount of released NP to estimate the specific infectivity per virus (Figure 1B, grey areas). Specific infectivity was also at a maximum between day one and two p.i.. Virus morphology in the supernatants was monitored by EM and revealed that at the peak of viral protein production and specific infectivity, 80% of the virus particles displayed the characteristic filamentous morphology of the filoviruses, with the remainder appearing bent or round (Figure 1B). Later in infection, the specific infectivity was lower and correlated with fewer filamentous particles and more round or bent particles (Figure 1B), suggesting that infectivity can predominantly be attributed to filamentous virus. There was also an increase in the release of cellular vesicular material over time (Table 1).
Quantitative EM analysis of infected adherent cells that were fixed and embedded in situ revealed that cell morphology also changed over time. At day 1 p.i. 95% of the observed cells appeared intact and displayed filopodia-like PM protrusions (Figure 1C, D). The cytoplasm often contained densely stained viral inclusions (Figure 1D) and virions were readily seen in the immediate periphery of cell profiles; they were easily identified by means of their electron-dense stain and a rod-shaped NC and were frequently found ‘trapped’ between or underneath adherent cells (Figure 1E). Over 90% of all observed virions around the cells were filamentous or hooked (Figure 1F). In contrast, when the monolayer was fixed at 4 days p.i., only 17% of the cell profiles were intact and 83% appeared vesiculated and resembled apoptotic cells (Figure 1C, G). Most of the virus particles around such vesiculated cells were bent or round (Figure 1F, H).
The above characterisation indicates that production of fully infectious, filamentous virus is highest at 1–2 days p.i., when the cells still have intact membrane profiles. This was therefore selected as an appropriate time point for studying assembly and budding of filamentous particles.
To study the assembly and budding of MARV we performed ET on MARV-infected cells. ET allows the different steps of assembly and budding to be visualized in 3D and the dimensions of viral structures to be measured and analysed. Infected cell monolayers were fixed at 1 day p.i., removed from the BSL-4 facility, processed for EM and cut into 300 nm thick sections in their in situ orientation. Dual-axis tilt series of the sections were acquired in the electron microscope, as described in Materials and Methods, and 3D reconstructions were computationally generated. Viral NCs could be readily identified in 3D reconstructions. They were found around viral inclusions, within the cytoplasm, associated with the PM, incorporated into budding viruses, and in released virus particles (Figure 2). We therefore used the NC itself as a convenient marker for identifying virus assembly intermediates.
The 3D data allowed us to measure the length of complete viral NCs even when they were bent or tilted with respect to the sectioning plane. In the periphery of viral inclusions in the cytoplasm individual NCs could be found. They appeared as rod-shaped, striated structures that were more densely stained than the cytoplasm and were on average 711 nm long (Figure 2A, Table 2). NCs were frequently seen at the PM or in filopodia-like membrane protrusions, where they were associated with the membrane along their whole length. The length of PM-associated NCs was 707 nm (Figure 2B, Table 2), the same as that of cytoplasmic NCs. Viral budding structures localized predominantly to filopodia-like protrusions of infected cells, in agreement with previous data [23]. They appeared as filamentous finger-like extensions emerging either from the tip or the sides of filopodia-like protrusions (Figure 2C-E). Each bud accommodated a rod-shaped NC that was surrounded by densely stained material and had the same length as intracellular and PM-associated NCs (Table 2). Of the budding structures reconstructed in 3D, 13% appeared as extrusion intermediates (Table 2). These structures contained a full length NC (729 nm) that was only partially extruded (Figure 2C): one end of the NC was tightly wrapped on all sides by the PM, whereas the other was attached to the PM along one side (Figure 2C). The majority (87%) of budding structures had the NC completely inserted into the finger-like membrane extension (Figure 2D, E). Scission at the base of these fully extruded buds would lead to the release of filamentous virions (Figure 2F and Video S1).
Most released viruses in the periphery of infected cells appeared as straight filaments in 3D reconstructions (Figure 2F). Some filamentous viruses displayed one bent or buckled end giving rise to hooked or six-shaped particles (Figure 2G and Video S1). Filamentous viruses were on average 789 nm long and had a diameter of 88 nm (Table 2), in agreement with previous studies [20], [24]. All virions had a membrane envelope and contained a rod-shaped NC that displayed the regular striated pattern previously described [12], [20]. The volume between membrane envelope and NC was filled with a densely stained material, most likely the VP40 matrix protein, which appeared to link the envelope to the NC on all sides and along the whole length (Figure 2F). 3D analysis revealed that NCs in all released virions had the same length, on average 735 nm, and appeared bent or kinked when particles were hooked or six-shaped (Figure 2G, Table 2). These length averages exclude a single outlier, which was a double-length filamentous virus with a double-length NC (Table 2). The presence of NCs of the same length in the cytoplasm, at the PM, in filopodia, in all budding structures and in filamentous viruses suggests that the NC assembles into a helix of a defined and final length prior to being transported to the PM.
The 3D data also allowed us to closely examine both ends of reconstructed filamentous virions. This revealed morphological differences between the two virus ends: 19 out of 21 3D-reconstructed filamentous virions displayed a membrane bulge or hook at one end (Figure 3A), while the membrane formed a round, intact hemisphere at the opposite tip (Figure 3B and Table 2). Only 2 out of 21 filamentous viruses had two round, intact tips. Virions with membrane distortions at both ends were never observed. Although we cannot exclude that such membrane distortions are due to chemical fixation, they suggest that a local instability of the viral membrane is specifically induced on one virus end. Thus, we made use of the fact that cells and extracellular viruses were fixed and examined in their in situ orientation. Measurement of the average distance of intact and distorted virus ends from the nearest cellular membrane revealed that intact virus ends were predominantly found at distances >250 nm away from a cellular membrane, whereas distorted virus ends were more frequently localized within 50 nm distance (Figure 3C). This suggests that the membrane distortions are found at the rear end of filamentous viruses where scission took place (also shown in Figure 2G and Video S1).
We also carried out ET of infected cells fixed at 4 days p.i.. At this time point the majority of cells exhibited heavily vesiculated membranes and produced rounded viruses of lower infectivity. The 3D analysis revealed that most viruses were roughly spherical in shape. They were generally found in close proximity to convoluted, vesiculating areas of the PM (figure 4A) and often surrounded by large numbers of cell-derived vesicles (Figure 4A and Video S2). This suggests that release of spherical viruses occurs rapidly and simultaneously with the shedding of other cell-derived vesicles. NCs within spherical virions were bent or kinked but never broken or segmented, as might be suggested by 2D EM of thin sections (see, for example the 2D and 3D views of the spherical particle shown in Figure 4B). None of the virus particles were toroidal, as has previously been suggested [39]. Interestingly, kinked NCs in spherical viruses had a total length of 734 nm, the same length as NCs in filamentous virions (Figure 4B, Table 2) and they were associated along one side with the viral membrane (Figure 4B). These findings demonstrate that preassembled full-length NCs are packaged into each virion, irrespective of virus shape.
The analysis of MARV-infected cells by ET allowed us to take 3D measurements of NCs and revealed that NCs have uniform length at all stages in the virus assembly and budding process described here, from intracellular NCs to released virus particles. This suggests that the length of the complete viral genome and the number of nucleoproteins required to encapsidate the genome dictate the length of the intracellular NC prior to its transport to the PM.
Once the NC has reached the PM, two contrasting mechanisms have been proposed for filovirus budding. In the “submarine-like” model, budding proceeds via lateral association of the NC with the membrane, followed by horizontal budding. In the “rocket-like” model, the NC is extruded vertically from the PM. The analysis of budding MARV particles presented here allows the sequence of steps resulting in filamentous virus budding and release to be described in 3D. Viral NCs are assembled in the cytoplasm (Figure 5A) and delivered in their full-length form to the PM, with which they associate laterally, along one side, for their entire length (Figure 5B). Envelopment of the PM-associated NC is initiated at one end, and proceeds along the length of the NC (Figure 5C) until the nascent virion protrudes from the membrane, remaining attached at only one end (Figure 5D). Only very few particles are seen which appear to be partly extruded, whereas larger numbers of NCs are either associated with the PM prior to extrusion or are fully extruded and protruding from the PM. This strongly suggests that extrusion is a rapid process in comparison with its initiation or the subsequent membrane scission event.
Scission of the filamentous virus particle from the PM then takes place with a bud-neck shape which can have a circular cross-section. This is the same shape as the bud-neck which would be present in the budding of spherical enveloped virions, or in the budding of vesicles into a multi-vesicular body [40]. Release of a horizontally budding particle would require scission of a membrane neck with a non-circular cross-section. Our observations indicate that no such unusual scission mechanism needs to be proposed.
Scission of the protruding bud leaves local membrane instability at the rear end of the virus particle (Figure 5E), which may be exaggerated during sample preparation. In contrast, the front end of the filamentous particle is a well-defined hemisphere. The difference between the two ends could reflect a destabilization at the rear end of the particle induced by scission. Alternatively, the hemispherical front end of the particle may be stabilized by a structure involved in initiating envelopment, similar to the front end of bullet-shaped rhabdovirus particles [41], [42].
The budding of filamentous MARV therefore proceeds via an NC that is laterally associated with the PM, and a vertically protruding bud. Any changes in the rates of individual steps in the budding process could dramatically alter the appearance of the budding structures in EM. For example, if the rate of scission were significantly increased, or the rate at which extrusion is initiated were dramatically decreased, then large numbers of NCs might be expected to collect horizontally under the PM, giving the appearance of a predominant “submarine” mode of budding [36]. If the rate of initiation of extrusion were significantly increased, or the rate of scission decreased (for example by inhibiting recruitment or function of the cellular endosomal protein sorting system), then large numbers of protruding buds would accumulate, giving the appearance of a predominant “rocket-like” mode of budding. We suggest the contrasting appearance of Ebola virus and MARV infected cells in EM does not reflect different budding mechanisms, but rather different rates for the individual steps of the process presented here. These rates are likely to be both virus and cell type dependent.
The sequence of steps in budding described here results in a very different mechanistic understanding of filovirus budding. Firstly, the proposed two budding modes, one vertical and one horizontal, can be reconciled as representing different snapshots of a single budding mechanism. Secondly, rather than filoviruses adopting a unique horizontal mode of budding not seen in other systems, the budding process can now be placed within the established framework of other cellular budding events. In the new model, budding is initiated by wrapping of one end of the NC with a hemispherical membrane, and completed by scission at a bud-neck with a classic round cross-section. These are the membrane shapes present during budding and scission events of cellular vesicles and spherical viruses. Thirdly, comparison of our data with published electron micrographs of filamentous virus budding structures suggests that other viruses may also follow the sequence of steps described here, and that rather than being unique, filovirus budding belongs within a more general budding mechanism also adopted by other rod-shaped viruses. For example, budding VSV, a rhabdovirus, can be found protruding perpendicular to cell membranes in infected cells, but the NC can also be seen underneath the cell membrane, with which it associates laterally along one side [35]. The rounded end of the virus, easily distinguished in VSV, seems to associate more tightly with the membrane [43], and buds first. After budding, under certain preparation conditions, the rear of the virus is seen to show small membrane blebs or instabilities [41] similar in appearance to those described here for MARV. These striking observations are consistent with VSV following the same sequence of budding steps as MARV. A general assembly and budding mechanism for filoviruses and rhabdoviruses might represent a target for future antiviral drugs.
After a prolonged period of infection, the released MARV particles have lower specific infectivity and the majority is roughly spherical in form. Like the filamentous virions, the spherical particles contain a single full-length NC, which is kinked in a number of places, but not broken, suggesting that these particles still contain a full-length viral genome. The NC is not tightly wrapped on all sides as it is in filamentous virions, but is associated with the membrane along one side for its entire length. This lateral association is also seen for intracellular NCs prior to extrusion from the PM. Since formation of spherical particles is paralleled by convolution of the PM and shedding of cellular vesicular material into the supernatant, we propose that spherical viruses are released by large-scale membrane instability at sites of budding, leading to vesiculation of viruses prior to extrusion. In this model, kinking of the NC might be induced by the forces which lead to invagination and vesiculation of the PM, though other factors could contribute, such as lack of a viral or cellular factor which rigidifies the NC. The interplay between the budding process and the dynamics or composition of the cellular membrane, as well as possible changes in the relative rates of the different steps of the budding process in different virus strains [44], mutants [45], cell types [46] or stages of infection, may also contribute to the variable morphology observed in other viruses.
In summary, our observations reconcile the contrasting models of filovirus budding and allow us to describe the sequence of events taking place during budding and release of MARV virus. We propose that this represents a general sequence of events also followed by other filamentous and rod-shaped viruses. Furthermore, we demonstrate that virus shape is determined both by viral and cellular factors. The model for filamentous virus budding presented here raises a number of new questions. Do elements of the cytoskeleton or other cellular components play a specific role in mediating envelopment or budding of filamentous or spherical virus particles? Is one end of the PM-associated NC specifically able to initiate the extrusion of the filament, as in rhabdoviruses, or is the directionality of the extrusion process random? What is the organisation of the NC and VP40 matrix layers in the released virion? What kind of arrangement of interactions between NC and VP40 occurs in the wrapping process during bud extrusion? These and other questions must be addressed in future structural studies, and may have wider implications for the budding and assembly of filamentous viruses.
The HUH-7 human hepatoma cell line and Vero cells were maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal calf serum, l-glutamine and penicillin–streptomycin at 37°C under 5% CO2. All work with infectious MARV was performed under BSL-4 conditions at the Institute of Virology in Marburg. The MARV Leiden strain, isolated in 2008 in Leiden, the Netherlands [47], was propagated in Vero E6 cells and purified as described previously [48]. HUH-7 cells were infected with MARV with a multiplicity of infection of approximately 1 plaque-forming unit per cell for one to four days. At the indicated time points, cell culture supernatants were collected and used for viral infectivity assays (see below), or viruses in the supernatants were purified by centrifugation over a sucrose cushion and fixed for 48 hours (h) with 4% paraformaldehyde in PBS for analysis of particle morphology (see below). For EM and ET of infected cells, HUH-7 cells were grown on carbon-coated sapphire disks, infected as above and cell monolayers were fixed at the indicated time points in the cell culture dish with 4% paraformaldehyde/0.1% glutaraldehyde in 0.1M PHEM buffer (60 mM PIPES, 25 mM HEPES, 2 mM MgCl2, 10 mM EGTA), pH 6.9 for 30 min, after which the fixative was replaced with 4% paraformaldehyde in 0.1M PHEM. Fixed cells were removed from the BSL-4 lab after 48 h of inactivation.
Infectivity of MARV particles released into the supernatant of cells 1 to 4 days p.i. was assayed by a 50% tissue culture infective dose (TCID50) assay: Vero cells were grown in 96-well plates to 30 to 40% confluence. Cells were inoculated in quadruplicate with 10-fold serial dilutions of supernatants of HUH-7 cells infected with the MARV Leiden strain for one to four days as described above. The assays were evaluated at 10 days p.i.. TCID50 values were calculated using the Spearman-Karber method [49]. Equal volumes of each supernatant were separated by SDS-PAGE followed by quantitative immunoblotting on a LiCor Odyssey system using a mouse monoclonal anti-NP antibody, and secondary antibodies, protocols and software (Odyssey version 2.0) provided by the manufacturer. TCID50 values were normalized to NP levels detected in each supernatant, and the TCID50 value in the supernatant collected at day 2 p.i. was set to 100%.
For quantification of virus morphology at different time points p.i., fixed cell culture supernatants were purified by centrifugation over a sucrose cushion, pelleted for 30 min at 40,000 g, 4°C in a Beckman ultracentrifuge using a TLA-55 rotor. Pellets were embedded in 12% gelatin and prepared for EM as described previously [50]. 70 nm cryosections were obtained with a Leica EM UC6 microtome, FC6 cryochamber (Leica Microsystems, Wetzlar, Germany) and a diamond knife (Diatome, DiS-Galetzka Weinheim, Germany). Thawed cryosections were counterstained with uranyl acetate as described elsewhere [51]. For EM and ET of infected cells, HUH-7 cells were grown on carbon-coated sapphire disks, infected and fixed as described above. Samples were high-pressure frozen with a BalTec HPM-010 and freeze substituted with 0.1% (w/v) uranyl acetate, 1% osmium tetroxide (w/v) and 5% (v/v) water in glass distilled acetone in a temperature-controlling device (Leica EM AFS I). Cells were kept for 40 h at −90°C and warmed up to 0°C (slope 5°C/h) with an additional 3 h infiltration period at −30°C. Samples were washed three times with glass distilled acetone, infiltrated at room temperature with increasing concentrations of epoxy resin (Glycidether 100, Roth, Karlsruhe, Germany) in acetone over 12 h, and polymerized at 60°C for 48 h. 150 nm and 300 nm sections were obtained with a Leica Ultracut UCT microtome and a diamond knife.
Thin sections of virus and infected cells were examined on a FEI Morgagni 268 TEM equipped with a 1K side mounted CCD camera (SIS, Muenster, Germany). Quantification of cell morphology and cell-associated virus morphology was carried out on 150 nm resin-embedded sections in a systematic random sampling manner [52]. Starting points for sampling were chosen randomly on each grid, and all grids were examined in the same systematic manner. At least three grid squares per EM grid and three EM grids per time point were sampled. Morphology of viruses from cell culture supernatants was quantified by EM in the same systematic random sampling manner on thawed 70 nm cryosections of virus pellets, purified and embedded as described above. At least 200 particles from three different grids were evaluated per sample. ET was carried out essentially as described elsewhere [53]. Dual axis tilt series from 300 nm sections were recorded on a FEI TECNAI TF30 microscope operated at 300kV (4K FEI Eagle camera; binned pixel size 0.77 nm or 1 nm on the specimen level) over a −60° to 60° tilt range (increment 1°) and at a defocus of −0.2 µm. Tomograms were reconstructed using the IMOD software package (version 3.12.20) [54]. 3D measurements of virus end distance, analysis of virus end morphology and 3D surface renderings were carried out using the AMIRA Visualisation Package (version 5.2.0, Visage Imaging, Berlin, Germany). In total, 68 3D reconstructions from three independent infection experiments were analysed, of which 48 of high image quality were used for detailed analysis and measurements. NCs in different samples displayed variable contrast, which resulted from differential access of electron-dense stain to individual cells and different subcellular regions during sample preparation. NC representations in Figures 2 and 4 and Video S1 and Video S2 were generated and displayed using the AMIRA EM package [55] and MatLab (version 7.4.287).
|
10.1371/journal.pgen.1005631 | GBStools: A Statistical Method for Estimating Allelic Dropout in Reduced Representation Sequencing Data | Reduced representation sequencing methods such as genotyping-by-sequencing (GBS) enable low-cost measurement of genetic variation without the need for a reference genome assembly. These methods are widely used in genetic mapping and population genetics studies, especially with non-model organisms. Variant calling error rates, however, are higher in GBS than in standard sequencing, in particular due to restriction site polymorphisms, and few computational tools exist that specifically model and correct these errors. We developed a statistical method to remove errors caused by restriction site polymorphisms, implemented in the software package GBStools. We evaluated it in several simulated data sets, varying in number of samples, mean coverage and population mutation rate, and in two empirical human data sets (N = 8 and N = 63 samples). In our simulations, GBStools improved genotype accuracy more than commonly used filters such as Hardy-Weinberg equilibrium p-values. GBStools is most effective at removing genotype errors in data sets over 100 samples when coverage is 40X or higher, and the improvement is most pronounced in species with high genomic diversity. We also demonstrate the utility of GBS and GBStools for human population genetic inference in Argentine populations and reveal widely varying individual ancestry proportions and an excess of singletons, consistent with recent population growth.
| Eukaryotic genomes range from millions to billions of base pairs in size, but for many genetic experiments it is sufficient to gather information from just a fraction of these sites. In practice, selecting a consistent set of sites can be achieved by cutting genomic DNA with enzymes that recognize DNA sequence motifs, and then sequencing the ends of the resulting fragments. The advantages of this well-known approach are its low cost relative to whole-genome sequencing (WGS) and that it does not require a sequenced genome. These methods, for example genotyping-by-sequencing (GBS), are popular for mapping genes and studying population genetics, particularly in non-model organisms. Here we demonstrate, however, that computational tools designed for WGS are insufficient for handling certain error types that arise in GBS and other similar methods. We present a modified protocol for GBS and a statistical method for detecting these errors, implemented in the software package GBStools. We tested our methods on human DNA samples from Argentine populations. Our results reveal widely varying degrees of European and Native American ancestry, and that rare genetic variants are more numerous than would be expected in a population with constant size.
| High-throughput reduced-representation sequencing methods[1] are inexpensive, suffer little from ascertainment bias, and generate genetic markers that are approximately randomly distributed throughout the genome. These methods have been successfully used in trait mapping[2,3], linkage map construction[1,4], selection scans[5,6], and estimating genetic diversity[7]. One such method is genotyping-by-sequencing[8] (GBS). In GBS, the sequencing target is reduced to < 5% of the genome by ligating sequencing adapters only to restriction enzyme cut sites (Fig 1A). GBS reads can also be assembled into short contigs, which enables single nucleotide variant (SNV) calling without the aid of a genome sequence[9]. Hence, GBS is a popular approach in non-model systems, which typically lack resources such as genome assemblies and microarrays.
Unlike whole genome sequencing (WGS), GBS is prone to variant calling errors due to restriction site polymorphisms[7,10–14] (‘allelic dropout’, Fig 1B). Allelic dropout in GBS can confound applications that rely on accurate calling of rare variation, such site frequency spectrum estimation in population genetics. Here, we present a modified GBS protocol, similar to ddRAD-seq[15], and quantify its error rate. In addition, we present a systematic statistical approach to detect allelic dropout in GBS sequence data, implemented in the open-source software package GBStools.
This approach is based on the fact that allelic dropout reduces a sample's read coverage at a particular site in proportion to the number of non-cut restriction site alleles it carries there (Fig 1C). Therefore GBStools models coverage of each sample at a particular site as an overdispersed Poisson random variable drawn from either a distribution with mean λ (zero non-cut alleles carried), a distribution with mean ½λ (one non-cut allele), or with mean zero (two non-cut alleles). GBStools calculates the maximum-likelihood estimate of the parameter λ by expectation-maximization (EM), with the true number of non-cut alleles per sample serving as latent (unobserved) variables (S1 Appendix). The expected values of these latent variables can be used to estimate which samples carry a non-cut allele (see "Expected non-cut alleles" in Fig 1C). Simultaneously, GBStools estimates the site frequency of the observable reference and alternative SNP alleles, ϕ1 and ϕ2 (for example see Fig 1B), and the non-cut allele, ϕ3, where ϕ1 + ϕ2 + ϕ3 = 1. Finally, it performs a likelihood ratio test comparing the null hypothesis ϕ3 = 0 to the alternative hypothesis ϕ3 > 0. In its current implementation GBStools cannot infer the true genotypes obscured by allelic dropout, but it can be used to remove errors by filtering out sites where a high likelihood ratio indicates the presence of restriction site polymorphism.
Lastly, we describe the application of these methods to an extant mixed ancestry population from Argentina to test the performance of GBS in ancestry estimation and demographic inference.
We estimated the magnitude of GBS errors caused by restriction site polymorphisms from both simulated and real data. We chose human as a model system for GBS methods development due to the availability of a high-quality reference genome assembly, high-coverage whole-genome sequencing data,[16,17] and dense SNP array data.
First, we prepared modified GBS libraries from eight HapMap samples from a diverse range of populations and sequenced them on a single HiSeq lane (S1 Table, methods). We used the methylation-insensitive enzymes BpuEI, BsaXI, and CspCI, which cut away from their recognition site. Although a well-balanced mix of different sequencing adapters is commonly used to ensure that restriction enzyme recognition sequences are not over-represented at the start of the sequencing reads [3,4,8,15], our method tolerates low-diversity mixes of adapters, which is convenient when working with smaller sample sets. We quantified each sample by bioanalyzer after PCR, but before pooling, with the goal of reducing variance in the number of reads per sample in the final library. We found, however, that errors at this stage, particularly those caused by incorrect quantification of the bioanalyzer internal standard, can in fact lead to the opposite effect (S1 Fig). More careful quantification by bioanalyzer, or quantification by fluorimetry, should correct this problem and lead to the desired effect. The HapMap samples had 16.7X mean coverage in a 128 Mb target region (S2A and S3A Figs). We used GATK[18] to call SNPs in the target regions, and found 483,381 segregating sites that passed variant quality score recalibration. After applying hard filters (coverage ≥ 8X in 8/8 samples, mapping quality ≥ 57, SNP quality ≥ 30), these GBS genotype calls were 98.0% concordant with heterozygous calls from whole-genome sequencing data gathered from the same set of samples (Fig 2A and S2A Table). We found the error rate dropped as sequencing coverage increased up to 30-40X, after which further increases in coverage had little effect (Fig 2B). Furthermore, the error rate for singletons was roughly two-fold higher than for non-singletons (Fig 2C). A filter for known restriction site polymorphisms in the 1000 Genomes Project[19] data set also had a strong effect on concordance (Fig 2A–2C). These three factors appeared to be the major determinants of genotype calling accuracy.
The fact that hard filters resulted in a fairly low error rate (2%) suggested that this is a sensible approach for species with genetic diversity similar to humans. But many non-model organisms have higher levels of genetic diversity, which may lead to an error rate that is high enough to necessitate a more sophisticated approach. To explore this possibility, we simulated GBS data under a neutral coalescent model[20] with population mutation rates (θ = 4Nμ) between 1×10−3–2×10−2. In a preliminary filtering step, we removed SNVs with > 10% missing genotypes, which reduced the genotype error rate to 1.2% for data simulated with θ = 1×10−3 (typical of human data), and 4.7% for data simulated with θ = 5×10−3, which is typical of high-diversity species such as Drosophila (S4A Fig). We simulated 40X GBS coverage for these same genotype data, and found that the GBStools likelihood ratio test reduced the error more than 10-fold, for instance down to 0.3% in the case of the high-diversity (θ = 5×10−3) data set (S4A Fig). Although normalized site frequency spectra (SFS) were not substantially affected by restriction site polymorphisms (S4D and S4E Fig), errors in the genotypes themselves may cause problems in some applications. In these cases, particularly in studies of high diversity species, GBStools is expected to improve genotyping accuracy more than hard filters.
As a preliminary step in testing the utility of GBStools, it was necessary to confirm the theoretical prediction that samples with one non-cut restriction site allele (restriction site genotype +/–) have on average half the coverage of samples with two intact restriction site alleles (restriction site genotype +/+). To test this, we measured GBS coverage at known polymorphic restriction sites in the HapMap data (Fig 3). We applied a normalization to account for variation in total read numbers between libraries (methods), and binned the individual sample coverages according to the mean coverage of +/+ samples at each site. Within each bin, we observed two distinct, but overlapping, coverage distributions for samples with restriction site genotypes +/+ and +/–, suggesting that the prediction holds true. The proportion of the +/–distribution that does not overlap the +/+ distribution provides a rough measure of the potential power of a statistical test for restriction site polymorphism based on read coverage, and it is evident from the extensive overlap of the two distributions in the 5-15X and 25-35X bins that higher coverage is necessary to achieve substantial power. If the goal of a particular study were to estimate population-level summary statistics such as Fst, or to map traits in an experimental cross, the added accuracy afforded by such a test might not be worth the extra sequencing effort to achieve > 35X coverage. If the goal, however, were to estimate the site frequency spectrum, then high genotype accuracy would be necessary, and in such cases (e.g. exome sequencing) coverage in the > 35X range is not uncommon. Thus the conditions for high-sensitivity detection of restriction site polymorphisms might already exist in many experimental designs.
To better define the experimental conditions under which it is possible to use GBStools effectively, we applied GBStools to data simulated with different numbers of samples (from N = 8 to N = 500), and read coverages (10-100X). Since the proportion of homozygotes at a SNV observed by GBS is sometimes inflated by restriction site polymorphism, we also used an exact test to assess the chance of observing the given genotypes (or a worse-fitting set of genotypes) at each site under Hardy-Weinberg equilibrium. We then calculated the sensitivity and specificity of the GBStools likelihood ratio, or the Hardy-Weinberg p-values, as classifiers of incorrect vs correct genotype calls under varying thresholds (Fig 4A), and measured the area under curve (AUC) of the response operator characteristic (ROC) curves as indicators of the test's performance. In theory, an uninformative (random) classifier has AUC = 0.5, whereas a perfect classifier has AUC = 1.0. The GBStools test outperformed the Hardy-Weinberg test as measured by area under the curve (AUC), particularly at high-coverage sites (Fig 4A, S3 Table). We noted that the ROC curves for the GBStools test at low coverage (10X) and the Hardy-Weinberg test have a similar shape, which may be due to the assumption of Hardy-Weinberg genotype proportions in the GBStools model (S1 Appendix). Aside from the already-established benefit of high coverage, we also found that large sample sizes were beneficial to GBStools performance. For example, power to detect non-cut restriction site alleles of frequencies between 0.01–0.02 was 25% for 30 samples at 40X coverage, but was 94% for 500 samples at the same coverage (Fig 4B). For 40X sites in the 100- and 500-sample data sets, AUC was at least 0.96, suggesting that this is the ideal coverage and sample size range for using GBStools.
At lower coverage (10-20X) and with smaller sample sets (N = 8) GBStools did not perform as well in simulations (Fig 4), and this may explain the modest increase in concordance from 98.0% to 98.5% when the GBStools filter was applied to the HapMap data set (N = 8), which led to the removal of 9% of segregating sites (Fig 2A). For comparison, a filter for known restriction site polymorphisms in the 1000 Genomes Project[19] data set improved concordance to 99.0% (Fig 2A, S2C Table), suggesting that the power of GBStools was no higher than 50%. Indeed, power to detect common restriction site polymorphisms in the HapMap GBS data (non-cut allele frequency 0.25) was 56% for sites covered to 30-40X, but for singleton sites covered to 30-40X it was only 13%, which was lower than predicted by simulation (Fig 4B). In addition, AUC values for the HapMap ROC curves were lower than the values obtained in simulations with matching coverage levels (Fig 5A, S3 Table). This is possibly due to the model's assumption of a constant value for the index of dispersion in depth of coverage between samples, whereas the empirical data exhibit variation in dispersion from site to site (S5 Fig). It should be possible to relax this assumption by estimating dispersion on a site-by-site basis, or by calculating a joint estimate from genome-wide data, but these methods are currently not implemented. Joint modeling of genotypes at multiple closely-linked SNPs should also offer an increase in power over the single-marker model currently implemented. This would be particularly useful in the case of long reads, where each "stack" of reads mapped to a particular restriction site would contain more SNPs on average than a stack of shorter reads. For the present time, however, our simulations suggest the easiest way to improve the low empirical power observed here is to increase the number of samples.
We investigated whether it is possible to accurately estimate which particular genotypes are likely to be affected by allelic dropout. As mentioned in the introduction, the true numbers of non-cut alleles per sample are latent variables in the GBStools likelihood model, and the expected values of these variables are output by GBStools in VCF format. We compared these expected non-cut allele counts to the true counts inferred from whole-genome sequencing data to gain an idea of their predictive value (Fig 6). Although samples with a true allele count of one (i.e. restriction site genotype +/–) had higher average expected non-cut allele counts than samples with true allele count of zero (genotype +/+), it is clear that this is not a very sensitive predictor. For instance, +/–samples at sites with non-cut allele frequency 0.25 and 30-40X coverage had a median expected non-cut allele count of 0.01 (Fig 6), far from the true value of 1.0. Yet power to detect restriction site variants in these same data was 56% (Fig 4B). This indicates that the true utility of GBStools is in determining whether or not any samples at a site carry non-cut alleles rather than determining which particular samples carry them, although in some cases (Fig 1C) there is diagnostic value in the latter approach.
The site frequency spectrum derived from our filtered GBS data was similar to the spectrum from whole-genome sequencing data, with 2.3% fewer singletons (Fig 7A). This suggested that GBS data can be useful in population genetic studies, for example demographic inference based on the site frequency spectrum.
To explore this further, we sequenced 89 admixed Argentine individuals to test for signatures of mixed ancestry and demographic changes (S4 Table). The Argentine samples had 7.5X mean coverage in a 177 Mb target region (S3B Fig, S4 Table). Argentine samples with < 30% of reads mapped to restriction sites (26/89 samples) were excluded from further analyses, as it is likely that these samples were not digested to completion. A total of 1,013,785 segregating sites were called in the remaining samples and concordance with exome array data was 99.7% after filtering with GBStools, which led to removal of 25% of sites (Fig 2D, S3H and S3K Table). A filter for Hardy-Weinberg equilibrium showed similar sensitivity and specificity (Fig 5B), although fewer segregating sites were removed (Fig 2D), indicating the GBStools critical value we used was more conservative. Both tests performed better than expected in simulations with a similar number of samples (N = 100). This is probably due to the small number of errors that remained after applying basic filters (15 in total, see S3 Table), and the fact that over half of these errors originated from a single SNP (rs6861689) that is near a common restriction site polymorphism (BsaXI site overlapping rs6861731).
We calculated the expected SFS from the Argentine GBS data and compared it to the SFS under a neutral coalescent model, and to the SFS from 386 Argentine individuals genotyped on an exome SNP array (Fig 7B). The excess of singletons in the GBS spectrum is consistent with recent population growth,[21] but was not observed in the array data, most likely due to ascertainment bias. Another potential area where GBS can be useful is in ancestry estimation. We joined the Argentine GBS data set with SNP data from Yoruban, European, and Mexican individuals from the 1000 Genomes Project[19] phase 1 data set, and from Mayan individuals from the Human Genome Diversity Project, and performed principal components analysis (Fig 7C, methods). As expected, individuals from the admixed Argentine populations fell between the European and Native American populations in PC space.
In summary, we have used high-quality human SNP chip and whole-genome sequencing resources to test several different methods for reducing genotype errors in GBS data, including commonly-used hard filters, and a new GBS-specific statistical method implemented in our open-source program GBStools. These methodological improvements enable GBS to nearly match whole-genome sequencing in accuracy, as we have demonstrated, but at a fraction of the cost. Furthermore, our simulations suggested that GBStools has substantially better performance than hard filters in high diversity species with extensive restriction site polymorphism. Since GBStools is designed to accept data in the standard VCF format (and can optionally use read data in the standard SAM/BAM format), it can supplement many pre-existing GBS variant calling pipelines, for example the one implemented in the program Stacks[22]. We anticipate that this approach may enable many GBS-based analyses beyond high-throughput trait mapping, in particular population genetics studies such as detecting signatures of hitchhiking and selection, and estimating demographic history.
All sequencing data have been deposited in the Short Read Archive (SRA) under accessions PRJNA300277 and PRJNA303086. Exome array data have been deposited at the European Genome-phenome Archive (EGA) under accession EGAS00001001663.
Genomic DNA from eight HapMap individuals, including six samples sequenced by Complete Genomics[16] and two samples sequenced with SOLiD technology[17], was obtained from Coriell Cell Repositories. The Argentine samples were collected from 15 geographical regions in Argentina in multiple sampling efforts between 2007–2012. Under local IRB approval, blood samples were collected from participants who gave informed consent. Both HapMap and Argentine samples were de-identified and analyzed anonymously.
We used Hudson's ms[20] to generate 1×107 random samples of 200 haplotypes at a 500 bp-long locus with a population mutation rate of 1×10−3 (θ = 4Neμ) without recombination. The position of each segregating site within the locus was drawn from a uniform distribution. The first and last 6 bp of the locus represented two 6 bp-long restriction enzyme recognition sites. If any segregating site fell within these two sites, a restriction site polymorphism resulted, and either the derived or ancestral allele was randomly chosen to represent the non-cut restriction site allele. Segregating sites in the interior of the fragment, but farther than 6 bp from the ends, were chosen to represent restriction site polymorphisms with probability 0.0074 (the frequency of bases that are part of BpuEI, BsaXI, and CspCI recognition sites in the human genome). Segregating sites within 101 bp of the fragment ends represented sites sequenced by GBS with paired-end 101 bp reads. We randomly paired the 2N haplotypes to create a set of N diplotypes. Heterozygous genotypes within the 'read' portion of diplotypes that were heterozygous for one of the restriction sites were counted as genotyping errors. Simulations with population mutation rates of 5×10−3, 1×10−2, and 2×10−2 were also carried out. As most loci simulated in this manner do not carry restriction site polymorphisms it is an inefficient way to simulate large numbers of them. Thus to simulate GBS data for estimating the power of the GBStools likelihood ratio test we randomly chose one segregating site per locus to represent a restriction site polymorphism, irrespective of its location, and randomly chose either the derived or ancestral allele to be the non-cut allele. Depth of coverage was drawn from a negative binomial distribution with mean μ and scale parameter μ / 1.5 (dispersion index = 2.5). Read likelihoods were then calculated[18], assuming a constant sequencing error rate of 1×10−3.
Data for Illumina Human Exome Beadchip v1.0 (HumanExome-12v1_A) were generated for the Argentine samples at the Hussman Institute for Human Genomics, University of Miami. Genotypes were called with Illumina’s Genome Studio V2011.1 with a no-call threshold of 0.15. A minimum call rate of 99.3% was required for each sample and 386 of the 391 Argentinean samples passed this filter. Per-SNP quality filters included: mapping to a unique genomic location, and minimum per-SNP call rate of 99% (245,937 SNPs met these criteria). Of these sites, 8 were excluded from the concordance analysis for the reason that more than one sample had an exome array call of homozygous reference and a GBS call of homozygous non-reference (or vice versa).
Variation data files (masterVar) for samples NA18505, NA18508, NA19648, NA19704, NA21732, and NA21733 were downloaded from the Complete Genomics ftp site. We generated a vcf file with the mkvcf utility (v1.6.0 build 43). Before calculating concordance with GBS calls, we removed low confidence and hemizygous genotype calls, and excluded 10 sites that exhibited discordance with the GBS calls across the majority of samples. We used the unfiltered variant calls for site frequency spectrum estimation, but split multi-nucleotide polymorphisms into their component SNPs with a custom python script. We used another custom python script to predict BpuEI, BsaXI, and CspCI restriction site variants caused by bi-allelic SNPs and indels in the unfiltered calls. The sequencing of samples NA19740 and NA19836 was described previously[17]. We predicted restriction site polymorphisms caused by SNPs in these samples in the same manner.
Genomic DNA (50 ng) was digested with BpuEI (2.5 U), BsaXI (2 U), and CspCI (2.5 U) (NEB) at 37° for 90–120 min in buffer containing 20 μM S-adenosylmethionine. The digestion product was purified on a DNA Clean and Concentrate column (Zymo Research). DNA end repair, 3' monoadenylation, and ligation of sequencing adapters were performed as described in the Illumina TruSeq DNA Sample Preparation Guide. We designed a custom set of sequencing adapters, derived from the TruSeq adapters, with 65 six-bp barcodes (S5 Table). We used a standard protocol to anneal the common adapter to each of the 65 barcode adapters[8]. The ligation product was amplified by 10 cycles of PCR. For the HapMap samples, inserts between 350–650 bp were size selected on a Caliper Labchip, with one sample per gel lane. For the Argentine samples, inserts between 350–650 bp were size-selected in batches of 9–11 samples per gel lane. Bioanalyzer quantification was used to pool in equimolar amounts before and after size selection. For the 89 Argentine samples, two pools were prepared and sequenced separately, the first with 24 samples and the second with 65. Because of the high variance in read numbers per sample we observed in the Argentine libraries, we later re-analyzed the bioanalyzer data from the first set of 24 samples (S1 Fig).
Libraries were sequenced on the Illumina HiSeq 2000 in 2 x 101 bp mode following the standard TruSeq SBS protocol. The eight HapMap samples were sequenced on a single lane, with a mean of 18.3 M paired end reads per sample. In the Argentine study, the two pooled libraries were sequenced on four and five separate lanes respectively, with a mean of 17.5 M reads per sample. Reads were mapped to the human reference genome (build 37) with BWA[23] with the -q 20 parameter to include soft clipping of low quality bases. Local realignment of reads around known indels and base quality recalibration were performed with GATK[18]. We defined the target region for the HapMap samples by taking the union of predicted restriction site fragments between 400–700 bp that had ≥ 3X mean coverage, and where ≥ 10% of reads had a mate pair mapped to a restriction site (S2A Fig). The target region for the Argentine samples was defined in the same way, but with predicted fragments between 200–600 bp. Argentine samples with < 30% of reads mapped to restriction sites (26/89 samples) were excluded from further analyses.
For each of the HapMap samples in our GBS data set we inferred the number of cut and non-cut alleles at each restriction site in the genome from the Complete Genomics and SOLiD data. We then calculated depth of coverage and median insert size at each site. For this analysis we kept only sites where the median insert sizes were between 350–625 bp for each sample, and where ≤ 4 samples had zero depth of coverage. We normalized the depth of coverage for each sample by multiplying by the following normalization factor:
normij=1n∑k=1nrkjrij
(1)
Here n is the total number of samples, and rij is the total number of library inserts of size j for individual i. In calculating normij for a particular site we took j to be the median insert size of reads from individual i at that site. We then binned each site according to the mean coverage of samples that had two restriction site copies. Then, aggregating the coverage data across samples, we plotted the coverage distributions for each bin.
