A prominent member of the healthy skin microbiome, and a focus of Dr. Belkaid’s research, is the gram-positive bacteria Staphylococcus epidermidis. And then presented to naive T cells by the dendritic cells using a type of protein called MHC. In the context of bacterial pathogens, class II MHCs are used and the resulting peptide-MHC II. MHC (major histocompatibility complex) molecules are glycoproteins that present antigens to T cells to discriminate between self (our cells) and non-self (the invaders or modified self).
The presentation of protein antigens on the cell surface by major histocompatibility complex (MHC) molecules coordinates vertebrate adaptive immune responses, thereby mediating susceptibility to a variety of autoimmune and infectious diseases. The composition of symbiotic microbial communities (the microbiota) is influenced by host immunity and can have a profound impact on host physiology. Here we use an MHC congenic mouse model to test the hypothesis that genetic variation at MHC genes among individuals mediates susceptibility to disease by controlling microbiota composition. We find that MHC genotype significantly influences antibody responses against commensals in the gut, and that these responses are correlated with the establishment of unique microbial communities. Transplantation experiments in germfree mice indicate that MHC-mediated differences in microbiota composition are sufficient to explain susceptibility to enteric infection. Our findings indicate that MHC polymorphisms contribute to defining an individual's unique microbial fingerprint that influences health.
Classical major histocompatibility complex (MHC) genes encode cell-surface glycoproteins that form the basis of antigenic self versus non-self discrimination, by presenting protein antigens (peptides) to circulating T lymphocytes,. Functionally, recognition of self peptides presented by MHC to T cells results in a tolerant immune response, while inflammation is mounted towards non-self antigens. MHC genes are also some of the most polymorphic loci found in vertebrates6, and alleles have been linked to most known infectious and autoimmune diseases of man7. The central role MHC molecules play in vertebrate adaptive immunity has led to intense research spanning several decades on the functional significance of their extreme diversity.
The physiological relevance of MHC polymorphisms has classically been appreciated from the perspective of host-pathogen interactions, where certain MHC alleles bias susceptibility to infection by virtue of their ability to present different pathogenic epitopes. However, in contrast to the transient nature of most infections, individuals are colonized from birth with their microbiota, which is known to have a pervasive influence on host physiology. Studies in knockout mouse models have shown that immune-mediated dysregulation of microbiota composition is a predisposing factor for multiple diseases,. In addition, multiple studies in mice, rats, fish and humans, have demonstrated correlations between MHC variation and microbiota composition, though the physiologic relevance of these relationships were not determined. Together, these observations suggest that an individual's MHC genotype might exert its most profound effect on host fitness by influencing the relationship between hosts and their symbiotic microbiota. Whether MHC genotype impacts host health by functioning to sculpt an individual's microbiota has not been tested.
Antibody-mediated (that is, humoral) immunity is facilitated in the gut by interactions between MHC class II restricted CD4+ T-follicular (TFH) helper cells and naive B cells that instigate germinal centre formation and the production of high-affinity immunoglobulin A (IgA). IgA controls the abundance of extracellular microbes by tagging organisms for destruction by the immune system, by regulating bacterial epitope expression, and by aggregating and eliminating them from the gut via peristalsis. Thus, antibody-mediated selection is a key means by which hosts are capable of controlling microbial community composition in the gut. In support of this, activation-induced cytidine deaminase (AID)-deficient animals (whose B cells do not undergo somatic hypermutation and affinity maturation) have severe alterations to their gut microbiota. In addition, defects in the interaction between TFH cells and germinal centre B cells alters the host IgA antibody repertoire, which is associated with differences in the community of organisms that develop within these animals,. Given the role of MHC class II molecules in driving humoral immune responses, this is a likely mechanism through which MHC polymorphisms could shape microbiota composition.
Previous research has demonstrated differential patterns of susceptibility among MHC congenic mouse strains against a wide variety of enteric pathogens,. This is generally assumed to reflect variability in an individuals' suite of MHC molecules that differentially stimulate the immune system to clear infection and limit disease. However, differences in the composition of resident microbial communities can influence disease susceptibility associated with pathogenic infection. Colonization resistance is a phenomenon that occurs when members of the microbiota inhibit the establishment of environmentally acquired pathogens, thus limiting their potential to infect and cause disease. Moreover, specific members of a microbiota are more important than others in conferring colonization resistance,. Based on this, we tested the hypothesis that MHC polymorphisms could dictate susceptibility to enteric infection and its associated disease by influencing microbial community architecture.