We called SNPs in the target regions described above with the GATK Unified Genotyper, emitting both variant and invariant sites. We also used the GATK Haplotype Caller to call SNPs in the HapMap data set. We found that specificity was higher for Haplotype Caller, with fewer true homozygous reference genotype called heterozygous (S2F Table), but also found that sensitivity was lower, with fewer true SNPs called. It is possible that this was because we used Haplotype Caller parameters that are optimal for whole-genome sequencing but not for GBS. We did not explore this point further, however, and instead used the SNP calls from Unified Genotyper for the remainder of the analyses. We performed variant quality score recalibration on segregating sites with GATK with the following training data sets (downloaded from the Broad Insitute ftp server): hapmap_3.3.b37.sites.vcf 1000G_omni2.5.b37.sites.vcf. For the HapMap samples we also trained with known variants from previous whole-genome sequencing studies[16,17]. We trained VQSR with the annotations HaplotypeScore, QD, ReadPosRankSum and HRun, and kept sites in the 99% sensitivity tranche. Invariant sites were not subjected to the VQSR filter. We applied the following hard filters (labeled as 'basic filters' in figures and tables): mapping quality ≥ 57, SNP quality ≥ 30, coverage ≥ 8X in all samples (HapMap samples) or coverage ≥ 8X in ≥ 40/63 of samples (Argentine samples). We also filtered out sites that fell within the 1000 Genomes Project callability masks for depth of coverage and mapping quality. In addition, we applied a filter for sites where the observed genotypes differ significantly from those predicted under Hardy-Weinberg equilibrium (p < 0.05), with the software package vcftools[24].
We used a custom python script to predict BpuEI, BsaXI, and CspCI restriction site variants caused by SNPs and indels in the 1000 Genomes Project data set. For each sample we created a set of genomic intervals where more than five read pairs spanned a restriction site that was polymorphic with a minor allele frequency of > 0.01. We then filtered out all sites that fell within the interval set of more than one sample.
The calculation of frequency estimates for non-cut restriction site alleles, and the calculation of the likelihood ratio test statistic for restriction site polymorphism are described in S1 Appendix. We implemented these algorithms in the python package GBStools (http://med.stanford.edu/bustamantelab/software.html). Frequency estimates for a non-cut restriction site allele are expected to be zero under the null hypothesis (no polymorphism). Since this is on the boundary of the parameter space (0, 1], the parameter estimate is expected to have a half-normal distribution. Therefore, the test statistic is expected to have an approximately one-half chi-squared distribution with one degree of freedom[25], which has a critical value of 2.71 (p = 0.05). We applied the likelihood ratio test to simulated GBS data and found that at high coverage the test statistic was equal to zero more often than expected (S6 Fig). In the 20-50X coverage range, however, it agreed well with the expected distribution. The departure from the expected null distribution at high coverage was related to the fact that more than half of the allele frequency estimates were zero (S7 Fig) and suggested that in general 2.71 is a lenient critical value (p < 0.05) for detecting restriction site polymorphisms. We performed the likelihood ratio test for SNPs where the median insert size was between 450–625 bp (HapMap individuals) or 300–500 bp (Argentine individuals) and where the median absolute deviation in insert size was less than 60 bp (S8 Fig). For the 'GBStools filter' listed in the figures and tables, we kept only SNPs that had a likelihood ratio < 2.71 and an estimated frequency of the non-cut restriction site allele < 0.05. In addition, we excluded the region spanned by the two restriction sites nearest to any site that did not meet these criteria.
We applied the likelihood ratio test described above to GBS data from the HapMap samples. We restricted the power analysis to autosomal sites that were segregating in the Complete Genomics data set, where the median GBS insert size was between 450–625 bp, and the median absolute deviation for insert sizes was ≤ 60 bp (331,861 sites). We binned the sites according to mean depth of coverage, and for each bin we calculated the power to detect known polymorphic restriction sites at a conservative critical value of 2.71 (empirical p = 0.05 critical values were slightly lower).
We calculated the expected site frequency spectrum from GBS data and Complete Genomics data for HapMap samples NA18505, NA18508, NA19648, NA19704, NA21732, and NA21733 as a subsample of size five in order to allow for missing data[26,27]. We used 1000 Genomes inferred ancestral alleles, and discarded sites where the ancestral allele was not consistent with the observed alleles. We kept sites that passed variant quality score recalibration and passed the hard filters ('basic filters'), the 1000 Genomes Project restriction site polymorphism filter, and the GBStools filter (29.2 Mb of total unmasked sites). The whole-genome sequencing (Complete Genomics) site frequency spectrum was calculated based on segregating sites in this same region. We calculated the expected site frequency spectrum for the Argentine samples as a subsample of size 40 after applying the filters shown in S2 Fig (12.7 Mb of total unmasked sites). We also calculated the expected site frequency spectrum for 386 Argentine individuals genotyped on the Illumina exome chip, as described above. We used exome chip genotypes located in both filtered and unfiltered regions.
We merged the Argentine GBS data with 1000 Genomes Project SNP data (CEU, YRI, and MXL populations), and with HGDP SNP data from sequenced Mayan individuals[28]. Of the segregating sites in the merged data set, 715,082 were present in each of the original data sets. We kept Argentine individuals that had > 25% of these sites sequenced to ≥ 7X (42/63 samples were kept). We then filtered out sites where < 90% of all samples had called genotypes. We then applied the hard filters listed previously, and pruned SNPs for linkage disequilibrium (r2 < 0.8 in 50 bp windows with 5 bp step size) with PLINK[29], resulting in a final set of 45,630 SNPs. We performed principal components analysis on this set of SNPs with smartpca[30].
|
10.1371/journal.pntd.0003765 | Multiplex Real-Time PCR Assay Using TaqMan Probes for the Identification of Trypanosoma cruzi DTUs in Biological and Clinical Samples | Trypanosoma cruzi has been classified into six Discrete Typing Units (DTUs), designated as TcI–TcVI. In order to effectively use this standardized nomenclature, a reproducible genotyping strategy is imperative. Several typing schemes have been developed with variable levels of complexity, selectivity and analytical sensitivity. Most of them can be only applied to cultured stocks. In this context, we aimed to develop a multiplex Real-Time PCR method to identify the six T. cruzi DTUs using TaqMan probes (MTq-PCR).
The MTq-PCR has been evaluated in 39 cultured stocks and 307 biological samples from vectors, reservoirs and patients from different geographical regions and transmission cycles in comparison with a multi-locus conventional PCR algorithm. The MTq-PCR was inclusive for laboratory stocks and natural isolates and sensitive for direct typing of different biological samples from vectors, reservoirs and patients with acute, congenital infection or Chagas reactivation. The first round SL-IR MTq-PCR detected 1 fg DNA/reaction tube of TcI, TcII and TcIII and 1 pg DNA/reaction tube of TcIV, TcV and TcVI reference strains. The MTq-PCR was able to characterize DTUs in 83% of triatomine and 96% of reservoir samples that had been typed by conventional PCR methods. Regarding clinical samples, 100% of those derived from acute infected patients, 62.5% from congenitally infected children and 50% from patients with clinical reactivation could be genotyped. Sensitivity for direct typing of blood samples from chronic Chagas disease patients (32.8% from asymptomatic and 22.2% from symptomatic patients) and mixed infections was lower than that of the conventional PCR algorithm.
Typing is resolved after a single or a second round of Real-Time PCR, depending on the DTU. This format reduces carryover contamination and is amenable to quantification, automation and kit production.
| Chagas disease, caused by the protozoan Trypanosoma cruzi, represents a health and social threat to an estimated number of eight million people, affecting mainly neglected populations in endemic areas and emerging in non endemic countries by migratory movements. Parasite genetic diversity is related to geographical distribution and transmission cycles and might play a role in clinical manifestations as well as in anti-parasitic chemotherapy response. T. cruzi has been classified into six Discrete Typing Units (DTUs), after consensus reached among experts in the field. In order to effectively use this standardized nomenclature, a reproducible genotyping strategy is needed. Available typing schemes are usually applied to cultured parasite stocks, because they are not sensitive enough to be used in biological specimens. Only nested PCR procedures could directly type biological samples, but are prompt to contamination and require a high number of reactions. Thus, we developed a multiplex Real-Time PCR using TaqMan probes (MTq-PCR) for DTU typing in a single or a second round of amplification. It proved useful to determine DTUs in cultured stocks, vector and reservoir specimens, as well as in patients´samples, especially in those from individuals with acute, congenital infection or Chagas reactivation. It is amenable to quantification and automation for kit production.
| Infection with Trypanosoma cruzi is a complex zoonosis, transmitted by more than 130 triatomine species and sustained by over 70 genera of mammalian reservoir hosts. T. cruzi has a broad endemic range that extends from the Southern United States to Argentinean Patagonia. The human infection, which may lead to Chagas disease, is the most important parasitic infection in Latin America with serious consequences for public health and national economies.
The diversity of the T. cruzi genome is well recognized [1–3]. Designation of ecologically and epidemiologically relevant groups for T. cruzi has oscillated between a few discrete groups [4] and many [5]. Currently, six Discrete Typing Units (DTUs) are defined [2]. In 2009, these DTUs were renamed by consensus as TcI–TcVI [6]. Several reviews already describe how these DTUs correspond with former nomenclatures and with prospective biological and host associations [6–8]. All six DTUs are known to be infective to humans and to cause Chagas disease. Further, in patients infected with DTU mixtures, different tissue distribution has been detected [9–11]. Recently a new genotype associated with anthropogenic bats (TcBat) has been detected in Brazil, Panama and Colombia and awaits further characterization for definitive DTU assignment [12–14].
The standardized nomenclature for T. cruzi DTUs should improve scientific communication and guide future research on comparative epidemiology and pathology. However, a straightforward and reproducible DTU genotyping strategy is still required. Numerous approaches have been proposed to characterize the biochemical and genetic diversity of T. cruzi isolates [15–23] with variable levels of complexity, selectivity and analytical sensitivity. Due to sensitivity constraints, most of these strategies have been applied only to cultured stocks and not directly to biological or clinical samples. Thus, their results may have underestimated parasite diversity due to possible strain selection during culture expansion [24–25]. Some methods require multiple sequential conventional PCR reactions, PCR-RFLP, hybridization or post-PCR sequencing steps; these tests are cumbersome and time-consuming, and their results are often difficult to interpret. Accordingly, we aimed to develop a novel multiplex Real-Time PCR method using TaqMan probes, allowing distinction of the six DTUs in a few steps not only from cultured stocks but also from a high proportion of biological and clinical samples.
Reference strains: Genomic DNA from a panel of reference stocks representative of the 6 T. cruzi DTUs, Trypanosoma rangeli and Leishmania spp. was used for analytical validation of the assay (Table 1).
Clinical specimens: A total of 132 clinical samples were included in the study: one tissue sample and 131 peripheral blood samples obtained from acute T. cruzi infected patients (AI, n = 13), asymptomatic (ACD, n = 64) and symptomatic (SCD, n = 27, 19 cardiac, 5 digestive and 3 mixed disease patients) chronic Chagas disease patients, congenitally infected children (CI, n = 16), and from adult patients with clinical reactivation in the context of immunosuppression (RCD, n = 11) (S1 Table).
Triatomine samples: A total of 104 triatomine derived samples were included in the study: 16 culture isolates and 88 direct samples (38 abdomen/midgut samples and 50 feces/urine samples collected on filter paper) from infected bugs (S2 Table).
Mammalian reservoir samples: A total of 71 samples obtained from T. cruzi reservoirs were included in the study: 27 culture isolates and 44 direct samples (38 peripheral blood samples and 6 heart explants) from mammalian reservoirs (S3 Table).
The study with human samples was approved by the ethical committees of the participating institutions (Comité de Bioseguridad del INLASA, Ministerio de Salud de Bolivia; Comité de Ética en Investigación de la Universidad de Granada; Comité de Ética de Investigación del Instituto Nacional de Salud Pública de México; Comité de Ética del Hospital Italiano; Comité de Bioética del Hospital Universitario Fundación Favaloro; Comité de Bioética del Instituto de Medicina Regional de la Universidad Nacional del Nordeste; Comité de Bioética de la provincia de Jujuy), following the principles expressed in the Declaration of Helsinki. Written informed consents were obtained from the adult patients and from parents/guardians on behalf of all children participants.
Preparation of DNA from biological specimens was done according to the type of sample and the operating procedures followed by the laboratories from which DNA aliquots were obtained (S1–S3 Tables). At our laboratory, peripheral blood and tissue samples were processed using High Pure PCR Template Preparation Kit (Roche, Germany) following the recommendations of the manufacturer. Triatomine feces impregnated on filter paper and abdomen samples were processed as reported [42].
Identification of T. cruzi DTUs was assessed using a conventional PCR algorithm for DTU genotyping, based on the amplification of three nuclear loci, the spliced leader intergenic region (SL-IR), the 24Sα-ribosomal DNA (24Sα-rDNA) and the A10 fragment, as reported [11,17]. Analytical sensitivity for these methods was described in Burgos et al. (2007) [17]: SL-IRac PCR: 1 pg, SL-IR I PCR: 5 pg, SL-IR II PCR: 5 pg, 24Sα-rDNA PCR: 100 fg, and A10 PCR: 1–10 pg DNA per reaction tube.
Multiple sequence alignments of the T. cruzi SL-IR, cytochrome oxidase subunit II (COII), 18S ribosomal DNA (18S rDNA) and 24Sα-rDNA genes were performed using the ClustalW algorithm in MEGA 5.2 software [43]. Reference sequences were retrieved from the GenBank database. The PrimerQuest and OligoAnalyzer tools (provided online at the website http://www.idtdna.com) were used for the final design of specific primers and probes (Table 2). To minimize nonspecific detection, the oligonucleotides were compared with all relevant sequences using the BLAST database search program (provided online from the National Center for Biotechnology Information [NCBI]).
A Real-Time PCR flowchart for identification of T. cruzi DTUs in biological samples using TaqMan probes (MTq-PCR) is shown in Fig 1. Oligonucleotide concentration and sequence information is detailed in Table 2. TaqMan probes were purchased from Integrated DNA Technologies, Inc. (USA). SL-IR and 18S-COII MTq-PCR assays were carried out using 1X QIAGEN Multiplex PCR Kit (QIAGEN, USA), while the 24Sα-III/IV MTq-PCR used 1X FastStart Universal Probe Master (Roche, Germany). All PCR reactions were carried out with 2 μL of resuspended DNA in a final volume of 20 μL. Optimal cycling conditions for the SL-IR and 18S-COII MTq-PCR assays were initially 15 min at 95°C followed by 40 cycles at 95°C for 30 sec and 60°C for 1 min in an Applied Biosystems (ABI 7500, USA) device. In turn, optimal cycling condition for the 24Sα-III/IV reaction was an initial cycle of 10 min at 95°C followed by 40 cycles at 95°C for 30 sec and 57°C for 1 min in a Rotor-Gene 6000 (Corbett, UK) device.
In order to characterize the performance of the MTq-PCR, several analytical parameters were determined [40].
The inclusivity of the assays was evaluated using 0.05–5 ng/μL of genomic DNA obtained from a panel of 39 T. cruzi stocks belonging to the six DTUs from different geographic origins (Table 1). On the other hand, 1–5 ng/μL of genomic DNA obtained from T. rangeli, L. major, L. amazonensis, L. brasiliensis and L. mexicana, was used to assess the specificity of the assays. Specificity was also tested using human DNA from a seronegative patient as template.
Analytical sensitivity and reaction efficiency were evaluated using 2-fold, 10-fold and 100-fold serial dilutions spanning 1 μg to 1 fg of genomic DNA per reaction tube obtained from T. cruzi stocks belonging to different DTUs, depending on the assay. Moreover, in the case of TcI, four stocks representing TcIa, TcIb, TcId and TcIe genotypes based on the polymorphism of the SL-IR gene were analyzed [30]. In addition, in the case of TcIV, DNA from strains representing populations from South America (TcIV-SA) and North America (TcIV-NA) were used [46]. Each concentration was tested in duplicate.
Fig 1 illustrates the MTq-PCR flowchart designed to distinguish among the six T. cruzi DTUs.
Inclusivity and specificity results are shown in Table 3. T. cruzi I, including stocks representing SL-IR genotypes TcIa, TcIb, TcId and TcIe, were detected by the FAM fluorescence signal in the SL-IR MTq-PCR assay and did not amplify in the downstream reactions of the flowchart. The TcII/V/VI group was detected with the HEX-labeled probe in the SL-IR MTq-PCR. The 18S-COII MTq-PCR assay distinguished TcII (FAM + Cy5 signals) from TcV (HEX signal) and TcVI (FAM signal only). There were two groups of TcIII strains, one group reacted only with the SL-IR TcIII-Quasar670 probe, and the other one composed by three strains (from Brazil, Paraguay and Argentina), reacted with both TcIII-Quasar670 and TcIV-CAL Fluor Red610 SL-IR probes. Thus, the latter group of strains was identified as TcIII after a second round of amplification using the 24Sα-FAM probe. CAL Fluor Red610 and HEX fluorescence signals were detected when the assay contained DNA from TcIV-SA and TcIV-NA strains in the SL-IR and the 24Sα-III/IV MTq-PCR assays, respectively.
TcV was amplified and detected with the FAM probe in the 24Sα-III/IV MTq-PCR assay. Besides, TcIII and TcIV were also detected with the 18S-HEX probe in the 18S-COII MTq-PCR. Specificity of the MTq-PCR was not affected since all these DTUs are confirmed in a previous stage.
On the other hand, MTq-PCR was tested with purified DNA from T. rangeli, L. amazonensis, L. major and L. mexicana stocks and from a seronegative patient. No detectable fluorescence signals were obtained for any of them, indicating the specificity of the assays (Table 3).
Analytical sensitivity and reaction efficiency were estimated separately for each of the three MTq-PCR reactions using genomic DNA from reference stocks representing the six T. cruzi DTUs: TcIa (K98), TcIb (Cas16), TcId (G), TcIe (PALV1 cl1), TcII (Tu18), TcIII (M5631), TcIV-SA (CanIII), TcIV-NA (Griffin), TcV (PAH265) and TcVI (CL-Brener). The SL-IR MTq-PCR yielded a positive result starting from 1 fg DNA/reaction tube of TcI reference strains with an efficiency (Eff) of 108% (TcIa), 104% (TcIb), 99% (TcId) and 98% (TcIe). Similar sensitivity was obtained for strains representing TcII (Eff: 90%) and TcIII (Eff: 97%). In the cases of TcIV-SA, TcV and TcVI, sensitivity was lower (1 pg DNA/reaction tube) with Eff of 80%, 88% and 86%, respectively (Fig 2). The 18S-COII MTq-PCR reaction rendered a sensitivity of 100 fg DNA/reaction tube for strains representing TcV (Eff: 82%) and TcVI (Eff: 83%) and 1 pg DNA/reaction tube for TcII (Eff: 77% and 70% using the 18S-FAM and the COII-Cy5, respectively) (Fig 3A). The 24Sα-III/IV MTq-PCR method was capable of detecting 100 fg DNA/reaction tube of the TcIII (Eff: 92%) and TcIV-SA (Eff: 81%) stocks, whereas TcIV-NA was detected at concentrations ≥ 1 ng/reaction tube (Eff: 78%) (Fig 3B).
A total of 307 biological specimens, including clinical samples (n = 132) as well as samples obtained from different species of vectors (n = 104) and mammal reservoirs (n = 71) from different endemic regions were evaluated using MTq-PCR and a conventional PCR based strategy [11, 17].
As a consequence of the standardized nomenclature for the six T. cruzi DTUs having been ratified by a committee of experts [6], it became imperative to develop a reliable genotyping strategy that could be adopted by the research community [8]. Throughout the past years, several typing schemes have been developed. A PCR assay system based on the amplification of particular regions of the SL gene and 24Sα-rDNA [44] and 18S rDNA [48] was first proposed [15] in which the size polymorphisms of the amplification products were suitable for T. cruzi assignment into each of the six DTUs. A multilocus PCR-RFLP analysis of genetic polymorphism of 12 loci also was proposed for DTU genotyping [16]. Additionally, a three-marker sequential typing strategy was proposed consisting of PCR amplification of the 24Sα-rDNA and PCR-RFLP of the heat shock protein 60 and glucose-6-phosphate isomerase loci [18]. Yeo et al. (2011) and Lauthier et al. (2012) designed Multilocus Sequence Typing (MLST) schemes in which sequence information of 4 to 10 single copy housekeeping genes allowed the resolution of the six DTUs [21–22]. A recent assay that uses a single copy gene (TcSC5D) followed by two RFLP reactions has been reported [23]. However, most of the above mentioned assays are complex to perform and have been applied only to cultured parasites. Another scheme using nested-hot-start PCR assays allows direct DTU typing in biological [25, 49] and clinical [11, 17] samples but requires between 3 and 9 sequential PCR reactions.
To overcome these difficulties we developed a novel MTq-PCR approach that identifies the six T. cruzi DTUs in a single or two sequential reactions with adequate sensitivity to analyze different types of biological samples, such as those derived from triatomine vectors and different type of wildlife, livestock, pets and human tissues. The Real-Time format reduces PCR associated contamination and is amenable to quantification, automation and kit production. A first round allows distinction of TcI strains from those belonging to TcIII/IV or TcII/V/VI groups, which are discriminated after a second MTq-PCR round. The method was inclusive for a panel of 39 T. cruzi stocks. In particular, the TcI primer/probe set was inclusive for all TcI SL-IR genotypes [30], and the TcIV primer/probe set was inclusive for TcIV strains from South and North America [46]. Besides, the test did not recognize human, T. rangeli and Leishmania spp. DNAs.
MTq-PCR methods showed an analytical sensitivity ranging from 1 fg to 1 pg DNA per reaction tube depending on the DTU being analyzed. As an exception, TcIV-NA was detected at concentrations ≥ 1 ng/reaction tube by the 24Sα-III/IV MTq-PCR. The analytical sensitivity for the conventional PCR scheme used in this study was reported in Burgos et al. (2007) [17] and ranged from 100 fg to 10 pg DNA per reaction tube depending on the reaction and the DTU under analysis. Thus, both PCR algorithms used in the present study showed similar ranges of sensitivity when compared at analytical levels.Out of 210 biological samples that could be typed by both algorithms, 24 (11.4%) gave inconclusive TcII/V/VI, TcII/VI, TcV (or TcV plus TcVI) and TcIII (or TcIII plus TcI) results by either conventional or MTq-PCR. In nine samples, conventional PCR was not able to discriminate between single TcV infection and a mixture of TcV plus TcVI. However, MTq-PCR confirmed TcV in seven of these samples thanks to specific detection of the 18S-HEX probe. One sample, typed as TcII/V/VI by conventional PCR, could be resolved as TcII/VI by MTq-PCR. Furthermore, an indeterminate TcII/VI and 2 TcIII (or TcIII plus TcI) samples were confirmed as TcII, TcI and TcIII, respectively, by MTq-PCR (S1–S3 Tables). On the other hand, 7 TcVI and one TcII samples typed by the conventional PCR algorithm were classified as indeterminate TcII/VI by the MTq-PCR (S1–S3 Tables).
Finally, both algorithms confirmed mixed infections in one patient from Jujuy, Argentina (TcV plus TcVI), in one cat from Mexico (TcI plus TcII) and in several sylvatic vector species, such as TcI plus TcIII, TcI plus TcIV and TcIII plus TcIV (S1–S3 Tables). In general the MTq-PCR detected mixed infections in a lesser extent than the conventional PCR scheme. Oligonucleotide interactions, competition for reagents, different amplification efficiency of the targets, and accumulation of amplicons of the predominant target that inhibit Taq polymerase are factors that might be involved.
The MTq-PCR test was less sensitive than conventional PCR algorithm for direct typing of peripheral blood samples of a proportion of chronic Chagas disease patients harboring low parasite loads. We have evaluated the analytical sensitivity of the assay using mixtures of T. cruzi DNA with DNA extracted from human blood from non-infected subjects and no differences in analytical sensitivity were found (S1 Fig). This suggests that the lower clinical sensitivity of the assay in blood samples would not be due to inhibitory substances present in the samples. In some human cases tested in this study, we can not discard some DNA degradation with respect to the period where the extracts were analyzed using conventional PCR algorithm [50].
The findings herein obtained, promote MTq-PCR as a valuable laboratory tool for distinction of T. cruzi DTUs. It appears adequate in surveillance and identification of outbreaks sources [51] or to follow-up acute infections of seronegative recipients that receive infected organs from seropositive donors [52].
|
10.1371/journal.pntd.0000582 | Differential Cytokine Gene Expression According to Outcome in a Hamster Model of Leptospirosis | Parameters predicting the evolution of leptospirosis would be useful for clinicians, as well as to better understand severe leptospirosis, but are scarce and rarely validated. Because severe leptospirosis includes septic shock, similarities with predictors evidenced for sepsis and septic shock were studied in a hamster model.
Using an LD50 model of leptospirosis in hamsters, we first determined that 3 days post-infection was a time-point that allowed studying the regulation of immune gene expression and represented the onset of the clinical signs of the disease. In the absence of tools to assess serum concentrations of immune effectors in hamsters, we determined mRNA levels of various immune genes, especially cytokines, together with leptospiraemia at this particular time-point. We found differential expression of both pro- and anti-inflammatory mediators, with significantly higher expression levels of tumor necrosis factor α, interleukin 1α, cyclo-oxygenase 2 and interleukin 10 genes in nonsurvivors compared to survivors. Higher leptospiraemia was also observed in nonsurvivors. Lastly, we demonstrated the relevance of these results by comparing their respective expression levels using a LD100 model or an isogenic high-passage nonvirulent variant.
Up-regulated gene expression of both pro- and anti-inflammatory immune effectors in hamsters with fatal outcome in an LD50 model of leptospirosis, together with a higher Leptospira burden, suggest that these gene expression levels could be predictors of adverse outcome in leptospirosis.
| Leptospirosis is a widespread bacterial infection that is transmitted by soil or water contaminated by the urine of infected animals, or directly from these animals. It has highly diverse clinical presentations, making its differential diagnosis difficult. Though most cases are minor and self-resolving, there are also severe forms that include a sepsis pattern and multiple organ failure, and have possible fatal outcomes. Predictors of disease evolution and outcome are scarce, yet they would be very valuable to clinicians as well as to better decipher disease pathogenesis. In this study, we used a hamster model of leptospirosis to evaluate if immune genes were differentially expressed between individuals and if their expression levels could help forecast the outcome of the disease. We found that hamsters that later died from leptospirosis had significantly higher expression levels of both pro- and anti-inflammatory mediators compared to survivors. These results suggest that expression levels of these immune effectors might be helpful predictors of outcome in leptospirosis and that septic shock contributes to fatal leptospirosis.
| Leptospirosis is the most widespread zoonosis occurring worldwide with possible fatal outcomes [1]. Though most often an endemic disease, epidemics have been associated with particular meteorological events [2]–[4] or clusters of cases related to occupations or leisure activities [5]–[8]. It is notably highly prevalent in tropical areas, but some of the clusters of cases have been reported in temperate countries [7],[8]. Its clinical presentation is highly variable and is often initially suggestive of influenza, malaria or dengue fever, making the differential diagnosis more hazardous in tropical countries, during dengue or influenza epidemics, or in areas of high malaria incidence [9]. However, because of a relatively high fatality rate in leptospirosis, increased medical care must be provided to some of the patients suffering leptospirosis. Validated prognostic factors to help forecast the evolution of a leptospirosis are scarce. Yet, they would be valuable for clinicians to decide whether their patients should only be treated with antibiotics, kept at the hospital in a standard unit or directed to an intensive care unit. Few data have been published addressing this question; furthermore, contrasting observations were obtained [10],[11].
Severe leptospirosis manifestations include acute renal failure, caused by acute interstitial nephritis and pulmonary haemorrhage. Spirochete invasion and toxicity of outer membrane components cause robust inflammatory host responses [12] leading to clinical manifestations reflecting a sepsis syndrome. This latter condition has been characterized as a dysregulation of the inflammatory response, including a massive release of pro-inflammatory cytokines that induces multiple organ dysfunctions. Concomitantly, compensatory mechanisms, mostly regulatory cytokine-mediated (although having protective effects to prevent overwhelming inflammation) may become deleterious and have been associated with an immune paralysis and poor outcome [13],[14].
Cytokines are potent, pleiotropic, non-antigen-binding polypeptides secreted by cells of the immune system and are responsible for cell activation, differentiation and proliferation after they act on their target cells via specific receptors primarily through autocrine and paracrine stimulation. Their expression levels are largely studied in the context of septic shock and severe sepsis both for their possible prognosis value [13], [15]–[17], as well as for a better understanding of sepsis physio-pathology [18]–[20]. Because of high similarities in clinical presentation of septic shock and severe leptospirosis, we hypothesized that strong similarities in immune gene expression between severe sepsis and severe leptospirosis could help predict the evolution of leptospirosis towards multiple organ failure or recovery.
Tumor Necrosis Factor α (TNFα) is a cytokine involved in early systemic inflammation that stimulates the acute phase reaction, in synergy with interleukin -1 (IL-1) and interleukin-6 (IL-6) [21]. Elevated plasma concentrations of TNFα have been associated with poor prognosis in sepsis, but also in patients with leptospirosis [10]. IL-6, one of the most important pro-inflammatory mediators of the acute phase response to pathogens, has been suggested to be a downstream mediator of TNFα and IL-1. However, it also regulates anti-inflammatory effectors by controlling the level of pro-inflammatory cytokines. Many studies were conducted to evaluate the value of circulating IL-6 concentrations as indicators of clinical outcome in patients with severe sepsis, correlating high levels with fatal outcomes. Interferon-γ (IFN-γ) is a pluripotent pro-inflammatory cytokine [22]. Its production was shown as dependent on IL-12p40 in human blood stimulated by L. interrogans [23] notably inhibiting Th2 cell activity. Cox-2, one of the two forms of cyclooxygenase (COX), is highly induced and rapidly produced in macrophages and endothelial cells in response to proinflammatory cytokines and may be responsible for the oedema and vasodilatation associated with inflammation. It is recognized that inflammatory mediators such as COX-2 but also nitric oxide, a derived product of inducible nitric-oxide synthase (iNOS), are responsible for the symptoms of many inflammatory diseases [24],[25]. Increased level of nitric oxide have notably been evidenced in the sera of patients with severe leptospirosis [26].
Anti-inflammatory effectors play an important role to counter-regulate the effects of pro-inflammatory cytokines. Interleukin-10 (IL-10) is classically described as an anti-inflammatory cytokine with pleiotropic effects in immunoregulation and inflammation by down-regulating the expression of Th1 cytokines [27]. An early imbalance of IL-10 in sepsis was shown to be associated with death despite TNFα production. [17]. Transforming Growth Factor β (TGF-β) is believed to be important in the regulation of the immune system by regulatory T cells; it notably acts by blocking the activation of lymphocyte- and monocyte-derived phagocytes and by controlling iNOS expression [28]. Together with IL-10, TGFβ is considered as contributing to the immunosuppression observed in septic shock [13],[14].
The hamster is considered as the most valuable animal model for human leptospirosis [29],[30]. This animal model is notably used to maintain virulence of Leptospira strains or isolates. It was also used in studies aiming at better deciphering the virulence and pathogenesis mechanisms or the host immune response to leptospirosis or to vaccine candidates [30]–[37]. Using this animal model, our group [38] notably demonstrated in vivo the expression of Th1 cytokines during acute leptospirosis.
The aim of our study was, using a LD50 model of leptospirosis in hamsters, to evaluate gene expression levels in individual animals. Additionally, the Leptospira burden in blood was also assessed because it was shown to have a prognostic value [39]. The immune gene expression levels and Leptospira burdens were compared according to the spontaneous outcome of the leptospirosis. Differential expression levels were observed that related to the outcome of the infection.
The virulent Leptospira interrogans serovar Icterohaemorrhagiae strain Verdun, was obtained from the Reference Collection of the Institut Pasteur in Paris, France. Virulence was maintained by passages in Syrian golden hamsters (Mesocricetus auratus) and was regularly tested by lethal injection of 2×108 leptospires intraperitoneally. An avirulent variant corresponding to an isogenic clone of this strain was derived from the virulent strain by in vitro high-passage.
Leptospires were cultivated in liquid EMJH (Ellinghausen McCullough Johnson and Harris) medium at 30°C under aerobic conditions [40]. The bacterial cell density of the cultures was assessed in a Petroff-Hausser counting chamber.
Specific pathogen-free animals which parents were initially purchased from Charles River Laboratories (Charles River Wiga GmbH, Sulzfeld, Germany) were bred at the Institut Pasteur of New Caledonia. All in vivo studies were carried out using five- to six-week-old outbred golden hamsters handled in individual cages. During preliminary experiments, we infected hamsters by intraperitoneal injection of various doses of live virulent Leptospira ranging from 2×107 to 2×108 per hamster. Hamsters were checked four times a day to evaluate the appearance of clinical signs, deep unconsciousness or recovery. Deeply unconscious animals that did not react to a tactile stimulus were considered dead and euthanized. Additional preliminary experiments included the determination of the time course of gene expression by sampling three individual infected hamsters at 0, 4, 8, 14 hrs then day 1 (D1), D2, D3, and D4 post-infection to determine the most relevant time point allowing to evaluate the expression level of as many relevant genes as possible for future experiments. Each LD50 experimental set was composed of six- to eighteen animals intraperitoneally infected with 108 leptospires of a virulent culture in EMJH medium, and three or four negative controls injected with sterile EMJH medium. The study was made up of three independent experiments. Whole blood (400 µl) was collected on PAXgene blood RNA tubes (PreAnalytiX, Qiagen, Australia) by cardiac puncture under non lethal gas anaesthesia [29] on D3 after infection. Clinical symptoms and/or death were monitored four times daily for 21 days. Surviving animals at D21 were considered as spontaneously recovering. Negative controls and surviving animals were euthanized at D21. In order to evaluate the effect of the infective dose on the gene expression patterns, we used two other experimental infection models. We first injected five hamsters with a LD100 of the same virulent Leptospira, with 2×108 leptospires injected per hamster using the same intraperitoneal route. Secondly, we injected hamsters with a similarly high dose (2×108 leptospires per hamster) of the high-passage isogenic Leptospira variant, known not to induce any mortality. Control animals were similar to the LD50 experiments and were injected with an equal volume of sterile EMJH. Protocols for animal experiments were prepared and conducted according to the guidelines of the Animal Care and Use Committees of the Institut Pasteur and followed European Recommendation 2007/526/EC that provides “guidelines for the accommodation and care of animals used for experimental and other scientific purposes”. The protocol was validated before the start of the experiments by a scientific committee and an animal care committee of the Institute Pasteur in New Caledonia.
Total RNA was isolated from whole blood not later than 24 hours post-collection using the PAXgene Blood RNA system (PreAnalytiX) according to manufacturer's instructions, then immediately frozen at −80°C until use. DNase-treated RNAs were used to synthesize cDNA with the Transcriptor First Strand cDNA Synthesis Kit using random hexamers as specified by the manufacturer (Roche Applied Science). To minimize variation in the reverse transcription reaction, all RNA samples from a single experimental setup were reverse transcribed simultaneously and in duplicate.
The sequences of all primers used in this study are listed in table 1. They were designed with the LightCycler Primer Probe Design Software 2.0 (Roche Applied Science), selected according to intron spanning and GC%, and synthesized by Proligo Singapore Pte Ltd (Biopolis way, Singapore).
External standard curves either for household or effector genes consisted of serial dilutions of specific purified DNA ranging from 107 to 1 copies as described previously [38]. The copy number of each standard was calculated by standard methods using the Avogadro constant and the size of the amplified target as described [41]. Each standard curve was validated using established criteria (specific melting temperature, size of the PCR product, a mean error ≤0.03 and a slope near −3.3).
PCR amplifications and analysis were achieved using a LightCycler 2.0 instrument (Roche Applied Science) with software version 4.05. All reactions were performed in duplicates with the LightCycler FastStart DNA Master SYBR Green I kit (Roche Applied Science) in a final 20 µl volume with 4 mM MgCl2, 0.5 µM of each primer and 2 µL cDNA or 2 µL DNA standard dilution. Cycle conditions were optimized for each target, either immune mediator or β-actin. Amplification conditions consisted of an initial pre-incubation at 95° for 10 min (polymerase activation) followed by amplification of the target cDNA for 45 cycles (95°C for 8 s, 60°C for 5 s and a variable extension time at 72°C). Extension periods varied for each PCR depending on the length of the expected amplicon (∼1 s/25 bp) as shown in table 1.
Leptospiraemia was also determined after cDNA amplification with a PCR specific of a 331 bp sequence of the L. interrogans rrs (16S) gene [42] using a LightCycler 480 II instrument with software version 1.5.0. Amplification reactions were performed in duplicates with the LightCycler 480 SYBR Green I Master kit in a final 10 µL volume with 0.5 µM of each primer, 1 µL cDNA as follow: a 10 min enzyme activation at 95°C then 50 amplification cycles, each made of 8 s at 95°C, 5 s at 62°C and 12 s at 72°C. A negative control with PCR-grade water instead of cDNA was included in each run. With either instrument, melting peaks were automatically plotted by the software and used to assess the specificity of the amplified product.
Absolute quantification of each target was done using the comparative cycle threshold (CT) method: the concentration of a given target mRNA in any unknown sample was calculated by comparing its CT with the corresponding standard curve. Relative expression was calculated as the ratio of the target mRNA copy number to β-actin mRNA copy number. This ratio was then normalized using the same ratio calculated in uninfected controls (used as calibrators). This expression of the results allows directly providing an n-fold change ratio in gene expression of the experimental animals compared to their control counterparts. In this study, β−actin was used as the household reference gene since former work demonstrated that no significant difference was observed using either HPRT or β−actin [38].
The outcomes were defined as the spontaneous outcome of the infection and were either death (“nonsurvivors”) or spontaneous recovery (“survivors”). The results of three independent LD50 experiments were pooled and compared according to the outcome using Student's t test and Kruskal-Wallis test on Stata SE/8.0 for Windows (Stata Corporation, Texas, USA). The overall survival curve for all LD50 experiments was also plotted with 95% confidence intervals using Stata SE/8.0.
Preliminary experiments demonstrated that a dose of 108 live virulent leptospires per hamster injected by the intra-peritoneal route led to ca. 50% mortality, a dose therefore used for our LD50 experiments. The first signs of illness (prostration and anorexia) were observed at day 3 post-infection. This dose was confirmed in further experiments as being a relevant and reproducible model of LD50 (figure 1A). The relative normalized gene expression levels at various time points are summarized in figure 1B. After a rapid and intense rise to its maximum, TNFα expression rapidly decreased before to slowly and regularly increase again up to D3. A similar pattern was observed for IL-1 and IL-6 with much a higher amplitude of regulation. After no significant modulation during the first 14 hrs post-infection, IL-10 and COX-2 were expressed at a maximum level around D3 after a steady increase notably for IL-10. IFNγ expression levels were poorly modulated along this time-course. However its maximum expression level became relatively stable around D3. Reproducible results were obtained for all these effectors with RNA extracted from 400 µL whole blood, except IL-2 and IL-4, because of their low mRNA copy numbers. Additionally, the peak of IL-12p40 gene expression was observed very early at 4 hours post-infection. Therefore these 3 latter effectors (IL-2, IL-4 and IL12p40) were not analysed in further studies.
Taken together, these results led to determine D3 as a relevant time point for future studies, a consensus when most of the relevant effectors could be efficiently monitored and the appearance of the first clinical signs before any mortality occurs.
In the LD100 experiment, all 5 infected hamsters died at D5. As expected, all hamsters infected with a similarly high dose (2×108 per hamster) of the high-passage isogenic variant survived until D21 and were euthanized.
In total, 36 infected hamsters were included from our three independent LD50 infection challenges. Twenty two of them died at 6.1 (range 4.04–6.92) days post-infection (nonsurvivors), whereas 14 were considered as spontaneously recovering being alive at D21 (survivors). Gene expression levels evidenced that the pro-inflammatory cytokines IL-1α and TNFα but also the enzyme Cyclooxygenase-2 and the cytokine IL-10 were expressed at significantly higher levels (p<0.01, see table 2) in nonsurvivors when compared to survivors (Figure 2A).
Considering the basic criterion that a 2-fold change in transcript abundance represents differential expression [43],[44], the gene expression levels in survivors were not significantly different from controls (i.e. relative normalized gene expression levels in the range 0.5–2, see table 2).
As expected, the live Leptospira burdens, as evaluated by the ratio of Leptospira 16S rRNA to hamster β-actin, were nil in controls and were also significantly (p<0.01, see table 2) lower in spontaneously recovering survivors (figure 2B).
Contrastingly, the expression of the cytokines IFNγ and TGFβ appeared poorly modulated, their expression levels in infected animals being not significantly different from that in control animals (ratio not different from 1). Additionally, no difference in expression levels was observed between survivors and nonsurvivors after the Leptospira LD50 challenge (table 2 and figure 3). Though IL-6 is notably induced as a response to the LD50 infectious challenge (i.e. ratio significantly higher than 1.0), similar levels (p>0.1, see table 2) were observed in hamsters whatever the outcome of the LD50 infection. However and interestingly, two out of our three independent experiments, higher IL-6 expression levels were observed in nonsurvivors in two out of our 3 independent experiments and a very high expression level in a few survivors accounted for the similar average expression (see figure 3 insert).
Hamsters infected with a high (LD100) dose of virulent Leptospira displayed a gene expression pattern very similar to the one observed in nonsurvivors after the LD50 infection challenge. They displayed a similarly increased gene expression of IL-1α, TNFα and Cox-2 and a very similar Leptospira burden (figure 2). Their mean IL10 gene expression level was higher than in nonsurvivors after the LD50 challenge but this difference was not significant, due to high inter-individual variability. IFNγ and TGFβ expression levels were very poorly modulated, again not significantly different from uninfected controls (figure 3). Similar to IL-10, IL-6 gene expression was largely increased compared to animals infected with a lower dose but a high inter-individual variability was also noted.
In hamsters infected with a similarly high dose of the high-passage non-virulent Leptospira variant, the gene expression pattern was similar to the one displayed by survivors of the LD50 challenge. Surprisingly, a leptospiraemia was still observed at D3 in most of the animals, though no clinical sign was noted and no mortality occurred.
We first developed a LD50 model of leptospirosis in hamsters. Used together with a non-lethal blood sampling technique, it allowed the acquisition of individual gene expression patterns during the course of acute leptospirosis. Using this model, we demonstrated that the expression of some immune genes in blood, together with the Leptospira burden in blood of infected animals could be correlated with the outcome of the infection.
The hamster is recognized as a good animal model for severe human leptospirosis [29]. Using the virulent Leptospira interrogans Icterohaemorragiae strain Verdun [45],[46], we determined the dose of 108 live leptospires injected via the intra-peritoneal route as leading to ca. 50% mortality. This challenge technique proved to be reproducible in 5 to 6-week old Syrian hamsters and was used for our study. When infected this way, the first clinical signs in hamsters held in individual cages (anorexia, then prostration and ruffled fur) were observed from 3 days post-infection on. This 3-day post-infection time point was also shown, in another experimental hamster model of leptospirosis, to be the time point for the first detection of Leptospira mRNA in target organs and a relevant time point for immune gene expression studies [47]. The use of a non-lethal sampling technique together with the follow up of individual hamsters allowed relating the gene expression levels observed with the individual outcome. We additionally conducted two experimental infection experiments for comparison purpose, one using a high dose of the virulent strain (2×108 live leptospires via the intra-peritoneal route) leading to 100% mortality and a similarly high dose of an isogenic avirulent variant causing no mortality.
During these preliminary experiments, we determined that a 400 µL blood collection under gas anesthesia at this time point would not be too deleterious to the animals and was sufficient to allow the extraction of mRNA in adequate quantity and quality for gene expression studies. Based on our previous knowledge [38] and with additional preliminary experiments, we determined that TNFα, IL-1α, IL-6, IL-10, IFNγ, TGFβ and Cox2 gene expression levels could successfully be quantified using this experimental procedure. These targets were chosen for their relevance in studying our hypothesis of similarities between severe leptospirosis and septic shock. Only those evidencing a significant number of mRNA copy numbers at this time point, therefore allowing accurate determination of the gene expression levels, were studied.
The Leptospira burden was reported to be of prognostic value in human leptospirosis [39]. In our study, it was evaluated by q-RT-PCR targeting the 16S-rRNA allowing to evaluating the burden of mostly live leptospires, bacterial rRNAs being very short-lived when cells have reduced activity or die [48],[49]. Using a ratio of Leptospira 16S rRNA copy number to host β-actin mRNA copy number allowed a comparison between individuals. As expected and observed in humans [39], significantly higher Leptospira burdens were noticed in nonsurvivors. Interestingly, a leptospiraemia was still observed at D3 in animals infected with a high dose of the avirulent Leptospira variant, suggesting that this high-passage variant, though not lethal, has retained some degree of pathogenicity. This also suggests that leptospiraemia might be strain- and virulence-dependent, possibly jeopardizing its use as a tool for prognosis, when the virulence of the infecting strain is not known.
The immune response to an infection is nowadays considered as precisely modulated rather than simply induced. Cytokines expression levels are largely studied in the context of septic shock and severe sepsis both for their possible prognostic value [13], [15]–[17] and as a way to improve our understanding of host-pathogen interactions. Actually, sepsis is now recognized as associated with an exacerbated production of both pro- and anti-inflammatory cytokines and the prognostic value of some of these is widely recognized [20].
Using reverse transcription-real-time PCR, the transcripts can be quantified directly in a biological sample, providing information about the in vivo immune response mechanisms of the individual. Studying the immune response is only possible at the transcriptional level in our animal model, due to the lack of tools for assessing serum conentrations of immune efectors. however, it is also probably more sensitive to evaluate the fine-tuning of the immune response because the amount of circulating cytokines only represents a minor part of the total amount of cytokines produced [50].
Our results in the LD50 model evidenced differential gene expression according to the outcome. TNFα, IL-1α, Cox-2 were expressed at significantly higher levels in nonsurvivors than in spontaneously recovering hamsters. Using the other two infection models, the results demonstrated similar gene expression levels in animals challenged with a LD100 and in nonsurvivors after a LD50 challenge on one hand, and on the other hand in survivors after a LD50 and animals infected with a high dose of the avirulent variant, both actually surviving. These similarities using different doses and strains confirm the validity of our results.
Our IL-1 RT-PCR targets IL1α, one of the two main active forms in the IL-1 family. IL-1β is most often considered as the prototypic IL-1 effector because it is released in the bloodstream, whereas Il-1α mostly remains cytosolic with an autocrine activity or is bound to the cell surface. Though IL-1β was more frequently considered as an indicator of Il-1 activity, IL-1α was shown to have an action very similar to the action of the more largely studied IL-1β. Furthermore, it was shown that its gene expression is quite similarly regulated [51]–[53]. TNFα and IL-1 are prototypic pro-inflammatory mediators that have been reported to have a prognosis value in sepsis [16],[54], even if their clinical relevance was also questioned [55]. Interestingly, TNFα was also reported to have a similar prognosis value in leptospirosis [10], though these results were not confirmed in other studies [11]. Cox-2 is a highly inducible enzyme involved in the early phase of the inflammatory response. Notably induced by IL-1 and TNFα through the NFκB pathway, its induction can be considered as an end-result of the initial pro-inflammatory response. Interestingly, a significant induction of Cox-2 was observed only in nonsurvivors whatever the infective dose. This further suggests the probable contribution of a sepsis-like mechanism in severe leptospirosis. IL-10 is expressed at higher levels in nonsurvivors compared to the survivors in the LD50 model. These results are in agreement with several studies showing an exacerbated production of anti-inflammatory cytokines resulting in aggravation of a systemic disease and adverse outcome in febrile patients [14],[17]. The decreased production of Th1 cytokines in many cellular types was reported as an IL-10-induced adaptative immune response, by interfering on antigen-presenting cells and T cells, possibly via inhibition of NFκB nuclear translocation [14],[56]. This immunoregulatory role of IL-10 was clearly established after an anergy was observed in stimulated T cells in the presence of IL-10. Moreover, Il-10 production in innate immune response to a stimulus promotes the expansion of regulatory T cells, amplifying the anti-inflammatory effect of IL-10 [13],[14],[56].
However, the results observed with the other two infection models suggest a possible effect of the infective dose on IL-10 expression levels, higher levels being noted in animals infected with a high Leptospira dose (either virulent or not) when compared to their respective counterparts with the same outcome in the LD50 model. The IL-10/TNFα ratio has been proposed as a prognosis indicator in sepsis [17],[54] and in leptospirosis [57]. A high IL-10/TNFα ratio was reported as correlated with poor outcome in septic patients, contrary to the opposite results obtained in one limited study reported in leptospirosis [57]. This ratio was generally regarded as reflecting a persistent secretion of IL-10 at later time points, when concomitant down-regulation of TNFα occurs. From our experiments, this ratio, at least at the transcriptional level, does not appear relevant, both cytokines being induced in animals with fatal outcome, at least on the basis of our D3 time-point.
In our LD50 experiments, IL-6, often reported as the best marker of the severity of infectious (or even non-infectious) stress in humans and to have a prognosis value in sepsis [20], was not differentially expressed according to the outcome. However, our results merely reflect a very high inter-individual variability in IL-6 expression. Contrastingly, IL-6 expression in hamsters infected with (and dying from) a LD100 of Leptospira actually showed a highly increased expression, though again with a very high variability. Similarly, high fluctuations of bioactive IL-6 levels were reported in serum from septic patients [58]. This high variability would also be limiting the value of IL-6 as a predictor in leptospirosis.
In our model, IFNγ and TGFβ had no prognostic value and were similarly expressed in infected hamsters whatever the dose or the outcome. Interestingly, the gene expression of IFNγ and TGFβ appeared not being significantly regulated based on the common 2-fold variation criterion [43],[44]. Because IFN-γ gene expression is antigen-presenting cells dependent, we could hypothesize that high Il-10 levels limited its expression, though it could have been beneficial for an optimal defense against infection.
However, some cytokines have been shown not to be regulated at a transcriptional level and post-transcriptional regulation also plays a major role in cytokine cascades even after a transcriptional regulation has occurred. Unfortunately in the hamster, our animal model, no tool is available to evaluate the serum concentrations of the effectors studied. On one hand, gene expression techniques allow a rapid and highly sensitive study of immune gene transcriptional regulation, notably because the circulating part of cytokines is considered as being the “tip of the iceberg” [50]. On the other hand, some immune mediators of the septic shock are known to be poorly regulated at the transcriptional level and can therefore not be studied in our model. As an example, High-mobility group box (HMGB)-1 is primarily known as a nuclear DNA-binding protein with a transcription regulatory activity, but can also be excreted by stimulated macrophages, then displaying a cytokine activity, notably inducing the release of TNFα and IL-6 [59]. It has been proposed as a prototypic late mediator of inflammation in severe sepsis [60], its delayed and prolonged release in established sepsis rising an increasing interest as a prognostic indicator or a therapeutic target in late-phase inflammation processes. Its usefulness as a prognostic indicator or as a therapeutic target remains unexplored in leptospirosis.
Though obtained in an animal model with Leptospira strains of known and relatively low virulence, these encouraging results prompted us to initiate a clinical study aiming at investigating the prognostic value of these effectors in patients with confirmed leptospirosis. Cytokines will similarly be studied at the gene expression level, but also by measuring their serum concentration levels, a technique much easier transmissible to health centers. This study is currently underway.
|
10.1371/journal.pgen.1000788 | Feedback Inhibition in the PhoQ/PhoP Signaling System by a Membrane Peptide | The PhoQ/PhoP signaling system responds to low magnesium and the presence of certain cationic antimicrobial peptides. It regulates genes important for growth under these conditions, as well as additional genes important for virulence in many gram-negative pathogens. PhoQ is a sensor kinase that phosphorylates and activates the transcription factor PhoP. Since feedback inhibition is a common theme in stress-response circuits, we hypothesized that some members of the PhoP regulon may play such a role in the PhoQ/PhoP pathway. We therefore screened for PhoP-regulated genes that mediate feedback in this system. We found that deletion of mgrB (yobG), which encodes a 47 amino acid peptide, results in a potent increase in PhoP-regulated transcription. In addition, over-expression of mgrB decreased transcription at both high and low concentrations of magnesium. Localization and bacterial two-hybrid studies suggest that MgrB resides in the inner-membrane and interacts directly with PhoQ. We further show that MgrB homologs from Salmonella typhimurium and Yersinia pestis also repress PhoP-regulated transcription in these organisms. In cell regulatory circuits, feedback has been associated with modulating the induction kinetics and/or the cell-to-cell variability in response to stimulus. Interestingly, we found that elimination of MgrB-mediated feedback did not have a significant effect on the kinetics of reporter protein production and did not decrease the variability in expression among cells. Our results indicate MgrB is a broadly conserved membrane peptide that is a critical mediator of negative feedback in the PhoQ/PhoP circuit. This new regulator may function as a point of control that integrates additional input signals to modulate the activity of this important signaling system.
| The proteins PhoQ and PhoP comprise an environmental sensing system that has been extensively studied in numerous bacteria, including Salmonella typhimurium and Escherichia coli. The PhoQ/PhoP system is stimulated by conditions of low extracellular magnesium or the presence of certain cationic antimicrobial peptides; and it controls genes, whose protein products protect the cell under these conditions or play other critical roles in regulating the virulence of pathogens. The functions of many members of the PhoP regulon, however, remain uncharacterized. This leaves open the possibility that some PhoP-regulated genes may mediate feedback in this system. Regulatory circuits that allow adaptation to environmental change often make use of negative feedback to achieve the appropriate level of response. To look for negative feedback, we screened knockouts of PhoP-regulated genes in E. coli. We have identified a remarkably small membrane protein of just 47 amino acids that mediates potent negative feedback on the PhoQ/PhoP circuit in E. coli, S. typhimurium, Yersinia pestis, and likely other related bacteria. This represents a striking example of a small, easily-overlooked open reading frame that plays a critical role in regulating a broadly conserved signal transduction pathway.
| Many prokaryotes inhabit multiple niches with disparate environmental conditions and challenges for proliferation. Not surprisingly, they have evolved a plethora of regulatory circuits that enable them to adapt to these environments. One important and extensively studied example is the signaling system controlled by the sensor kinase PhoQ and response regulator PhoP, which is found in enterics such as Escherichia coli, Salmonella typhimurium, and related bacteria. This two-component system is activated by signals such as low Mg2+ [1], low pH [2], or the presence of antimicrobial peptides [3], and leads to expression of genes that encode, among others, Mg2+ transporters, enzymes that modify the cell envelope and confer resistance to cationic antimicrobial peptides [4]–[8], enzymes that alleviate stress associated with low pH [9]–[11], and other factors that regulate virulence in numerous gram-negative pathogens of humans, insects, and plants [12]–[18].
The sensor kinase PhoQ is an integral membrane protein whose periplasmic domain is involved in signal detection. Signal transduction occurs via PhoQ autophosphorylation, phosphotransfer to PhoP, and PhoQ-mediated dephosphorylation of phospho-PhoP. Magnesium sensing appears to be mediated by an acidic patch in the periplasmic domain of PhoQ, which structural data suggests is proximal to the membrane [19]–[22]. This acid pocket may also play a role in antimicrobial peptide sensing via displacement of divalent cations by these positively charged peptides [3].
In many cases, adaptive regulatory systems have some form of negative feedback to modulate the cellular response. Indeed, negative feedback is a common theme in cell regulation and its role in maintaining homeostatic control is well-established [23],[24]. Negative feedback can also play a similar role in reducing cell-to-cell variability within a population [25] and can increase the activation kinetics in some circuits [26],[27]. More generally, negative feedback, when combined with additional layers of regulation, may produce complex dynamics or process multiple input signals.
Since the PhoQ/PhoP system functions as a critical stress response circuit for survival under conditions of low magnesium or in the presence of antimicrobial peptides, we hypothesized that there may be sources of negative feedback in this circuit. We would expect that such a negative feedback loop would have at least one component that is regulated by PhoP. Microarray and sequence analysis indicate PhoP influences transcription of a large set of genes [11],[28],[29]. However, relatively few genes have been shown to be directly regulated by PhoP [11],[28],[30], and few of these have known functions. From this short list, none appeared to be obvious candidates for mediators of negative feedback. We therefore screened a reporter strain containing deletions of different PhoP-regulated genes for evidence of increased PhoQ/PhoP-dependent transcription. From an analysis of seven deletion strains, we found that mgrB (yobG), which is predicted to encode a 47 amino acid peptide of unknown function, plays a critical role in regulating the PhoQ/PhoP pathway.
To identify factors that mediate negative feedback in the PhoQ/PhoP circuit, a set of genes that have been shown to be directly regulated by PhoP in E. coli [28],[30] were individually deleted in a PhoP transcriptional reporter strain. The reporter contained the PhoP-regulated mgrB promoter driving YFP expression (PmgrB-yfp), which was integrated at the phage lambda attachment site. The strains also expressed CFP from a constitutive promoter, which served as an internal control for protein expression and fluorescence intensity. We chose the mgrB transcriptional reporter because it gives the highest level of fluorescence from among a collection of fluorescent reporters of PhoP-regulated transcription [31]. As seen in cells growing on LB agar plates, which are moderate inducing conditions for the E. coli PhoQ/PhoP system [30], deletion of mgtA, rstA, nagA, slyB, vboR, and yrbL had no visible effect on YFP fluorescence (Figure 1A). Deletion of mgrB, however, resulted in a clear increase in YFP with no visible difference in CFP fluorescence when compared with the wild-type strain (Figure 1A). To obtain a quantitative measure of the effects of the gene deletions on transcription levels, we measured the ratio of YFP to CFP fluorescence in single cells by fluorescence microscopy and image analysis [32]. Cells were grown in medium containing either 100 µM or 10 mM Mg2+. Consistent with the results observed on agar plates, deletions of mgtA, rstA, nagA, slyB, vboR, and yrbL showed no effect on the YFP/CFP fluorescence (the example of slyB is shown in Figure 1B, other examples are shown in Figure S1). Deletion of mgrB, on the other hand, resulted in roughly a 9-fold increase in YFP/CFP fluorescence at 10 mM Mg2+ and a 3-fold increase at 100 µM Mg2+, relative to the corresponding wild-type strain (Figure 1B). The observed increase in fluorescence was strictly dependent on PhoQ (Figure 1B). Deletion of mgrB in fluorescent transcriptional reporter strains for three other PhoP-regulated promoters, PmgtA, PphoPQ, and PhemL, resulted in similar increases in fluorescence (Figure 1C and 1D, and data not shown), suggesting that the effect of the deletion is likely to be at a point upstream in the pathway common to all of these genes, i.e. at some point in PhoQ/PhoP signaling. We also verified that the mgrB deletion could be complemented by a plasmid containing mgrB cloned downstream of the trc promoter (Figure 2). When this same plasmid was put in an mgrB+ strain, it resulted in a further repression of reporter gene transcription (Figure 2), presumably due to increased expression of MgrB above the wild-type level. Taken together, the above results suggest that the mgrB gene product acts as a repressive factor in PhoQ/PhoP signaling.
The mgrB gene consists of a small open reading frame of 141 base pairs. A recent study of small open reading frames in E. coli verified protein expression from the wild-type locus of an epitope-tagged mgrB [33]. The 47 amino acid MgrB peptide could potentially have a type I (secretion) or type II (lipidation) signal sequence [34],[35]. Alternatively, the N-terminal hydrophobic stretch may function as a transmembrane domain [36] (Figure 3A). To determine the localization of MgrB, we analyzed cell envelope and soluble protein fractions by Western blotting with rabbit antisera raised against a peptide from the C-terminus of the protein. We were able to detect a protein between 4 and 7 kilodaltons in the whole-cell lysate and envelope fraction of cells expressing mgrB, but not in the soluble protein fraction (Figure 3B). The envelope fraction of cells expressing MgrB showed no detectable contamination with cytoplasmic or periplasmic proteins as assessed by Western blots for the cytoplasmic proteins CFP and YFP and the periplasmic protein beta-lactamase (top two panels of Figure 3B).
To further confirm membrane association, we constructed a fusion of GFP to the N-terminus of MgrB. The fusion is similar to wild-type MgrB in its ability to complement a deletion, as indicated by repression of PhoQ/PhoP signaling (Figure S2). Western blots show no evidence of cleavage or degradation of the fusion protein (data not shown). The fact that GFP is not cleaved from MgrB by a signal peptidase suggests that nascent MgrB is not secreted into the periplasm or lipidated. Fluorescence microscopy of cells expressing GFP-MgrB revealed a halo-like fluorescence at the cell boundary, suggesting localization to the envelope, whereas cells expressing GFP alone showed uniform fluorescence indicative of cytoplasmic localization (Figure 3C). When GFP is targeted to the periplasm through the Sec pathway, either as a secreted protein or as part of a membrane protein, it does not fold properly and fails to fluoresce [37],[38]. The only reported pathway for producing fluorescent GFP in the periplasm is through the Tat secretion system [39],[40]. However MgrB is not predicted to have a signal sequence that would target it for secretion through this pathway [41]. Taken together, the above results indicate that MgrB is associated with the inner-membrane with its N-terminus in the cytoplasm. The size and hydrophobicity profile of the peptide further suggest that it spans the membrane with a single transmembrane domain.
Given the localization of MgrB, we hypothesized that it exerts its effects through PhoQ. We therefore tested the action of MgrB on a chimera in which the periplasmic domain of PhoQ from E. coli was replaced with the highly divergent periplasmic domain of PhoQ from Pseudomonas aeruginosa [42], a bacterium that does not appear to possess an mgrB ortholog. We measured fluorescence of a phoQ− mgrB− reporter strain for PhoP-dependent transcription that was transformed with a plasmid expressing either E. coli PhoQ or the PhoQ chimera (PhoQchim), or a control plasmid. The strain also contained either a compatible plasmid expressing mgrB or a compatible control plasmid (Figure 4). We note that the fluorescence levels of the MgrB− PhoQ+chim strain were lower than the corresponding levels for the MgrB− PhoQ+ strain, but they were significantly higher than the fluorescence levels of the PhoQ− strain. As with the complementation experiments above, introduction of the MgrB expression plasmid resulted in decreased fluorescence of PhoQ+ cells. In contrast, no change in fluorescence levels was observed when the plasmid was introduced into the PhoQ+chim strain. That MgrB does not decrease PhoP-regulated transcription when the periplasmic sensor domain of PhoQ is modified suggests MgrB acts at or upstream of PhoQ in the signaling pathway.
To look for evidence of an interaction between MgrB and PhoQ in E. coli, we used a bacterial two-hybrid assay based on split adenylyl cyclase [43]. In this system, adenylyl cyclase activity is reconstituted when the T18 and T25 fragments of Bordetella pertussis CyaA are brought into close proximity. The resulting increase in cAMP levels is detected through expression of beta-galactosidase from the lac promoter. We fused the T18 and T25 fragments to the N-termini of MgrB and PhoQ, respectively. Based on the known topology of PhoQ, and the topology of MgrB (discussed above), both CyaA fragments should be in the cytoplasm. A strain expressing T18-MgrB and T25-PhoQ showed a significantly higher level of beta-galactosidase activity when compared with strains expressing the T18 and T25 fragments alone (Figure 5) or expressing the fusions to either MgrB or PhoQ individually (data not shown). These results suggest there is a physical interaction between MgrB and PhoQ. Furthermore, a strain expressing T18-MgrB and the T25- fragment fused to the N-terminus of PhoQchim showed a minimal increase in beta-galactosidase activity relative to the controls. This is unlikely to be due to a defect in PhoQchim expression because we were able to detect PhoQchim–PhoQchim interactions at levels comparable to those for (wild-type) PhoQ-PhoQ interactions, consistent with previous reports that both PhoQ and PhoQchim form functional complexes [42] (Figure S3). Taken together, these results suggest that the periplasmic domain of E. coli PhoQ is important for the interaction with MgrB. Interestingly, co-expression of T18-MgrB and a fusion of T25 to the N-terminus of MgrB also showed significant beta-galactosidase activity relative to the controls (Figure 5), suggesting that MgrB may form dimers or higher order complexes.
The mgrB gene, also known as yobG, has been identified in Salmonella enterica, where it has been shown to be activated by PhoP in microarray and transcriptional reporter experiments [11],[44]. Like many other small open reading frames, mgrB is frequently missed in genome annotations. However we have been able to identify putative mgrB homologs in numerous species among several genera of Enterobacteriaceae. Specific examples and their alignments are shown in Figure 6. All of the genomes for which we could identify a candidate mgrB also contained phoP and phoQ orthologs. However, the converse does not appear to be the case: we were unable to find an mgrB homolog in Erwinia or Pseudomonas species, though members of both of these genera possess phoP and phoQ. In addition, although the endosymbiont Sodalis glossinidus has phoP and phoQ genes, its mgrB is unlikely to be functional due to the presence of two internal stop codons (Figure 6).
To test whether the action of MgrB on PhoQ/PhoP signaling is conserved among several genera, we over-expressed the native MgrB from Salmonella enterica serovar typhimurium and Yesinia pestis in their respective host strains and measured transcription of the promoter for phoN, a PhoP-regulated gene from Salmonella. For both Y. pestis and S. typhimurium, MgrB over-expression resulted in a significant repression of reporter activity (Figure 7). We also verified that the mgrB genes from both of these organisms complement an mgrB deletion in E. coli (data not shown).
Based on the results presented here, we propose that the peptide MgrB spans the inner-membrane and represses PhoP phosphorylation by inhibiting PhoQ kinase activity, stimulating phosphatase activity, or both (Figure 8). Our results further suggest that an interaction between MgrB and the periplasmic domain of PhoQ plays a critical role. Since mgrB transcription is activated by phosphorylated PhoP, MgrB is part of a negative feedback loop in the PhoQ/PhoP signaling circuit. It is particularly striking that deletion of mgrB results in a strong increase in PhoP-regulated transcription even for growth in 10 mM Mg2+–a condition that strongly represses PhoQ/PhoP signaling in wild-type cells. The four PhoP-regulated promoters that we tested (promoters for mgrB, phoPQ, mgtA, and hemL) show at least some level of magnesium responsiveness in an mgrB− strain (Figure 1B–1D and data not shown). The significant decrease in magnesium sensitivity of mgtA transcription in the absence of MgrB (Figure 1D) most likely indicates that mgtA has reached near-maximal levels of PhoP-activated transcription for the mgrB− strain growing in 10 mM Mg2+. Indeed, previous studies suggest that the E. coli mgtA promoter saturates at lower levels of PhoQ/PhoP stimulation when compared to the other promoters considered here [31]. We have also found that PhoP-regulated transcription remains responsive to the antimicrobial peptide LL37 and to acidic pH in mgrB− strains (Figure S4 and Figure S5, respectively). We note, however, that mgrB deletion affects the fold-change in reporter gene expression for all three stimuli (Mg2+, pH, and LL37). Thus, MgrB modulates the magnitude and sensitivity of PhoQ/PhoP signaling but is not strictly required for magnesium, pH, and antimicrobial peptide responsiveness.
Negative feedback has been shown to increase activation kinetics and minimize cell-to-cell variability in some regulatory circuits [25]–[27]. Such behavior can be readily understood for simple examples of feedback arising from negative autogenous control. However it is difficult to predict the effects of negative feedback in more complex circuits with additional regulators, which have the potential for time delays and additional sources of cell-to-cell variability. Indeed, deletion of MgrB has relatively little effect on the kinetics of fluorescent protein reporter accumulation (Figure S6) and on the cell-to-cell variability in fluorescent protein expression (Figure S7), despite its strong effect on the magnitude of PhoQ/PhoP signaling. We hypothesize that MgrB may provide a point of control for integrating additional input signals that modulate PhoQ/PhoP activity. For example, such signals could act by regulating the expression level or repressive action of MgrB (Figure 8).
We have shown that for bacteria from at least three different genera, the native MgrB homolog represses PhoQ/PhoP signaling, suggesting that this mode of negative regulation is broadly conserved. At least two other proteins have been reported to regulate PhoQ/PhoP activity, however the level of functional conservation appears to be more limited for these cases. YneN (B1500) is a small integral-membrane protein that stimulates PhoQ in E. coli [45]. This protein is a member of the EvgS/EvgA regulon and thus mediates cross-regulation between two signaling circuits. This mechanism for PhoQ regulation does not appear to be widely conserved; we have been unable to identify YneN orthologs in sequenced genomes other than those of various E. coli and Shigella isolates. Another protein, SlyB, has been shown to mediate some negative repression of the PhoQ/PhoP system in Salmonella typhimurium. Deletion of slyB resulted in a roughly 1.5-fold increase in transcription of multiple PhoP-regulated genes in this bacterium [44]. However, we did not detect any difference in transcription of PhoP-regulated genes when we compared wild-type and slyB− strains of E. coli (Figure 1B). This may indicate another example of divergence in the PhoQ/PhoP regulatory circuit, even for these closely related species [29],[46].
MgrB-mediated regulation of PhoQ is part of an emerging theme of modulation of membrane proteins by small hydrophobic peptides [47]. It is a striking example of a small, easily overlooked open reading frame that plays a critical role in regulating a signal transduction pathway. Sequence alignments reveal several conserved residues, which may be critical for MgrB function (Figure 6). Further genetic and biochemical work will be required to understand the mechanism by which MgrB represses PhoQ, and to determine whether MgrB has additional roles as part of this important two-component signaling system.
See Supporting Information for tables of strains (Table S1), plasmids (Table S2), and PCR primers (Table S3). A description of plasmid construction is given in Text S1.
Deletions of PhoP-regulated genes in E. coli were transduced from strains in the Keio Collection [48] by P1vir transduction. The deletions were confirmed by PCR using primers that flank the gene. When necessary, kanamycin resistance markers were removed with FLP recombinase by transforming with pCP20 [49] and subsequent curing of the plasmid.
E. coli strains were grown at 37°C, unless otherwise indicated, in Luria-Bertani (LB) medium (Difco - BD, Franklin Lakes, NJ) or minimal A medium [50] supplemented with 0.2% glucose, 0.1% casamino acids (Difco) and with the indicated concentration of MgSO4. Salmonella strains were grown in LB at 37°C. Yersinia pestis strains were grown at 26°C in brain-heart infusion (BHI) broth (Difco). The lac and trc promoters were induced with isopropyl β-D-1-thiogalactopyranoside (IPTG) at a final concentration of 1 mM when indicated. When IPTG was not mentioned in the description of the culture conditions, the basal transcription from the trc promoter was used to drive expression.
Strains were streaked onto LB Agar plates (Difco) and incubated overnight at 37°C. Images of YFP and CFP fluorescence were acquired with a home-built fluorescence illuminator as previously described [51].
Single-cell measurements were performed essentially as described previously [31]. Briefly, overnight cultures, grown in minimal A medium with 1 mM MgSO4, and 50 µg/mL ampicillin with or without 50 µg/mL spectinomycin to maintain plasmids when necessary, were diluted back 1∶1000 in pre-warmed minimal medium containing 100 µM, 1 mM, or 10 mM MgSO4 and grown to an OD600 between 0.2 and 0.3. Cultures were cooled quickly with an ice water slurry and streptomycin was added to a final concentration of 250 µg/mL to inhibit protein synthesis. Images were acquired and analyzed as previously described [31],[32].
Envelopes were prepared with a protocol modified from [52]. An overnight culture in LB medium with 50 µg/mL ampicillin was diluted 1∶1000 into the same medium supplemented with 1mM IPTG and shaken at 37 degrees for 5.5 hours at 250 rpm. The culture was split in half, chilled on ice, and then spun at 3,300 g for 10 minutes at 4°C. One pellet was saved at −20°C for the total lysate. The second pellet was resuspended in a 10 mL of 30mM Tris pH 7.8 and pelleted again. The pellet was then resuspended by vortexing in 200 µL of a cold 20% sucrose/30mM Tris solution. 20 µL of freshly prepared 10 mg/mL lysozyme in 0.1 M EDTA pH 7.0 was added and the solution was mixed by inverting at 4°C for 20 minutes. 3 mL of 3 mM EDTA pH 7.5 was then added, the solution was sonicated on ice, and then centrifuged at 47,000 g at 4°C for 75 minutes. The supernatant, which is the soluble fraction, was removed and saved. The pellet, which is the envelope fraction, was resuspended in 2 mM Tris pH 8.0. To bring the volume of supernatant down to the same concentration of cell equivalents as the envelope fraction, the supernatant was concentrated with a Speed Vac (Thermo Fisher, Waltham, MA).
Samples were boiled in Tris-Tricine loading buffer with 0.2 M dithiothreitol for 5–7 minutes and loaded on 16.5% Tris tricine gels (ReadyGel - BioRad, Hercules, CA). Proteins were transferred to Immobilon-P PVDF (Millipore, Billerica, MA) followed by western blot analysis. A rabbit polyclonal MgrB antiserum was generated by using a KLH-conjugated synthetic peptide corresponding to residues 27–40 of the predicted MgrB protein sequence (GenScript, Piscataway, NJ). The peptide was synthesized with a C-to-S substitution for the first cysteine residue. Beta-lactamase and CFP/YFP were detected with rabbit polyclonal anti-beta-lactamase (Millipore,) and anti-GFP (A.v. Peptide Antibody – Clontech, Mountain View, CA) antibodies. A horseradish peroxidase-conjugated anti-rabbit antiserum (GE Healthcare, Piscataway, NJ) was used as the secondary antibody.
Cultures of BTH101 bearing combinations of pKT25- and pUT18-derived plasmids were grown for 9 hours in LB supplemented with 100 µg/mL ampicillin and 50 µg/mL kanamycin. The cultures were then diluted 1∶1000 into medium supplemented with 1 mM IPTG and allowed to grow at 30°C for 14 hours. Cultures were grown at 30°C in order to increase complementation efficiency [53]. Cells were cooled quickly in an ice slurry and kept on ice for 30 minutes. Beta-galactosidase assays were performed as described in [50] using chloroform and SDS for permeabilization.
Cultures of Salmonella were grown overnight in LB with 50 µg/mL ampicillin and 50 µg/mL spectinomycin. The next day, they were diluted 1∶1000 into the same medium and grown to OD600 0.1–0.3. Cells were permeabilized with chloroform and SDS and assayed as in [50]. Cultures of Yersina were grown and assayed in the same fashion except the culture medium was BHI with 200 µg/mL ampicillin and 50 µg/mL spectinomycin and the overnight culture was diluted 1∶50.
|
10.1371/journal.ppat.1001341 | The Antiviral Efficacy of HIV-Specific CD8+ T-Cells to a Conserved Epitope Is Heavily Dependent on the Infecting HIV-1 Isolate | A major challenge to developing a successful HIV vaccine is the vast diversity of viral sequences, yet it is generally assumed that an epitope conserved between different strains will be recognised by responding T-cells. We examined whether an invariant HLA-B8 restricted Nef90–97 epitope FL8 shared between five high titre viruses and eight recombinant vaccinia viruses expressing Nef from different viral isolates (clades A–H) could activate antiviral activity in FL8-specific cytotoxic T-lymphocytes (CTL). Surprisingly, despite epitope conservation, we found that CTL antiviral efficacy is dependent on the infecting viral isolate. Only 23% of Nef proteins, expressed by HIV-1 isolates or as recombinant vaccinia-Nef, were optimally recognised by CTL. Recognition of the HIV-1 isolates by CTL was independent of clade-grouping but correlated with virus-specific polymorphisms in the epitope flanking region, which altered immunoproteasomal cleavage resulting in enhanced or impaired epitope generation. The finding that the majority of virus isolates failed to present this conserved epitope highlights the importance of viral variance in CTL epitope flanking regions on the efficiency of antigen processing, which has been considerably underestimated previously. This has important implications for future vaccine design strategies since efficient presentation of conserved viral epitopes is necessary to promote enhanced anti-viral immune responses.
| One of the greatest challenges to developing an effective HIV vaccine is the ability of HIV to rapidly alter its viral sequence. Such variation in viral sequence enables the virus to frequently evade recognition by the host immune system. To counteract this problem, there has been increasing interest in developing HIV vaccines that target T-cell responses to the regions of the virus that are highly conserved between strains of HIV. However, previous studies have focused on identifying amino acid variation predominantly within a single viral isolate, or have focused on classical within-epitope escape mutation. Our study assessed T-cell recognition of a conserved epitope shared by a total of 13 HIV strains. Strikingly, we show that only a small proportion of the viral strains were effectively recognised and targeted by the T-cells. In contrast, differences in amino acid sequence in the region flanking the epitope impaired the intracellular processing and presentation of epitope in the majority of HIV strains tested. Thus, our findings highlight that a large proportion of HIV strains may evade epitope-specific T-cell recognition despite absolute epitope conservation. This has important implications for both vaccine design and evaluation of vaccine efficacy.
| One of the greatest challenges in developing an effective T-cell based vaccine against HIV-1 is its high genetic variability [1]. Group M HIV-1 has expanded globally into 15 major clades, sub-clades and several interclade circulating recombinant forms. These continually evolving HIV-1 clades differ by over 30% in envelope amino acid sequences and viral isolates within the same clade may also differ by up to 15% [2]. The high rate of mutation from error-prone reverse transcription combined with replicative ability enables HIV-1 to adapt rapidly to immune and drug pressure, with the generation of multiple genetically distinct quasispecies within infected individuals. HIV-1 vaccines must overcome these obstacles to induce protective immunity against heterologous viral variants [3].
A critical component of HIV-1 control during the acute phase is the cytotoxic T-lymphocyte (CTL) response [4], [5]. Therefore, many current vaccine strategies focus on identifying immunogens that elicit effective T-cell immunity against a diverse range of viral variants and characterising HIV-1 specific CTL responses in order to define the immune correlates of protection [5]. To counteract antigenic diversity, there is an increasing interest in developing HIV vaccines which elicit CTL responses to conserved epitopes, centralised sequences or immunogenic regions of high inter-clade homology [6], [7], [8], [9], [10]. Currently, the interferon gamma (IFNγ) producing ELISpot assay is frequently used to quantify the breadth and magnitude of CTL responses [11], using peptides matched to consensus virus sequence or occasionally to autologous infecting virus sequence [12] [13]. Most data show that CTL recognition of epitope peptides is very sensitive to any change in the epitope peptide [13], [14]. Thus the HLA type of the patient imprints changes on the sequence of the infecting virus, generally thought to be within the epitopes that have stimulated CTL responses [15], [16]. However, while CTL may efficiently recognise exogenously loaded synthetic peptide matched to HIV-1 clade variants, it has been found that this does not necessarily correlate with CTL antiviral activity against HIV-1 infected cells displaying endogenously derived peptides [17]; for example, the artificial peptides may be added at non-physiological concentrations. Therefore, conventional peptide-based assays may over-estimate the ability of CTL to cross-recognise variant epitopes [18]. The use of exogenous synthetic peptides to quantify CTL responses may also fail to detect differences in the antiviral efficacy of CD8 T-cells that reflect variation in antigen processing efficiency within HIV-infected cells [19], [20]. Remarkably, whilst much research has focused on recognition of exogenously added peptide epitopes, CTL recognition of virus-infected cells has been examined relatively rarely, and there has been no analysis of CTL recognition of invariant epitopes shared by diverse viral isolates and clades.
The present study arose from the observation that CD8+ T cells specific for the highly conserved HLA B8 restricted Nef epitope FLKEKGGL (FL8) failed to recognize HLA B8 positive cells infected with several HIV-1 isolates. Previously it has been shown that escape mutations can occur in the epitope flanking regions through impaired processing and presentation [21], [22], [23], [24], [25], [26], [27], however, such studies have focused predominantly on a single viral isolate, mostly in circulating virus rather than selected viral strains in individual patients, or focused on classical escape mutation [28]. Therefore a range of HIV-1 isolates and vaccinia viruses expressing different Nef proteins, each of which shared this conserved epitope, were used to test responses from a set of CTL clones isolated from HLA B8+ patients. Overall, we evaluated CTL recognition and antiviral efficacy induced by a total of thirteen viral isolates containing the same conserved epitope. Surprisingly, only a small proportion (23%) of these HIV-1 isolates induced optimal CTL recognition and antiviral efficacy. We found that variations in the flanking region had a profound effect on the presentation of this epitope, and viral isolates within the same HIV-1 clade were differentially recognized by FL8-specific CTL clones. Furthermore, we identified a phenylalanine motif in the FL8-epitope flanking regions of four HIV-1 isolates that led to an altered pattern of cleavage by the immunoproteasome that correlated with loss of CTL recognition.
In conventional IFNγ ELISPOT and chromium release assays, we assessed three Nef FL8-specific CTL clones and four control Gag EI8-specific CTL clones for their recognition of HLA-B8+ matched C8166 target cells pulsed with peptides at different concentrations to measure functional avidity (the peptide concentration that gives 50% maximum effect). The FL8-specific and EI8-specific CTL clones had comparable levels of functional avidity, measured by IFNγ release (Figure 1A) and in a lytic assay (Figure 1B). In addition, there were no significant differences (p>0.05) observed between FL8- and EI8- specific responses in the lysis assay at all peptide concentrations tested. This suggested that both peptides would be equally recognised in HLA-B8 target cells infected with HIV-1.
We then compared CTL recognition of synthetic Nef FL8 peptide pulsed exogenously onto the surface of uninfected C8166 targets with CTL antiviral efficacy against endogenously derived FL8 epitope presented on the cell surface of HIV-1HXB2 infected C8166 targets. Surprisingly, FL8-specific CTL did not recognise or mount an antiviral response against the virus-infected target cells.
In a Viral Suppression Assay (VSA), the panel of FL8- and EI8- specific CTL clones were co-cultured for four days with C8166 target cells infected with HIV-1HXB2 at five E∶T ratios, after which suppression of viral replication in the supernatant was quantified by measuring the level of Gag p24 by ELISA; and suppression of HIV-infected targets (via CTL lysis or non-cytolytic inhibition) was measured using intracellular p24 staining. Despite the similar functional avidity in peptide based assays, at a low infectious titre of virus, EI8-specific CTL exhibited superior suppression of viral replication compared to FL8-specific CTL when p24 was measured in supernatant by ELISA (Figure 2A). At all E∶T ratios tested, co-culture with EI8-specific CTL resulted in undetectable levels of p24 Gag (below the ELISA threshold of 10 pg/ml) whilst viral suppression by FL8-specific CTL was negligible when compared to wells of virus-infected targets in the absence of CTL. Analysis of suppression at a high infectious titre showed similar results, with negligible suppression of virus-infected cells by FL8-specific T-cells at all E∶T ratios tested (ranging from 37–89 ng/ml with a mean p24 of 62 ng/ml in the absence of CTL) (Figure 2B). Subsequent analysis of the p24 stained cells showed a similar pattern (Figure 2C). At the 1∶1 and 1∶4 ratios, the control EI8 specific CTL clones demonstrated effective inhibition of viral replication, although they were unable completely to abrogate infection, whilst FL8-specific CTL did not differ from the controls in their suppression of viral infection at all three ratios tested. 2-way ANOVA with bonferroni post-test confirmed a statistically significant difference (p<0.01 or p<0.001) between Nef-specific and Gag-specific clones at each E∶T ratio in the ELISA.
We also developed a Live Virus Elispot (LVE) assay to assess IFNγ release by CTL exposed to HIV-infected target cells, as a marker of CTL antiviral activity over time. C8166 target cells were infected with HIV-1HXB2 and incubated for a period of 24, 48, 72 and 96 hours at 37°C. At each respective time point, the HIV-infected cells were co-cultured with HLA-matched FL8- and EI8-specific CTL clones on a pre-coated IFNγ ELISpot plate. Again, FL8-specific CTL did not mount an IFNγ response at any of the time points tested (Figure 2D). In contrast, the EI8 epitope was recognised by the EI8-specific CTL clones, which generated a detectable IFNγ response that was significantly different (p<0.001) between FL8- and EI8-specific clones at 48, 72 and 96 hours post-infection in a 2-way ANOVA with bonferroni post-test.