Results from our experiments demonstrate that MHC polymorphisms influence gut mucosal immunity by driving differential IgA responses that develop against commensal microbes. MHC-mediated differences in gut immunity were correlated with the establishment of unique microbiota communities among individuals. Importantly, microbiota transplant experiments in germfree mice demonstrated that the unique microbiotas formed in mice of different MHC genotypes impacted host health by controlling susceptibility to enteric infection independent of the immune response. In addition, microbiota from an MHC heterozygous genotype conferred resistance to infection similarly to the microbiota derived from the most resistant MHC homozygous genotype. Thus, results from our experiments indicate that MHC-mediated patterns of disease susceptibility, including heterozygote advantage, may partially be explained by how MHC sculpts microbiota composition in the gut. This study also establishes MHC genes as primary host immunogenetic factors driving the high degree of individuality in microbiota composition observed among humans.
Humoral immunity is a key means by which hosts regulate microbial composition in the gut in an antigen-specific manner, primarily through production of secretory IgA,. MHC class II genes are central to this process, so we first sought to determine whether an individual's MHC genotype influenced steady-state development of the gut immune response. We focused on characterizing immune parameters within the Peyer's patches of mice as these are the primary inductive sites of T-cell-dependent IgA responses in the gut. MHC (called the H2 complex in mice) congenic mice have been bred to possess unique suites of alleles at MHC class I and class II genes, while sharing the same genetic background (BALB/c), making them an ideal model to investigate the role of MHC genotype on animal physiology (Supplementary Table 1). Hereafter our model BALB/c H2 congenic mouse strains will be designated as H2bb, H2dd and H2kk, where superscripts denote homozygous H2 genotypes derived from C57BL/6, BALB/c and C3H/He mice, respectively. Immune phenotyping experiments using flow cytometry revealed significant differences among MHC genotypes in a variety of immune parameters related to the IgA response (Fig. 1a,b). Specifically, compared with animals possessing the H2dd and H2kk genotypes, congenic mice possessing the H2bb region had significantly increased abundance of TFH cells (CD4+B220−CXCR5+PD-1+), germinal centre B cells (B220+IgDloFAS+GL7+), IgA+ B cells (B220+CD138−IgA+) and IgA+ plasmablasts (B220−CD138+IgA+; Fig. 1b). Interestingly, H2kk congenic mice had significantly reduced MHC class II surface expression on conventional dendritic cells (CD40+CD86+CD11c+MHCII+) as well as on naive B cells (B220+IgDhiMHCII+) compared with H2dd and H2bb mice (Supplementary Fig. 1a). Finally, epistatic interactions between MHC polymorphisms and non-H2 genes in the BALB/c genetic background could potentially explain the observed differences in immune phenotypes between mice possessing the natural H2dd region and congenic animals possessing the recombined H2bb and H2kk regions. To address this, we compared the same set of immune parameters in C57BL/6 and C3H/He mouse strains that were H2-matched with the H2bb and H2kk congenic strains, respectively. Parallel effects were observed between H2-matched genotypes for all the observed phenotypes. Similar to H2bb congenic BALB/c mice, C57BL/6 (H2bb) had higher levels of TFH, GC B cells, IgA+ B cells and IgA+ plasmablasts compared with the H2dd genotype (Fig. 1b), and C3H/He (H2kk) animals had similarly reduced surface expression of MHC on conventional dendritic cells and naive B cells (Supplementary Fig. 1a). These data strongly support that observed phenotypic differences among BALB/c H2 congenic mice are a result of MHC-intrinsic effects, and demonstrate that variability in MHC genes leads to unique steady-state immune development in the gut.