The three Nef FL8-specific clones were generated from three separate HLA-B8+ long-term non-progressors (LTNP) and control Gag EI8-specific clones were generated from one HLA-B8+ LTNP patient. The use of this pre-screened panel of T-cell clones removes the complexity of different TCR affinities, and thus variation in TCR/pMHC interactions, of polyclonal T-cell responses.
To investigate whether CTL recognition and antiviral activity to a conserved epitope may depend on the infecting HIV-1 isolate, we chose five high titre HIV viruses (three clade B laboratory strains and two clade A isolates) that share the invariant Nef FL8 epitope for testing via in vitro Viral Suppression Assays (VSA) and Live Virus ELISPOTS (LVE). FL8 is an immunodominant epitope that is highly conserved amongst HIV-1 Group M isolates in the Los Alamos National Laboratory (LANL) HIV sequence database. Proviral DNA for each virus was isolated from control wells containing infected C8166 targets in the absence of CTL, PCR amplified and sequenced to confirm the presence of the invariant FL8 epitope. The results from both assays and viral sequencing are summarised in Table 1.
Data from the viral suppression assay for each virus shows differing FL8-specific CTL antiviral activity for the different virus isolates, despite sharing the conserved FL8 epitope (Figure 3). FL8 specific CTL suppressed both clade A viral isolates HIV-192UG029 and HIV-193RW024, as well as Clade B HIV-1MN, to below the threshold of detection (10 pg/ml) in the p24 ELISA (Figure 3C). However, clade B HIV-1HXB2 (as characterised before in Figure 2A) and clade B HIV-189.6 were not suppressed by FL8-specific CTL at any of the effector∶target (E∶T) ratios tested, and were not statistically different (p>0.05) when compared to wells of virus-infected cells in the absence of CTL. Similar trends were obtained when analysing the corresponding infected co-cultures from the VSA (Figure 3B). In HIV-192RW024 infected target cells, FL8 specific CTL reduced infection from 9% to <1% whilst a reduction was also observed in HIV-192UG029 infected cells at all three E∶T ratios tested, which were significantly different (p<0.01) from control wells of infected cells in the absence of CTL. The clade B virus HIV-1MN was also efficiently suppressed by FL8 and EI8 CTL. However, FL8-specific CTL exerted no significant antiviral efficacy against HIV-1HXB2 or HIV-189.6 infected target cells, suggestive of impaired intracellular FL8 epitope processing and presentation. Similar result was observed while using HIV-1HXB2 and MN infected HLA B8+ PBMCs as target cells in VSA (Figure S2).
Differential CTL antiviral activity to the FL8 epitope between viral isolates was also observed in the Live Virus Elispots (LVE) assay (Figure 3C). FL8-specific CTL demonstrated a strong IFNγ response with an average magnitude of 46–101 SFUs at 96 hours post-infection against endogenously presented FL8 peptide by cells infected with clade A isolates HIV-192UG029 and HIV-192RW024 respectively. In contrast, a varied FL8-specific CTL response was observed against clade B infected cells, with a mean of 65 SFUs recorded for HIV-1MN, but 0 SFUs at 96 hrs post-infection for both HIV-1HXB2 and HIV-189.6. Further LVE studies were conducted at 6 and 12 hours post-infection to assess ‘early’ CTL recognition of endogenous Nef and Gag epitopes but no IFNγ responses were observed.
All high titre viruses were tested in parallel with the same panel of three highly avid FL8-specific clones and four control EI8-specific clones with the same results. All five viruses were CXCR4-tropic, able to replicate efficiently and were pathogenic, as confirmed by syncytia formation with 6–20% of C8166 cells (p24+/CD4+) infected by Day 4 in the absence of CTL. A low MOI was chosen for physiologic relevance to an in vivo setting, with <20% infection observed, comparable to other SIV and HIV- based viral suppression assays [21], [29]. However, similar results were also obtained at a higher MOI, so viral titre does not appear to impact upon the patterns observed (as observed in figure 2B). Furthermore, since HIV-1HXB2, MN and 93RW024 contained a conserved HLA-A24 restricted RW8 Nef epitope, we were also able to utilise CTL clones specific for RW8 to confirm that these viruses were able to present Nef derived peptides to CTL and were not simply Nef-deficient (Figure S1). Interestingly, although there is a general trend towards higher p24 at the lower E∶T ratios with the control gag clones, and with the Nef clones to a lesser extent, overall the E∶T ratio appears to have limited influence in the VSA. This is in accordance with the results from a similar assay [30] and may be attributable to the use of CTL clones that were pre-selected for their high avidity and ability to suppress viral replication at low E∶T ratios (in comparison to polyclonal populations with differing avidity). It should also be noted that our control Gag EI8 clones specific to EIYKRWII were chosen as an internal control as they had similar functional avidity for the clade A intra-epitopic variant DI8 (DIYKRWII) when tested with peptides in the IFNγ ELISPOT and chromium release assays (data not shown). However, these Gag-specific CTL were unable to respond to DI8 targets within the clade A viruses HIV-193RW024 and HIV-192UG029 in Figure 3, due to more efficient immunoproteasomal processing for EI8 than DI8 (data not shown); however they still acted as good internal controls for clade B infections.
Overall, our data from the viral suppression assays and live virus ELISPOTS clearly demonstrate that efficient CTL antiviral activity is heavily dependent on the infecting HIV-1 isolate, which cannot be accurately predicted from epitope sequence conservation alone or by clade-grouping.
We also infected HLA-B8 B-cells with recombinant vaccinia virus constructs expressing HIV-1 Nef (rVV-Nef) from eight viral isolates, each from a different group M clade (for simplicity abbreviated by their clade reference): HIV-192UG037.1(A), HIV-1MN (B), HIV-196ZM651 (C), HIV-194UG114.1 (D), HIV-1CM235-32 (AE), HIV-193BR020 (F), HIV-192NG83.2 (G) and HIV-190CF056 (H). The rVV-Nef expressing cells were co-cultured with FL8-specific CTL in two assays; a 51Cr-release assay and IFNγ ELISPOT. The results from both assays and the pre-determined rVV-Nef sequences are summarised in Table 1.
In the 51Cr-release assay, striking differences were observed in the level of CTL-induced lysis of cells infected with rVV-Nef from different viral isolates, which could be categorised into three distinct groups based upon their mean percentage lysis. Only in cells infected with rVV-Nef-D was the proportion of lysed cells (70%) comparable to peptide-pulsed targets (75%) indicative of optimal abundance of endogenously derived FL8 for efficient recognition and lysis (Figure 4B). Subsequent analysis confirmed that there was no significant difference between rVVNef-D and the control in an ANOVA (p>0.05). In contrast, infection with viral isolates B, C, F and H resulted in reduced CTL recognition, with only 20–70% lysis compared to the peptide control, whilst no significant CTL killing (>20% lysis) was observed for isolates A, AE and G, suggestive of impaired or abolished FL8 processing and presentation. Concordant results were observed in the IFNγ ELISPOT (Figure 4A). Only isolate D elicited levels of IFNγ release (132 SFCs) similar to the peptide control (135 SFCs). In contrast, infection with B, C, F and H showed intermediate levels of 67, 23, 57 and 34 SFCs respectively and less than 10 SFCs for A, AE and G. In total, seven rVV-Nefs elicited CTL antiviral responses that were significantly different (p<0.01 or p<0.001) from the CTL response to the FL8 peptide control. These data support the isolate-specific nature of CTL antiviral efficacy, despite epitope conservation between viruses.
To ensure similar levels of Nef protein expression in these assays, rVV infection used the same plaque forming unit (pfu) titre and a CTL clone specific to a conserved Nef epitope HLA-A3 QK10 was utilised as an internal control. The strong CTL response to endogenous QK10 peptide presented by rVV-A, AE and G in the chromium release assay confirmed that these rVV efficiently infected the BCL targets and were not Nef-defective (Figure S1).
To verify that the striking differences in CTL antiviral activity to a conserved epitope observed in Figure 3 and 4 were not attributable to differing Nef-induced down-regulation of surface MHC class I expression by the various virus isolates [31], [32], we infected target cells with virus and assessed HLA-B8 surface expression using a HLA-B8 specific monoclonal antibody. We compared HLA-B8 expression in HLA-B8+ targets infected with our five high titre HIV-1 isolates (48 hours post-infection) and eight rVV-Nef constructs (4 hours post-infection), and also utilised three HLA-B8− negative targets as a negative control.
Our results showed that although down-regulation of HLA-B8 surface expression was observed in the high titre virus infected cells when compared with uninfected or vaccinia-infected targets, there was no marked variation in HLA-B8 between the five high titre HIV-infected target cells (Figure 5A). Furthermore, there was no marked variation in HLA-B8 expression between the eight rVV-infected target cells (Figure 5B). However, we acknowledge that the direct measurement of HLA-B8 expression may vary between experiments. Whilst Lewis et al controlled for this with internal standards [33], Nef-deleted viruses were not available for comparison with our wild-type viruses. Alternatively, we utilised B8-restricted Gag CTL controls in our viral suppression assays and live virus Elispots, since any effect of Nef mediated down-regulation of HLA-B8 ought to apply equally to this epitope and not be restricted to Nef epitopes. The CTL recognition of the Gag epitope was the same for the viruses tested, even when Nef recognition differed markedly. Together, these data therefore demonstrate that Nef-mediated down-regulation of HLA-B8 is not responsible for diminished CTL recognition of the conserved FL8 epitope on cells infected with HIV-1HXB2 and 89.6 isolates, or with rVV-Nef constructs A, AE and G.
From the above results it seemed likely that differences in antigen processing might explain the failure of CTL to recognise cells infected with some of the virus isolates and rVV-Nef. Intracellular peptide processing is a multi-step pathway and amino acid variations within epitopes and their flanking regions can affect any step of this complex cascade [34], which includes proteasomal or immunoproteasomal cleavage, TAP mediated transport to the ER lumen and N-terminal trimming by aminopeptidases. Alternatively, other proteases such as tripeptidyl peptidase II (TPPII) may act independently or in combination with the proteasome to generate epitope precursors. TPPII is known to play a pivotal role in the generation of the immunodominant HLA-A3 Nef QK10 epitope [35], which is located 8 amino acid residues downstream from the HLA-B8 Nef FL8 epitope.
To test the importance of different components of the antigen processing machinery on the generation and presentation of FL8 on the infected-cell surface, we chose two viruses that elicited a dominant FL8 response; high titre HIV-192UG029 and rVV-D (HIV-194UG114). Infected target cells were treated with inhibitors to block proteasome and immunoproteasome (Epoxomicin), aminopeptidase activity (Bestatin) and tripeptidyl-peptidase II activity (AAF-CMK) at appropriate concentrations. Controls included inhibitor treated cells pulsed with peptide (both HIV-infected and uninfected) to ensure that the read out was not altered by inhibitor-induced cell death.
The addition of epoxomicin to high titre HIV-192UG029 infected cells in a modified Live Virus ELISPOT intracellular antigen processing inhibition assay (IAPIA) completely abolished recognition of FL8 at 10 µM (Figure 6A). In contrast, addition of bestatin and AAF-CMK, even at high concentrations of 10–100 µM, had little impact on CTL recognition. A similar pattern was observed when the same three inhibitors were added to BCL infected with rVV-D (HIV-194UG114) in modified chromium release IAPIA (Figure 6B). The addition of epoxomicin at 10 µM and 1 µM reduced CTL lysis by 82% and 53% respectively, whilst addition of Bestatin and AAF-CMK at 10 µM had minimal effect. The marked difference in CTL lysis at 100 µM is likely to represent partial inhibition of proteasomal activity when used at high concentrations. Overall, in contrast to the Nef QK10 epitope [35], the result with AAF-CMK showed that TPPII is not involved in the FL8 processing pathway for the two viruses tested.
Together, these inhibition assays clearly demonstrate that proteasomes and immunoproteasomes play a pivotal role in the endogenous processing of the HLA-B8 FL8 epitope which is required to initiate efficient CTL antiviral activity against HIV-192UG029 and HIV-194UG114 infected targets. Therefore, as a first key step in the processing pathway and in accordance with the antigen processing literature (reviewed in [36], [37]), the cleavage specificities of the immunoproteasomes and proteasomes in particular are likely to have a significant impact on epitope generation correlating with subsequent epitope abundance on the cell surface [38].
Since the FL8 epitope is conserved between all HIV-1 isolates tested in previous assays, we next investigated whether viral isolate-specific polymorphisms flanking FL8 could modulate the antigen processing efficiency of this epitope. We hypothesised that processing and production of the FL8 epitope may be more efficient in cells infected with clade A isolates HIV-192UG029 and HIV-193RW024, and clade B HIV-1MN, whilst impaired antigen processing could account for poor CTL recognition of clade B HIV-1HXB2 and HIV-189.6 viruses in Figure 3. Two 25-mer oligopeptides were therefore synthesised to span FL8 (Nef90–97) and its flanking region (Nef82–106), one corresponding to HIV-1HXB2 and the other corresponding to HIV-192UG029. The two oligopeptides differed in the flanking region by 3 amino acids including position 83 (glycine/alanine), 85 (valine/leucine), and 104 (arginine/glutamine). Of these amino acid polymorphisms, the former two can be considered conservative variations, while replacement of arginine by glutamine leads to loss of a basic residue and replacement by an uncharged side chain. To test whether these residues affected antigen processing by the immunoproteasome, which is most commonly involved in CTL epitope generation, each oligopeptide was incubated with purified immuno-20S-proteasome (i20S) for 0, 10, 40 and 70 minutes. The peptide fragments resulting from immunoproteasomal digestion were identified using tandem mass spectrometry. Since immunoproteasomes and proteasomes rarely produce the exact 8–11mer peptide, but instead generate longer epitope pre-cursors that are often correctly cleaved at the C-terminus, we defined an ‘epitope precursor’ as a peptide fragment containing the intact FL8 epitope that is extended at the amino (N) terminus and carboxyl (C) terminus, and a ‘correct epitope pre-cursor’ as a peptide that is extended only at the N-terminus and correctly cleaved at the C-terminus.
Both oligopeptides showed strikingly different digestion patterns (Figure 7) when digested at these time points. Oligopeptide HIV-192UG029 produced three dominant FL8-containing precursor peptides, of which one was a correct C-terminally cleaved peptide (KGAVDLSHFLKEKGGL) (Figure 8A). Although mass spectrometry is only semi-quantitative, UPLC-MSE analysis at each time point showed that this C-terminally cleaved peptide was present as early as 10 minutes post i20S digestion and increased in quantity at 40 and 70 minutes post digestion (Figure 8B). In contrast, digestions of oligopeptide HIV-1HXB2 produced several FL8 containing precursor peptides, but these contained substantial N- and C-terminal extensions, and were only 3 amino acids shorter at most than the original 25-mer oligopeptide. None of the precursor peptides generated from digestion of oligopeptide HIV-1HXB2 contained the correct C-terminal cleavage for FL8.
Because of the large number of viral isolate-specific polymorphisms in the different viruses tested, it is difficult to define accurately which amino acid polymorphism(s) in the flanking regions present in different HIV-1 isolates critically impair immunoproteasomal cleavage of FL8. We speculated that one or several amino acid polymorphisms may act in tandem to alter cleavage patterns. Since most of the flanking amino acid differences were in the N-terminal part of the FL8 epitope, we “swapped” the N-terminal sequence of the FL8 epitope region from the HIV-192UG029 virus that is recognized by CTL with the sequence derived from the rVV-Nef-A(HIV-192UG037) virus that is not recognized (see Table 1 and Figure 9A) to create a “hybrid” FL8 epitope precursor. When digested by proteasomal proteolysis and analysed by UPLC-MSE, no correct C-terminal cleavage was observed, indicating that the N-terminal region of the FL8 epitope is critical for this cleavage step. We therefore further examined the polymorphisms observed in the PCR-derived sequence ‘tracked’ to either superior or abolished antiviral efficacy in our assays (Table 1). Interestingly, polymorphisms in our live viral sequences did not track to CTL antiviral efficacy. Conversely, we noted that the presence of a phenylalanine at position 89 immediately adjacent to the N-terminus of the FL8 epitope and an additional phenylalanine at position 85, correlated with abolished CTL antiviral efficacy in three rVV isolates; A, AE and G. Variation of N- and C-terminal epitope flanking residues can influence the length and nature of epitope precursors that are generated [39] [40] [41] and the immunoproteasome is known to have a preference for cleaving after hydrophobic residues [42]. Thus, this motif is likely to have a pronounced impact on FL8 cleavage patterns. We therefore repeated the in vitro proteasomal digestion assay to test whether this phenylalanine motif may play a key role in modulating epitope processing. We chose two previously studied viruses that exhibited contrasting epitope generation in digests (associated with contrasting CTL antiviral efficacy); HIV-192UG029 and rVV-Nef-A(HIV-192UG037). Previous digestion of the HIV-192UG029 derived precursor peptide generated a correct C-terminally cleaved FL8 pre-cursor that was present in large quantities. In contrast, digestion of the HIV-192UG037 derived precursor peptide containing this phenylalanine motif generated no FL8-containing pre-cursors at any of the time points tested (Figure 8). Therefore, we took the HIV-192UG029 sequence (KGAVDLSHFLKEKGGLDGLIYSRKR) and designed two new 25-mer oligopeptides; the first in which we substituted the basic histidine residue with a large hydrophobic phenylalanine at position 89 immediately adjacent to FL8 (HIV-192UG029+1: KGAVDLSFFLKEKGGLDGLIYSRKR) and an additional substitution at position 85 (HIV-192UG029+2: KGAFDLSFFLKEKGGLDGLIYSRKR) in which hydrophobic valine was replaced with a hydrophobic phenylalanine. The original oligopeptide and two new oligopeptides were then digested with immunoproteasome for 0, 10, 40, and 70 minutes and peptide fragments were identified using tandem mass spectrometry (Figure 9). The insertion of the phenylalanine motif markedly altered the pattern of immunoproteasomal cleavage. Whilst the digestion of the original HIV-192UG029 sequence generated the correct C-terminally cleaved FL8 pre-cursor in high quantity, the +1 and +2 oligopeptide sequences did not generate any C-terminally cleaved pre-cursor peptides. For the +1 oligopeptide, the breadth of intra-epitope cleavage was enhanced in comparison to the original peptide (in which intra-epitope cleavage site was predominantly focused after the F). For the +2 oligopeptide, despite repeat analyses, only three fragments were identified in addition to the mother peptide. This is indicative that the motif markedly alters both cleavage patterns and the overall quantity of pre-cursors generated. Collectively, these data demonstrate that the phenylalanine motif in the N-terminal sequence can diminish epitope processing, and highlight that even a single virus-isolate polymorphism (H89P) in the flanking region can substantially alter epitope production.
We have shown that striking differences exist in CTL antiviral efficacy to a conserved epitope shared between diverse viral isolates, and that from the panel of HIV-1 isolates only 23% were recognized by CTL. In experiments with five high titre HIV-laboratory strains in vitro, FL8-specific CTL demonstrated efficient viral suppression and a strong IFNγ response against endogenous FL8 peptide presented by cells infected with clade isolate HIV-193RW024, and clade B HIV-1MN, and sub-optimal antiviral activity to clade A isolate HIV-192UG029. However, no CTL antiviral activity was detected against a further two clade B strains, HIV-1HXB2 and HIV-189.6. We also utilised a recombinant vaccinia virus system expressing Nef from eight viral isolates, each from a different group M clade (A–H), to evaluate whether FL8 epitope-specific CTL recognition and antiviral activity differs between viral isolates. Again, we observed significant differences in CTL lysis and IFNγ secretion, with correct epitope processing and CTL recognition completely abolished in three viral isolates (A, AE and G) and impaired in a further four isolates (B, C, F, H). Collectively, we demonstrate that a surprisingly large proportion (77%) of Nef proteins with a conserved FL8 epitope, expressed by HIV-1 isolates or as recombinant vaccinia-Nef were suboptimally recognised by FL8- specific T-cells, despite the presence of the same epitope. Both endogenous expression systems clearly show that the antiviral efficacy of CD8+ T cells to an invariant epitope is heavily dependent on the infecting viral isolate, which occurs independently of clade-grouping. We have shown that inter-clade and intra-clade polymorphisms in the FL8 flanking region modulate epitope processing by the immunoproteasome, to enhance or impair epitope generation, which is associated with altered CTL recognition and antiviral activity in the infecting HIV-1 strains tested. Furthermore, we demonstrate that the ‘swapping’ of the flanking regions between viruses that are recognised and not recognised can appreciably modulate epitope processing, and we identify a N-terminal phenylalanine motif that can diminish epitope generation. When modified, this can help to optimize CTL responses elicited by appropriately designed vaccine vectors containing critical flanking sequences in addition to the epitope that enhance epitope antigenicity.
Although it has been previously shown that flanking residues can impact epitope presentation by MHC, also for HIV-1 derived epitopes [19], [23] such studies have focused on identifying amino acid variation predominantly within a single viral isolate. In contrast, our study evaluated the reason for variable CTL responses despite absolute conservation of a HIV Nef epitope shared by a panel of 13 HIV-1 whole viral isolates and recombinant vaccinia-Nef. Strikingly, our results strongly suggest that the impact of viral variation on efficient antigen processing has been seriously underestimated, and this has been reinforced by a reliance on peptide-based assays to measure T-cell responses in natural history and vaccine studies. Due to the current focus on developing HIV vaccines that elicit T-cells to conserved regions of the viral genome, this is an important finding that has implications for both vaccine design and evaluation of vaccine efficacy.
Our results emphasise that the antiviral efficacy and cross-reactive potential of CD8+ T-cells should be assessed by their ability against cells infected with virus, and it cannot be accurately predicted solely on the basis of epitope conservation or based on the results of CTL assays using exogenously-loaded peptides. Since infecting viral isolates can exploit impaired antigen processing and presentation to hide from immune surveillance, epitope conservation between viruses may not accurately predict the cross-reactive potential and antiviral efficacy of CTL. The lack of epitope presentation in particular viral backbones may be one of the reasons that high levels of CTL elicited in people who become super-infected, either after stable HIV-1 infection [43] or after vaccination [44], appear to be ineffective even when the infecting virus contains the identical epitope. Our data suggest that potential immunogens for cross-clade vaccine design should not be based solely upon invariant epitopes, but should focus upon conserved regions that include similar epitope flanking regions which have been tested via functional assays for endogenous presentation prior to use in vaccine constructs. Similarly, it will be important to measure the ability of vaccine elicited T cells to recognize a range of different virus isolates in whole-virus assays as well as the more conventional peptide-based studies. The use of whole-virus assays is not only informative on the efficacy of antigen processing and presentation, but can also be adapted to detect additional changes in CTL antiviral efficacy attributable to viral functions; which include differential protein expression kinetics, variation in protein expression levels by Tat, and the down-regulation of MHC class I by Nef [31]. For future vaccine design, it may also be necessary to reassess the relative value of utilising particular highly conserved and immunogenic regions of the virus, such as this FL8 epitope or the Gag p24 region [10], [45], since sequence conservation may mask a lack of presentation that undermines the efficacy of vaccine-induced responses. Interestingly, since Gag is much less variable between HIV-1 primary isolates and also during the natural course of infection due to fitness costs associated with changes in sequence [46], the probability of gag epitopes being processed more efficiently among diverse viral isolates may be higher than for epitopes from other HIV proteins. This could potentially contribute to the efficacy of Gag-specific CTL observed in chronic HIV-infected patients [47].
Inter-clade and intra-clade virus specific polymorphims may also shape immunodominance as the targeting of HIV-specific CTL epitopes in a hierarchical pattern is sensitive to alterations in antigen processing and presentation [48]. FL8 is a well characterised immunodominant epitope during acute infection, with FL8-specific T-cells detected in over 70% of HIV-infected HLA-B8 participants tested [49]. Yet, immunodominance does not necessarily mean that the FL8 epitope is optimally processed in the majority of viral strains within each infected individual. The low antigen load generated by a viral isolate with sub-optimal or impaired epitope processing may be sufficient to prime T-cells, but insufficient to trigger cytotoxic killing in vivo, especially by low avidity CTL. In addition to the impact of viral isolate-specific polymorphisms, the high mutation rate of HIV may give rise to potential intra-epitopic and epitope-flanking escape mutations during acute infection [50] that subsequently diminish processing. Therefore, the T cells primed by the infecting virus might not be able to control mutated viruses that emerge during the course of infection. Also, strong immune pressure upon neighbouring epitopes such as B57-KF9 Nef in the first few weeks post-infection [13] or on epitopes overlapping with FL8, such as B60-KL9 [51], A2-FL11, A3-AK9, A3-DK9, A24-HL9, may alter the generation and consequent immunogenicity of FL8. Furthermore, in the absence of sufficient antigen, where cells are infected by non-infectious virus or epitope processing is abolished, dendritic cells may acquire these HIV-infected targets and successfully cross-present HIV antigens to prime the expansion of HIV-specific T-cells [52]. Collectively, these mechanisms may explain why FL8-specific T-cells are immunodominant in acute infection, but unlike other responses, FL8 is an outlier that does not correlate with viral control at set point [53].
In our study, we have used rVV in addition to whole HIV to demonstrate our major conclusion that the anti-viral efficacy of B8 FL8 restricted responses are heavily dependent on viral isolates. Since the backbones for rVV and whole virus are different, it is possibile that rVV may not fully represent whole virus, and vaccinia expression levels and HIV expression levels are not equivalent. Ideally, direct HIV to HIV comparison would be very useful for the hypothesis. However, it is worth noting that the ELISPOT by using live virus (MN) infected cells or Vaccinia virus(MN) both give detectable FL8-specific responses which is indicative that the two assays are comparable. Processing data by using the sequences derived from rVV further confirmed the initial observation utilising rVV in IFNγ ELISPOT and lytic assays. Interestingly, analysis of viral amino acid sequence (arising from PCR amplification and sequencing of our infected targets) consistently differed from the expected Los Alamos National Laboratory (LANL) HIV Sequence Database for each of the five viruses, although the FL8 epitope remained unchanged. Whilst some of the viruses used in our assays were molecular clones, others were clinical isolate swarms, and therefore we could not exclude the possibility that there is heterogeneity in the FL8 epitope or flanking sequences. Yet consistent sequencing results were obtained when multiple PCR were performed, suggesting sequences showed in table 1 were dominant sequences. Considering that HIV undergoes an average of 1 mutation per genome per replicative cycle due to the error-prone nature of the reverse transcriptase, some mutations may become ‘fixed’ or continue to evolve in in vitro assays in which CD4+ T-cells are infected with HIV-1 laboratory strains, even under no selection pressure. Due to this high mutation rate, the LANL database represents a ‘snap shot’ of viral genomes and proteomes at a single time point, and therefore repeated sequencing is necessary.
Our in vitro digests utilised immunoproteasomes as they are typically induced within the first day of viral infection by exposure to IFNγ or TNFα [54] [55]. Although immunoproteasomes can generate a different spectrum of epitopes from standard 20S proteasomes, the effects are thought to be relatively subtle, with more pronounced quantitative rather than qualitative differences [56]. Current estimates suggest that only 15% of peptides are the appropriate 8–11mer length for MHC-loading after proteasomal or immunoproteasomal digestion [57], therefore the generation of extended FL8 precursors is not unusual. The N-terminal extensions are usually trimmed by aminopeptidases in the cytosol or after TAP-mediated transfer into the ER [58], although successful CTL recognition of an endogenously derived 15-mer N-terminally-extended HIV-1 gp160 peptide was recently observed [59]. Since more than 99% of peptides are thought to be destroyed by cytosolic peptidases before encountering TAP [60], a limitation to this technique is that the liberation of FL8-containing precursors in our in vitro assays cannot guarantee efficiency in subsequent processing and presentation steps of FL8. However, the importance of the proteasome in liberating Nef peptides, and the strong correlation between the identities of extended epitopic precursors generated via in vitro proteasomal cleavage and naturally processed peptides acid-eluted from the surface of Nef-transfected cells, indicate that in vitro digests are a reliable tool [38], [61]. These studies on Nef, together with our in vitro inhibition assays and immunoproteasomal digestions, strongly support the central role of the proteasome and immunoproteasome in altering peptide generation. In addition, since the specificity of TAP transport, ERAAP trimming and also HLA-binding affinity are dictated by internal epitope composition, particularly the C-terminal residue [37], they are likely to be similar between viral strains with a conserved epitope (post-proteasomal proteolysis). Therefore, these steps further along in the processing pathway are expected to have limited influence on the generation of the conserved FL8 epitope studied here, although this warrants further investigation. Interestingly, the overlapping HLA-B*60 KL9 epitope (KEKGGLEGL) is flanked by the amino acids FL at the N-terminus [49], and the FL8 epitope (FLKEKGGL) is flanked by SH. These differences can be sufficient to determine a completely different outcome in their processing efficiency by the proteasome.
In conclusion, our findings show that striking differences exist in the antiviral efficacy of CTL to an invariant epitope shared between viral isolates. Only a small proportion (23%) of the HIV-1 Nef proteins, expressed by HIV-1 isolates or as recombinant vaccinia-Nef, elicited optimal FL8-specific CTL antiviral responses, whilst the majority demonstrated impaired or completely abolished epitope processing. We found that virus-specific polymorphisms in the flanking region to a conserved epitope substantially alters immunoproteasomal proteolysis, favouring, impairing or completely abolishing epitope generation, which is correlated with the efficiency of CTL antiviral activity in vitro and may play an important role in determining epitope immunogenicity and immunodominance required to prime T-cells in vivo. CTL antiviral efficacy is heavily dependent on the infecting viral strain; this occurs independently of clade-grouping. This is of major relevance for the design of future vaccines protective against genetically diverse strains of HIV-1, and should be carefully evaluated when assessing the effectiveness of vaccine-induced T-cell responses.
HIV-1 viral isolates 92UG029, 93RW024 (clade A) were obtained from The UNAIDS Network for HIV Isolation and Characterization, and the DAIDS, NIAID. HIV-1 laboratory strains IIIB, MN, 89.6 (clade B) were obtained from the AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH, in addition to eight recombinant vaccinia viruses expressing Nef protein from different group M clades (specified within brackets): HIV-192UG037.1(A), HIV-1MN (B), HIV-196ZM651 (C), HIV-194UG114.1 (D), HIV-1CM235-32 (AE), HIV-193BR020 (F), HIV-192NG83.2 (G) and HIV-190CF056 (H). Details and acknowledgment of each reagent are listed in supplemental material.
CTL clones specific for HLA-B8 epitopes were generated by limiting dilution from the PBMCs of HIV-infected patients responding to the FL8 and EI8 epitope and maintained as described in Dong et al 2004 [62]. The Human T cell leukaemia C8166 line stably expressing CD4+ and HLA-B8 was maintained in R10 media (RPMI 1640 with 100 U/ml of penicillin, 100 U/ml of streptomycin, 2 mM L-glutamine and 10% heat-inactivated fetal calf serum from Sigma-Aldrich).
CTL lysis assays were performed using standard 51Chromium release assays as described elsewhere [63]. Slight modifications were made to test infection with recombinant vaccinia viruses (rVV). In brief, a B-cell line (BCL) expressing HLA-B8 was incubated with 150 µCi of chromium for one hour at 37°C/5% CO2, then washed extensively prior to infection with rVV encoding HIV-nef genes at 2×106 pfu/million cells for one hour in serum-free RPMI. Infected cells were allowed to recover for a further two hours in R10. The labeled target cells were transferred to a round bottom 96-well plate and co-cultured with HLA-matched CTL clones in triplicate at an E∶T ratio of 40000∶5000 per well for a further 4 hours at 37°C. Supernatants were harvested and radioactivity was assessed using a beta-plate counter (Wallac). Spontaneous chromium release was determined by analysing the release from target cells incubated in R10 media and maximum release of incorporated chromium was obtained from target cells treated with 5% Triton X-100 detergent. Specific lysis was calculated = 100 × (experimental lysis − spontaneous lysis)/(maximum lysis − spontaneous lysis).
Standard Human IFNγ ELISPOT assays were performed as described elsewhere [62]. Slight modifications were made when testing recombinant vaccinia viruses (rVV). In brief, a HLA-B08 expressing BCL was infected with rVV encoding HIV-nef genes at 2×106 pfu/million cells for one hour in 1 ml RPMI and recovered as described above. Subsequently, the infected target cells were washed twice and added to pre-coated IFNγ ELISPOT plates with a HLA-matched CTL clone at an E∶T ratio of 400∶20,000, in triplicate, in a final volume of 100 µ/well. Control wells included uninfected BCL (negative) and peptide pulsed uninfected BCL (positive) plus peptide pulsed rVV-infected BCL co-cultured with CTL. Assays were incubated for 6 hours at 37°C/5% CO2 and developed as normal.
Differing HIV-1 strains (2-fold TCID50) were used to infect C8166 cells. Infected cells were washed twice, split with at least 1×106 cells/T25 flask cultured in a total volume of 2 ml R10 for a period of 6, 24, 48, 72 and 96 hours post-infection at 37°C/5% CO2. At each time point, cells were washed, counted and co-cultured with the panel of HLA-matched CTL clones in triplicate at one E∶T ratio of 400∶20000 on the pre-coated interferon gamma IFNγ ELISPOT plates at a final volume of 100 µl/well. Negative controls included the individual CTL clones co-cultured with uninfected target cell line in triplicate, and positive control included each CTL clone co-cultured with uninfected target cells pulsed with 2 µM of specific peptide. ELISPOT plates were incubated for 6 hrs at 37°C/5% CO2 and subsequently washed and developed as described previously. Spot forming units (SFUs) were counted using the ELISPOT reader system AID ELIspot 4.0.
Differing high titre HIV-1 strains (TCID50) were used to infect the C8166 cell pellets during a 90 minute incubation at 37°C/5% CO2 and were subsequently washed (2×) to remove free virus. 5×104 infected cells were co-cultured with HLA-matched HIV-1 specific CTL clones at differing E∶T ratios of 1∶1, 1∶2, 1∶4, 1∶8 & 1∶16 in H10-IL2 (200 U/ml) on a flat bottom 96-well plate, in a final volume of 200 µl per well, at 37°C for 4 days. Each condition was performed in triplicate, including one HLA-mismatched clone as a negative control and virus-infected cells in the absence of CTL as a positive control. Suppression of infected cells by CTL on Day 4 was directly monitored using an intracellular anti-p24 gag mAb (KC57-RD1, Beckman Coulter). The extracellular p24 content in the supernatant was also assayed by quantitative p24 antigen ELISA (Immunodiagnostics) in accordance with the manufacturer's protocol.
Proviral DNA was isolated for each high titre virus from control wells containing only HIV-infected C8166 cells in the viral suppression assay using the PureGene DNA isolation Kit (Gentra Systems) as per the manufacturer's instructions. Nef was amplified from proviral DNA by nested PCR. The nested PCR amplification was carried out in a total volume of 50 µl using the Taq polymerase PCR reaction kit (QIAGEN) and AccuPrime Taq DNA Polymerase High Fidelity kit (Invitrogen), according to the manufacturer's instructions. GTA GCT GAG GGG ACA GAT AG and TGC TAG AGA TTT TCC ACA C. Initial denaturing at 94°C for 120 seconds was followed by 30 cycles of denaturing at 94°C for 30 seconds, annealing at 52°C for 30 seconds and extension at 72°C for 90 seconds. A final extension at 72°C was run for 300 seconds. The internal pair of primers used was GAA GAA TAA GAC AGG GCT and AGG CTC AGA TCT GGT CTA A. Initial denaturing at 94°C for 120 seconds was followed by 30 cycles of denaturing at 94°C for 30 seconds, annealing at 56°C for 30 seconds and extension at 72°C for 90 seconds, with a final extension at 72°C was run for 300 seconds. The PCR products were checked for size on a 1% agarose gel and sequencing was performed in the MRC HIU Sequencing Facility, Weatherall Institute of Molecular Medicine.
1 million C8166 were infected separately with the five high titre viruses (TCID50) for 48 hours whilst 1 million BCL were infected separately with the eight rVV (2×106 pfu/million targets) for four hours. Both sets of virus-infected cells were stained with a biotin-conjugated anti-HLA-Class I B8 monoclonal antibody (AB33716, Abcam) washed twice, and fixed with 1% paraformaldehyde. HLA-B8− negative cells were used to set the negative quadrants and to confirm negligible cross-reactivity of the antibody to other HLA-alleles expressed on target cells. Additional controls included the biotin-conjugated antibody in the presence and absence of streptavidin and uninfected cells. At least 105 live cells per infection were counted using a Cyan flow cytometer and analysed with FlowJo.
The C8166 cells were pre-treated with proteasome inhibitor Epoxomicin, TPP II inhibitor AAF-CMK and aminopeptidase inhibitor Bestatin (all Biomol) at varying concentrations at 37°C for one hour prior. The cells were then infected with HIV-192UG029 at TCID50 for 90 minutes at 37°C/5% CO2. Two controls comprising pre-treated cells not infected with HIV (to determine the affect of the inhibitor on cell survival in the absence of HIV-induced cell death), and HIV infected untreated cells (to determine the maximum response elicited in the absence of inhibitor treatment) were used. All cells were washed after 90 minutes and transferred to a T25 flask containing R10 medium for 36 hrs before being used as targets in a modified Live Virus Elispot with a FL8-specific CTL clone at an E∶T ratio of 400∶20000 as described previously. Slight modifications were made to test infection with recombinant vaccinia virus (rVV) expressing Nef from HIV-1 clade D isolate HIV-194UG114. In brief, BCL expressing HLA-B08 was incubated with 150 µCi of chromium for one hour at 37°C/5% CO2 and then washed extensively. The BCL were pre-treated with proteasome inhibitor Epoxomicin, TPPII inhibitor AAF-CMK and aminopeptidase inhibitor Bestatin (all Biomol) at varying concentrations in serum-free RPMI for one hour at 37°C. The cells were then infected with rVV-Nef-D at 2×106 pfu/million targets for one hour in serum-free RPMI, before recovery in R10 for a further one hour. The labelled target cells were then transferred to a round bottom 96-well plate and co-cultured with a HLA-matched CTL clone in triplicate at an E∶T ratio of 40000∶5000 per well for a further 4 hours at 37°C. Supernatants were harvested and radioactivity was assessed as described previously.
Four 25-mer oligopeptides were designed to cover the FL8 epitope and flanking regions corresponding to the amino acid sequence as determined by sequencing for HIV-1HXB2IIIB, HIV-192UG029 and HIV-192UG037 and hybrid viral sequences. Extended peptides
KAALDLSHFLKEKGGLDGLIYSQKR (HIV-1HXB2IIIB), KGAVDLSHFLKEKGGLDGLIYSRKR (HIV-192UG029), KAAFDLGFFLKEKGGLDGLIYSKKR (HIV-192UG037), KAAFDLGFFLKEKGGLDGLIYSRKR (HIV-192037N-029C), KGAVDLSFFLKEKGGLDGLIYSRKR) (HIV-192UG029+1), KGAFDLSFFLKEKGGLDGLIYSRKR) (HIV-192UG029+2): containing the HLA-B8 restricted FL8 epitope FLKEKGGL (indicated in bold) were synthesized by solid-phase F-moc chemistry on an automated peptide synthesizer (Advanced ChemTech) and purified to >98% by reversed-phase HPLC. In vitro proteasomal digestion of synthetic 25mer oligopeptides was conducted using immuno-20S (i20S) proteasome purchased from Biomol International, essentially as described [64]. For each oligopeptide, 1 µg of immuno-20S proteasome was incubated with 10 µg of peptide in a final volume of 100 µL of 20 mM Hepes, pH 7.8, 2 mM magnesium acetate and 2 mM dithiothreitol and incubated at 37°C/5% CO2. Sample aliquots were taken at several time points (0, 10, 40, 70 minutes) and reactions were terminated by the addition of 0.1 volume of formic acid (FA). Control reactions without the i20S-proteasome were analysed to evaluate non-specific peptide degradation. The i20S-proteasome was pelleted by ultracentrifugation at 100000 g for 5 hours, and supernatants containing digested peptides were analysed using nano UPLC-high/low collision energy switching MS (MSE) as described [65]. In brief, the peptide digests were subjected to chromatographic separation by a NanoAcquity UPLC system coupled to a QTof premier tandem mass spectrometer (Waters, Milford, MA, USA). For peptide precursor and fragment identification and simultaneous quantification, the instrument was run in MSE mode. Peptide peaks corresponding to the original undigested and shorter peptide fragments resulting from immunoproteasomal digestion were identified using a MASCOT search engine (version 2.2) and a custom made database containing HIV-peptide sequences. Quantitative information was obtained from extracted ion chromatograms using MassLynx 4.1 software.
|
10.1371/journal.ppat.1006859 | Infectious blood source alters early foregut infection and regurgitative transmission of Yersinia pestis by rodent fleas | Fleas can transmit Yersinia pestis by two mechanisms, early-phase transmission (EPT) and biofilm-dependent transmission (BDT). Transmission efficiency varies among flea species and the results from different studies have not always been consistent. One complicating variable is the species of rodent blood used for the infectious blood meal. To gain insight into the mechanism of EPT and the effect that host blood has on it, fleas were fed bacteremic mouse, rat, guinea pig, or gerbil blood; and the location and characteristics of the infection in the digestive tract and transmissibility of Y. pestis were assessed 1 to 3 days after infection. Surprisingly, 10–28% of two rodent flea species fed bacteremic rat or guinea pig blood refluxed a portion of the infected blood meal into the esophagus within 24 h of feeding. We term this phenomenon post-infection esophageal reflux (PIER). In contrast, PIER was rarely observed in rodent fleas fed bacteremic mouse or gerbil blood. PIER correlated with the accumulation of a dense mixed aggregate of Y. pestis, red blood cell stroma, and oxyhemoglobin crystals that filled the proventriculus. At their next feeding, fleas with PIER were 3–25 times more likely to appear partially blocked, with fresh blood retained within the esophagus, than were fleas without PIER. Three days after feeding on bacteremic rat blood, groups of Oropsylla montana transmitted significantly more CFU than did groups infected using mouse blood, and this enhanced transmission was biofilm-dependent. Our data support a model in which EPT results from regurgitation of Y. pestis from a partially obstructed flea foregut and that EPT and BDT can sometimes temporally overlap. The relative insolubility of the hemoglobin of rats and Sciurids and the slower digestion of their blood appears to promote regurgitative transmission, which may be one reason why these rodents are particularly prominent in plague ecology.
| Yersinia pestis, the bacterial agent of plague, is transmitted by fleas that feed on blood from rodents that carry this disease. The conclusions from studies comparing how efficiently fleas transmit plague after becoming infected have been inconsistent, possibly because a variety of rodent blood sources have been used. To investigate this, we infected three different flea species with Y. pestis using four different types of rodent blood and compared how well they could transmit three days later. The two rodent flea species that transmitted efficiently tended to reflux bacteria and blood into their esophagus when rat or guinea pig blood was used for the infections, but not when mouse or gerbil blood was used. This reflux phenomenon appears to be related to the solubility of the hemoglobin molecule of different rodent species. In contrast, cat fleas, inefficient transmitters, never refluxed their infected blood meal into the esophagus. Rodent fleas that were infected using reflux-inducing rat blood transmitted more Y. pestis than those that fed on infected mouse blood. These findings improve our understanding of how fleas transmit Y. pestis soon after becoming infected and suggest a reason why certain rodents figure more prominently in plague ecology than others.
| Y. pestis is transmitted by the bite of infected fleas, and two modes of transmission have been described: early-phase transmission (EPT) and biofilm-dependent transmission (BDT) [1–3]. Fleas that have taken a highly bacteremic infectious blood meal are capable of EPT on their next feeding attempt within 4 days of becoming infected [4]. An extrinsic incubation period, the time needed for a vector to become infective after acquiring a pathogen, is not required or is very short; fleas can transmit Y. pestis by 24 h after an infectious blood meal [1]. In contrast, BDT does not typically ensue until at least 5–7 days after infection, the time required for a mature biofilm to form in the proventriculus [5, 6]. The proventriculus is a valve in the flea foregut that regulates the ingress of blood and prevents its backflow into the esophagus [7]. BDT occurs when the Y. pestis biofilm begins to interfere with or block normal blood feeding. In partially blocked fleas, biofilm accumulation prevents the proventricular valve from closing completely. Blood can still pass through the partially obstructed valve, mix with bacteria in the midgut, and be forced back out through the incompetent valve and into the bite site due to midgut peristalsis and release of pressure from the bloodsucking pump muscles [7, 8]. The biofilm can eventually fill the proventriculus in some fleas, creating a physical barrier to ingestion. When such completely blocked fleas attempt to feed, incoming blood is unable to pass the obstruction, but dissociates bacteria from the biofilm and forces the esophagus to distend [2, 8]. When the flea stops trying to suck blood, elastic recoil of the esophageal wall causes regurgitation of the infectious mixture into the bite site. The physical and molecular mechanisms of BDT are well described; however, neither aspect has been determined for EPT.
The original descriptions of EPT came from experiments in which fleas were fed on a rodent with terminal bacteremia and then collected shortly after their infectious blood meal and transferred individually or in groups to a naïve rodent, which was monitored for plague morbidity and mortality [9–14]. These studies sometimes used different rodent species as the source of the infectious blood meal. When Oropsylla montana or Xenopsylla cheopis fleas were infected by feeding on bacteremic rats, guinea pigs, or squirrels, 10–100% of animals challenged by groups of 10–100 fleas developed plague [9–11]. However, EPT was rarely observed when individual infected fleas were used (0–1.5% transmission rate per flea), regardless of the flea species or infectious blood source [9, 12–14]. More recent evaluations of early-phase transmission have used an artificial feeder in which rat or mouse blood served as the infectious blood meal. Similar to the earlier studies, when O. montana or X. cheopis were infected using rat blood containing ≥108 CFU/ml Y. pestis, 0–60% and 0–100%, respectively, of Swiss-Webster mice challenged 1 to 4 days later with groups of 7–11 fleas either developed disease or seroconverted, an estimated transmission rate per flea of 0 to ~10% [1, 4, 15, 16]. Fleas apparently transmit few Y. pestis by the early phase mechanism. In early-phase mass transmission tests in which groups of more than 100 O. montana or X. cheopis infected using mouse blood were allowed to feed three days later from a sterile blood reservoir, relatively few CFUs were recovered (median = 3, range = 0–164), and in about half of the experiments no transmission was detected [6]. The total CFUs transmitted in those experiments was below the reported ID50 (250 CFU) of the California ground squirrel, a major host for O. montana [17].
One variable that may complicate interpretation of disparate vector competence data is the source of rodent blood used in the infectious blood meal [18]. In this study, we evaluated whether different sources of infectious rodent blood (mouse, rat, guinea pig, and gerbil) modulate early infection of the flea foregut by Y. pestis or EPT. The rodent blood sources used in this study, except for Mongolian gerbil blood, have been previously used in evaluations of flea vector competence for Y. pestis [1, 2, 6, 10, 12–14]. Gerbil (jird) blood was included in our analysis as Meriones spp. are known to be plague reservoir hosts in Asia [19].
We show that ~10–25% of rodent fleas that consumed rat blood or guinea pig blood, but not mouse or gerbil blood, containing high titers of Y. pestis refluxed a portion of the infectious blood meal from the midgut and proventriculus back into the esophagus shortly after infection, a phenomenon we term PIER (post-infection esophageal reflux). PIER frequently obstructed the passage of blood during the flea’s next blood meal and correlated with increased numbers of Y. pestis CFU transmitted by O. montana. Enhanced in vitro transmission at 3 days after infection was dependent on the ability of Y. pestis to produce the exopolysaccharide required for biofilm formation. Thus, when rat blood was used to infect O. montana, they can become partially or fully blocked and transmit Y. pestis within a time frame typically attributed to EPT.
When a flea feeds, muscles located in the head pump blood through the esophagus, the proventriculus, and into the midgut [20]. As feeding ends, the esophagus is cleared of residual blood while the proventriculus and hindgut confine the blood meal while it is digested in the midgut [21]. Unexpectedly, 24 h after X. cheopis fleas consumed rat blood containing ~1 x 109 Y. pestis KIM6+ (pAcGFP1) CFU/ml and 1 μm fluorescent beads, both blood and beads were observed in the esophagus in 21 of 80 (26%) of fleas (Fig 1A). We termed this post-infection esophageal reflux (PIER). The distance blood and beads refluxed through the esophagus due to PIER was variable, ranging from just above the proventricular-esophageal junction to within the head. In contrast, PIER was not observed when fleas were fed sterile rat blood or when mouse blood served as the infectious blood meal; the blood meal remained confined to the proventriculus and MG. Because PIER was not evident immediately after the infectious blood meal and the fleas had not fed subsequently, we concluded that a portion of the blood meal had refluxed from the proventriculus or MG back into the esophagus during the first 24 h of infection.
To determine the incidence of PIER in infected flea populations and confirm that the fluorescent beads did not induce PIER, groups of 82–296 rat fleas were fed a sterile or high-titer infectious blood meal (without fluorescent beads) from 1 of 4 different rodent species: Sprague-Dawley rat (Rattus norvegicus), RML Swiss-Webster mouse (Mus musculus), domestic guinea pig (Cavia porcellus), or Mongolian gerbil (Meriones unguiculatus). Immediately following feeding, and again 24 h later, fleas were immobilized, examined under a microscope, and the number of fleas with and without PIER was recorded (Fig 1B). One day after feeding on infected rat or guinea pig blood, 8–28% of fleas exhibited PIER (Fig 1C). In contrast, ≤1.2% of fleas fed bacteremic mouse or gerbil blood, or fleas fed sterile blood from any rodent species, developed PIER (Fig 1C). Induction of PIER did not require the Y. pestis hmsHFRS-dependent exopolysaccharide needed for biofilm formation, or the Pla protease (Fig 1C and Table 1). Adding live Escherichia coli S17-1 or DH5α (pCH16), or heat-killed Y. pestis to the blood meal did not induce PIER at rates comparable to live Y. pestis (0–6% vs. ~22%, respectively), indicating that metabolic activity and/or replication of Y. pestis in the digestive tract enhances development of PIER.
To verify that the method of anti-coagulation did not influence induction of PIER, fleas were infected with both defibrinated and sodium heparin treated mouse or rat blood (Fig 1C and S1 Fig). Identical rates of PIER were observed in infected fleas regardless of the treatment used to prevent the blood from clotting. Furthermore, PIER occurred in fleas infected using washed rat RBCs in PBS, indicating that coagulation factors or other plasma components are not essential for PIER (S1 Fig).
The data show that a portion of the infectious blood meal is often refluxed into the esophagus within 24 h of ingesting rat or guinea pig blood containing high titers of Y. pestis, but not when bacteremic mouse or gerbil blood is ingested.
A previous study indicated that the species of host blood affects Y. pestis infection rates in rodent fleas [18]. We found that in the short term (1 day after infection), X. cheopis that fed on blood containing ~5 x 108 Y. pestis/ml had similar infection rates and bacterial loads regardless of the rodent blood used for the infectious blood meal (Fig 1D). Flea groups infected with mouse, rat, or guinea-pig blood had 100% infection rates and equivalent numbers of Y. pestis (~5 x 105–1 x 106 CFU/flea) in the digestive tract for 24 h following the infectious blood meal. Fleas infected using gerbil blood had slightly lower infection rates (93%) and more variable bacterial burdens; however, the majority had bacterial burdens comparable to those infected using other rodent blood sources. These results indicate that differences in replication or clearance of Y. pestis in the flea digestive tract do not explain why some blood sources induce PIER while others do not.
Y. pestis is non-motile and does not adhere to the midgut epithelium during the early stages of flea infection, so it seemed likely that plague bacilli would also be present in the esophagus following PIER [30]. To confirm this, the digestive tract of X. cheopis fleas were scored for PIER 24 h after infection, dissected, and the foregut was screened for localization of GFP+ bacteria to the esophagus. Digestive tracts of fleas infected with rat or guinea pig blood containing Y. pestis had high rates (~40–90%) of bacterial localization to the esophagus (Fig 2A and 2B). Conversely, only 3 to 10% of fleas infected using mouse or gerbil blood contained Y. pestis in the esophagus, even though 88% (50 of 57) of them contained Y. pestis in the proventriculus.
Initially, the proventriculus of all fleas contained a portion of the blood meal regardless of host blood source. Placement of a coverslip on the dissected digestive tract of all 57 fleas infected using mouse or gerbil blood was sufficient to force most this out into the esophagus or back into the midgut, leaving the valve somewhat flattened and its spines clearly distinct (Fig 2A). In contrast, the proventriculus of 43 of 50 fleas (86%) infected using rat or guinea pig blood, and of 33 of 34 fleas diagnosed with PIER, remained bulbous and retained blood meal contents, which obscured visualization of the proventricular spines (Fig 2A and 2C). In addition, the proventricular spines of fleas infected with rat or guinea pig blood appeared to be spread further apart when compared to spines of fleas infected using mouse or gerbil blood (Fig 2A). Thus, the flea foregut is rapidly colonized following the infectious blood meal and certain rodent blood sources promote PIER, esophageal localization of Y. pestis, and enhanced retention of partially digested infected blood in the proventriculus.
Upon dissection of X. cheopis that fed on rat or guinea pig blood, the presence of large numbers of crystals in the digestive tract became apparent (Fig 3A, 3B, 3E, 3F, 3I and 3J). These crystals were not observed in the digestive tract of infected fleas that fed on mouse or gerbil blood (Fig 3C, 3D, 3G and 3H). The morphology and color of these crystals were identical to descriptions of oxyhemoglobin crystals of the rat (rhomboidal or trapezoidal) and guinea pig (pyramidal) [31, 32]. Most of the rat hemoglobin crystals in the flea guts formed rhomboids, but some were jagged and trapezoidal. Crystallization of oxyhemoglobin during digestion of sterile rat and guinea pig blood has been observed previously in the gut of several blood-feeding arthropods, including soft ticks, kissing bugs, lice, and X. cheopis fleas [33–36]. Consistent with previous observations, oxyhemoglobin crystal formation was not dependent on the presence of Y. pestis: all 10 fleas dissected 4 h after feeding on sterile rat blood contained numerous crystals in the gut. Crystals varied in size and concentration but were present in >70% of fleas fed rat or guinea pig blood containing Y. pestis for at least 24 h following the infectious blood meal (Fig 3K). Oxyhemoglobin crystals were prevalent in the midgut of 32 of 34 fleas diagnosed with PIER and for many of these, crystals were also observed in the proventricular-esophageal junction (Fig 3L and 3M). Even though the thick muscle layers surrounding the proventriculus precluded direct observation of crystals, the general abundance of oxyhemoglobin crystals in the gut and frequent observations of oxyhemoglobin crystals within the proventricular-esophageal junction and esophagus indicate that the material in the proventriculus also included oxyhemoglobin crystals.
Rattus norvegicus hemoglobin is poorly soluble and readily crystalizes after RBC hemolysis [32, 37, 38]. We generated hemoglobin crystals with similar morphology to those found in the flea gut by lysing rat RBCs in sterile water at room temperature for 10 minutes (S2A Fig). Lysis of mouse, guinea pig, or gerbil RBCs using an identical protocol did not result in the formation of hemoglobin crystals. Absorption spectra of a suspension of the crystals derived from rat RBCs showed peaks at 540 and 570 nm, which is characteristic of oxyhemoglobin (S2B Fig) [33, 37].
Rodent blood sources with poorly soluble hemoglobin were also digested more slowly in the flea gut than was mouse blood. Most fleas infected with rat, guinea pig, or gerbil blood contained abundant partially digested RBC stroma 24 h after infection, whereas most fleas that consumed infectious mouse blood had completely digested their blood meal–the gut primarily contained a clear, viscous sepia-colored liquid and little or no RBC stroma (3E-H).
To directly assess whether rat RBCs are critical for induction of PIER, X. cheopis fleas were infected with Y. pestis suspended in washed rat RBCs in PBS (S1 Fig). PIER was seen in 15% of these fleas, indicating that rat RBCs alone are sufficient to induce PIER. Furthermore, addition of lysed rat RBC components (stroma and oxyhemoglobin crystals) to infectious mouse blood (which would not normally induce PIER) was sufficient to cause PIER in some fleas that fed on the mixture (S1 Fig). PIER was only observed when rat RBCs were present in the infectious blood meal.
The observations collectively indicate that the proventriculus becomes congested with a combination of oxyhemoglobin crystals, bacteria, and RBC stroma when fleas feed on rat or guinea pig blood containing Y. pestis.
To determine whether other known vectors of Y. pestis develop PIER, we infected groups of O. montana (like X. cheopis, an efficient EPT vector) and C. felis (a poor EPT vector) using rat blood and screened them for PIER. PIER was not seen in infected C. felis 24 h after the infectious blood meal (Fig 4A and 4C). In contrast, PIER was observed in 11% of O. montana, an incidence rate that was about half that observed for infected X. cheopis (Figs 1C, 4B and 4C). Similarly to X. cheopis, ~50% of infected O. montana had bacteria localized to the esophagus and accumulated partially digested blood debris in the foregut, which frequently obscured visualization of the proventricular spines (Fig 4B, 4D and 4E). The percentage of infected C. felis with bacteria in the esophagus and partially digested red blood in the foregut was significantly lower, 16 and 10%, respectively. 80–95% of O. montana remained infected 24 h following the infectious blood meal whereas the C. felis infection rate was more variable, ranging from 50–100% (Fig 4F). Bacterial burdens were comparable between flea species. Oxyhemoglobin crystals were observed in the digestive tract of both flea species; however, they were less numerous in C. felis, correlating with reduced incidence of partially digested blood material in the cat flea foregut. Thus, the efficient EPT vectors O. montana and X. cheopis regularly develop PIER, but the poor EPT vector C. felis does not.
To discern whether PIER interfered with ingestion of blood, X. cheopis and O. montana infected using rat blood were subsequently fed a sterile blood meal containing fluorescent beads 3 days after the initial infection. Prior to being fed on the bead-laden blood, fleas with PIER were separated from those without PIER. Following the blood meal, fleas that appeared to have proventricular obstruction (presence of fresh blood in the esophagus) were imaged to track the location of fluorescent beads in the digestive tract. Most fleas diagnosed with obstruction appeared to be either partially blocked (23–52%; beads present in both the esophagus and the midgut) (Fig 5A and 5D) or fully blocked (46–48%; beads present in the esophagus and proventriculus only) (Fig 5B and 5E). However, 31% of X. cheopis diagnosed with proventricular obstruction were false positives, with no evidence of beads in the digestive tract (Fig 5C and 5F). These fleas had not actually fed; the initial diagnosis of obstruction was due to PIER-derived blood that remained red for at least 3 days post-infection. This was not observed for O. montana, in which PIER-derived blood in the esophagus was dark brown and easily distinguished from recently ingested blood, and those diagnosed with obstruction always had beads present in the foregut.
O. montana or X. cheopis diagnosed with PIER prior to feeding on the bead-laden blood were 3 to 25 times more likely to have evidence of at least partial obstruction of the proventriculus (fresh blood and beads trapped in the esophagus) compared to those without PIER (Fig 5G). Fresh red blood trapped in the esophagus after feeding is typically observed in fleas that transmit by the biofilm-dependent regurgitative mechanism [2, 8]. These data indicate that PIER often causes sufficient proventricular obstruction to interfere with blood feeding.
To test whether the source of the infectious blood meal affected transmission of Y. pestis, groups of ~200 O. montana or X. cheopis fleas, infected 3 days earlier with mouse or rat blood containing high concentrations of Y. pestis (>1 x 108 CFU/ml; Table 2), were allowed to feed on sterile blood. Immediately following the 1.5 h feeding period, the number of CFU transmitted was determined by plate count. In addition, fleas were screened for evidence of feeding and for foregut obstruction. O. montana and X. cheopis infected using rat blood transmitted ~420- and 3-fold more CFU, respectively, than those infected using mouse blood (Fig 6A, Table 2). Significantly more fleas infected using rat blood had foregut obstruction compared to those infected using mouse blood (Fig 6B). Surprisingly, the increase in CFU transmitted by O. montana infected with rat blood was biofilm-dependent; fleas infected with Y. pestis KIM6+ ΔhmsR, a strain unable to produce and export the exopolysaccharide needed for mature biofilm formation, transmitted 98-fold fewer CFUs than O. montana infected with the parental Hms+ strain. The number of CFU transmitted and the number of fleas with foregut obstruction correlated well for infected O. montana (R2 = 0.84), but not for X. cheopis (R2 = 0.16) (Fig 6C and 6D). False-positive misdiagnosis of a portion of the obstructed X. cheopis may partially explain the poor correlation between foregut obstruction and CFU transmitted in EPT assays (Fig 5F). The data suggest that fleas infected using rat blood develop proventricular obstruction at a faster rate, reducing the extrinsic incubation period required for BDT. Thus, in certain conditions BDT can play a role as soon as 3 days after infection, overlapping with the traditional EPT timeframe (1–7 days post-infection) [39].
As previously reported, flea infection rates following the transmission tests (3 day post-infection maintenance feed) were significantly lower in O. montana that had been infected using mouse blood rather than rat blood (38% vs. 96%, Table 2) [18]. Infection rates of fleas in the mouse blood group were comparable to those of fleas that had been infected with KIM6+ ΔhmsR in rat blood (43%), a strain that cannot permanently colonize the proventriculus [8, 40]. In addition, bacterial burdens of O. montana from the mouse blood group that remained infected were about 10-fold lower than the rat blood group (Table 2).
We considered the possibility that differences in feeding rate may explain why O. montana transmitted more CFU than X. cheopis infected using rat blood. On average, 129–138 (60–63%) infected X. cheopis fed during the EPT transmission assays, whereas 149–172 (72–88%) infected O. montana fed (Table 2). However, even when estimating CFU transmitted on a per flea basis under optimal conditions (rat blood infection), infected O. montana transmitted significantly more Y. pestis than X. cheopis (20 vs 0.8, respectively; Table 2).
Between 1905–1916, soon after fleas were proven to be the important vectors of plague, two temporally distinct phases of flea-borne transmission were described [3]. What became known as mass transmission, and more recently termed early-phase transmission (EPT), occurs when groups of fleas feed on a naïve animal during the first week after their infectious blood meal, and wanes thereafter. A later phase of transmission occurs after Y. pestis produces sufficient stable biofilm growth in the proventriculus to partially or completely block normal blood flow during feeding, and is routinely effected by a single flea during continuous feeding attempts as it gradually starves.
Recent work has shed new light on both modes of transmission. EPT of Y. pestis was long assumed to be an example of mechanical transmission, and EPT of Rickettsia felis by cat fleas occurs by this mechanism [41]. However, the fact that different fleas vary greatly in their EPT efficiency following infection with Y. pestis, and that early-phase mass transmission by fleas infected with Y. pseudotuberculosis is never detected, argue against a simple mechanical mechanism, because all fleas would have been expected to soil their mouthparts and transmit equivalently by this means [42, 43]. We recently proposed that EPT, like BDT, occurs via regurgitation of bacteria from the proventriculus [8]. This was based on the striking observation that the proventriculus is a primary site of infection that is often heavily colonized within a day after an infectious blood meal. This was unexpected, because the long-standing conception has been that the midgut is infected first and that the proventriculus is not colonized until a few days later in X. cheopis, and later and less often in other fleas [12, 14]. However, we found that in the majority of all three flea species examined (X. cheopis, O. montana, and C. felis), large aggregates of bacteria, surrounded by an amorphous, brown-colored matrix derived from digestive products of the blood meal, were present in both the proventriculus and the midgut 1–3 days after feeding on blood containing the high concentrations of Y. pestis required to sustain EPT [8].
The results presented here support the regurgitative mechanism model for EPT and provide new details. According to our model ([8]; Fig 7), those fleas with an initial heavy proventricular colonization are competent early-phase vectors, because blood flow through the proventriculus is sufficiently restricted to result in reflux of blood mixed with bacteria back into the bite site. Here we show that the makeup of the proventricular bacterial aggregates differs with different infectious blood sources, and that this can lead to an extension of initial proventricular infection forward into the esophagus and affect EPT efficiency.
Within 24 h after ingesting bacteremic rat or guinea pig blood, X. cheopis and O. montana fleas often regurgitated some blood meal material containing Y. pestis into the esophagus (post-infection esophageal reflux; PIER). Although primary proventricular colonization occurs in 67–100% of X. cheopis and O. montana fleas after they feed on bacteremic mouse blood [8], they rarely exhibit PIER. PIER-inducing blood sources affected early proventricular colonization in two ways. In addition to the presence of infected blood meal regurgitant in the esophagus, fleas with PIER exhibited a more prominent and tenacious early colonization of the proventriculus. This was evidenced by greater expansion of the valve (Fig 2A) and an increased incidence of obstruction of blood flow into the midgut upon the first maintenance feed 3 days after infection compared to fleas without PIER (Fig 5; Table 2). The extrusion of bacteria forward into the esophagus and the enhanced proventricular obstruction associated with PIER would both be predicted to enhance a regurgitative EPT mechanism. EPT efficiency, as assessed by the average CFU transmitted per flea, was in fact ~4-fold and ~100-fold higher for PIER+ than for PIER- X. cheopis and O. montana, respectively (Table 2, Fig 6). Interestingly, most of this increase observed in O. montana was dependent on an intact hmsHFRS operon, which is required for Y. pestis to produce the poly-ß-1,6-N-acetyl-D-glucosamine exopolysaccharide of a mature biofilm and for BDT [5]. This result indicates that, although PIER fortifies initial proventricular colonization and obstruction, even with PIER the early-phase foregut obstruction generates less regurgitative pressure than during later BDT. The CFU transmitted per flea by the later, biofilm-dependent mechanism can be orders of magnitude higher than by EPT [6, 42, 44].
Evidence that EPT is due to regurgitation from the flea foregut includes: 1) The proventriculus (or proventriculus and esophagus in the case of PIER) is often heavily colonized by Y. pestis within 24 h after a highly bacteremic blood meal in all flea species examined ([8]; Fig 2). 2) The complex aggregate of bacteria and digestive byproducts that accumulates in the foregut frequently obstructs ingestion during the next blood meal, and this correlates with increased EPT efficiency (Fig 6; Table 2). 3) Flea species whose behavior and physiology promotes retention of bacteria in the foregut and induction of PIER (O. montana and X. cheopis) are relatively efficient EPT vectors whereas those that do not (C. felis) are inefficient (Fig 4) [6, 43, 45]. 4) Y. pestis Hms mutants unable to synthesize and export the exopolysaccharide required for mature biofilm growth nevertheless form typical aggregates, transiently colonize the proventriculus, and are transmitted during the early-phase window (Fig 6 and Table 2) [8, 15]. 5) EPT efficiency is not affected by loss of the phospholipase D activity encoded by the Y. pestis ymt gene, which enhances bacterial survival in the midgut but is not required for infection of the proventriculus [40, 46].
PIER correlated strongly with the use of rodent blood sources that are characterized by poorly soluble hemoglobin and a relatively slow digestion rate in the flea. The major constituent of a blood meal is water, much of which is rapidly eliminated through diuresis, concentrating the formed elements of the blood into a bolus [21, 36]. Hemolysis with release of hemoglobin occurs within 3–6 h after a blood meal [21, 47]. Free oxyhemoglobin can crystalize in the digestive tract, depending on the solubility characteristics of the particular species of hemoglobin [36]. The rate that oxyhemoglobin crystals and RBC stroma are processed and eliminated varies in different flea species and for different blood types (Fig 2 and Fig 4). PIER-inducing rat and guinea pig blood is digested more slowly in the midgut and the hemoglobin of these two species is poorly soluble and crystalizes rapidly, whereas non-PIER-inducing mouse blood is typically digested within 12–24 h and its hemoglobin does not readily form crystals [31, 32, 36]. Notably, X. cheopis that feed on sterile rat blood digest their entire blood meal and the hemoglobin crystals that form disappear within 20–30 h of feeding [36]. This timeframe of RBC hemolysis and subsequent hemoglobin crystal formation roughly correlates with the earliest observations of PIER in infected fleas (8 h post-infection). Interestingly, fleas that consumed sterile blood of any type or those that ingested E. coli or heat-killed Y. pestis rarely developed PIER (Fig 1A and 1C), indicating that metabolic activity of Y. pestis is important for PIER induction.
We hypothesize that PIER occurs when accumulation of partially digested blood meal material and oxyhemoglobin crystal prevents the proventricular spines from overlapping and sealing off the midgut, and bacteria and blood debris are pushed into the esophagus by proventricular contractions and midgut peristalsis (Fig 7B). Typically, the backward directed proventricular spines that circle the lumen of the proventriculus are thought to partially or completely interlock, preventing backflow of blood into the esophagus [7]. The flea midgut and proventriculus musculature continues to contract long after a blood meal, churning and passaging the gut contents between the two compartments [48]. The proventriculus of fleas that develop PIER appears to be enlarged (Figs 1B and 5A–5E) because it is packed with bacteria, partially digested blood, and hemoglobin crystals that spread the proventricular spines farther apart (Figs 2A, 2C and 4B). Consistent with our hypothesis, PIER appeared to inhibit normal post-feeding proventricular contractions to some degree, although midgut peristalsis still occurred (S1 and S2 Videos). Taken together, our results indicate that metabolic activity of Y. pestis, in addition to bacterial aggregation and colonization of the proventriculus, is important for PIER induction.
We considered the possibility that differential agglutination of mouse and rat RBCs by Y. pestis could partially account for the differences in proventricular colonization and induction of PIER because Yersinia pH 6 antigen (Psa) is expressed at 37°C (the growth temperature of the bacteria added to the infectious blood meals) and is a known hemagluttinin [49]. However, comparative hemagglutination assays revealed that RBCs from mice and rats were agglutinated equivalently by KIM6+ (S3 Fig). It is also important to note that while fibrin clots were once hypothesized to be important for chronic infection of the flea proventriculus, this has been ruled out [50, 51]. As such, we used either defibrinated or heparinized blood from mice or rats for our in vitro flea infections and showed that the incidence rates of PIER were equivalent for both methods of anti-coagulation (Fig 1C and S1 Fig). The results indicate that addition of sodium heparin to infectious rodent blood does not affect PIER for rodent fleas. The fact that rat RBCs alone are sufficient for PIER induction (S1 Fig) also indicates that plasma components are not required.
Some differences were observed in the effect that blood source had on early infection characteristics in the three flea species examined. Unlike the two rodent flea species, PIER did not develop in cat fleas infected using rat blood. This may be because cat fleas process their blood meals more quickly, making it less likely that hemoglobin crystals and RBC stroma accumulate (Fig 4E). In addition, cat fleas have a thick band of musculature that narrows and constricts the esophagus at its intersection with the proventriculus, which may act as a sphincter that prevents backflow of material into the esophagus [43]. This musculature is diminished or absent in X. cheopis and O. montana [8, 43]. We hypothesize that these anatomical and physiological traits make it more difficult for Y. pestis to persistently colonize the cat flea proventriculus, less likely to develop PIER, and less efficient at regurgitative transmission, both in the early phase and later [43].
Differences were also observed between X. cheopis and O. montana. Consistent with a previous study [18], the infection rate and bacterial burden for O. montana at 3 days post-infection was significantly lower when the infectious blood meal was composed of mouse blood compared to rat blood (Table 2). In contrast, X. cheopis infection rates and bacterial loads were equivalent with the two blood sources. The rates were also lower for O. montana infected with the Y. pestis KIM6+ ΔhmsR mutant than with KIM6+ wild-type in rat blood.
A second difference between O. montana and X. cheopis was the much greater effect that rat blood had on the number of bacteria transmitted 3 days post-infection (Table 2; Fig 6A). This increased transmission efficiency was biofilm-dependent as well as blood-source-dependent, because it was not seen with the hmsR mutant. This indicates that, in some fleas, sufficient HmsHFRS-dependent exopolysaccharide is produced to stabilize the proventricular aggregate, thereby increasing regurgitative transmission efficiency. We have recently shown that ~1–3% of O. montana infected using mouse blood become partially blocked within 3 days of infection and ~25% are fully blocked by 6 days post-infection [6]. The great gerbil flea Xenopsylla skrjabini can also become fully blocked by 3 days after infection [52]. In the present study, 4–20% of O. montana infected using rat blood appear to become either partially or fully blocked within 3 days of infection (Table 2; Fig 5), generally sooner and with greater frequency than those infected with mouse blood (Fig 6B). Furthermore, biofilm-dependent transmission is not dependent on complete blockage of the foregut; partially blocked fleas can also transmit [7, 8]. Thus, although the extrinsic incubation period required for biofilm-dependent transmission is generally reported as 1–2 weeks after infection for X. cheopis, it can be as short as a few days, suggesting that early-phase and biofilm-dependent transmission do not always occur in two distinct temporal phases.
Infections using rat blood also increased the number of Y. pestis transmitted by X. cheopis during the early-phase, but not to the extent seen for O. montana (Table 2, Fig 6A). This was surprising, because more X. cheopis than O. montana infected using rat blood were diagnosed with foregut obstruction (partial or full blockage) following the day-3 EPT feeds. Given this, it might be expected that groups of X. cheopis would transmit more CFU than O. montana. One confounding variable was that up to ~30% of infected X. cheopis identified as being obstructed in our EPT experiments were likely false positives (Figs 5B, 6C and 6F). The portion of the blood meal refluxed into the esophagus (by PIER) appears to be digested more slowly than the contents of the midgut, causing it to remain bright red for some infected X. cheopis. This limited our ability to visually diagnose obstruction in X. cheopis infected using rat blood when fed within 3 days of the infectious blood meal. Additionally, we have shown that O. montana transmits more Y. pestis than X. cheopis by the BDT mechanism, despite the fact that X. cheopis has a higher blockage rate [6]. We have hypothesized that these differences in transmission efficiency may be related to anatomical features of the flea foregut; specifically, the O. montana foregut appears more susceptible to biofilm-induced distension at or near the esophageal-proventricular junction [6]. This may affect the hydrodynamics of transmission, in that a greater surface area of the infectious biofilm is exposed to incoming blood, increasing the likelihood that bacteria are dissociated and transmitted. Collectively, the data lead us to conclude that O. montana is a more efficient vector since it becomes blocked more quickly than X. cheopis and transmits greater numbers of Y. pestis CFU (Fig 6) [6]. Our data indicate that rat blood-based infections decrease the time required for O. montana to transition from EPT to BDT, resulting in temporal overlap of the two modes of transmission. Based on this study and recent findings indicating that O. montana readily become completely blocked shortly after infection (mean = 9.6 days), even when mouse blood is used, and survive for extended periods after becoming blocked (mean = 7 days, range = 1–16 days), we think it likely that both EPT and BDT are simultaneously operative during epizootics involving the ground squirrel hosts of O. montana [6]. The high vector competence of O. montana may in part account for the rapidity with which Y. pestis spread throughout the western United States after it was introduced shortly before the turn of the 20th century, particularly if ground squirrel blood is PIER-inducing.
A previous study that measured EPT efficiency by groups of ~10 infected O. montana using rat blood found that biofilm-deficient Y. pestis ΔhmsR or ΔhmsT strains were transmitted at least as well as the parental strain [15]. In partial agreement with these observations, biofilm production was not requisite for EPT in our mass transmission experiments; however, significantly more CFU of the parental Hms+ Y. pestis than the ΔhmsR mutant were transmitted (Fig 6A and Table 2), which would be predicted to increase disease incidence following EPT challenge. The two studies used dissimilar experimental models, however: small groups of infected fleas used to challenge Swiss-Webster mice, with disease or seroconversion as the readout; versus large groups of infected fleas feeding on sterile blood in an artificial feeding device, with the number of CFU recovered from the blood reservoir being the readout. Swiss-Webster mice are highly susceptible to plague by peripheral routes (<10 CFU), making that model of EPT highly sensitive for detecting transmission of few CFU [22, 53]. Mice which seroconverted but did not develop fulminant disease were included in calculations of transmission efficiency, which indicates that the number of CFU transmitted was sometimes below the LD50 [15]. An advantage of the artificial feeding model of transmission is that it allows for quantification of the total CFU transmitted by groups of fleas. We hypothesize that the large increase in CFU transmitted by O. montana infected with the Hms+ compared to the Hms- Y. pestis in our mass transmission assays came from a small number of the 157–199 fleas in which enough of the Hms-dependent extracellular matrix had been produced to effect BDT (Table 2). Flea transmission efficiency, in general, is low by both EPT and BDT mechanisms; however individual blocked fleas can transmit >1,000 CFU by BDT, much higher than is transmitted by EPT [44]. Our use of large pools of fleas increases the likelihood of detecting these relatively infrequent early BDT events.
Hemoglobin solubility is highly variable among the Muridae (mice, rats, and gerbils) [31]. Digestion of rat blood resulted in the formation of abundant oxyhemoglobin crystals in the flea digestive tract, but digestion of mouse or gerbil blood did not (Fig 3). Interestingly, the hemoglobin of the Sciuridae family of rodents, which includes squirrels, marmots, and prairie dogs that are susceptible to plague epizootics, has been described as poorly soluble and prone to crystal formation, suggesting that fleas that feed on ground squirrels or prairie dogs would develop PIER after feeding on highly bacteremic blood [31]. Our findings provide a possible explanation as to why the Sciuridae and their fleas are prominent in plague ecology and epizootics spread rapidly in their populations: fleas that ingest bacteremic blood with poorly soluble hemoglobin can rapidly develop a foregut obstruction that is resistant to dislodgement and promotes regurgitative transmission of Y. pestis. Future studies of vector competence using rodent blood obtained from other hosts susceptible to Y. pestis such as great gerbils, prairie dogs, and ground squirrels will provide valuable information for generating hypotheses about Y. pestis transmission dynamics during plague outbreaks.
Yersinia pestis KIM strains and Escherichia coli cloning strains were grown at 37°C in brain-heart infusion (BHI) broth supplemented with hemin (10 μg/ml) as described previously [43] (Table 1). Unless otherwise noted, KIM strains contained pAcGFP1 (Clontech; Mountain View, CA). E. coli DH5α was transformed with pCH16, a plasmid that encodes Yersinia murine toxin (Ymt), a phospholipase D enzyme needed for bacterial colonization of the flea midgut [28] (Table 1). C. felis, O. montana, and X. cheopis fleas from colonies established at Rocky Mountain Laboratories (RML) were starved for 4–5 days prior to infection. Groups of about 150–300 fleas were allowed to feed from an artificial feeding device containing 5–6 ml of sterile blood or blood that contained 1 x 108–1 x 109 bacterial CFU/ml [54]. For other infection experiments, the bacteria were added to 5 ml of washed rat RBCs in PBS or to 5 ml of defibrinated mouse blood to which the RBC stroma and oxyhemoglobin crystals derived from 5 ml of rat blood (prepared as described below) had been added. For experiments using heat-killed bacteria, bacteria were concentrated, suspended in 1 ml of sterile phosphate-buffered saline (PBS), and incubated at 65°C for 20 minutes prior to addition to the blood meal. Loss of viability was confirmed by plating 50 μl of the heat-killed bacterial suspension on blood agar. Blood sources for the infections were defibrinated or heparinized Sprague-Dawley rat (Rattus norvegicus), defibrinated or heparinized Swiss-Webster mouse (Mus musculus), defibrinated Hartley guinea pig (Cavia porcellus), or defibrinated Mongolian gerbil (Meriones unguiculatus). Most blood was purchased from Bioreclamation IVT (New York). Sodium heparin treated mouse blood was collected on site. After a 1.5 h (O. montana and X. cheopis) or 4 h (C. felis) feeding period, fleas were screened for evidence of feeding and only those that had taken an infectious blood meal were used in experiments. When groups of fleas were not being screened for PIER or used in EPT assays, they were housed in plastic capsules and kept at 21°C, 75% relative humidity [54].
The Y. pestis hmsR mutant, which lacks 681 base pairs of the open reading frame (amino acids 63–290) including the entirety of the glycosyltransferase domain, was generated by allelic exchange using the pCVD442hmsR suicide vector (Table 1). An in-frame deletion (described above) was generated by inverse PCR of the pDONR221hmsR Gateway vector (Pathogen Functional Genomics Resources Center, J. Craig Venter Institute), which contains the Y. pestis KIM hmsR ORF as well as ~500 base pairs up- and downstream, using 5’ phosphorylated primers GGAAACAGCGTCTCCGCTGGGCGCAAGGCGGTGCGG and CCCACGGCCAGTGCCTCTCTCGGCGTAGCCAGAAATAGC. The deletion fragment was amplified from the cloning vector by PCR using primers CGCTCTAGAAGCGGTGGACTCGTTACAAG and GCGGAGCTCACTGGAGCAACTTCTGGCAG and was ligated into pCVD442 using restriction sites XbaI and SacI. Cointegrants were generated in Y. pestis KIM6+ by conjugation with E. coli S17-1 (pCVD442hmsR). Following sucrose selection, the hmsR mutant was validated by plating on Congo red agar and by PCR.
Constituents of whole blood were separated by centrifugation at 1500–2000 x g, plasma was removed, and RBCs were washed 3 times with sterile PBS. For hemolysis, RBCs were resuspended in dH2O equal to the volume of packed RBCs for 10 minutes at room temperature or were frozen and thawed (20 minutes, -80°C). Following hemolysis, RBC stroma and oxyhemoglobin crystals were isolated by centrifugation at 10,000 rpm for 10 minutes. Absorbance of oxyhemoglobin crystals was measured using a TECAN Safire2 plate reader.
For hemagglutination assays, Y. pestis was cultured at 37°C in BHI, resuspended in sterile PBS to a concentration of ~5x107 CFU/ml; and 2-fold serial dilutions were made in a 96-well round-bottomed polystyrene dish. Washed mouse or rat RBCs were added to each well to a final concentration of 1% and thoroughly mixed with bacteria. Wells that contained unsedimented RBCs after 3h at room temperature (persistence of a hazy suspension) were scored as positive for agglutination.
24 h following infection, fleas were immobilized by placing them on ice and screened using a dissecting microscope for signs of post-infection esophageal reflux (PIER). PIER was defined as the presence of red blood in the esophagus noticeably beyond the esophageal-proventricular junction following an infectious blood meal but prior to a subsequent maintenance feed on sterile blood. The number of fleas positive for PIER was recorded and fleas were returned to the capsule. For fluorescent tracking of PIER, 1 μm red fluorescent (580/605 nm) sulfate FluoSpheres (Thermo Fisher Scientific; Waltham, MA) was added to the infectious blood meal to a final concentration of ~3.6 x 108 beads/ml. To evaluate and verify proventricular obstruction, fleas were fed on sterile rat blood containing fluorescent beads 3 days after being infected using rat blood without beads.
Samples of 10–20 female fleas were collected immediately after their infectious blood meal and at 24 or 72 h post-infection and stored at -80°C. Subsequently, fleas were mechanically disrupted using a FastPrep24 bead beater (MP Biomedicals, Santa Ana, California), as previously described [43]. Aliquots of triturated fleas were plated in BHI soft agar overlays, supplemented with 100 μg/ml carbenicillin and 10 μg/ml hemin, to determine the prevalence of infection and average CFU per infected flea. Fleas that ingested fluorescent beads were not used for infection rate or CFU assays.
Mass transmission experiments were performed as described previously [6]. Briefly, 3 days following infection, ~200 fleas (approximately equal numbers of males and females) were allowed to feed for 90 minutes on sterile defibrinated rat blood in the artificial feeding system. After feeding, fleas were collected to determine how many had fed, and of those, how many had evidence of digestive tract obstruction (fresh red blood in the esophagus). Blood was removed from the feeding apparatus and the interior of the feeder was washed ten times with 3 ml PBS. The entire volume of blood, as well as resuspended material concentrated by centrifugation from the PBS wash step, were distributively plated on blood agar plates. The external surface of the mouse skin membrane was disinfected with ethanol, cut into small pieces, and homogenized using a bead beating apparatus (MP Biomedicals, Santa Ana, CA). Skin samples were concentrated by centrifugation and plated separately on blood agar. Fleas that ingested fluorescent beads were never used for EPT assays. Blood agar plates were incubated at 28° C for 48 h prior to colony counts.