MHC mediates IgA response against commensals.(a) Flow cytometry (F.C.) was used to quantify multiple immune parameters within the PPs of MHC congenic animals. A heatmap is used to summarize results of immune phenotyping experiments. Heatmap depicting relative abundances of cell subsets (red=more abundant, green=less abundant). Gray boxes represent no significant differences among genotypes for a given cell subset based on the results of a one-way ANOVA (P>0.05). (b) Data sets comparing discriminating immune phenotypes (H2bbn=15; H2ddn=14; H2kkn=15). Data represent pooled results of three replicate experiments. Data from H2-matched C57BL/6 (H2bb) mice (n=6) are provided for comparison against H2bb BALB/c congenics. (c) Abundance of IgA+ plasmablasts in the siLP was enumerated via flow cytometry (H2bbn=7; H2ddn=6; H2kkn=8). (b,c) Representative F.C. plots are provided to illustrate gating strategy. (d) The abundance of faecal IgA antibodies was measured by ELISA (H2bbn=11; H2ddn=11; H2kkn=11). Data represent pooled results of two replicate experiments. (e) The relative amounts of faecal bacteria bound by IgA was enumerated via flow cytometry among cohorts of H2bb (n=8), H2dd (n=8), and H2kk (n=7) animals. Representative F.C. plots illustrate assay specificity by demonstrating a low incidence of false-positive events recorded in IgA-deficient RAG1−/− control animals. Data represents pooled results from two replicate experiments. (f) The number of unique PP-derived CDR3 amino acid sequences (from IgH chain) observed per sample in each immunoglobulin repertoire rarified to a depth of 1,250 sequence observations per sample. Plots represent pooled data from all three genotypes. (g) Heatmap showing IgH variable family usage among genotypes. For each repertoire, sequence abundance per variable region family is normalized across row via Z-score. Gray cells indicate that a variable family was absent in all three genotypes within a given Ig repertoire. (h) The number of unique IgA and IgG CDR3 amino acid sequences (pooled) observed among genotypes rarified to a depth of 1,250 sequence observations per sample (H2bbn=4; H2ddn=4; H2kkn=4). (i) The mean affinity maturation index for all unique IgA and IgG sequences within a given genotype. Error bars represent s.e.m. (b,c,d,e,f,h,i) Bars represent means. Asterisks denote results of two-tailed unpaired Student's t-tests (****P<0.00001; ***P<0.001; **P<0.01; *P<0.05). NS, not significant.
IgA+ B cells and plasmablasts migrate to the small intestinal lamina propria for final maturation into Ig-producing plasma cells, where secreted antibody is then transported and deposited onto the luminal surface of the gut epithelium. Consistent with data from Peyer's patches, we found that H2bb animals had a significantly higher abundance of IgA+ plasmablasts in their small intestinal lamina propria compared with the other two genotypes (Fig. 1c). This was associated with significantly greater amounts of IgA in the faeces of H2bb animals (Fig. 1d). To ascertain whether differences in humoral immune responses resulted in a quantifiable difference in antibody binding to commensal bacteria, we employed a flow cytometry-based assay. A significantly greater fraction of commensal bacteria were bound to IgA in both H2bb and H2dd animals compared with H2kk animals (Fig. 1e). This is not due to differences in bacterial abundance as faecal bacterial loads are equivalent among genotypes (Supplementary Fig. 1b).
Selection on B cells to differentiate into high-affinity antibody-producing plasma cells is MHC restricted,. To address whether MHC polymorphisms influenced the quality of the antibody response against commensals by influencing the nature of selection on the antibody repertoire, we performed Illumina sequencing to characterize developing antibody repertoires in the Peyer's patches of H2 congenic animals. Somatic recombination produces a naive Ig repertoire with the potential capacity to recognize any epitope. However, during T-cell-dependent Ig responses, signals from T cells instruct selected B-cell clones to expand and undergo the process of affinity maturation and class-switch recombination. Therefore, only a fraction of the diversity within the naive antibody repertoire should be represented in the mature class-switched antibodies. Consistent with this process of negative selection, we observed a significant reduction in sequence diversity in the mature IgA+ and IgG+ repertoires compared with the naive IgM+ repertoire (Fig. 1f). In addition, differences in V-region gene usage were also observed among H2 genotypes in all three Ig repertoires (Fig. 1g), consistent with an influence of MHC polymorphisms on selection of the developing Ig repertoires. Moreover, consistent with the stronger humoral response seen in H2bb animals, this genotype is also associated with a significant increase in the accumulation of novel class-switched IgA and IgG sequences (Fig. 1h) and has the highest IgA/IgG affinity maturation scores of the three genotypes (Fig. 1i). Importantly, no differences in repertoire diversity or affinity maturation are observed among H2 genotypes when naive IgM repertoires are compared (Supplementary Fig. 1c,d), which further supports the argument that observed differences arise as a consequence of differential MHC-mediated selection on maturing IgA and IgG antibody repertoires. Collectively, results from our phenotyping, IgA-binding and Ig repertoire experiments indicate that MHC polymorphisms lead to an individualized antibody response that develops in the gut towards commensals.