Fleas were placed in PBS on a glass microscope slide and dissected with a set of fine forceps. The flea exoskeleton was removed and a glass cover slip was placed over the top of the digestive tract. For scoring esophageal localization of Y. pestis, digestive tracts in which ~25% or more of the length of the esophagus contained GFP+ bacteria were considered positive. Digestive tracts in which the proventricular valve contained red material that obscured visualization of at least 50% of the proventricular spines in bright field images were counted as positive for accumulation of partially digested blood in the proventriculus. For presence or absence of oxyhemoglobin crystals, 8 fields (half at 20x and half at 40x magnification) of digestive debris that leaked out of the midgut were screened per flea, and if any rectangular, pyramidal, rhomboidal, or trapezoidal objects were observed, the flea was considered positive for hemoglobin crystals. Intact fleas were immobilized by placing them on ice for 5–10 minutes, transferred to a drop of PBS on a microscope slide, overlaid with a glass cover slip, and PBS was added beneath the coverslip to completely submerge the flea. After 1 minute, the fleas were observed for 3–5 minutes for proventricular contraction and midgut peristalsis. To determine the location of ingested fluorescent beads in the digestive tract, fleas were examined by fluorescence microscopy. Images and videos of flea digestive tracts, bacterial biofilms, and fluorescent bead-laden fleas were obtained using a Nikon Eclipse E800 microscope or a Nikon SMZ 1500 dissecting microscope. The B-2A and G-2E/C fluorescent filter sets (Nikon) were used to acquire images of fleas containing GFP+ bacteria and fluorescent beads. Pictures and videos were obtained with an Olympus DP72 camera and cellSens software.
All analyses were performed using GraphPad Prism 7 (GraphPad Software Inc., La Jolla, Ca.). Statistical tests used and relevant p values are indicated in the figure legends.
All experiments involving animals were approved by the Rocky Mountain Laboratories, National Institute of Allergy and Infectious Diseases, National Institutes of Health Animal Care and Use Committee (protocols 16–011-E and 16–058) and were conducted in accordance with all National Institutes of Health guidelines.
|
10.1371/journal.pgen.1003831 | PKA Controls Calcium Influx into Motor Neurons during a Rhythmic Behavior | Cyclic adenosine monophosphate (cAMP) has been implicated in the execution of diverse rhythmic behaviors, but how cAMP functions in neurons to generate behavioral outputs remains unclear. During the defecation motor program in C. elegans, a peptide released from the pacemaker (the intestine) rhythmically excites the GABAergic neurons that control enteric muscle contractions by activating a G protein-coupled receptor (GPCR) signaling pathway that is dependent on cAMP. Here, we show that the C. elegans PKA catalytic subunit, KIN-1, is the sole cAMP target in this pathway and that PKA is essential for enteric muscle contractions. Genetic analysis using cell-specific expression of dominant negative or constitutively active PKA transgenes reveals that knockdown of PKA activity in the GABAergic neurons blocks enteric muscle contractions, whereas constitutive PKA activation restores enteric muscle contractions to mutants defective in the peptidergic signaling pathway. Using real-time, in vivo calcium imaging, we find that PKA activity in the GABAergic neurons is essential for the generation of synaptic calcium transients that drive GABA release. In addition, constitutively active PKA increases the duration of calcium transients and causes ectopic calcium transients that can trigger out-of-phase enteric muscle contractions. Finally, we show that the voltage-gated calcium channels UNC-2 and EGL-19, but not CCA-1 function downstream of PKA to promote enteric muscle contractions and rhythmic calcium influx in the GABAergic neurons. Thus, our results suggest that PKA activates neurons during a rhythmic behavior by promoting presynaptic calcium influx through specific voltage-gated calcium channels.
| Breathing, walking and sleeping, are examples of rhythmic behaviors that occur at regular time intervals. The time intervals are determined by pacemakers, which generate the rhythms, and the behaviors are carried out by different tissues such as neurons and muscles. How do timing signals from pacemakers get delivered to target tissues to ensure proper execution of these behaviors? To begin to address this question, we study a simple rhythmic behavior in the nematode C. elegans called the defecation motor program. In this behavior, enteric muscles contract every 50 seconds, allowing digested food to be expelled from the gut. The pacemaker is the gut itself, and here we identify a specific protein, PKA, that responds to the signal from the pacemaker by activating certain neurons that trigger enteric muscle contraction. We further demonstrate that PKA activates these neurons by controlling the entry of calcium into these neurons. We also identify two calcium channels that allow calcium to enter the neurons when PKA is activated by the signal from the pacemaker. Our results raise the possibility that PKA-mediated calcium entry might be a mechanism used in other organisms to regulate rhythmic behaviors.
| Cyclic adenosine monophosphate (cAMP) is a potent second messenger that plays an important role in cellular responses to extracellular signals to regulate a wide array of biological processes. In the nervous system, cAMP has been implicated in controlling axon guidance, axonal regeneration, sensory function, learning and memory [1]–[4]. cAMP signaling is also critical for the execution of rhythmic physiological processes such as heart beating and circadian rhythm in a variety of organisms [5]–[8]. However, the mechanism by which cAMP controls rhythmic outputs remains unclear.
cAMP is synthesized by adenylyl cyclases (ACs), which are activated by G protein-coupled receptors (GPCRs) that are coupled to the heterotrimeric G protein α subunit, Gαs [9]. Work in a variety of cell types has shown that cAMP has three major molecular targets: cyclic nucleotide-gated (CNG) channels, exchange proteins directly activated by cAMP (Epac) and cAMP-dependent protein kinase (PKA) (Figure 1A and [9]). CNG channels are non-selective cation channels that are critical for the excitability of certain sensory neurons [10]. Epac proteins are guanine exchange factors for the small G protein Rap, and have been shown to regulate cardiac function and insulin secretion [11]. PKA is a conserved serine/threonine kinase that has been implicated in a wide array of biological processes, including cell growth, neural function, cell differentiation and metabolism [12]. In neurons and neurosecretory cells, PKA regulates the release of neurotransmitter and neuropeptides [13]. PKA activity has also been implicated in the execution of rhythmic behaviors, such as sleep and circadian locomotor activity in the fly [14], [15].
PKA phosphorylates many substrates in excitable cells. For example, in cardiac muscles, PKA phosphorylates the ryanodine receptor and the L-type calcium channel to regulate heart beating [8], [16]. In neurons, several synaptic proteins, such as RIM-1α, synapsin and tomosyn are reported as PKA substrates that regulate neurotransmitter release [13], [17]. In addition, it has been shown that PKA can phosphorylate calcium channels in hippocampal neurons, which may account for PKA-dependent modulation of neurotransmitter release and gene expression [18]. However, it is unclear how PKA impacts the physiology of neurons to regulate rhythmic behavioral outputs.
The C. elegans defecation motor program is a simple rhythmic behavior that occurs about every 50 seconds [19]. Each cycle contains three sequential muscle contractions: the posterior body wall muscle contraction (pBoc), anterior body wall muscle contraction (aBoc), and enteric muscle contraction (Exp, which leads to the expulsion of the gut contents). The period of the defecation cycle is controlled by a pacemaker in the intestine, and the Exp step is initiated by the release of a neuropeptide-like protein NLP-40 from the intestine. NLP-40 instructs the excitation of a pair of GABAergic neurons (AVL and DVB), which in turn release the neurotransmitter GABA to trigger the Exp step [20]–[23]. NLP-40 activates the GPCR AEX-2 on the GABAergic neurons, which is coupled to the heterotrimeric G protein α subunit GSA-1/Gαs, leading to the activation of adenylyl cyclase and the production of cAMP [5], [23]. However, the molecular targets of cAMP in the GABAergic neurons are not known and how cAMP signaling impacts GABA release to mediate the Exp step is unclear.
In this study, we find that KIN-1, the C. elegans homolog of the PKA catalytic subunit, is essential for the rhythmic contraction of enteric muscles. By genetically manipulating the activity of PKA specifically in the GABAergic neurons, we establish that PKA is the downstream target of cAMP in the peptidergic signaling pathway that controls enteric muscle contraction. Furthermore, using in vivo calcium imaging, we find that PKA activates the DVB neuron by promoting calcium influx at presynaptic terminals, and that the voltage-gated calcium channels, UNC-2 and EGL-19, partially mediate PKA-dependent calcium influx. Thus, our results suggest that PKA signaling can control rhythmic behaviors by regulating presynaptic calcium entry.
To identify cAMP effectors that mediate enteric muscle contraction, we first examined C. elegans mutant worms of putative cAMP targets for defects in the Exp step. The C. elegans genome encodes six CNG channels: cng-1, cng-2, cng-3, cng-4/che-6, tax-2 and tax-4 [24], some of which have well characterized roles in sensory transduction [10], whereas the functions of the other channels are largely unknown. Putative null or loss-of-function mutations in any of these CNGs caused no obvious defects in the defecation motor program period or in the execution of pBoc or Exp (Figure 1B and data not shown). The C. elegans genome encodes a single Epac homolog, epac-1. Putative null epac-1 mutants had no defects in the defecation motor program, including in the Exp step (Figure 1B and data not shown). Thus, CNG channels and Epac are unlikely to be the cAMP targets that control the execution of the Exp step.
The PKA catalytic subunit is encoded by a single gene, kin-1, in C. elegans. kin-1(ok338) loss-of-function mutants, which contain a 763 bp deletion in kin-1 that removes part of the catalytic domain, die during embryogenesis. The lethality of kin-1(ok338) animals can be partially rescued by mosaic expression of wild type kin-1 transgenes [25]. We found that a fraction of mosaic kin-1(ok338) animals that survive to adulthood had distended intestinal lumens indicative of constipation. These animals lacked the Exp step in nearly all defecation cycles but both execution of pBoc and cycle length were normal (Figure 1B and 1C and data not shown). The Exp defects seen in constipated kin-1 mosaic worms were almost identical to those of animals lacking nlp-40 or aex-2/GPCR, which are components of the peptidergic signaling pathway that activates the GABAergic neurons (Figure 1B and 1C and [23]). Taken together, we conclude that PKA activity is absolutely required for the generation of rhythmic enteric muscle contractions and that PKA is likely to be a major, if not the only, downstream target of cAMP during the Exp step of the defecation motor program.
To determine whether PKA functions in GABAergic neurons to control the Exp step, we generated worms expressing dominant negative PKA transgenes specifically in these neurons. The PKA holoenzyme is a tetramer composed of two regulatory (R) subunits and two catalytic (C) subunits (Figure 1A). When cAMP levels are low, R subunits bind to and inhibit the activity of C subunits; when cAMP levels increase, two cAMP molecules bind to each of the R subunits at two distinct sites (site A and site B), leading to the dissociation of the PKA holoenzyme and activation of the C subunits [12]. It has been shown that a single amino acid substitution (G324D) in site B of the mouse regulatory subunit (RIα) abolishes cAMP binding and prevents the dissociation of the PKA holoenzyme in vitro [26]. Expression of this mutant R subunit in mice generates a dominant negative effect on PKA activity in vivo [27]. kin-2 encodes the sole R subunit in C. elegans, and KIN-2 shares 74% overall sequence similarity with mouse RIα, and 97% similarity in the cAMP binding sites (Figure S1). We mutated the corresponding Gly residue to Asp (G310D) in the B site of the KIN-2a isoform (referred to as PKA[DN], Figure 2A and S1) and expressed this construct specifically in GABAergic motor neurons (using the full length unc-47 promoter) [5], [23]. Two independently generated PKA[DN] transgenic lines (vjIs76 and vjIs77) displayed distended intestinal lumens, and dramatically reduced cycles in which Exp occurred (Figure 2B and 2D and data not shown). This phenotype was similar to, but not as severe as that of the mosaic kin-1 mutants, likely due to variable expression of the unc-47 promoter in DVB neurons (data not shown). Thus, PKA activity is required in GABAergic neurons to promote the Exp step.
To determine whether PKA functions in the peptidergic signaling pathway activated in the GABAergic neurons that control the Exp step, we examined whether PKA[DN] transgenes could block Exp in animals in which this pathway is constitutively active. gsa-1/Gαs gain-of-function (gsa-1(gf)) mutations restore Exp to aex-2/GPCR mutants, consistent with a role for gsa-1/Gαs downstream of aex-2/GPCR [5]. However, gsa-1(gf) mutations failed to restore Exp to animals expressing PKA[DN] (Figure 2C and 2D). These data show that the effects of the AEX-2/GPCR peptidergic signaling pathway on Exp are dependent on PKA in the GABAergic neurons. For reasons that are unclear, the gsa-1(ce81); PKA[DN] double mutants had a slightly shorter cycle length compared with either single mutant (Figure 2D and data not shown).
To independently confirm that PKA functions in the AEX-2/GPCR peptidergic signaling pathway to regulate the Exp step, we generated a constitutively active kin-1/PKA variant (referred to as PKA[CA]) and examined whether PKA[CA] transgenes could bypass the requirement of nlp-40 or aex-2/GPCR. The His87 and Trp196 residues in the mouse PKA catalytic α subunit lie at the interface between the R and C subunits and are necessary for the interaction between them [28]. H87Q, W196R substitutions disrupt this interaction resulting in a constitutively active catalytic subunit that increases PKA activity in vivo [29], [30]. The C. elegans KIN-1a isoform shares 91% amino acid identity with the mouse PKA catalytic α subunit (Figure S2). We made the corresponding substitutions in the KIN-1a isoform (H96Q, W205R) (Figure 3A and S2), and generated two independent transgenic lines (vjIs102 and vjIs103) expressing KIN-1a(H96Q, W205R) specifically in GABAergic neurons (using the full length unc-47 promoter).
Several pieces of evidence support the notion that PKA[CA] transgenes confer constitutive PKA activity. First, animals expressing PKA[CA] transgenes had grossly normal defecation motor programs (Figure 3B and 3C) but occasionally displayed “ectopic” Exp following normal Exp steps (4 out of 100 cycles in vjIs102 and 10 out of 100 cycles vjIs103. p = 0.11 and 0.03, respectively, n = 10, one tail t-test, compared to wild type) (Figure 3C). Ectopic Exp steps were also observed in kin-2(ce179) mutants (5 out of 100 cycles, p = 0.09, n = 10), which carry a mutation in the PKA regulatory subunit that is predicted to increase PKA catalytic activity [31]. Second, ectopic Exp steps are also observed in gsa-1/Gαs gain-of-function mutants, which mimic constitutive peptidergic activation of the pathway in GABAergic neurons [5]. Finally, PKA[CA] transgenes completely suppressed the Exp defects caused by PKA[DN] expression (Figure 3B and 3C), demonstrating that PKA[CA] is not inhibited by the PKA regulatory subunit.
We found that PKA[CA] transgenes partially restored Exp to aex-2/GPCR and nlp-40 mutants (Figure 3B, 3C, and S3). Similarly, kin-2 loss-of-function mutations restored Exp to nlp-40 mutants to a similar extent as PKA[CA] transgenes (Figure S3). Constitutively active acy-1/adenylyl cyclase transgenes in GABAergic neurons have also been reported to partially rescue the Exp defects of aex-2/GPCR mutants [5]. Taken together, these results are consistent with the idea that cAMP generated by the NLP-40-AEX-2/GPCR peptidergic signaling pathway in the GABAergic neurons activates PKA, which drives Exp by promoting GABA release.
We previously found that during the defecation motor program, the axons of DVB neurons display robust, rapid calcium transients that peak just before each Exp step, suggesting that rhythmic presynaptic calcium influx in DVB drives rhythmic GABA release from DVB neurons. Using a genetically-encoded calcium indicator, GCaMP3 [32], we found that fluorescent spikes in DVB axons began about 3 seconds following the pBoc step, reached maximal intensity about 1 second later (immediately before each Exp step), and returned to baseline within 2 seconds (Figure 4A, 4B and 4C and Video S1 and [23]). In most cases, calcium transients appeared to initiate in the synaptic region of DVB neurons and often would spread along the axon to the cell body. In wild type animals the “normal” pattern of pBoc-calcium transient-Exp was highly reproducible, occurring 100% of the time (23 cycles, 11 animals, Figure 4C and 4D). In mutants lacking aex-2/GPCR, only about 9% of cycles adopted a normal pattern (4 out of 44 cycles, 11 animals), and the remaining 91% of cycles lacked both the calcium transients and Exp (40 out of 44 cycles, 11 animals, Figure 4C and 4D and Video S2). In contrast, in unc-25 mutants, which lack the GABA biosynthetic enzyme glutamic acid decarboxylase (GAD) [33], most of the cycles without Exp steps still produced a calcium spike (31 out of 33 cycles without Exp, 11 animals, Figure 4C and 4D and Video S3). These results show that the generation of calcium transients is dependent on the peptidergic signaling pathway but occurs independently of GABA release from DVB neurons, and are consistent with the idea that the calcium spikes in the synaptic region of DVB neurons drive the Exp step by triggering GABA release.
Our behavioral results show that PKA activity may be necessary and sufficient for GABA release from the GABAergic neurons. In principal, PKA could control GABA release directly by acting on the synaptic vesicle machinery at presynaptic terminals or indirectly by regulating the excitability of the GABAergic neurons. To distinguish between these two possibilities, we examined whether PKA[DN] or PKA[CA] transgenes impacted GCaMP3 fluorescence spikes in DVB neurons during the defecation motor program. PKA[DN] transgenes produced a calcium pattern that was similar to that observed in aex-2/GPCR mutants: 80% of cycles lacked both calcium transients and the Exp steps (31 cycles, 12 animals, Figure 4C and 4D and Video S4). On the other hand, PKA[CA] transgenes caused “ectopic” fluorescent transients in DVB neurons, which occurred once or twice in between cycles on average (22 regular calcium spikes and 26 ectopic calcium spikes observed during 24 cycles, 11 animals, Figure 5A, 5B and 5C and Video S6). These “ectopic” fluorescent transients were occasionally associated with an ectopic Exp (6 out of the 26 ectopic calcium spikes, 11 animals, see Discussion). The duration of both the “normal” and “ectopic” fluorescent transients in PKA[CA] transgenic animals was significantly longer compared to wild type controls (Figure 5D and Video S5 and S6). However, the amplitude of calcium transients was not significantly different (Figure 5E, wild type: ΔF/F0 = 1.83±0.17, n = 30; PKA[CA]: normal calcium transients ΔF/F0 = 2.37±0.22, n = 21, p = 0.06; ectopic calcium transients, ΔF/F0 = 2.08±0.22, n = 26, p = 0.39, two tail t-test, compared to wild type). Taken together, we conclude that PKA is essential for the generation of calcium transients in DVB neurons during the defecation cycle. In addition, PKA activity is sufficient to generate calcium transients and may positively regulate the duration but not the amplitude of calcium transients.
Voltage-gated calcium channels (VGCCs) are critical for presynaptic calcium influx during regulated neurotransmitter release [34]. unc-2 encodes the α1 subunit of the C. elegans P/Q type VGCC. UNC-2 localizes to presynaptic terminals, and promotes calcium influx and neurotransmitter release [35], [36]. unc-2 has been reported to be expressed in the DVB neuron, and to regulate the Exp step [37]. Consistently, we observed the Exp step in only about 40% of the defecation cycles in unc-2(lj1) null mutants (Figure 6A). In unc-2 mutants expressing PKA[CA], the Exp step was observed in about 60% of cycles (Figure 6A). The incomplete restoration of the Exp step by PKA[CA] is consistent with the idea that UNC-2 functions downstream of or in parallel to PKA to regulate Exp.
To test whether the rhythmic calcium transients are mediated by UNC-2, we examined GCaMP3 fluorescence in synaptic regions of DVB neurons in unc-2 mutants. Surprisingly, fluorescent transients were observed in all cycles including the normal cycles that had the Exp steps and the incomplete cycles without the Exp steps (Figure 6B and 6C and Video S7 and S8). However, there was a significant reduction in the average amplitude of fluorescent transients in the incomplete cycles (that lacked Exp, n = 18) compared to the fluorescent transients in the normal cycles (those with Exp, n = 22) (Figure 6B, 6C and 6D). To determine if the reduction in fluorescence amplitude in incomplete cycles was a secondary effect of compromised Exp in these mutants, we examined fluorescence amplitudes in mutants lacking unc-25/GAD. We found no difference in the average amplitude between normal and incomplete cycles in unc-25/GAD mutants (n = 10 and 31 cycles, respectively, and Figure 6D). These results indicate that UNC-2 is not absolutely required for the generation of calcium transients but rather regulates the size of calcium transients, possibly allowing them to reach the threshold required for triggering Exp.
Because unc-2 does not appear be to the only channel to mediate the calcium transients in DVB neurons for the Exp step, we next sought to identify other calcium channels that might function in this process. The C. elegans genome encodes two additional VGCC α1 subunits: egl-19, the L-type α1 subunit, and cca-1, the T-type α1 subunit [38], [39]. We found that loss-of-function cca-1/VGCC mutants had normal Exp steps, whereas loss-of-function egl-19/VGCC mutants displayed a modest reduction in the number of cycles with Exp (Figure 7A). Interestingly, in double mutants lacking both unc-2 and egl-19, the Exp step was nearly eliminated (Figure 7B). Furthermore, PKA[CA]; egl-19; unc-2 triple mutants had the same, severely reduced Exp frequency as the egl-19; unc-2 double mutants (Figure 7B). Thus, two VGCCs, egl-19 and unc-2, function downstream of PKA to control the Exp step.
To directly test whether egl-19/VGCC mediates calcium influx in DVB neurons, we next examined the effects of egl-19 mutations on GCaMP3 fluorescence in DVB neurons. We were unable to analyze calcium influx in egl-19 mutants due to an unexpected genetic interaction between egl-19 and unc-13(s69) (used to paralyze animals for imaging without stopping the defecation cycle) that suppressed the Exp defects of egl-19 mutants (see Materials and Methods). Instead, we examined calcium influx in egl-19; unc-2 double mutants (which were sufficiently paralyzed for imaging without the unc-13(s69) mutation). We found that the fraction of incomplete cycles increased to 80% (from 40% in unc-2 mutants, Figure 7C and Video S9). Finally, no calcium influx was observed at all in 35% of the incomplete cycles, a defect that was not observed in the unc-2 mutants alone (Figure 7C and Video S10). In egl-19; unc-2 double mutants expressing PKA[CA] transgenes, these defects in calcium influx were as severe as those observed in egl-19; unc-2 mutants (Figure 7C), suggesting that egl-19 and unc-2 function downstream of PKA to promote calcium influx into DVB.
egl-19 and unc-2 mutations were not able to block all effects of PKA[CA] transgenes, including the increased frequency of ectopic calcium transients and the increased duration of normal and ectopic calcium transients (Figures 7D and S4 and Video S11 and S12). In addition, the amplitude of regular calcium spikes of the incomplete cycles in egl-19; unc-2 mutants expressing PKA[CA] transgenic was similar to those in PKA[CA] transgenic animals, but significantly larger than those observe in egl-19; unc-2 double mutants (Figure 7E). Together these results suggest that the effects of egl-19 and unc-2 mutations on Exp can in part be attributed to their requirement for calcium influx in DVB neurons and that additional calcium channels must act downstream of PKA to mediate calcium influx in DVB.
In this study, we identify PKA as the major downstream target of cAMP in the NLP-40-AEX-2/GPCR peptidergic signaling pathway that functions in the GABAergic neurons to regulate the Exp step during the defecation motor program in C. elegans. PKA controls rhythmic activation of the GABAergic neurons by promoting presynaptic calcium influx in these neurons. We find that the mechanism by which PKA functions to promote rhythmic calcium influx is partially dependent on the P/Q-type VGCC, UNC-2 and the L-type VGCC, EGL-19. These results suggest that PKA promotes rhythmic neurotransmitter release by controlling calcium influx in neurons during a rhythmic behavior.
Our results suggest that PKA functions as an essential cue for calcium influx in neurons for their activation to control rhythmic behaviors, because both calcium spikes in DVB neurons and Exp steps are abolished in most of the defecation cycles in animals expressing PKA[DN] transgenes (Figure 2B, 2D, 4C and 4D). To our knowledge, this is the first in vivo example showing that PKA in neurons is absolutely required for the execution of a rhythmic behavior. In many preparations, PKA has been shown to modulate biological processes by regulating intracellular calcium concentration. One classic example is that PKA mediates the enhancement of calcium influx in cardiac myocytes to modulate the rate of heart beating in response to norepinephrine from the sympathetic nervous system [8]. In neurons, PKA also has a modulatory role in regulation of synaptic transmission and synaptic plasticity by phosphorylating several synaptic vesicle proteins that function downstream of calcium influx [13].
How does PKA control calcium influx in the GABAergic neurons? Our results suggest that PKA acts at least partially through UNC-2, the P/Q-type VGCC, and EGL-19, the L-type VGCC. unc-2 has been reported to be expressed in the DVB neuron and egl-19 is expressed in many neurons as well [37], [38], . Our behavioral and calcium imaging analysis of egl-19; unc-2 double mutants expressing PKA[CA] transgenes suggests that one or both of these channels must also function in enteric muscles to promote Exp since PKA[CA] can restore normal calcium spike amplitudes but it fails to restore the Exp step in egl-19; unc-2 double mutants (Figure 7B, 7C and 7E). Consistent with this idea, egl-19 has been reported to be expressed in muscles, including some of the enteric muscles [38]. Our data indicates that other non voltage-gated calcium channels must be also required for calcium influx in DVB neurons for the Exp step, since egl-19 and unc-2 mutations together could not completely block the calcium influx in DVB neurons or the effects of PKA[CA] transgenes on calcium influx (Figure 7C, 7D and 7E). The identification of these channels will further our understanding of the mechanism underlying PKA-dependent calcium influx in these neurons.
PKA may regulate calcium influx either by direct phosphorylation of UNC-2 and/or EGL-19 or by an indirect mechanism involving, for example, the regulation of membrane potential. Indeed, both mechanisms have been reported. The fight-or-flight response is controlled by PKA-dependent phosphorylation of the L-type calcium channels Cav1.1 and Cav1.2 which mediate calcium influx in skeletal and cardiac muscles, respectively, to enhance their contraction [8], [41]. EGL-19 does not have the homologous phosphorylation site. In pancreatic beta cells, PKA has been reported to phosphorylate ATP-sensitive potassium channels to regulate the membrane potential [42]. Interestingly, the egl-36/Shaw type potassium channel functions in DVB to regulate the Exp step [43], raising the possibility that it may be a target for PKA.
Our results also show that PKA[CA] transgenes increases the duration of calcium spikes in DVB neurons in both wild type and egl-19; unc-2 mutants, suggesting that PKA regulates calcium spike dynamics independently of egl-19 and unc-2. PKA might regulate the open time or the inactivation of other calcium channels or reduce the rate of calcium clearance from the synaptic region in DVB neurons. Consistent with our observation, injection of the PKA catalytic subunits in cells has been reported to increase the duration of calcium currents [44]–[46].
Our previous work indicates that NLP-40 functions as the timing signal from the intestine and it delivers the timing information to the GABAergic neurons by instructing their rhythmic activation [23]. In this study, we show that the downstream effector PKA is also instructive for the activation of the GABAergic neurons, because the PKA[CA] transgenes can elicit ectopic calcium spikes in between cycles. The observation that ectopic calcium spikes in DVB neurons do not occur in wild type animals suggests that endogenous PKA activity in these neurons must somehow be turned down rapidly following each Exp step. Thus, we propose a model in which rhythmic activation of PKA by the NLP-40-AEX-2/GPCR peptidergic pathway stimulates rhythmic calcium influx in the GABAergic neurons to drive the Exp step. It will be interesting to directly determine whether cAMP levels and/or PKA activity oscillate in DVB neurons and whether these oscillations correspond with the calcium oscillations in vivo. Several negative feedback mechanisms following PKA activation that have been reported in other preparations may also help establish rhythmic PKA activity in DVB neurons. These include the activation of by phosphodiesterases (PDEs) that break down cAMP, calcium-mediated inhibition of adenylyl cyclases and activation of specific phosphatases that counteract PKA activity [9], [47], [48]. PKA activity has been reported to be essential for the initiation of the cAMP-PKA-Ca2+ oscillation circuit in insulin-secreting MIN6 beta cells (Ni et al., 2010). Thus, the interplay between cAMP, PKA and calcium may be a general mechanism by which PKA generates oscillatory signaling circuits in neurons to control rhythmic behaviors.
Rhythmic PKA activation may be essential for the reliability of the Exp step during the defecation cycle, because PKA[CA] transgenes cannot fully restore Exp to animals that lack the NLP-40-AEX-2/GPCR peptidergic signaling pathway components (Figure 3B). Similarly, constitutive acy-1/adenylyl cyclase expression in the GABAergic neurons also only partially rescues the Exp defects of aex-2 mutants [5]. Interestingly, gsa-1(gf)/Gαs mutations almost fully rescue the Exp defects in aex-2/GPCR mutants [5]. Since GSA-1/Gαs functions upstream of both adenylyl cyclase and PKA, it is possible that gsa-1(gf) mutants retain the proper negative feedback mechanisms that allow PKA activity to remain rhythmic.
Rhythmic PKA activation may work together with other mechanisms to ensure the proper execution of the Exp step. Indeed, it has been postulated that a permissive signal may control the refractory period of enteric muscles by allowing Exp to occur only within a small window of time following the beginning of each cycle (the pBoc step) [5], [23]. We speculate that this permissive signal may be entrained to the pacemaker activity in the intestine, because most of the Exp steps observed in the aex-2 mutants expressing PKA[CA] did not happen at random times, but rather occurred a few seconds after the pBoc step. The absence of this permissive signal in between cycles may also explain why PKA[CA] elicits ectopic calcium spikes in DVB neurons more frequently than ectopic Exp steps. The identification of this permissive signal or the nature of the refractory period of enteric muscles would provide more comprehensive understanding on how the rhythmic Exp is reliably generated.
In mammals, several genes encode different PKA regulatory and catalytic subunits, and each of these genes may have several different isoforms due to alternative splicing [12]. Although many PKA isoforms have been knocked out and knocked in in mice to dissect the role of PKA in vivo, compensation by remaining isoforms has complicated the interpretation of these studies [27], [49], [50]. Unlike mammals, C. elegans has a single gene (kin-2) for the PKA regulatory subunit and a single gene (kin-1) for the catalytic subunit, both of which share high similarities with their counterparts in higher mammals. Thus, C. elegans has the potential to be a good genetic model to dissect the physiological roles of PKA. However, studies on the contribution of PKA signaling in C. elegans have been limited since null mutants of either kin-1/catalytic subunit or kin-2/regulatory subunit are lethal. Various strategies using weak alleles of kin-2, pharmacological treatments with PKA inhibitors, and RNAi-mediated knockdown, have revealed roles for PKA in regulating neurotransmitter release, axon regeneration and behaviors [2], [31], [51]. However, non-specific effects of pharmacological PKA inhibitors and spreading of RNAi limit the utility of these approaches [52], [53]. The ability to manipulate PKA activity in a tissue specific manner using the PKA variants developed in this study represents a powerful approach for probing the function of PKA signaling in vivo.
Strains were maintained at 20°C on NGM plates with E. coli strain OP50 as food. The Bristol strain N2 was used as reference strain. The strains used in this study: OJ1672 cng-2(tm4267) IV, FX05036 cng-4/che-6 (tm5036) IV, OJ1748 tax-2(p671) I, KJ5562 tax-4(p678) III; cng-3(jh113) IV; cng-1(jh111) V, RB830 epac-1(ok655) III, DG3393 tnEx109; kin-1(ok338) I, OJ1540 aex-2(sa3) X, OJ1603 vjIs76 [Pttx-3::RFP, 40 ng/µl; Punc-47(FL):: kin-2a(G310D), 50 ng/µl] V, OJ1601 vjIs77 [Pttx-3::RFP, 40 ng/µl; Punc-47(FL):: kin-2a(G310D), 50 ng/µl] IV, KG421 gsa-1(ce81) I, OJ1908 gsa-1(ce81) I; vjIs77 IV, OJ1854 vjIs102 [Pmyo-2::NLS::GFP, 10 ng/µl; Punc-47(FL)::kin-1a(H96R, W205Q), 50 ng/µl] V, OJ1858 vjIs103 [Pmyo-2::NLS::GFP, 10 ng/µl; Punc-47(FL)::kin-1a(H96R, W205Q, 50 ng/µl)] I, OJ1896 vjIs103 I; vjIs77 IV, OJ1909 vjIs102 V; aex-2(sa3) X, OJ1910 vjIs103 I; aex-2(sa3) X, OJ680 unc-13(s69) I, OJ1213 vjIs58 [Pmyo-2::NLS::mCherry, 10 ng/µl; Punc-47(mini)::GCaMP3, 125 ng/µl] IV, OJ1443 unc-13(s69) I; vjIS58 IV, OJ1468 unc-13(s69) I; vjIS58 IV; aex-2(sa3) X, OJ1759 unc-13(s69) I; vjIs58 IV; vjIs76 V, OJ1859 unc-13(s69) I; unc-25(e156) III; vjIs58 IV, OJ1917 unc-13(s69) I; vjIs58 IV; vjIs102 V, OJ1526 unc-2(lj1) X; OJ1899 vjIs103 I; unc-2(lj1) X; OJ1919 unc-13(s69) I; vjIs58 IV; unc-2(lj1) X, JD21 cca-1(ad1650) X, OJ1911 egl-19(n582) IV, OJ1925 egl-19(n582) IV; unc-2(lj1) X, OJ1924 vjIs103 I; egl-19(n582) IV; unc-2(lj1) X, OJ1351 vjIs64 [Pmyo-2::NLS::mCherry, 10 ng/µl; Punc-47(mini)::GCaMP3, 125 ng/µl] II, OJ1905 unc-13(s69) I; vjIs64 II, OJ1918 vjIs64 II; egl-19(n582) IV; unc-2(lj1) X, OJ1923 unc-13(s69) I; vjIs64 II; vjIs102 V, OJ1949 vjIs64 II; egl-19(n582) IV; vjIs102 V; unc-2(lj1) X, OJ794 nlp-40(tm4085) I, OJ1524 kin-2(ce179) X, OJ1525 nlp-40(tm4085) I; kin-2(ce179) X, OJ1855 nlp-40(tm4085) I; vjIs102 V, OJ1856 nlp-40(tm4085) vjIs103 I.
cng-2(tm4267) mutants contain a 330 bp deletion and a single nucleotide (g) insertion (www.wormbase.org), which is predicted to generate a frame shift that truncates CNG-2 at the C-terminus. cng-2(tm4267) was originally reported to be sterile and lethal (Mitani Lab). However, we found that after outcross cng-2(tm4267) mutants were viable.
The defecation motor program was analyzed as previously described [19], [54]. Briefly, L4 stage hermaphrodites were transferred to a new plate. After about 20–24 hours, each worm was transferred to a fresh NMG plate and let to settle down for at least five minutes. Ten consecutive defecation cycles were scored for each worm using the Etho program software (http://depts.washington.edu/jtlab/software/otherSoftware.html) [54]. Only the pBoc and the Exp steps were scored, omitting the aBoc step. 8–10 worms were assayed for each genotype. “Exp per cycle” for each worm was calculated as the ratio of Exp over pBoc. The results were present as mean ± sem for each genotype. Unpaired two-tail Student's t-test with unequal variance was used to examine the significant difference between two different genotypes.
The backbone of the plasmids constructed below was pPD49.26 (A. Fire).
Punc-47(FL), the 1444 bp full length unc-47 promoter, which was expressed in all GABAergic neurons in C. elegans [5], was amplified from N2 genomic DNA using 5′ primer: ccccccGCATGCatgttgtcatcacttcaaactt and 3′ primer ccccccGGATCCctgtaatgaaataaatgtgacgctg. The PCR product was partially digested with SphI and BamHI, and cloned into the MCSI of a derivative plasmid of pPD49.26.
Punc-47(mini), the 215 bp unc-47 mini promoter, which was only expressed in a subset of GABAergic neurons (4 RME, RIS, AVL and DVB neurons) [55], was amplified from N2 genomic DNA with 5′ primer ccccccGCATGCCTGCAGctttcggtttggagagtag and 3′ primer ccccccGGATCCctgtaatgaaataaatgtgacgctg. The PCR product was digested with SphI and BamHI, and cloned into the MCSI of a derivative plasmid of pPD49.26 (with AsiSI and NotI inserted between the NheI and KpnI in MCS II).
GCaMP3 (1353 bp), the genetically-encoded calcium indicator [32], was cloned from a plasmid with GCaMP3 sequence (a gift from Robert Chow, USC) by PCR with 5′ primer ccccccGCGATCGCAAAAatgggttctcatcatcatcatc and 3′ primer ccccccGCGGCCGCttacttcgctgtcatcatttg. The PCR product was digested with AsiSI and NotI, and was cloned into a derivative of pPD49.26 that had unc-47 mini promoter in MCS I and had AsiSI and NotI sites inserted between NheI and KpnI in MCS II to generate the plasmid pHW100: Punc-47(mini)::GCaMP3.
The wild type cDNA of kin-2a (1101 bp) was cloned from an N2 cDNA library (synthesized using NEB Protoscript RT-PCR kit) with 5′ primer ccccccGCTAGCAAAAatgtcgggtggaaacgaagag and 3′ primer ccccccGGTACCttaggtcatcagtttgacgtatgag. The Gly310 residue in KIN-2a was corresponding to the Gly324 residue in the site B of the mouse PKA regulatory subunit that will generate dominate negative PKA when it is mutated to Asp [26]. The kin-2a (G310D) variant was generated by two-step overlapping PCR using the following primers: pair 1 (5′ primer ccccccGCTAGCAAAAatgtcgggtggaaacgaagag and 3′ primer gaagaagagcgatttcGTcgaaatagtccgacattccaag and pair 2 (5′ primer cttggaatgtcggactatttcgACgaaatcgctcttcttc and 3′ primer ccccccGGTACCttaggtcatcagtttgacgtatgag).The final overlapping PCR product was digested with NheI and KpnI, and then cloned into the MCS II in a derivative plasmid from pPD49.26 that contained the unc-47 full length promoter in MCS I to generate the plasmid pHW154: Punc-47(FL)::kin-2a(G310D).
The wild type cDNA of kin-1a (1080 bp) was cloned from N2 cDNA library with 5′ primer ccccccGCTAGCAAAAatgctcaagtttctgaaacc and 3′ primer ccccccGGTACCttaaaactcggcaaactctttg. The His96 and Trp205 residues in KIN-1a are corresponding to the His87 and Trp196 in the mouse PKA catalytic subunits which would generate constitutively active PKA when they are mutated to Gln and Arg, respectively [30]. The KIN-1a (H96R, W205Q) variant was created by two-step overlapping PCR using the following primers: pair 1(5′ primer ccccccgctagcAAAAatgctcaagtttctgaaacc and 3′ primer gaatgcgcttttcgttcaacgtTtgctccacttgcttgagttttac), pair 2(5′ primer gtaaaactcaagcaagtggagcaAacgttgaacgaaaagcgcattc and 3′ primer tctggtgtgccgcacaatgtccTcgttcgtcctttgacacgtttc) and pair 3(5′ primer gaaacgtgtcaaaggacgaacgAggacattgtgcggcacaccaga and 3′ primer ccccccGGTACCttaaaactcggcaaactctttg).The final overlapping PCR product was digested with NheI and KpnI, was cloned into the MCS II in a derivative plasmid from pPD49.26 that contained the unc-47 full length promoter in MCS I to generate the plasmid pHW173: Punc-47(FL)::kin-1a(H96R, W205Q).
Sequencing was performed to confirm the mutations kin-2a(G310D) and kin-1a(H96R, W205Q) in the pHW154 and pHW173, respectively.
Microinjection of expression plasmids into the gonad of C. elegans was performed to generate transgenic animals with extrachromosomal arrays, according to the standard procedure [56]. Generally, total DNA concentration of the injection solution was 100 ng/µl (using the plasmid pBluescript to fill up if needed). The extrachromosomal arrays were integrated into the genome to generate stable transgenic worms using UV irradiation. The integrated transgenic lines were outcrossed at least 6 times.
The plasmid pHW154 (Punc-47(FL)::kin-2a(G310D)) was injected at 50 ng/µl, together with the co-injection marker plasmid KP708 (Pttx-3::RFP) at 40 ng/µl to generate the array vjEx582 [Pttx-3::RFP, 40 ng/µl; Punc-47(FL):: kin-2a(G310D), 50 ng/µl]. This array was integrated into genome to generate the PKA[DN] transgenic strains (vjIs76 and vjIs77) with dominant negative PKA specifically expressed in GABAergic neurons.
The plasmid pHW173 (Punc-47(FL)::kin-1a(H96R, W205Q)) was injected at 50/µl, together with the co-injection maker KP1106 (Pmyo-2::NLS::GFP) at 10 ng/µl to generate the array vjEx709 [Pmyo-2::NLS::mCherry, 10 ng/µl; Punc-47(FL):: kin-1a(H96R, W205Q), 50 ng/µl]. This extrachomosomal array was integrated into genome to generate the PKA[CA] transgenic strains (vjIs102 and vjIs103) with constitutively active PKA specifically expressed in GABAergic neurons.
The plasmid pHW100 (Punc-47(mini)::GCaMP3) was injected at 125 ng/µl, together with co-injection marker plasmid KP1368 (Pmyo-2::NLS::mCherry) at 10 ng/µl to generate the array vjEx429 [Pmyo-2::NLS::mCherry,10 ng/µl; Punc-47(mini)::GCaMP3,125 ng/µl]. This array was integrated into genome to generate transgenic strains vjIs58 and vjIs64, which expressed GCaMP3 in AVL and DVB neurons.