The IgA response against commensals has been shown to directly influence microbial community composition in the gut,. Therefore, we next sought to determine whether differences in the immune response between H2 congenic mice translated into alterations in microbiota architecture. We used 16S ribosomal RNA (rRNA) gene sequencing to characterize the influence of MHC-mediated antigen presentation and MHC polymorphisms on microbial community architecture in the gut. Differences in microbial communities can be quantified based on both presence/absence and relative abundance of OTUs (Operational Taxonomic Units at 97% similarity; approximately species equivalents). We hereafter use the terms ‘composition' and ‘structure', respectively, to refer to these different quantifications. Unweighted and weighted UniFrac analyses reflect the relative importance of host MHC genotype on the composition or structure of the microbiota, respectively. Lack of class I- and class II-mediated antigen presentation (B2M−/− and MHCII−/− mice, respectively) was associated with significant shifts in faecal microbial composition and structure compared with wild-type (WT) controls (Fig. 2a, Supplementary Fig. 2a), with significant differences in the relative abundance of a variety of specific genera (Supplementary Fig. 2b). Notably, the genus Lactobacillus was significantly reduced in MHCII−/− compared with WT animals, while segmented filamentous bacteria and Helicobacter were enriched (Supplementary Fig. 2b). Faecal microbial communities from MHCII−/− mice were significantly more dissimilar to those from WT animals than were communities from B2M−/− mice (Fig. 2b), implying that class II antigen presentation plays a stronger role in mediating microbial community composition in the gut. In addition, communities from MHCII−/− animals were more similar to one another (Fig. 2c) and had fewer observed species (Fig. 2d) compared with WT and B2M−/− animals, suggesting that class II antigen presentation promotes inter-individual variability in microbiota composition and microbial community diversity, respectively.
MHC influences microbial composition in the gut.(a) PcoA plot based on unweighted UniFrac of faecal communities from male WT (n=4), B2M−/−(n=6), and MHCII−/−(n=6) C57BL/6 animals. (b) Distance boxplots of community similarity between WT animals and each of the two KO mouse strains. (c) Distance boxplots of community similarity among WT animals versus similarity among MHCII−/− animals. (d) Comparison of the observed number of species in WT, B2M−/−, and MHCII−/− communities. (male and female animals pooled per genotype (WT n=9; B2M−/−n=10; MHCII−/−n=9) (e) PcoA plot depicting relationships between faecal communities among female MHC congenic animals (H2bb=blue dots (n=5); H2dd=green dots (n=5); H2kk=red dots (n=5)). (f) Distance boxplots of community similarity among individuals within each MHC genotype. (g) Comparison of observed number of species among H2 congenics. (h) Comparison of Shannon diversity estimates among communities from H2 congenic animals. (i) PcoA plot illustrating differences in community composition between faecal and mucosal communities, and among MHC genotypes within each of these communities (H2bb=blue dots (n=5); H2dd=green dots (n=5); H2kk=red dots (n=5))(squares=mucosal samples, circles=faecal samples). (j,k) Comparison of observed number of species (j), Shannon diversity (k) between faecal and mucosal communities. (l) Distance boxplots based on between-genotype unweighted UniFrac distance estimates reflecting the degree of dissimilarity in community composition between genotypes by site (feces versus Mucosa-associated). (a,e,i) Results of PERMANOVAs for PcoA plots provided under respective plot. (b,c,d,f,g,h,j,k,l) Results of two-tailed unpaired Student's t-test (****P<0.00001; ***P<0.001; **P<0.01; *P<0.05). (d,g,h,j,k) Bars represent means. (b,c,f,l) Boxplots represent medians, the 25 and 75% quartiles, and whiskers represent range. NS, not significant.