The calcium imaging experiment was performed as previously described [23]. We used two independent transgenic strains (vjIs58 and vjIs64) with AVL and DVB neurons expressing GCaMP3 to perform in vivo calcium imaging on DVB neurons. Both strains had normal Exp steps (data not shown). Although unc-13(s69) mutants had a longer cycle length (78.5±11.4 s, mean ±SD, n = 8), they were also most completely paralyzed and still had normal Exp steps [23]. Thus, unc-13(s69) mutation was included in all strains to immobilize animals for calcium imaging, except for those strains with elg-19(n582);unc-2(lj1) double mutations, because the elg-19(n582);unc-2(lj1) double mutants were almost completely paralyzed. Young adult worms were transferred to NGM-agarose plate seeded with the food OP50. These agarose plates were topped with cover slides and imaged under a Nikon eclipse 90i microscope equipped with a Nikon Plan Apo 40× oil objective (N.A. = 1.0), a standard GFP filter and a Photometrics Coolsnap ES2 camera. Only those worms kept continuous pumping and positioned laterally with the left side pointing to the objective were selected for imaging. Time lapse imaging was obtained using Metamorph 7.0 software (Universal Imaging). Each worm was recorded for 250 s at 4 frames per second (3×3 binning, exposure time ranging from 5 ms to 80 ms dependent on the baseline GCaMP3 fluorescence in DVB neuron in each individual worm). Unlike DVB, AVL neuron is located in the head, so we could not routinely perform the calcium imaging on AVL and observe the Exp step in the same field. However, during some experiments, we did observe that AVL fired at the same time as DVB in coiled worms in which we could see AVL and Exp in the same field.
The quantification of the GCaMP3 fluorescence intensity in the synaptic region of DVB neurons using Metamorph 7.0 software (Universal Imaging) was performed as previously described [23]. The synaptic region of DVB neurons was manually selected as region of interest (ROI) and the average of the GCaMP3 fluorescence intensity for each frame was recorded. Meanwhile, a similar area near the tail region was used as background fluorescence for each frame. The GCaMP3 fluorescence (F) was defined as the (ROI - background). The average of the GCaMP3 fluorescence in the first 10 frames was used as baseline fluorescence F0. For each frame, the change of the GCaMP3 was presented as ΔF/F0 = (F-F0)/F0. The duration of each calcium spike was defined as the time differences between the first frame in which the GCaMP3 fluorescence increased and the frame where the GCaMP3.0 fluorescence returned the baseline. Ectopic calcium spikes describe those calcium spikes that did not happen within 10 seconds after pBoc.
|
10.1371/journal.pbio.1001661 | Generalization and Dilution of Association Results from European GWAS in Populations of Non-European Ancestry: The PAGE Study | The vast majority of genome-wide association study (GWAS) findings reported to date are from populations with European Ancestry (EA), and it is not yet clear how broadly the genetic associations described will generalize to populations of diverse ancestry. The Population Architecture Using Genomics and Epidemiology (PAGE) study is a consortium of multi-ancestry, population-based studies formed with the objective of refining our understanding of the genetic architecture of common traits emerging from GWAS. In the present analysis of five common diseases and traits, including body mass index, type 2 diabetes, and lipid levels, we compare direction and magnitude of effects for GWAS-identified variants in multiple non-EA populations against EA findings. We demonstrate that, in all populations analyzed, a significant majority of GWAS-identified variants have allelic associations in the same direction as in EA, with none showing a statistically significant effect in the opposite direction, after adjustment for multiple testing. However, 25% of tagSNPs identified in EA GWAS have significantly different effect sizes in at least one non-EA population, and these differential effects were most frequent in African Americans where all differential effects were diluted toward the null. We demonstrate that differential LD between tagSNPs and functional variants within populations contributes significantly to dilute effect sizes in this population. Although most variants identified from GWAS in EA populations generalize to all non-EA populations assessed, genetic models derived from GWAS findings in EA may generate spurious results in non-EA populations due to differential effect sizes. Regardless of the origin of the differential effects, caution should be exercised in applying any genetic risk prediction model based on tagSNPs outside of the ancestry group in which it was derived. Models based directly on functional variation may generalize more robustly, but the identification of functional variants remains challenging.
| The number of known associations between human diseases and common genetic variants has grown dramatically in the past decade, most being identified in large-scale genetic studies of people of Western European origin. But because the frequencies of genetic variants can differ substantially between continental populations, it's important to assess how well these associations can be extended to populations with different continental ancestry. Are the correlations between genetic variants, disease endpoints, and risk factors consistent enough for genetic risk models to be reliably applied across different ancestries? Here we describe a systematic analysis of disease outcome and risk-factor–associated variants (tagSNPs) identified in European populations, in which we test whether the effect size of a tagSNP is consistent across six populations with significant non-European ancestry. We demonstrate that although nearly all such tagSNPs have effects in the same direction across all ancestries (i.e., variants associated with higher risk in Europeans will also be associated with higher risk in other populations), roughly a quarter of the variants tested have significantly different magnitude of effect (usually lower) in at least one non-European population. We therefore advise caution in the use of tagSNP-based genetic disease risk models in populations that have a different genetic ancestry from the population in which original associations were first made. We then show that this differential strength of association can be attributed to population-dependent variations in the correlation between tagSNPs and the variant that actually determines risk—the so-called functional variant. Risk models based on functional variants are therefore likely to be more robust than tagSNP-based models.
| In the past six years, genome-wide association studies (GWAS) have revealed thousands of common polymorphisms (tagSNPs) associated with a wide variety of traits and diseases, particularly as study sample sizes have increased from thousands to hundreds of thousands of subjects. Typically GWAS analyses stratify on genetic ancestry, because many polymorphism allele frequencies differ by ancestral group, easily producing false positive associations for traits that also correlate with genetic ancestry. The large majority of GWAS results reported to date derive from analyses in populations of European ancestry (EA) [1],[2]. Although GWAS in Asian populations in particular are becoming more common [3]–[6], it remains important to understand the degree to which the magnitude and direction of allelic effects generalize across diverse populations [7]–[10]. The multi-ethnic PAGE consortium [11] provides a unique opportunity to assess GWAS generalization across multiple non-EA populations and multiple traits.
Subject and genotyping panel selection for the PAGE consortium have been described elsewhere [11],[12]. In brief, a panel of 68 common polymorphisms previously reported to associate with body mass index (BMI) [13], type 2 diabetes (T2D) [14], or lipid levels [15] was genotyped in up to 14,492 self-reported African Americans (AA), 8,202 Hispanic Americans (HA), 5,425 Asian Americans (AS), 6,186 Native Americans (NA), 1,801 Pacific Islanders (PI), and 37,061 EA (for details, see Materials and Methods, Table S1 and Table S2). We also analyzed a subset of 5863 AA from PAGE who were genotyped on the Illumina Metabochip, which contains approximately 200,000 SNPs densely focused on 257 regions with reported GWAS associations to traits that include lipids, BMI, and T2D [16].
For a replication analysis it would be overly conservative to use the Bonferroni correction, so the Benjamini-Hochberg method [17] was applied to assess replication of previous EA reports in the PAGE EA population. Reported effects in EA were replicated for 51 out of the 68 index SNPs at a 5% FDR. Power to replicate at most of these 68 SNPs far exceeded 80%; 16 of the 17 SNPs that did not replicate exceeded 80% power to replicate the reported effect size, and the 17th exceeded 70% power, as described previously [13]–[15]. The originally reported effect sizes tend to be less extreme for these seventeen index SNPs, but in 63 out of 79 comparisons between non-EA and EA populations involving these 17 SNPs, the direction of effect was the same in EA and non-EA groups (p<10−5 for the null hypothesis of random effects in either direction, data in Table 1 column “Index SNPs Not Replicated in EA”). Only 79 of the 85 possible pairwise comparisons against EA were assessed, because some of the 17 SNPs were not genotyped in all five non-EA populations. Thus, it appears likely that most of the 17 failures to replicate represent weak effects that were underpowered in PAGE EA, rather than false-positive primary reports. Therefore, all 68 index SNPs were carried forward in the generalization analysis.
In all non-EA groups, we observe significantly more effects in the same direction as in EA than expected under the null hypothesis, ranging from 68% in Asians to 88% in Hispanics (p<0.001 in all non-EA groups, Table 1 and Figure 1). Even in the relatively small Pacific Islander population (N = 1801), where only four index SNPs were significantly associated with reported traits, 48 out of 62 effects were in the same direction as EA (p<0.001), so in larger samples from this population we would expect additional loci to generalize. Although a higher proportion of effects in the opposite direction of EA was observed in Asians and Pacific Islanders, the opposite effects were neither significantly different from no effect, nor significantly different from the observed effect in the EA population. This suggests that the greater number of effects in the opposite direction observed in these smallest groups simply reflects greater uncertainty in estimating effect sizes for these populations, rather than any true trend toward opposite effects. The proportion of effects in the same direction as EA was similar across all non-EA populations, suggesting that for at least 70% of index SNPs, a significant effect in a consistent direction will ultimately be observed in non-EA populations of adequate size.
Whereas the direction of effect was consistent between EA and non-EA populations, the magnitude of effect varied considerably, consistent with prior meta-analyses of generalization [18]. Because effect sizes were correlated among non-EA populations, we applied the Benjamini-Hochberg method within each population to identify index SNPs with significantly inconsistent effects between EA and non-EA populations. Inconsistent effects (βpop≠βEA at 5% FDR) were observed for 17 of 68 index SNPs in at least one non-EA population (Table 2 and Table S2, see Box 1 for definitions). Inconsistent effects were most frequent in the AA population (12 out of 68 loci), but examples were also observed in Pacific Islanders and Native Americans. Although most effects were consistent between EA and non-EA populations, the relatively high frequency with which differential effects were observed in non-EA populations suggests that genetic risk models derived from GWAS in EA will predict risk less reliably in non-EA populations, particularly AA. Consequently, caution should be exercised in applying risk models based upon risk variants genotyped outside of the ethnic background in which they were derived [19], regardless of the factors causing the observed variation between populations,.
Four index SNPs showed differentially generalized effects (ßpop≠ßEA and ßpop≠0). Two of these did not replicate in EA (rs7578597 and rs7961581 for T2D in NA) so consistency of direction cannot accurately be inferred. Direction of effect in EA and non-EA was the same for the remaining two index SNPs; rs3764261 was significantly weaker for HDL in AA, and rs28927680 was significantly stronger for TG in Pacific Islanders. There were no observations of opposite effects where both the EA effect and the non-EA effect were significant.
Considering only the 15 SNPs with a significantly inconsistent effect between EA and at least one non-EA population, 14 of 15 diluted toward the null (p<0.01, Table 2), a trend driven by the AA population, where all 12 out of 12 significant inconsistencies were diluted. Expanding analysis to all 51 loci replicated in EA, regardless of whether a significant difference was observed between EA and non-EA at a given SNP, we observed a significant excess of effects diluted toward the null (ßpop/ßEA<1) in AA, HA, and NA populations (Table S5). Comparisons between non-EA populations revealed that diluted effect sizes were significantly more likely in AA than in any other non-EA population.
Given that differential effect sizes were observed for many tagSNPs, we sought to leverage the data in order to assess the relative contributions of several factors that might contribute to the significant trend toward diluted effects, including gene–environment interaction with an exposure that varies across populations (differential environment), differences in the correlation between the index SNP and the functional variant across populations (differential tagging), modulation of the index SNP effect by additional, population-specific polymorphism (differential genetic background), population-specific synthetic alleles (combinations of rare, functional alleles tagged by a single common tagSNP [20]), or some combination of these factors. It seems unlikely that differential environments would be much more frequent in AA than other non-EA populations, or that differential environment would consistently bias toward the null within AA. Differential tagging is consistent with differentially diluted effects in AA; because linkage disequilibrium extends over significantly shorter distances in African populations than in non-African populations [21],[22], common functional variants (or synthetic alleles) are likely to be less strongly tagged by the index tagSNPs in AA. Differential genetic background effects in AA would also be consistent with the high nucleotide diversity known to exist in this population. The rare functional variants contributing to synthetic alleles will tend to be younger than common variants, and therefore are more likely to be population-specific, so synthetic alleles are compatible with the trend toward dilution. Thus, although differential environmental effects cannot be excluded, the observed data are more consistent with differential tagging and/or differential genetic background effects, and synthetic alleles cannot be excluded.
Genetic background effects can be subdivided into modifying effects, where variants elsewhere in the genome directly alter the effect associated with a given index SNP, and interference effects, where secondary variants change the proportion of variance explained by the index SNP. Interfering functional variants with effects in the same direction as the index SNP would tend to dilute the apparent effect size at the index variant. The most likely source of such variants is the region surrounding an index SNP, as demonstrably functional variants already exist in that region. Although examples have been described of genes carrying both risk and protective mutations [23]–[25], others clearly exhibit trends toward risk alleles with similar effects (e.g., preferentially toward breast cancer risk alleles at BRCA1 [26]). If the direction of effect for functional variants in a given region is consistently biased, then an increase in the number of interfering variants within a given population would be consistent with a trend toward dilution of index effects. The higher nucleotide diversity observed in African populations relative to non-African populations [27],[28] would be consistent with a greater burden of secondary functional variants in AA than other populations.
In order to assess contribution of the factors outlined above to differential effect sizes between EA and AA in the index tagSNP associations, high density genotype data were collected from a subset of the PAGE African American sample. The number of AA individuals used for index tagSNP analyses varied by phenotype, with an average of 7501 (Table S3). Similar data on other populations are currently unavailable, so only loci showing differential effects between EA and AA could be analyzed. Genotype data were collected using the Metabochip, a high density genotyping array commercially available from Illumina. Detailed methods for the Metabochip genotype data collection, calling, and quality control are available elsewhere [12].
In order to measure the contribution of differential LD to dilution, we need a model of how changes in LD between tagSNP and a functional variant would be expected to alter the observed effect size at the tagSNP, assuming that the effect size at the functional variant is the same in both populations. Given a functional SNP (fSNP) and an associated tagSNP, linkage disequilibrium between the two SNPs can be described as the measurement error introduced by genotyping the tagSNP, rather than genotyping the fSNP directly. As such, by appealing to prior work on regression dilution bias, it can be shown that the effect size β′ at the tagSNP is related to the effect size β at the fSNP by the following equation: (see Text S1 for details). Thus, assuming that the effect size at the fSNP is constant between populations, when linkage disequilibrium between tagSNP and fSNP is weaker in a given population, we expect to see a greater degree of dilution bias for the estimated tagSNP effect size. Rearranging this equation, . Extrapolating to compare the degree of dilution bias between AA and EA populations, we expect changes in linkage disequilibrium across populations to be reflected by changes in relative effect size:Assuming the effect size of the functional variant is the same in both populations, this reduces to:
The above equation allows us to directly compare the observed distribution of relative effect sizes at the tagSNPs in AA and EA () against the relative strength of tagging in AA and EA (). Considering the subset of index tagSNPs in regions that were present on the Metabochip, we observed 51 index tagSNPs that fell into 47 independent loci on the Metabochip. We identified the set of SNPs tagged by each index tagSNP at r2>0 .8 in an EA population [29],[30], yielding a total of 1,093 tagged SNPs for the 51 index tagSNPs. For each of these 1,144 SNPs, we then calculated . Let this represent the expected distribution of differential LD between AA and EA. Next, we calculated for the subset of 40 of the 51 index tagSNPs that replicated at q = 0.05 in EA, truncating at 0 if the signs were opposite between populations. These two distributions ( in all 1,144 SNPs versus for the 40 index tagSNPs) were not significantly different by two-tailed t test. Thus, we cannot reject the hypothesis that the observed dilution bias in AA effect sizes at the index tagSNPs is consistent with the observed distribution of differential LD between the two populations. A single-locus example of the potential for differential LD to contribute to diluted effect sizes is shown in Figure 2.
Considering the 12 SNPs showing differential effect size in AA, regions spanning 11 were present on the Metabochip (Table S3). Before comparison with EA, we compared the observed effect sizes at the index tagSNPs in the full AA sample and the subsample of AAs genotyped on the Metabochip (AAmchip). Three of the index tagSNPs failed to genotype on the Metabochip, leaving eight index tagSNPs for this direct comparison (Table S4). No significant allele frequency differences were observed between the AAmchip subset and the full AA population, consistent with AAmchip being a representative subsample. A significantly inconsistent and diluted effect size in AAmchip compared to EA was still observed for five of these eight tagSNPs (p<0.05, Table S4). The index tagSNPs without a significant difference likely reflect reduced power to detect the differential effect size in the AAmchip subsample, as these three index tagSNPs also had the least significant differential effect when comparing the full PAGE AA subpopulation against EA.
The Metabochip genotype data allowed us to evaluate regions spanning each of the 11 variants for the underlying contributions of population-specific alleles, differential tagging, and secondary alleles to differential effect sizes. Detailed discussion of each locus is provided in Text S1. In summary, the 11 SNPs fell in 10 Metabochip regions, so all SNPs in each of the 10 regions were assessed for association with the reported trait in AAmchip. The threshold level for significance within each region was conservatively adjusted for multiple testing by Bonferroni adjustment for the number of SNPs successfully genotyped on the Metabochip within the region, with minor allele frequency greater than 1% in the AAmchip sample. For example, the Metabochip region spanning CETP contained 84 SNPs, so our significance threshold for that region was p<0.05/84 = 1.1*10−4. One locus (APOE) could not be dissected confidently as LD data for the index tagSNP were not available in EA, and two loci were underpowered to draw strong conclusions, as evidenced by the failure of any variant in the region to show a significantly inconsistent effect with the index tagSNP effect in EA. Among the remaining seven loci, we observed one clear example of a diluted signal consistent with EA-specific functional alleles, either common or synthetic (Figure 3a), and five loci showed patterns consistent with fine-mapping of the index tagSNP bin (Figure 3d–f, Figure 4a, 4d). One of these fine-mapped the EA association to a variant that was not strongly associated with the index tagSNP in EA (r2<0.5, Figure 3f), potentially consistent with a synthetic allele in EA. We also observed statistically significant secondary functional alleles at three loci (Figure 4).
Thus, although the overall pattern of effect dilution in AA is consistent with expectations on the basis of differential LD patterns between AA and EA populations, putative examples of EA-specific alleles and secondary alleles in AA were also observed. A contribution from synthetic alleles cannot be excluded, and may well account for the EA-specific allele at CILP2 (Figure 3a). However, at half of the 10 loci we observed at least one of the tagged SNPs in EA that showed an effect size in AA consistent with the effect size at the tagSNP in EA. These examples of fine-mapping EA signal suggest that at least half of EA GWAS signals tag a common, functional variant. The observed excess of dilution effects in AA (as compared to other non-EA populations) suggests that African-descended populations will be the most useful single subpopulation for fine-mapping of EA GWAS associations, although the significant trend toward excess dilution in HA and NA populations (Table S5) suggests that trans-ethnic fine-mapping may prove more powerful than fine-mapping with any single non-EA population.
In conclusion, we have assessed the generalization of GWAS associations from EA populations across five clinically relevant traits, in five non-EA populations. Our results demonstrate that although most EA GWAS findings can be expected to show an effect in the same direction for non-EA populations, a significant fraction of GWAS-identified variants from EA will exhibit differential effect sizes in at least one non-EA population, and these differential results will be far more frequent in the AA population. These findings suggest that expanded GWAS and fine-mapping efforts focused on non-EA populations, especially AA, will substantially enhance our understanding of the genetic architecture of common traits within non-EA populations. It will be particularly important to extend GWAS discovery efforts to non-EA populations if genetic risk prediction models using tagSNP genotypes demonstrate clinical utility, because risk estimates derived from European GWAS clearly generalize imperfectly to non-EA populations. Our analyses suggest that variable LD in its many guises accounts for much of the heterogeneity of effect size at index tagSNPs, rather than any “true” differences in effect size between populations for the functional variants that were tagged. Thus, risk models derived directly from genotypes at functional variants (rather than tagSNPs) may generalize more effectively to non-EA populations.
Traits considered were those for which more than 10 GWAS-identified variants were genotyped in the first year of PAGE. Variants considered for this analysis included 13 previously reported to associate with body mass index, 20 for type 2 diabetes, 27 for HDL, 19 for LDL, and 14 for triglycerides. Eleven of these GWAS-identified variants were previously reported to associate with more than one trait in EA (Table S1), so we constrained the analysis of each such SNP to whichever trait had the most significant association (smallest p value) in the PAGE EA population, leaving a panel of 82 unique variants.
Because highly correlated SNPs might overweight specific results toward a specific trait or gene, we extracted a subset of minimally correlated index GWAS-identified variants from this panel of 82. At each step, we added the SNP with the most significant association in PAGE EA to a list of index SNPs, and then filtered the remaining SNPs not yet in the index list to exclude those exceeding r2 = 0.2 in the PAGE EA population with any index tagSNP. The panel of 82 SNPs was recursively filtered in this manner, leaving a final panel of 69 index SNPs, each of which was minimally associated with any other index SNP in the PAGE EA (r2<0.2). One additional SNP (rs11084753) was removed from the analysis due to concerns regarding power to replicate, leaving a final panel of 68 index SNPs for analysis, including seven index SNPs for BMI, 18 for HDL, 15 for LDL, nine for triglycerides, and 19 for T2D (Table S2).
Power estimates are taken directly from Fesinmeyer et al. [13] for BMI, Dumitrescu et al. [15] for lipids, and Haiman et al. [14] for T2D. For details, see the original publications.
In order to assess the generalization of effects to each population, we used effect sizes (β) and standard errors derived from minimally adjusted (age, sex, and study), ancestry-specific meta-analyses described in the primary PAGE publications [15],[13],[14]. Using these data, we tested two hypotheses: first, that the GWAS-identified variant has no effect in the non-EA population (i.e., the coefficient ßpop = 0 in a linear or logistic regression model), and second, that the effect size in the non-EA population is the same as the effect size in EA (). The first hypothesis was tested by assuming the estimate is normally distributed (which is reasonable as sample sizes exceeded 500 for all populations) and calculating the probability that given and the standard error of . The second hypothesis was tested by defining and calculating the probability that , again assuming to be normally distributed. These tests are all carried out at a nominal significance level of 0.05, as we see them as the (single) test that an investigator may carry out to validate a result first observed in EA in another ethnic group, and then significance was assigned using the Benjamini-Hochberg method at a false discovery rate of 5%.
A reasonable concern in these analyses is that population stratification can distort effect size estimates in some circumstances. Some of the effect sizes from trait-specific PAGE manuscripts were not adjusted for genetic ancestry, due to either availability of data [15] or informed consent in specific populations [13],[14]; where available and allowed we have used the ancestry adjusted effect sizes. Both ancestry adjusted and unadjusted data were available for the PAGE obesity analysis [13], where ancestry adjustment did not significantly alter effect size estimates.
|
10.1371/journal.pgen.1003195 | Genome-Wide Screens for In Vivo Tinman Binding Sites Identify Cardiac Enhancers with Diverse Functional Architectures | The NK homeodomain factor Tinman is a crucial regulator of early mesoderm patterning and, together with the GATA factor Pannier and the Dorsocross T-box factors, serves as one of the key cardiogenic factors during specification and differentiation of heart cells. Although the basic framework of regulatory interactions driving heart development has been worked out, only about a dozen genes involved in heart development have been designated as direct Tinman target genes to date, and detailed information about the functional architectures of their cardiac enhancers is lacking. We have used immunoprecipitation of chromatin (ChIP) from embryos at two different stages of early cardiogenesis to obtain a global overview of the sequences bound by Tinman in vivo and their linked genes. Our data from the analysis of ∼50 sequences with high Tinman occupancy show that the majority of such sequences act as enhancers in various mesodermal tissues in which Tinman is active. All of the dorsal mesodermal and cardiac enhancers, but not some of the others, require tinman function. The cardiac enhancers feature diverse arrangements of binding motifs for Tinman, Pannier, and Dorsocross. By employing these cardiac and non-cardiac enhancers in machine learning approaches, we identify a novel motif, termed CEE, as a classifier for cardiac enhancers. In vivo assays for the requirement of the binding motifs of Tinman, Pannier, and Dorsocross, as well as the CEE motifs in a set of cardiac enhancers, show that the Tinman sites are essential in all but one of the tested enhancers; although on occasion they can be functionally redundant with Dorsocross sites. The enhancers differ widely with respect to their requirement for Pannier, Dorsocross, and CEE sites, which we ascribe to their different position in the regulatory circuitry, their distinct temporal and spatial activities during cardiogenesis, and functional redundancies among different factor binding sites.
| The Drosophila homeodomain protein Tinman was the first transcription factor found to control the development and differentiation of the heart in any species. In spite of that, our knowledge of the number, identities, and mode of regulation of the downstream target genes of Tinman that are necessary to exert its cardiogenic functions is still very incomplete. To address these issues, we have performed a genome-wide analysis of DNA regions associated with Tinman-binding in embryos and the genes linked to them. The combined data from our in-depth in vivo assays of sequence elements with high Tinman occupancy allow the following general conclusions: (1) The majority of such sequences are active as regulatory elements (called enhancers) in mesodermal tissues that include Tinman-expressing cells. (2) The enhancers active in the heart progenitor cells and the heart generally are dependent on tinman gene activity, whereas those active in non-cardiac mesoderm are often bound neutrally by Tinman. (3) Tinman binding motifs in most cases are essential for cardiac enhancer activity, but in some cases they can be functionally-redundant with those of other cardiogenic factors. (4) Tinman-occupied cardiac enhancers are enriched for a newly discovered binding motif for an unknown factor that is functional in vivo.
| Early cardiogenesis in Drosophila relies on a regulatory network of evolutionarily conserved transcription factors and closely integrated signalling events. The key cardiogenic transcription factors include the NK homeodomain factor Tinman (Tin), the GATA factor Pannier (Pnr), and the Dorsocross T-box factors (Doc1, -2 and -3). Loss of function of either of these factors results in a complete failure of heart formation (reviewed in [1], [2]). In vertebrates, closely related factors such as the NK homeodomain factor Nkx2-5, GATA4, -5, and -6, and Tbx2, -3, and -5 are likewise known to play prominent roles during early development of the normal heart and in congenital heart disease (reviewed in [3]). In Drosophila, tinman is active as the earliest among these three types of cardiogenic genes as it is initially activated during gastrulation by the bHLH factor Twist in almost the entire mesoderm [4], [5]. After gastrulation and mesoderm spreading, Tin protein is then required in conjunction with Dpp signals from the dorsal ectoderm for inducing the expression of pnr and maintaining the expression of its own gene, tin, in the dorsal mesoderm [6], [7]. During the same time, the expression of the Doc genes is induced by combinatorial Dpp and Wg signals, independently of tin and pnr, in segmentally-repeated areas within the dorsal mesoderm [8]. Within these areas, the three cardiogenic genes then maintain each others' expression via cross-regulatory interactions in the presence of perduring Dpp signals. In these cells of the cardiogenic mesoderm, each of the three classes of cardiogenic factors is critically required for the induction of cardiomyocyte progenitors (also known as cardioblasts). In the developing heart, the expression of tin and Doc, but not pnr, is maintained in cardioblasts and, in the case of tin, also in a subset of non-contractile pericardial cells. Unlike during the earlier stages, tin and Doc are expressed in mutually-exclusive subsets of cardioblasts within the heart, which involves cross-repressive interactions [9]. tin is expressed in four out of a total of six bilateral pairs of cardioblasts of the heart in each segment and is needed for specifying them as “working” cardiomyocytes, whereas Doc is expressed in the remaining two pairs and regulates their differentiation into inflow valves (ostia).
In accordance with its early and more widespread expression, first in almost the entire mesoderm and then in the whole dorsal mesoderm, tin fulfills functions in addition to cardiogenesis during mesoderm patterning and mesodermal tissue development (reviewed in [10]). In the dorsal mesoderm, tin is required for the formation of all somatic muscle founder cells and their corresponding muscles. Moreover, in the dorsal segmental areas between the segmental cardiac primordia tin is essential for the formation of the primordia of the trunk visceral mesoderm. In ventral-lateral regions of the somatic mesoderm tin is also required for the formation of muscle founders, albeit only for a specific subset of them. Lastly, in mesodermal areas along the ventral midline, tin is required for the formation of glia-like cells, termed Dorsal Median cells. Importantly, during all these events including cardiogenesis, tin has an obligatory requirement for additional, spatially-restricted factors and signalling effectors to fulfill its functions. This requirement is most evident from the observations that, in the normal situation, defined responses to tin only occur in specific subareas within Tin-expressing domains, and that forced uniform expression of tin in the mesoderm produces only minor alterations during mesodermal tissue development [11]. The molecular basis for these combinatorial and synergistic requirements has been clarified most extensively for the activation of the homeobox genes even-skipped (eve) in specific dorsal muscle founders as well as pericardial progenitors and bagpipe (bap) in the trunk visceral mesoderm primordia. Both eve and bap are direct targets of tin that contain essential Tin binding sites within their respective mesodermal enhancers. In addition, however, the eve enhancer (MHE/EME) requires sites for the pan-mesodermal regulator Twist, Smad binding sites targeted by Dpp, TCF/Pan sites that mediate both derepression and activation in cells receiving Wg signals, and Ets domain binding motifs likely mediating inputs from receptor tyrosine kinases [12]–[14]. Likewise, the bap enhancer additionally requires Smad sites for its activation (in cells receiving Dpp but not Wg signals) and sites for the forkhead domain repressor Sloppy paired (Slp) for blocking its activation by Tin and Smads (in cells receiving both Dpp and Wg signals) [15].
A number of additional direct tin target genes have been defined by using candidate approaches based upon genetic and expression data, particularly with respect to heart development. These include genes encoding various cardiac transcription factors (tin, pnr, Hand, svp, mef2, mid) [6], [7], [16]–[19], an ion channel subunit (Sur) [20], [21], a cytoskeletal component (b3-tubulin) [22], and a transmembrane receptor (Toll) [23]. All these genes fulfill stringent criteria for cardiac Tin target genes, i.e., loss (or strong reduction) of gene expression in tin mutant backgrounds, Tin binding to a linked cardiac enhancer element, and loss (or strong reduction) of cardiac enhancer activity upon mutation of the Tin binding motifs. In most of these cases, only the role of Tin has been defined at the level of the respective enhancers, but their detailed functional architectures have not been examined. However, in a few cases additional direct and obligatory co-regulators were described, in particular Smads (for the tinD enhancer of tin) [7], Pnr (for a heart enhancer of Hand) [16], and Doc (for a heart enhancer of Toll) [23].
Although the candidate approach-based studies were successful in providing a basic framework of the regulatory circuits during heart development downstream of tin, for a more complete picture of the roles of tin it is necessary to recover functionally important tin target genes at a more global level. A profitable starting point towards this goal can be the genome-wide identification of the DNA sequences to which Tin is bound in vivo. In our present study, we have taken this directed approach by performing chromatin immunoprecipitations (ChIPs) of Tin-bound sequences from embryos during early cardiogenesis, analyzing them globally, and dissecting selected examples of them functionally. Similar approaches have been initiated by others in parallel to ours [24], [25], [26], but overall, information about the functionality of Tin binding to distinct sequences is still rather limited.
Herein, we present an analysis of the Tin-bound regions and their linked genes in embryos during mesoderm patterning (3–5.5 h after egg laying) and during specification of heart, visceral and somatic muscle progenitors (5–8 h AEL). After their global characterization, which shows that the linked genes are associated preferentially with expression and functions in various mesodermal tissues and, unexpectedly, also in ectodermal tissues, we present data from our functional in vivo analysis of ∼50 Tin-bound sequences. The overwhelming majority of these sequences, which were selected largely among the ∼250 regions with highest levels of Tin occupancy (with the omission of sequences linked to known genes with ectodermal activities), showed enhancer activities in various mesodermal tissues, including the entire early mesoderm, dorsal and cardiac mesoderm, heart, visceral mesoderm, somatic mesoderm, and Dorsal Median cells. To test whether Tin binding is functionally relevant, we focused on the subset of enhancers active in the dorsal and cardiac mesoderm and/or the heart and demonstrate that all enhancers of this class require functional tin for being active. For a representative subset, we tested the in vivo requirements for their Tin binding motifs, as well as the requirements for the Pnr and Doc binding motifs that were also present in these enhancers. Further, by using machine learning approaches we identified a novel motif as a classifier of cardiac enhancers and show that in three out of eight tested enhancers this motif is essential for their cardiac activity. Altogether, our data indicate that high-level Tin binding signifies mesodermal enhancer activity, and at least with respect to highly-occupied cardiac enhancers, in vivo binding of Tin is almost always essential, although it can be functionally-redundant with other bound cardiogenic factors. We discuss the large architectural diversities of the Tin-bound cardiac enhancers in terms of the arrangement and requirement of the binding motifs for the various cardiogenic factors as well as more general aspects regarding the interpretation of genome-wide binding data from ChIP analyses.
Genome-wide binding sites for Tin in embryos were determined for two developmental windows, 3–5.5 h and 5–8 h after egg laying. During the 3–5.5 h window (herein termed “Early”), Tin is expressed initially in the entire mesoderm (except for the hematopoetic mesoderm) and then becomes restricted to bilateral dorsal mesodermal areas under the influence of Dpp signals (Figure 1A). During this stage, Tin is known to fulfill key roles in the early specification of heart progenitors (both cardioblasts and pericardial cells), the trunk visceral mesoderm, all dorsal somatic muscle founder cells and specific ventrolateral founders, as well as ventrally-located glia-like mesodermal cells, termed Dorsal Median (DM) cells. During the 5–8 h window (termed “Late”), the expression of Tin narrows from the broad dorsal mesodermal domain to the cardiogenic mesoderm and developing cardioblasts along the dorsal margin of the mesoderm, as well as to segmental subsets of trunk visceral mesoderm cells (Figure 1A). At this stage, Tin continues to act in the determination and diversification of cardiac cells (reviewed in [1]). The only non-mesodermal expression of Tin, which is present during both time windows, occurs in the ectodermal anlagen of the foregut (Figure 1A).
Chromatin immunoprecipitations were performed with our high quality, immunopurified Tin antibody [4], assayed by microarray hybridizations (“ChIP-chip”), and binding peaks were called using MAT (see Materials and Methods). The 3–5.5 h dataset included 2536 genome-wide binding peaks. The 5–8 h dataset yielded 983 binding peaks, of which 846 overlapped with peaks from the 3–5.5 h dataset. The lower number of peaks in the “Late” dataset above threshold is possibly due to the smaller proportion of Tin-expressing cells in embryos of this time window, which results in lower signal-to-noise ratios. In both datasets, almost half of the Tin-occupied regions were found in introns, ∼40% in intergenic regions, and most of the remaining ones in 5′UTR-encoding portions of genes (Figure S1A, S1B; Table S1). 16% of the peak summits were found in promoter regions (0 kb to −2 kb) in both datasets. If the binding regions are assigned to the nearest gene (in either direction) or exons (for intronic peaks), there are 1530 putative Tin target genes in the 3–5.5 h dataset and 679 in the 5–8 h dataset, of which 613 are shared between both sets (Table S1).
We compared our binding regions with those obtained in an independent study of Tin in comparable developmental windows (Zinzen et al. 2009, [25]). About 34% of the binding regions in our 3–5.5 h dataset overlapped with those in the 4–6 h dataset from Zinzen et al., and ∼60% of our binding regions in the 5–8 h dataset overlapped with those in their 6–8 h dataset. Conversely, ∼92% of their 4–6 h binding regions and ∼63% of their 6–8 h binding regions overlapped with those from our “Early” and “Late” datasets, respectively (Figure S1C, S1D).
We also compared our Tin-bound regions with highly occupied target (HOT) regions, which are regions displaying a binding complexity of >8 transcription factors of diverse functions during blastoderm stages, as determined by the ModENCODE Consortium [27]. Notably, a significant portion of binding regions from both of our datasets, namely ∼38%, overlapped with HOT regions (Figure S1D).
To obtain a picture of the type of genes that are bound by Tin preferentially we searched for over-represented Gene Ontology (GO) terms. As shown in Table S2A, genes encoding transcription factors and developmental genes were strongly enriched among the Tin-bound genes, both in the 3–5.5 h and the 5–8 h datasets. Of note, the GO terms “heart development”, “mesoderm development” and “gastrulation” that are particularly relevant to the known functions of tin ranked among the top 25 in both lists. An analysis for encoded protein domains that are enriched among the Tin-bound genes showed a preponderance of various types of DNA-binding domains at the top of both datasets. In addition, genes likely encoding components of signalling cascades such as protein kinases, immunoglobulin domains, EGF-like regions, and leucine-rich domains ranked near the top of the lists (Table S2B).
We also addressed the question of whether the Tin-bound genes were expressed preferentially in the tissues in which tin is expressed and is known to exert its genetic functions. For this purpose, we used the temporal and spatial annotations from the 3833 genes in the BDGP embryonic in situ hybridization database that had yielded “acceptable quality” expression data [28]. Indeed, there was a significant enrichment of the terms “trunk mesoderm primordium”, “visceral mesoderm primordium”, and “cardiac mesoderm primordium” among the Tin-bound genes both from the “Early” and “Late” datasets. In addition, terms reflecting the ectodermal and mesodermal expression domains of Tin in the head region (ectodermal foregut primordia and head mesoderm) were highly enriched (Table S3A). Unexpectedly, when all expression terms were considered, terms related to ectodermal and neuronal expression domains turned out to be among the most highly enriched in both of our datasets (Table S3B). This was not only the case for ectodermal domains of the head, which could overlap with Tin expression, but also for ectodermal domains of the trunk including the ventral nerve cord that exclude Tin expression. As this enrichment also holds true for the subset of genes containing intronic Tin-binding regions it is unlikely to be a result of incorrect assignments of flanking genes to the obtained binding peaks. It is also unlikely that this enrichment is an artefact due to biases in gene lengths [29], as we do not detect any major length differences between mesodermally and ectodermally/neuronally expressed genes, and terms encompassing genes with much longer average lengths (e.g., “tracheal primordium”), were not enriched (data not shown). Another possible explanation of these results could be that expression within Tin+ tissues is linked preferentially to expression in ectodermal and neuronal tissues, but upon omission of the genes with Tin-overlapping expression domains there is still a strong bias for ectodermal and neuronal expression. Likewise, the omission of the genes associated with Tin binding in HOT regions, which may potentially reflect non-functional (neutral) binding [50] and could have produced an artificial bias, did not change the enrichment for ectodermal tissues (data not shown). It is possible that the observed enrichment for Tin-bound genes with ectodermal domains reflects a role for Tin in repressing ectodermal genes but currently, genetic evidence for such a role is lacking. Alternatively, for unknown reasons non-functional binding of Tin could occur preferentially in ectodermally-expressed genes. One hypothesis is that Tin can occupy potential binding sites for the NK-2 type transcription factor Ventral nervous system defective (Vnd), with which it shares a very similar binding motif [30] and which may be bound by Tin if the chromatin is not masked in mesodermal tissues. Interestingly, the majority of Tin peaks in both early and late conditions are localized in open chromatin when compared to the genome-wide DNase I footprint profile generated by the BDTNP for whole embryos [31], that is, 91% of Tin Early peaks overlap DNA accessibility sites at stage 9 and 93% of Tin Late peaks overlap DNA accessibility sites at stage 11.
To determine whether the Tin-bound regions contained any prevalent sequence motifs we performed a de novo motif search using Regulatory Sequence Analysis Tools (RSAT) [32] on repeat-masked DNA [33] from 3–5.5 h and 5–8 h Tin peak regions. With both the top 100 peaks and the total datasets from each developmental window, we retrieved the known Tin binding motif, which closely resembled the motif obtained with functionally-tested Tin target sites that were published at the beginning of our study (Figure 1B; Table S5). In addition, the consistently most enriched six-mer core motif AGATAC closely resembles the motif bound by various GATA factors, including that of the vertebrate cardiogenic factor GATA-6 (Figure 1C). In Drosophila, the GATA-4/-5/-6-related factor Pannier (Pnr) is known to play a key role in early heart development [34]–[36]. We found that the de novo motifs presumed to bind Tin and GATA (Pnr) are located preferentially in the center of enriched binding regions (Figure 1B and 1C). This preference was less pronounced for the motifs of the cardiac T-box factor Dorsocross 2 (Doc2) (determined by SELEX; Figure 1D; Table S5). Closely related Tin and GATA motifs were also derived de novo from Tin and Pnr-bound sequences, respectively, by Junion et al. [26], whereas their Doc de novo motif deviates significantly from the SELEX-derived motifs for Doc (Figure 1D) and the related Tbx6 [37].
In a first step towards testing the observed Tin-bound regions functionally we ranked the binding peaks according to their size (as defined by the area underneath the contour), with the assumption that on average the regions bound more strongly (e.g., because of the presence of multiple binding sites, high affinity sites, or of binding sites for other Tin-binding factors that increase occupancy by Tin) would be more likely candidates for functional Tin targets [38]. As shown in the ranking lists for the 3–5.5 h and 5–8 h datasets of the top ∼450 peaks (Table S4A, S4B), the majority of known Tin downstream genes and proven Tin targets genes, defined previously via genetic and/or molecular data, were present among the genes associated with the top-ranking binding peaks. These included, in the “Early” (E) and “Late” (L) lists, respectively: tup (rank #E9, L34), bap (#E10, E198, L352), mid (#E19, L298, L350, L402, L411), H15 (#E22, L5), mir-1 (#E43, E119, E155, L100), , eya (#E91, E128, L57, L165), Doc genes (#E105, E122, E282, L33), Poxm (#E138, L276), zfh1 (#E141, L173, L227, L348), mef2 (#E170, E181, L341), jeb (#E177), odd (#E81, L181; #E254, L164), Six4 (#E255), pnr (#E261, L162), slou (#E273, E425, L140), lbe (#E308, L170), Him (#E318, L47), Tl (#E359), disco (#E420, L10, L217, L286), btn (#E455), apt (#L133, L336), svp (#L216, L330), htl (#L383), lbl (#L443), and tin (#L424). Apart from these, the top-ranking genes included many that are expressed in mesodermal tissues with patterns that overlap spatially and temporally with Tin (Table S4A, S4B).