We next sought to determine whether MHC polymorphisms lead to distinct microbial communities. To control for effects of isolation among congenic strains (legacy effects), homozygous animals were purchased from Jackson Laboratories and bred to create F1 heterozygotes. F1 animals were bred and homozygous offspring were identified via genotyping using microsatellite markers that distinguish the respective H2 haplotypes. F2 homozygotes were subsequently bred pure and housed under identical conditions in our specific pathogen-free (SPF) facility. Unweighted UniFrac analysis revealed significant divergence of faecal microbial communities among H2 congenic animals (Fig. 2e), while a marginal difference among genotypes was observed based on weighted UniFrac analysis (Supplementary Fig. 2c). Differences among MHC genotypes in constraining community membership were also observed, as comparisons between individuals of the same genotype revealed greater dissimilarity within H2bb and H2kk communities than within H2dd communities (Fig. 2f). Finally, microbial communities from H2dd animals were also observed to be less species rich (Fig. 2g) and less diverse (Fig. 2h) than H2bb and H2kk communities.
IgA antibodies are transported across mucosal epithelium and concentrate within the mucus layer where these molecules provide a first line of antigen-specific defense of underlying host tissues,. Indeed, IgA tends to target organisms present within the mucus. To test whether MHC genotype exerted its strongest effect at this site, we sequenced mucosally associated and faecal communities in MHC congenic mice. In agreement with previously published findings, faecal and mucosally associated communities were significantly different (Fig. 2i). Mucosally associated communities were less species rich (Fig. 2j) and more phylogenetically diverse (Fig. 2k) than faecal communities. These data imply that the mucosa represents an environment with more unique niches relative to faeces that are filled by a more phylogenetically diverse array of species. In addition, while MHC genotype had a significant effect on community composition at both sites (Fig. 2i, Supplementary Fig. 2d), compositional differences were more pronounced when mucosally associated versus faecal communities were compared (Fig. 2l). Thus, MHC polymorphisms are again associated with significant compositional differences among individuals and the effect of MHC genotype is most potent at the site where antibodies are at their highest concentrations; the mucosal surface. Collectively, results from three independent sequencing experiments provide strong support to the growing body of evidence indicating that MHC antigen presentation sculpts microbiota communities and contributes to the high degree of variability in microbiota composition among individuals. Finally, the vast majority of significantly different OTUs across all of our experiments fall within the same six bacterial families from the two dominant phyla found within the vertebrate gut; the Bacteroidetes (families S24-7, Prevotellaceae and Rikenellaceae) and Firmicutes (families Lactobacillaceae, Lachnospiraceae and Ruminococcaceae; Supplementary Fig. 2e, Supplementary Tables 4–6). This is noteworthy because, with the exception of the Rikenellaceae, all of these families of bacteria have previously been shown to either be significantly impacted by defective IgA responses, or to be preferentially targeted by IgA antibodies,. This further supports our argument that observed microbiota shifts are likely due to MHC-mediated variability in the IgA response generated against commensal microbes.
To investigate whether MHC genotype and the associated microbial communities that develop within these genotypes influence susceptibility to disease, we utilized the Salmonella enterica typhimurium (S. e. typhimurium) model of enteric infection to which BALB/c mice are susceptible. To first demonstrate that MHC genotype influences susceptibility to S. e. typhimurium infection cohorts of H2dd, H2bb and H2kk animals were infected orally with 104S. e. typhimurium colony-forming units (CFUs) to model the natural route of infection, and the severity of disease was compared 7 days later. In this model, orally gavaged S. e. typhimurium rapidly colonizes the gastrointestinal tract in as little as 1 h post infection, followed by rapid clearance from the faeces (Supplementary Fig. 3a). H2bb animals were significantly more susceptible to Salmonella-induced disease than were H2dd and H2kk BALB/c mice as measured by weight loss, splenomegaly and splenic S. e. typhimurium loads (Fig. 3a–c). Importantly, these data establish MHC-mediated patterns of susceptibility to lethal Salmonellosis.
MHC-mediated susceptibility to S. e. typhimurium infection is microbiota-dependent.(a–c) Susceptibility to systemic disease resulting from oral infection with 104S. e. typhimurium CFUs among MHC genotypes (H2bbn=10; H2ddn=9; H2kkn=9). (a) Weight loss by day (left panel) and trend in weight loss over time (right panel), (b) splenomegaly, and (c) S. e. typhimurium loads in the spleen on day 7 are shown. Data represents pooled results from two independent experiments. (d–f) Susceptibility to systemic disease resulting from oral infection with 104S. e. typhimurium CFUs among GF BALB/c animals that had been previously colonized with the microbiotas derived from either H2bb (blue open circles (n=16)) or H2dd (green open circles (n=10)) congenic animals. Data represent the pooled results of three independent replicate experiments. Dotted lines represent the mean scores of GF BALB/c animals for each of the respective disease parameters. (g) (left panel) Heatmap illustrating how colonization of GF animals with H2bb or H2dd microbiota influences the gut immune phenotype of animals (GF-n=9; H2bb transplant n=7; H2dd transplant n=7). (right panel) Results of MANOVA analysis demonstrating that colonization of GF animals results in a significant, but not differential, increase in immune investment by microbiota treatment. MANOVA analysis was based on cell abundance estimated as a percentage. MANOVA (****P<0.00001). Data represents results of two independent experiments. (a right panel, b,d right panel, e) Results of two-tailed unpaired Student's t-test (***P<0.001; **P<0.01; *P<0.05). (a and d (left panels)) Error bars represent s.e.m. Asterisks represent significant differences based on results of t-tests comparing weight loss by day and reflect significant differences compared to H2bb animals in a or GF animals receiving the microbiota of H2bb animals in d. (c,f) Asterisks represent significance based on results of Mann–Whitney U-Test, (**P<0.01; *P<0.05). NS, not significant.
To determine whether the reduced disease severity in H2dd and H2kk animals was associated with a more robust immune response against S. e. typhimurium, we analysed multiple immune parameters via flow cytometry at day 7 post infection. Multiple components of cell-mediated immunity were increased in response to infection, including activated and inflammatory CD4+ TH cells (Supplementary Fig. 3b,c). However, this was exclusively driven by animals possessing the susceptible H2bb genotype (Supplementary Fig. 3b,c). In addition, while phagocyte uptake is a means by which S. e. typhimurium can disseminate to the systemic compartment, there was no significant difference in phagocyte abundance among genotypes (Supplementary Fig. 3d–f). Results from these experiments indicate that resistance to infection in our model is not driven by a more robust immune response and that enhanced disease in H2bb animals is not simply a consequence of increased phagocyte trafficking of S. e. typhimurium into the systemic compartment.
To determine whether enhanced resistance to enteric S. e. typhimurium infection was influenced by the microbiota, we first treated H2dd animals with a cocktail of antibiotics for 3 days to deplete the microbiota and then assessed resulting patterns of susceptibility. Short-term antibiotic treatment, which reduced faecal microbiota loads by an order of magnitude, rendered resistant H2dd animals highly susceptible to Salmonella-induced disease and systemic invasion (Supplementary Fig. 4a–d). Thus, depletion of the microbiota facilitates systemic invasion and renders an otherwise resistant host genotype highly susceptible to lethal salmonellosis. Previous work has shown that germfree BALB/c animals orally infected with this pathogen quickly succumb to systemic salmonellosis, and that animals can be protected from systemic disease through mono-association with specific bacterial species or colonization by a complex microbiota,. Consistent with these studies, germfree BALB/c animals raised in our facility were highly susceptible to systemic S. e. typhimurium infection, whereas exGF BALB/c animals that had been conventionally housed since birth in our SPF facility were protected from disease (Supplementary Fig. 4e–g). Together, these results indicate that the microbiota provide strong colonization resistance against S. e. typhimurium infection. Finally, to directly test whether compositional differences in microbial communities are responsible for observed differences in patterns of disease among MHC genotypes, we performed microbial transplantation experiments into GF BALB/c recipients. For these experiments, animals were equally colonized with microbiota stocks derived from animals of the highly resistant H2dd or highly susceptible H2bb genotypes (Supplementary Fig. 5a–c) and they were subsequently challenged with an oral dose of 104S. e. typhimurium CFUs. Consistent with patterns of disease susceptibility observed between H2dd and H2bb animals, GF mice colonized with the microbiota derived from H2bb animals developed significantly worse disease than those animals colonized with microbiota derived from resistant H2dd animals (Fig. 3d–f). Importantly, the results shown in Fig. 3d–f are the pooled results of three independent replicate experiments, and for each replicate experiment a new microbiota stock was created from each H2 genotype. Each replicate experiment showed similar trends in disease susceptibility, which is inconsistent with a cage effect significantly biasing microbiota composition in our model. Collectively, these data demonstrate that the unique microbiotas formed in discrete MHC genotypes are both necessary and sufficient to explain observed patterns of disease susceptibility.