For functional tests of enhancer activity in vivo we selected 51 binding regions (Table S4A, S4B). The following criteria were employed for the selection: 1) Higher priority was given to regions ranking higher on the size-ranked lists. 2) Lower priority was given to fragments that did not include at least one high-scoring Tin binding motif (Materials and Methods, Table S5). 3) Excluded were fragments associated with genes with published expression patterns that do not include mesodermal tissues. Thus, potential Tin targets in the Tin+ ectodermal foregut primordia were also omitted. The initial fragments had an average length of 1.4 kb and the endpoints were chosen such that the fragments were centered around the peak maxima and included any Tin motifs underneath the peaks.
39 of the 51 fragments tested (76%) were active in mesodermal domains overlapping with Tin domains. Among these, 16 (41%) were active in the cardiogenic mesoderm and/or in the developing heart, which were of particular interest to our study, and one (associated with nau) was active in the early dorsal somatic mesoderm (Figure 2; see Figure S2 for zfh1E141, which exhibits very weak activity in the dorsal vessel after stage 14; Tables S3A, S3B and S4). Several of these putative cardiac target enhancers of Tin were active additionally in cells of the somatic or visceral mesoderm (Figure 2; Tables S3A, S3B and S4). The onset of reporter expression within dorsal mesodermal and cardiogenic regions with these fragments included the time windows assayed in the ChIP experiments and the corresponding enhancers showed peaks in both the 3–5.5 h and the 5–8 h windows. Notably, several enhancers only began showing cardiac activity during the 5–8 h window, even though Tin binding was already detected during the 3–5.5 h window (Figure 2; lin-28L64, midE19, unc-5L25, CG3638L6) (note that the enhancers are named after the presumed associated gene and their highest rank in the Early and Late lists, respectively; Table S4). Conversely, other enhancers were still occupied by Tin in the 5–8 h window, but their activity in the dorsal mesoderm and developing heart had ceased during this period (Figure 2, discoL10, nauL35, CG9973E15, and data not shown). However, most of the cardiac enhancers remained active in developing cardioblasts and/or pericardial cells until stages 14–16 (Figure 2, EgfrE1, fzL4, HimL47, hthE54, lin-28L64, mαL9, midE19, nocL7, tshL8, tupE9, unc-5L25).
14 of the 39 positive fragments (36%) were active in the somatic mesoderm but not in the cardiogenic mesoderm (Figure S2). Half of these showed activity in the whole somatic mesoderm. Among these were fragments associated with zfh1 (with additional weak expression in cardiac mesoderm), CG5522, CG32792 and, surprisingly, H15 and hh, two genes that are not known to be expressed in the somatic mesoderm. The other half showed activity in more restricted patterns in the somatic mesoderm, such as in large lateral cell clusters (lbeL170), segmental stripes (slpE35), and stripes restricted to ventral areas (PoxmE138, Six4E255).
6 of the positive fragments (15%) tested positive in the primordia of the trunk visceral mesoderm and one (oddE81) in fat body primordia (Figure S3). These included fragments near the genes Alk, Fas3, and H2.0, which are known to be expressed in the trunk visceral mesoderm. The enhancers of both Alk and Fas3 start being active in the entire segmental visceral mesoderm primordia during the 3–5.5 h window (Figure S3A, S3B), whereas those of Gukh, Nrt and H2.0 become active only during the 5–8 h window with their activity being restricted to the ventral rows of circular gut muscle founder cells (Figure S3C–S3E; wgnE1197-LacZ appears in portions of visceral mesoderm only at stage 14, Figure S3F). The remaining two positive fragments showed activity in the Dorsal Median (DM) cells, which form in the mesoderm along the ventral midline in each segment and later are attached dorsally to the CNS (Figure S4). The DM cells are known to depend on tin function, which must be needed during early stages when Tin expression is still present in ventral areas of the mesoderm (i.e., during our 3–5.5 h test window) [39]. In contrast to the above enhancer activities that overlap with Tin expression domains in the mesoderm, 12 additional enhancer fragments were active in a variety of non-mesodermal tissues and cell types that are mostly ectodermally-derived (Figure S5). A comparison with the genomic ChIP data published by Junion et al. [26] subsequent to our in vivo analysis shows that 50 of our 51 enhancer regions tested in vivo and all ten positive control enhancers used are also positive for Tin occupancy in their screens (the exception being the low ranking wgnE1197), and are often positive for various combinations of Doc, Pnr, pMad, and dTCF (Table S5). Perhaps unexpectedly, the 61 mesodermal and non-mesodermal enhancers can not be differentiated via their occupancy patterns by these factors, e.g., within each of the mesodermal (any tissue), cardiac, and non-mesodermal subset ∼1/3 of the enhancers are occupied by Tin+Doc+Pnr, ∼1/3 by Tin+Doc, and ∼1/3 by Tin only (Table S5). In addition, when we overlap our list of 61 enhancers with the ChIP data and predicted clusters of Junion et al., we find neither their individual ChIP signals nor any of their predicted ChIP cluster types to be significantly associated with cardiac enhancers (Fisher's exact test, P-value>0.10; a subtle association could have been missed because of the size of the dataset).
To get an overview of the involvement of tin in generating the observed patterns of mesodermal enhancer activity we tested a selection of the reporters from each class in tin mutant backgrounds. As shown in Figure S6A, S6A′, S6C, S6C′, S6H, S6H′, the enhancer activities of AlkE301, Fas3L254, and NrtL30 in the visceral mesoderm are severely reduced or absent in tin mutant embryos. Likewise, the activity of the btn enhancer in DM cells (Figure S6B, S6B′) and of the enhancers of hth (Figure S6D, S6D′, arrows) and noc (Figure S6G, S6G′) in cardiogenic regions is strictly tin-dependent. In contrast, the results for somatic mesodermal enhancers were more varied. Whereas the dorsal somatic mesoderm activities of the nau and noc enhancers were strictly dependent on tin (Figure S6F, S6F′, S6G, S6G′), the ventral somatic mesoderm activity of the Six4 enhancer was reduced only slightly in tin mutants (Figure S6I, S6I′) and the somatic mesoderm activity of the hth and zfh1 enhancers was completely tin-independent (Figure S6D, S6D′, S6J, S6J′). In summary, the selected fragments showing in vivo Tin binding were highly enriched for sequences containing enhancer activities in tissues with known genetic requirements for tin. Particularly the activities of the heart and visceral mesoderm enhancers appear to depend on tin, whereas the tin-dependency of somatic mesoderm enhancers seems less predictable.
For the remainder of the study we concentrated on the characterization of the identified enhancers that were active in the cardiogenic mesoderm and developing heart. We verified that all of the enhancers with activities assigned to the cardiogenic mesoderm and heart progenitors during the developmental windows assayed by ChIP were indeed active in Tin-positive cells (Figure 3A–3J). The exact patterns varied, with some enhancers being active in all cells of the segmental cardiogenic anlagen (e.g., HimL47, tupE9, Figure 3C, 3H) and others becoming active shortly afterwards in newly-specified, segmentally arranged heart progenitors (e.g., fzL4, midE19, tshL8, unc-5L25, CG3638L6; Figure 3A, 3E, 3G, 3I, 3J) or throughout the cardiogenic mesoderm (EgfrE1, lin-28L64, Figure 3A, 3D). When assayed in tin null mutant backgrounds, the activities of all these enhancers in cardiogenic regions and heart progenitors were completely lost (Figure 3A′–3J′). By contrast, if enhancers were also active in other mesodermal areas, such as the somatic mesoderm (EgfrE1, HimL47, RhoLE102, unc-5L25; Figure 3A′, 3C′, 3F′, 3I′) or in amnioserosa and ectodermal tissues (lin-28L64, tshL8; Figure 3D′, 3G′), these non-cardiac activities did not require tin.
Whereas the activities of some of the assayed cardiac enhancers were transient and ceased by germ band retraction, the activities of several others persisted in the dorsal vessel until late stages of embryogenesis. The midE19 and tupE9 enhancers were active only in cardioblasts (especially in Tin+ cardioblasts) and the unc-5L25 and CG3638L6 enhancers in both cardioblasts and pericardial cells (Figure 4A–4D). To test whether these enhancer activities within cells of the dorsal vessel are also tin dependent we used a “conditional” tin mutant background, in which Tin is still expressed during early mesoderm patterning and heart progenitor specification but not in cardioblasts and pericardial cells after they are formed [9]. As shown in Figure 4A′–4D′, the activities of all four of these enhancers in cardioblasts (marked by Doc) and pericardial cells are nearly absent if Tin is not present in these cells. In summary, we found that all identified enhancers that drive expression in the cardiogenic mesoderm, heart progenitors, and cells of the dorsal vessel are strongly dependent on tin. Hence, these enhancers were strong candidates for direct functional targets of tin.
Next, we tested the in vivo relevance of the Tin binding sites in selected cardiac enhancers under investigation. Because the binding motif for the cardiogenic GATA factor Pnr had been prevalent among the genome-wide Tin-bound regions (Figure 1), and because the T-box factors Doc1-3 are known to have key functions during early cardiogenesis and cardioblast diversification as well [8], we also included the binding motifs of Pnr and Doc in this analysis. We note that the Doc binding motifs are closely related to the binding motifs for the Tbx20 cardiac T-box factors Midline (Mid) and H15 (Figure 1D). Thus, from stage 12 when these Tbx20 factors are expressed in cardioblast progenitors, the T-box binding motifs may be occupied either by Doc or Tbx20 factors.
Figure 5A1–5F1 shows the distribution of the Tin, Pnr, and Doc binding motifs within the cardiac enhancers from Egfr, lin-28, mid, RhoL, tup, and unc-5. The Tin matrix from proven target genes (Figure 1B), the SELEX-based matrix from the Pnr-related vertebrate cardiogenic factor GATA-6 with GATA core sequence [40], and our SELEX-based Doc2 motif (Figure 1D) were used to locate and score binding motifs. Subsequent to our functional analysis, Junion et al. [26] found regions overlapping with our enhancers from Egfr, lin-28, RhoL, and unc-5 to be occupied by all three cardiogenic factors, whereas the mid and tup enhancers appeared to be solely occupied by Tin (Table S5). For all six chosen enhancers, shortened versions (∼600–800 bp) were used for functional analysis of binding motifs, named EgfrE1s, lin-28L64s, midE19s, RhoLE102s, tupE9s and unc-5L25s (Tables S4, S5). All enhancers examined included binding motifs for all three cardiogenic factors except for the EgfrE1s enhancer, which lacked Doc motifs. The numbers and spatial arrangements of these motifs within the different enhancers were quite variable (Figure 5A1–5F1).
In the EgfrE1s enhancer, the Tin motifs were essential for enhancer activity in the cardiogenic mesoderm, but not in the somatic mesoderm (Figure 5A3, cf. Figure 5A2). Mutation of the GATA (Pnr) motifs caused a weakening of the cardiac versus somatic mesodermal enhancer activity at early stage 12 (Figure 5A4), which became even more pronounced during stage 13 (Figure 5A6, cf. Figure 5A5).
In the lin-28L64s enhancer, the Tin motifs were essential for cardioblast activity, and the GATA sites were essential for both cardioblast and amnioserosa activity (Figure 5B3, B4, cf. Figure 5B2; note that Pnr expression includes both the cardiogenic mesoderm and the amnioserosa, as well as the dorsal ectoderm). By contrast, the Doc motifs were not essential for lin-28L64s enhancer activity (Figure 5B5).
In the midE19s enhancer, the Tin motifs were essential for its activity in cardioblasts (Figure 5C3, cf. Figure 5C2, see also [19]). Mutation of the GATA motifs did not have any effects, but mutation of the Doc motifs led to a strong reduction of enhancer activity in the Tin+ cardioblasts (Figure 5C5). It is conceivable that this effect is caused not only by the failure to bind Doc during heart progenitor specification but also by the inability to bind Mid, which may prevent maintenance of activity by autoregulation (in combination with Tin) in developing cardioblasts. A second effect of the Doc-site mutations was a moderate upregulation in the Tin−/Doc+ subset of cardioblasts, in which this particular enhancer is normally not active although mid is expressed there. Thus, the combined effects of the Doc site mutations lead to weak uniform enhancer activity in all cardioblasts. These observations point to an unexpected complexity of mid regulation through both positive and negative effects of cardiogenic regulators and to the existence of yet unidentified regulatory elements for expression in the Tin−/Doc+ cardioblasts.
The RhoLE102s enhancer was active in the cardiogenic mesoderm even after mutation of any decent Tin motifs, which was unexpected (Figure 5D3, cf. Figure 5D2). Although one low-scoring sequence motif, CCAAGGG is still present in the mutated version, this particular sequence deviates significantly from the motifs shown in Figure 1A and is not known to bind Tin in vitro. Mutations of the GATA motifs also did not have any effects (Figure 5D4). However, in the absence of the Doc motifs the RhoLE102s enhancer was inactive (Figure 5D5).
In the tupE9s enhancer, only the Tin motifs were required for its activity in the cardiogenic mesoderm, whereas the GATA and Doc motifs were not essential (Figure 5E3, cf. Figure 5E2, 5E4, 5E5). These results are compatible with the absence of in vivo binding by Pnr and Doc [26]. Upon mutation of the Tin binding sites, enhancer activity in the dorsal mesoderm was completely lost and instead, ectopic segmental activity in the dorsal ectoderm appeared. Of note, a similar switch in germ layer activities has also been observed upon mutation of the dorsal mesodermal enhancers of tin and bap. For these two examples, it was found that Tin acts synergistically with Smads bound to nearby sites during the activation of the enhancers in dorsal areas of the mesoderm. By contrast, in the dorsal ectoderm an unknown repressor with a recognition site that overlaps with the Tin motif was proposed to bind and prevent ectodermal enhancer activation by Dpp and activated Smads [7]. The tup enhancer appears to be a third example for this regulatory mechanism to achieve germ layer specificity of the Dpp response.
Surprisingly, for the unc-5L25s enhancer neither mutation of the Tin motifs nor of the GATA or Doc motifs had any measurable effect on its activity in cardioblasts and pericardial cells (Figure 5F3, 5F4, 5F5, cf. Figure 5F2). We can not exclude that Tin can still bind and provide full activity through three low-scoring candidate Tinman motifs that were not mutated (see Figure 5, legend). Alternatively, or in addition, there is the possibility of functional redundancy among these three factors and their binding sites within this enhancer, which we tested with derivatives containing combinatorial mutations in the different motifs. When the Tin and GATA motifs were mutated in combination, there was still no change in the cardiac activity of the enhancer (Figure 5F7). By contrast, simultaneous mutation of both the Tin and Doc motifs caused near absence of enhancer activity in cardioblasts, whereas its activity in pericardial cells was unaffected (Figure 5F8). Additional mutation of the GATA motifs had only minor additive effects on the reduction of enhancer activity, which showed only residual expression in Eve+ pericardial cells (Figure 5F9, cf. Figure 5F6, 5F8). Hence, we propose that in the case of the unc-5 enhancer, binding of either Tin or Doc (and perhaps Mid) to the Tin or T-box motifs, respectively, is functionally relevant but either factor alone is largely sufficient for activating the enhancer.
We used machine learning methods to test whether we can predict cardiac enhancers based on their motif content, as had previously been shown for human heart enhancers [74]. For this analysis, we added ten well-characterized Tin-dependent enhancers that are active in cardiogenic tissues, visceral mesoderm, or somatic mesodermal cells to the collection described herein, which yielded 24 cardiac enhancers and a total of 61 enhancers to be examined (Table S5). The enhancers were classified according to their tissue-specific activity patterns as denoted in the above table. First we undertook extensive de novo motif discovery using RSAT [32] on each experimentally verified enhancer class separately and between pairs of enhancer classes. This yielded 33 de novo motifs, which we put together with previously described motifs in Drosophila (see Materials and Methods). To prioritize motifs predictive in cardiac enhancers we constructed a LASSO classifier to predict the 24 cardiac enhancers from the set of 61 experimentally verified enhancers (Narlikar et al., 2010). Since our set of enhancers is small and we wanted to assign confidence values to predictive motifs, we used a bootstrap modification of LASSO called bolasso [41]. We found one de novo motif to be especially predictive of the cardiac class, ATT[TG]CC, which we termed “Cardiac Enhancer-Enriched (CEE) Motif”. This motif has a bootstrap confidence of 96% and its presence is the strongest predictor of cardiac enhancers followed by the GATA motif (Figure 6). In several sequences with cardiac activity, the CEE motif is particularly frequent, namely in midE19 (6.9 motif hits per kb), mef2 I-E (5.3 per kb), mef2 II-B[S] (4.3 per kb) [42], lin-28L64 (4.2 per kb), hthE54 (4.1 per kb), zfh1E141 (4.1 per kb) and sli (3.8 per kb). The enhancers with cardiac activity have an average density of 2.86 CEE motif hits per kb. In contrast, the experimentally verified enhancers with no cardiac activity have the average density of 1.52 per kb of sequence, which is almost identical to that in 2 kb Drosophila gene upstream regions (1.53 CEE motif hits per kb).
Because of the relatively small size of the enhancer dataset we needed to use all of the data for de novo motif finding and model selection, which can lead to an over-estimation of the AUC. For this reason, it was important to test the predictions of the classifier (namely the CEE motif) for their in vivo relevance in transgenic animals carrying reporter constructs in which the CEE motifs were mutated. We included all enhancer fragments that were also tested for functionality of the Tin, Doc, and GATA motifs and in addition, the published cardiac mef2 enhancers I-E and II-B[S]. As shown in Figure 7, in three of these fragments the CEE site mutations caused severe reductions of their cardiac enhancer activity. In EgfrE1s, enhancer activity in heart progenitors at stage 12 (Figure 7A3, cf. Figure 7A2) and particularly in cardioblasts following stage 13 (Figure 6A4, cf. Figure 6A3) was strongly reduced, whereas the activity in the somatic mesoderm was unaffected. In lin-28L64s, CEE site mutation caused loss of enhancer activity in cardiac cells in embryos until stage 14 (Figure 7B3, cf. Figure 7B2). Activity in the amnioserosa was also lost, whereas head expression was unaffected. In the maturing dorsal vessel, the CEE site-mutated lin-28L64s enhancer regained some activity, particularly in pericardial cells, but its activity in cardioblasts remained much lower than that of the unmutated control (Figure 7B5, cf. Figure 7B4). For the midE19s enhancer, CEE site mutations caused a complete loss of activity in the vast majority of Tin+ cardioblasts. In rare remaining positive cardioblasts, nearly full activity was seen (Figure 7C3, cf. Figure 7C2). These results were very similar to the results obtained with Tin site mutations for these three enhancers, and for EgfrE1s and lin-28L64s also with the GATA site mutations, even though the CEE motifs do not overlap with any of the Tin and GATA motifs (Figure 7A1, 7B1, 7C1). Altogether, we hypothesize that the CEE motifs bind a yet unknown co-factor of these cardiogenic factors. At least in the cases of EgfrE1s, lin-28L64s, and midE19s, the combination of this co-factor and one or several of the cardiogenic factors is needed for activating the respective enhancer in cardiac cells. Although the verification rate of three functional CEE motifs out of eight may seem modest, it is in the same range as found for the Doc motifs (2/5) and GATA motifs (2/6). It remains to be shown whether this factor is not functional in the other five enhancers tested, in which CEE site mutations did not cause any noticeable differences in cardiac activity. Alternatively, it is conceivable that in the negative cases additional, degenerate versions of this sequence motif were present and functional, or that there are additional binding sites for different co-factors that function redundantly with the putative ATT[TG]CC binding factor.
Although the developmental functions of tin during mesodermal tissue and particularly heart formation have been defined in quite some detail by genetic approaches, it is far from being clear how many target genes of tin are involved in executing these functions. Candidate approaches based on genetic observations have led to the identification of a relatively small number of essential Tin downstream targets, most of which correspond to members of the transcriptional network controlling the development of the heart, visceral muscles, and specific body wall muscles. The chromatin immunoprecipitation approach taken in our present study, as well as in studies performed in parallel by others [24], [25], provides a picture of the upper limit of potential Tin target genes that could be relevant biologically, although this number still depends on the specific cut-off used. With this latter approach, the challenge is to determine the particular fraction of genes associated with Tin binding in vivo that indeed require tin and are utilized for implementing the various tin dependent programs of mesodermal tissue development. In our present study we have begun to address this issue with a combination of genetic tests and enhancer dissections.
A comparison between the Tin-bound regions from our study and that of Zinzen et al. [25], which were done using very similar developmental time windows, shows strong overlaps. In addition, almost all enhancers known to be targeted directly by Tin during the tested time windows and the large majority of known tin downstream genes were found to be associated with in vivo Tin binding. The overall similarity of binding data obtained in different labs corroborates that a majority of reported peaks reflects authentic in vivo occupancy although there can be differences in sensitivities. For example, the number of Tin-peaks in our early window was ∼2.5 times larger than that in the early time window from Zinzen et al. [25], which we partially attribute to our use of affinity-purified anti Tin.
The observed prevalence of the GO terms “mesoderm development” and “heart development” and the prevalence of genes associated with Tin occupancy with expression patterns in “trunk mesoderm primordium”, “visceral mesoderm primordium”, and “cardiac mesoderm primordium” (BDGP; [28]) underscores the notion that Tin is bound to a large number of enhancers that are active in tissues requiring tin. This was further confirmed by the activity of ∼3/4 of our reporter constructs in various mesodermal domains that overlap with the presence of Tin protein (although this number may be somewhat skewed because we purposely omitted Tin-occupied sequences near known non-mesodermal genes in our analysis). Comparable results were also reported by the Furlong group [24]–[26]. Although at first glance, this could be taken as an indication that Tin is a direct regulator of most of these tested enhancers and globally-bound elements, there is an increasing body of evidence suggesting that in vivo binding of most factors is very promiscuous. According to this view (e.g., Biggin (2011) [43]), the high concentrations of nuclear Tin could simply drive binding to sequences containing its cognate binding motifs, as long as they are accessible as would be the case in mesodermally-active enhancers. Indeed, several observations indicate that many Tin-bound sequences do not function as Tin-dependent enhancers. For example, genes expressed in ectodermal including neuronal tissues but not in the mesoderm were also very prevalent globally among the genes linked to Tin-occupancy, and clearly are not activated by Tin. A prominent example of a gene with ectodermally-restricted expression associated with Tin-binding is vnd, which is flanked by one of the most highly Tin-occupied sequences in the entire genome. Of note, it has been shown that in mesodermal cells this particular region features the presence of repressing chromatin marks (H3K27me3) and the absence of activating marks and of PolII, but contains H3K4me1 which frequently marks potential enhancers [44]. This situation contrasts with the one found for most enhancers active in the mesoderm, which tend to show low or absent H3K27me3 but high activating marks and PolII. One possible interpretation could be that at the vnd locus, Tin functions in maintaining the repressed state of a potential enhancer and as a consequence excludes vnd expression from the mesoderm in order to restrict it to the ventral ectoderm (perhaps in cooperation or temporal succession with the Snail repressor [45]). In certain contexts tin does have repressing functions, particularly during the regulation of Doc within the dorsal vessel, although the mode of action of Tin in this situation is not yet known [9], [46]. Likewise, Tin might repress other ectodermal enhancers in the mesoderm, but genetic evidence for such a role of tin is currently lacking. An alternative explanation for Tin binding to vnd sequences would be that it is opportunistic and does not have any functional consequences. As both Tin and Vnd belong to the NK homeodomain family and share closely related binding motifs, Tin may associate neutrally with NK homeodomain motifs in the mesoderm that are actually destined for binding of Vnd and vnd autoregulation in the ectoderm, if they are accessible despite inactivating chromatin marks [47], [48]. Analogous situations could exist at other ectodermal target genes of vnd and related NK homeobox genes (e.g., Scarecrow [49]). For a deeper understanding of the genome-wide roles of Tin-associated sequences it will be necessary to include ectodermally-expressed genes linked to Tin-binding in functional dissections to distinguish between neutral and potentially repressive binding.
A significant portion (>20%) of the Tin-bound regions overlap with so-called HOT regions, which are characterized by the simultaneous binding of >8 different transcription factors during pregastrulation stages [27]. It was shown that the vast majority of HOT regions have enhancer activity, although these do not necessarily reflect the patterns and stages predicted by the factors bound to them during blastoderm stages [50]. It is thought that binding of ubiquitous DNA-binding proteins, particularly the TAGteam factor Zelda and the GAGA factor (Trithorax-like), to their binding motifs within HOT regions establishes open chromatin and that mass action then attracts a variety of other factors, which in many cases bind neutrally [50]; reviewed in [51]. The observation that many HOT regions can later serve as Tin-bound mesodermal enhancers may suggest that these enhancers are “primed” by the binding of chromatin-opening factors to facilitate binding of mesodermal factors during post-gastrulation stages. Again, in some cases mesodermal factors including Tin may bind functionally whereas in other cases they may bind neutrally. In agreement with this proposal, ∼50% of the Tin-bound mesodermal enhancers from our study overlapped with HOT regions. Seven among these (tupE9, unc-5L25, malphaL9, CG3638L6, EgfrE1, eveE428, sliL427) showed cardiac expression patterns and the others were active in the somatic and/or visceral mesoderm. In every case among the above HOT regions with cardiac activities where tested, Tin binding was shown to be functional.
The sizable yield of cardiac enhancers with functionally important Tin binding sites is consistent with the known key role of tin during various stages of cardiac development. In addition, in light of global data from other factors and the distributions of their binding motifs [38], [50], our exclusion of regions with low levels of Tin binding and without well-matching Tin binding motifs likely selected against regions with promiscuous and neutral Tin binding. Nevertheless, we did encounter several examples of Tin-bound mesodermal enhancers that were still active in tin mutant backgrounds, particularly among those active in the somatic mesoderm. In some cases shown herein, as well as in the case of the Fas3L254 enhancer (H.J. and M.F., unpublished), mutations of the Tin binding motifs did not affect enhancer activity in the somatic, cardiac, or visceral mesoderm, respectively. We infer that also at mesodermal enhancers, transcription factors such as Tin can bind neutrally, for example if a fortuitous binding motif is made accessible by other factors, including “chromatin-opening” factors or specific histone marks. In other instances, they may be attracted via mass action solely involving protein-protein interactions, without significantly contributing to the enhancer activity. Unfortunately, laborious tests are necessary in each case to firmly distinguish between different scenarios such as, 1) binding that is biologically essential; 2) binding that makes subtle contributions to the enhancer activity, which may either be biologically irrelevant or perhaps affect the robustness of the developmental process; 3) binding that is functionally redundant with that of other factor(s); 4) promiscuous binding with neutral effects due to the absence of necessary co-factors; 5) binding that is functionally neutralized by a co-bound repressor, as has been shown in the case of some visceral mesoderm enhancers that bind both Tin and the repressor Sloppy paired (Slp) [26]. More generally, it is becoming apparent that, in the absence of functional in vivo tests, global binding data and computational enhancer predictions need to be interpreted with great caution (as for example discussed in [43], [52], [53]). Thus, we propose to reserve the terms “target” and “CRM” for the sequences that have been verified as such in vivo.
In our present study we have performed such tests with a select number of Tin-bound enhancers active in the cardiogenic mesoderm and heart and also included the binding motifs of two other major cardiogenic factors, Pnr and Doc. In five of the six tested enhancers we could demonstrate that Tin binding is functional. This result suggests that, at least among enhancers with cardiac activity, high Tin occupancy, and well-matching Tin binding motifs, there is a high probability that Tin binding is functional rather than being promiscuous.
Each of the six dissected enhancers showed distinct properties with regard to its dependency on the binding sites for the different cardiogenic factors and featured very diverse spatial arrangements of these sites. Thus, EgfrE1s showed complete dependency on the Tin sites, but depended on Pnr sites only from stage 13 onwards and lacked any Doc sites. lin-28L64s required both Tin and Pnr sites independently, but not the Doc sites. midE19s required both the Tin and the Doc sites, but not the Pnr sites. RhoLE102s required the Doc sites but, surprisingly, not the Tin sites and also not the Pnr sites. tupE9s required only the Tin sites but not the Pnr and Doc sites. Finally, in unc-5L25s none of the sites of the three cardiogenic factors were required individually, but in cardioblasts the presence of either Tin or Doc sites (and, to a lesser degree, Pnr sites) was essential, which suggests full or partial functional redundancy between these factors when bound to this enhancer. In pericardial cells, enhancer activity was retained in the total absence of binding sites for all three factors, as was the case in cardiac-specific tin mutants. It is likely that in the wild type, early tin activates a different set of transcription factors in these pericardial cells, which in turn activate the unc-5 enhancer. Altogether, our findings illustrate the diverse functional architecture of each of these cardiac enhancers. At least in part, this reflects their diverse temporal, spatial, and cell type-specific activity patterns. Clearly, a lot more effort will be required before we can fully explain the differential requirement for individual factors and correlate it with the specific activity patterns of the respective enhancers.
In their recent study, Junion et al. [26] assayed global in vivo binding of Pnr and Doc, as well as the Dpp and Wg signaling effectors Mad and TCF, and compared their occupancies with that of Tin. They showed that a significant fraction (∼20–40%, depending on the cut-off) of the Tin-bound regions were co-occupied by all four of the other factors, which reinforces the notion that factors are often co-recruited to active enhancers in the tissues in which they are jointly expressed [43]. An analogous situation was found for the global occupancies of Nkx2-5, GATA4, and Tbx5 in a mouse cardiomyocyte cell line and of Nkx2-5, GATA, and Tbx3 in adult mouse heart, albeit with a lower degree of overlap as compared to Drosophila [54], [55]. The large majority of the tested Drosophila enhancers of this type were indeed active in the dorsal and cardiogenic mesoderm, in which these factors overlap [26]. While these authors did not examine the functionality of individual binding motifs in these enhancers, our present study suggests that for most enhancers the binding of only one or two cardiogenic factors among the three (Tin, Pnr, Doc) is essential and sufficient. Regardless of whether binding is functional or not, the tendency of co-recruitment of factors to active enhancers and the knowledge of their domains of spatial overlap facilitates machine learning approaches to predict spatial patterns of enhancer activities [25].
Largely based on their analysis of the global frequencies of the motifs for Tin, Pnr and Doc within the regions binding all five tested factors versus in those binding only Tin plus one other factor, Junion et al. [26] proposed that Pnr and Doc are recruited preferentially to these enhancers to promote the subsequent recruitment of the other factors including Tin. Although we agree that examples for such an activity of Pnr and Doc might exist, the global data may not be sufficiently reliable to deduct such a general rule. Moreover, the particular enhancer used to support this model appears to be active exclusively in ectodermal stripes and not in the mesoderm (which is consistent with the RNA expression pattern reported for the associated gene, CG14888; BDGP; [28]). Instead, the initial pan-mesodermal expression of Tin, which occurs prior to the dorsal mesodermal expression of Pnr and Doc and is needed upstream of pnr, may argue for the possibility that it is more commonly Tin rather than Pnr and Doc that functions like a pioneer to promote the recruitment of subsequent cardiogenic factors. Tin may be bound to co-repressors prior to the recruitment of subsequent activators, which could sharpen the on/off state of enhancers [46], [56]. In the future, these models can be tested via assaying the in vivo occupancies of cardiogenic factors on binding site-mutated enhancers or, globally, the occupancies of embryonic enhancers in purified mesodermal cells from tin, pnr, or Doc mutant embryos.
The CEE motif discovered by our machine learning approach using 24 enhancers with cardiac activities vs. 37 non-cardiac enhancers was the motif most highly predictive for the cardiac set of enhancers. The validity of the approach was supported by the fact that we also found the GATA motif as a classifier for cardiac enhancers whereas the forkhead domain binding motif was found as a classifier for somatic mesodermal enhancers. The latter was also reported in a recent study using a related approach [57]. It should be noted that the de novo CEE motif is both constructed on and used for the classification of the same dataset. This can lead to selection bias that can produce both a high false positive rate in discovering significant motifs and an overly optimistic estimate of the classifier performance [58]. Therefore, experimental validation of de novo motifs prioritized by the machine learning approach is required to verify their biological function.
In light of these caveats, our findings that three out of eight tested enhancers require the CEE motifs for normal cardiac activity do support the conclusion that these motifs correspond to binding sites of a crucial factor in cardioblasts and their progenitors. This putative factor likely cooperates with the other cardiogenic factors that are active at these enhancers. We can not exclude that this factor is also active when bound to the other cardiac enhancers that did not depend on its binding sites, where it may be functionally redundant. Based upon the sequence of the CEE motif it is difficult to predict the identity of the corresponding binding factor with confidence. However, we note that the CEE motif bears a striking similarity with the binding motifs of several vertebrate Ets domain factors (e.g., FEV, SPI1; [59]). In Drosophila, the Ets domain factor Pointed (Pnt) has been identified as an important regulator of heart patterning and the induction of tin-dependent muscle founder cells in the dorsal mesoderm [12], [34]. The eve MHE enhancer tested for Pnt binding does contain CEE motifs, although they were not included in the mutational analysis by Halfon et al. [12]. The CEE motif also has some resemblance to activating half sites of the NFκB factor Dorsal [60], [61], but Dorsal binds to half sites only poorly, if at all [62], and is present solely in the cytoplasm with undetectable nuclear levels at the stages when the CEE-dependent enhancers become active (A. O., unpublished data). For these reasons, we currently do not favor Dorsal as an essential CEE-binding trans-activator. Additional studies, including the determination of the regions bound by Pnt and Dorsal in vivo during the relevant stages, and genetic as well as biochemical assays, are needed to clarify whether either Pnt or Dorsal might correspond to the presumed CEE motif binding factor or whether another factor is involved. In the near future, the availability of an increasing number of enhancers active in the cardiac mesoderm and other mesodermal tissues will greatly improve the reliability of machine learning approaches to detect novel functional motifs in tissue-specific enhancers.
Embryo crosslinking, chromatin isolation, chromatin immunoprecipitation and DNA amplification were carried out according to the protocol described previously in detail [38], except that Protein A/G beads (Pierce) were used for the immunoprecipitation. Wild-type embryo populations were collected at 3–5.5 and 5–8 hrs after egg laying, respectively. For each developmental time period, two independent ChIP experiments were performed. The polyclonal anti-Tin antibody was obtained by affinity purification with recombinant Tin from rabbit anti-Tin serum [4]. Normal rabbit IgG was used as the mock control. The final amplified and labeled DNA samples were hybridized to GeneChip Drosophila Tiling 1.0R Arrays (Affymetrix) at the microarray core facility of Mount Sinai School of Medicine (New York, NY). ChIP peaks were called using Model-based Analysis of Tiling-arrays (MAT) [63] using a default P-value threshold of 1e-5. The accession number for the dataset in the Gene Expression Omnibus (GEO) database is GSE41628. The data were mapped to the UCSC dm3 genome assembly reference genome.
The strains M{eGFP.vas-int.Dm}ZH-2A;M{RFP.attP}ZH-51C and M{eGFP.vas-int.Dm}ZH-2A;M{RFP.attP}ZH-35B [64] were used for germline transformation of the reporter constructs. The strains tin346/TM3,eve-lacZ and tin-ABD;tin346/TM3,eve-lacZ [9] were used for analyzing the Tin dependency of enhancer activities.
Genomic fragments with the coordinates shown in Tables S4 and S5 were amplified from genomic DNA of Oregon R by PCR, and cloned into either the pH-Stinger-attB vector, which was constructed by inserting the attB sequence [65] into the AvrII site of the pH-Stinger vector, and injected into ZH-35B (at Erlangen or Duke University Model System Genomics, Durham, NC). For AlkE301, AopE53, AopL18, CG5522E11, EgfrE1, EgfrE63, gukhE135, meso18EE403, mipleE8, NrtL30, nvyE164, slpE35, tshL8 and wgnE1197, cloning was done into the eve.promoter-LacZ-attB vector [66] and injections into ZH-51C (at Rainbow Transgenic Flies, Inc., Camarillo, CA). Although located closer to the genes Her and Dot, respectively, the enhancer activities of HimL47 and oddE81 clearly reflected specific aspects of the neighboring genes Him and odd and were named accordingly. The Tin and Doc sites within the genomic fragments were predicted using the web-based tool Target Explorer [67]. Alignment scores >4.5 using the same score matrix for Tin as for Tin motif shown in Figure 1 (see also Table S5) were given higher priority for selection of candidate enhancer fragments. Mutations of these motifs and the GATA sequences were obtained via de novo DNA synthesis (Mr. Gene, Regensburg, DE; Biomatik, Cambridge ON, CA) (Table S6). For Target Explorer, the position weight matrix for Tin was built with Tin-binding sites verified experimentally and in vivo (Table S5) [4], [6], [18], [22], [68]–[70]. Construction of the position weight matrices for Doc2 and Mid was based on the results of SELEX experiments with recombinant Doc2_T-box-GST and Mid_T-box-GST, which were performed following a GST pull-down version of the SELEX method described previously (Table S5) [71].
Whole mount embryo in situ hybridization was carried out following standard protocol, as described previously [69]. The digoxigenin-labeled probe targeting the eGFP transcript was synthesized by in vitro transcription of eGFP sequence inserted into the TOPO cloning site of the pCRII-TOPO vector.
Immunostaining of embryos using DAB and double or triple fluorescent immunostaining were carried out as described previously [69]. The following primary antibodies were used: rabbit anti-GFP (1∶2000, Invitrogen), rabbit anti-β-gal (1∶1200, Promega), rabbit anti-Tin (1∶1000, this study), mouse anti-GFP monoclonal (1∶1000, Invitrogen), mouse anti-β-gal monoclonal (1∶50, Developmental Studies Hybridoma Bank), guinea pig anti-Doc2/3 (1∶400) [72], rabbit anti-Eve (1∶3000) [73]. For fluorescent detection, FITC-, Cy3-, or DyLight-conjugated secondary antibodies (Jackson Laboratories) were used. For monoclonal primary antibodies, Tyramide Signal Amplification (TSA) was performed using biotinylated secondary antibodies (1∶500, Vector Laboratories) in combination with the Vectastain ABC Kit (Vector Laboratories) and fluorescent Tyramide Reagent (PerkinElmer). The images of stainings were acquired using the Zeiss ApoTome/AxioImager.Z1 microscope system with Plan-Apochromat 20×/0,8.
We used RSAT tools oligo-analysis and oligo-diff to find de novo motifs [32]. For both Early and Late datasets we used oligo-analysis with default parameters on the top 100 peaks. On the datasets of all peaks we found that with default parameters the list of de novo motifs is dominated with repeats, thus we use Markov background model of 3rd order and performed Repeat Masking [33]. We used default parameters for oligo-analysis and oligo-diff to discover motifs in subset of experimentally verified enhancers. We ran RSAT on peak regions defined by MAT, but also on the full experimentally verified fragments with and without Repeat Masking. The resulting de novo motifs were curated by hand to remove duplicates and merge similar motifs (consensus sequences different in one nucleotide). The CEE motif was derived by merging a consensus motif enriched in Repeat Masked cardiac enhancers and a consensus motif discovered by oligo-diff when comparing cardiac with somatic enhancers. The somatic1 motif (Figure 6) was discovered as being enriched in somatic enhancers compared to cardiac using oligo-diff.
We used a classification approach similar to [74]. The set of 61 enhancers was scanned with a database of motifs we derived from: JASPAR insects collection [75], ChIP-chip for key mesodermal TFs [25], Dan Pollard's MEME motifs from FlyReg [76] and de novo motifs (as described above). We used a log-odds threshold of 4.6 to call significant motif hits with a Markov background of order 2. A LASSO classifier was trained to predict cardiac enhancers from the set of 61 experimentally verified enhancers based on the density of motif hits (defined as number of motif hits per kb of sequence). Motif hit density was used because the sequences were of different lengths. The classifier was trained on 1000 bootstrap samples of 61 enhancers and the regularization parameter picked so to minimize cross-validation error with the maximal of 50 features selected. We reasoned that if a classifier picked more than 50 features it would most certainly be over-fitting the data. The final classifier (Figure 6) was constructed by taking all features that were present in at least 65% of bootstrap runs. We picked this threshold by starting with 95% and then decreasing it as long as the AUC-ROC of the classifier improved. This yielded a classifier with Area Under the ROC Curve (AUC-ROC) score of 0.90 obtained from Leave-one-out cross validation (LOOCV).
We assigned each ChIP peak to the nearest gene boundary (5′ or 3′ exon) from the peak maximal enrichment point in either direction because a number of key mesodermal enhancers are known to be downstream of their respective target genes. If the peak was in an intron, we assigned it to the nearest transcription start site of the nested gene. We performed GO enrichment analysis using Ontologizer [77] using the Topology-weight algorithm to account for the hierarchical structure of GO ontologies [78]. We used FlyMine to perform protein domain enrichment analysis [79]. We used the BDGP in situ database release 3 [28] and we only retained those entries that were marked of “acceptable” quality. We performed enrichment analysis using a standard hypergeometic test with False Discovery Rate (FDR) correction.
We analysed motif location (Figure 1) by taking all defined ChIP regions and scanning them with the appropriate de novo motifs. Significant motif hits were called above log-odds threshold of 4.5 and their position recorded in percentages relative to peak region width. Thus, we used the original peak positions without trimming them to a fixed size. The motif counts were then binned in 5% bins and the resulting histogram smoothed using loess smoothing. We scaled the density so that number 1 represents the expected uniform distribution. By scanning with unrelated PWMs we verified that the expected distribution is indeed a flat line at 1 (data not shown).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.