The microbiota can promote defense against an invading pathogen by stimulating an immune response. Immune phenotyping experiments in infected MHC congenic animals suggested that resistance to infection in our model is not immune mediated (Supplementary Fig. 3b–f). However, our initial analysis of immune system development in MHC congenic mice demonstrated differences in steady-state levels of several immune parameters (Fig. 1a). Thus, it is possible that colonization with H2bb or H2dd microbiotas differentially influence the immune response that develops in the gut, which mediates susceptibility to infection. To test this we again performed faecal transplants into GF BALB/c mice and compared the immune phenotypes that emerged in the absence of Salmonella infection. Analysis of multiple immune parameters 3 weeks post colonization revealed a significant overall increase in immune response in the gut of colonized animals (Fig. 3g, left panel). Importantly, however, microbiota derived from H2bb and H2dd animals did not differentially influence the immune response that develops (Fig. 3g, right panel). These data indicate that microbiota-dependent patterns of disease susceptibility shown in Fig. 3d–f are not a consequence of these two microbiotas driving the development of significantly different immune responses in the gut. Thus, microbiota derived from H2bb or H2dd genotypes recapitulate patterns of disease susceptibility independently of the immune response. In addition, these data also suggest that the differential patterns of immune response observed in MHC congenic animals in Fig. 1 are not due to differences in microbiota composition, but rather are a consequence of host genetics.
MHC genes are highly polymorphic and most individuals possess a unique suite of alleles. Moreover, MHC alleles are co-dominantly expressed. Because of these features it is widely assumed that MHC heterozygosity benefits host health by allowing the immune system to present a wider variety of antigens, thereby enhancing resistance to disease. However, our results demonstrate that MHC polymorphisms lead to unique microbial communities and this can control disease associated with enteric infection. Thus, the advantage conferred by MHC heterozygosity might also be a result of how MHC influences the composition of the microbiota. To test this, we performed 16S rRNA gene sequencing to compare microbial communities between female H2bd heterozygotes and the two H2bb and H2dd parental strains used to derive this genotype. While on average H2bd heterozygotes do not have more diverse microbial communities compared with homozygotes (Supplementary Fig. 6a,b), MHC heterozygosity had a significant effect on community composition but not structure based on unweighted (Fig. 4a) and weighted (Supplementary Fig. 6c) UniFrac analysis, respectively. These results lend further support for a deterministic role of host MHC in sculpting the composition of microbial communities.
MHC heterozygote advantage is microbiota-dependent.(a) PcoA plots highlighting the significant difference in community composition based on unweighted UNIFRAC between H2bd (n=5) heterozygotes and animals from the respective homozygote H2bb (n=5) and H2dd (n=5) genotypes. P-value represents result of PERMANOVA significance test. Ellipses highlight relevant comparisons and are non-quantitative. H2bb and H2dd data points are from the same animals depicted in Fig. 2e–h. (b–d) Susceptibility to systemic disease resulting from oral infection with 104S. e. typhimurium CFUs among H2bd (n=7) animals compared to H2bb or H2dd animals. Data represents pooled results from two replicate experiments. Greyed H2bb and H2dd data points in b-d are the same as those shown in Fig. 3a–c. (e–g) Susceptibility to systemic disease resulting from oral infection with 104S. e. typhimurium CFUs among GF BALB/c animals that had been previously colonized with the microbiota from H2bd heterozygous animals (n=7). Data represents pooled results from two independent experiments. Greyed H2bb transplant and H2dd transplant data shown in e–g are the same as those shown in Fig. 3d–f. (b right panel, c,e right panel, f) Results of two-tailed unpaired Student's t-test (***P<0.001; **P<0.01). (b and e (left panels)) Error bars represent S.E.M. Asterisks represent significant differences based on results of t-tests comparing weight loss by day and reflect significant differences between H2bd heterozygotes compared to H2bb animals in b or GF animals receiving the microbiota of H2bb animals in e. (d,g) Asterisks represent significance based on results of Mann–Whitney U-test, (**P<0.01). NS, not significant.
To directly test heterozygote advantage in our model, we performed experimental S. e. typhimurium infections in H2bd animals and compared resulting patterns of disease susceptibility with that observed in the H2dd and H2bb parent strains. H2bd animals were observed to be similarly resistant to infection as the H2dd parent strain (Fig. 4b–d). These data indicate that resistance to S. e. typhimurium infection in the H2bd heterozygote genotype is dominant, which is consistent with heterozygote advantage in this model. To address whether the microbiota formed in H2bd heterozygotes conferred protection against S. e. typhimurium infection we again conducted microbial transplantation experiments in GF BALB/c mice. Transfer of the microbiota from H2bd donor mice into GF BALB/c recipients completely protected these animals from S. e. typhimurium invasion and associated disease (Fig. 4e–g). Therefore, in this model heterozygote advantage operates independently of the effect of heterozygosity on immunocompetence, and instead results from a bias in the composition of the microbiota formed within this genotype. Collectively, results from experiments with MHC heterozygotes indicate that MHC-mediated resistance to S. e. typhimurium infection is dominant, which is a commonly observed pattern. Our data indicate that this common pattern, which results in heterozygote advantage over the course of many infections, might be driven by how MHC influences microbiota architecture in addition to its influence on immunocompetence. Interestingly, overall microbiota composition and structure in H2bd animals is more similar to H2bb than H2dd animals (Supplementary Fig. 6d). These data suggest that the presence or relative abundance of a specific species shared between H2bd and H2dd animals might be the basis of colonization resistance against Salmonella.
Anaerobic bacteria are known to interfere with colonization of the murine gut by Salmonella, so enhanced susceptibility to Salmonellosis in H2bb animals, or GF animals colonized with H2bb-derived microbiota, could be due to reduced loads of anaerobic bacteria, or reduced loads of specific groups of anaerobes. Plating of faeces did not reveal a difference in total loads of anaerobic bacteria among these genotypes (Fig. 5a). To determine whether specific anaerobic species previously shown to be antagonistic to Salmonella colonization were differentially abundant between H2bb, H2dd and H2bd animals, we quantified the relative abundance of species known to be antagonistic to Salmonella colonization of the gut. Based on sequencing results, Lactobacillus spp. and Bacteroides spp. were enriched in H2dd animals over H2bb animals, but only Lactobacillus appeared to be enriched in the H2bd genotype as well (Fig. 5b). Quantitative PCR using Lactobacillus-specific primers confirmed this result (Fig. 5c). Also, during microbiota transplants significantly less Lactobacillus is transplanted into animals colonized with H2bb-derived microbiota than in animals colonized with H2dd- and H2bd-derived microbiota (Fig. 5d). These data imply that enrichment for anaerobes, and specifically Lactobacillus spp., a bacterial group known to be highly antagonistic to Salmonella colonization, might account for enhanced colonization resistance against S. e. typhimurium in our model.
Enrichment of Lactobacillus may explain enhanced colonization resistance.(a) Results of experiment quantifying faecal loads of anaerobic bacteria from the faeces of H2bb, H2dd and H2bd animals. Loads are standardized to faecal weight (per mg faeces). (b) OTU abundance plots of specific bacterial groups from H2bb (n=5), H2dd (n=5), and H2bd (n=5) animals. Asterisks denote significance based on Mann–Whitney U-test (*P<0.05). (c) Results of qPCR experiment quantifying faecal bacterial loads of Lactobacillus, Bacteroides, and Bifidobacteria in the faeces of H2bb (n=10), H2dd (n=9), and H2bd (n=7) animals. Loads are standardized to the amount of input DNA per reaction and are reported as copies per nanogram faecal DNA. Data represents pooled results from two replicate experiments. (d) Results of Q-PCR experiment quantifying faecal bacterial loads of Lactobacillus in the faeces of GF animals that had been previously colonized with the microbiotas of H2bb (n=5), H2dd (n=4), or H2bd (n=7) animals (day 7 of colonization). (c,d) Asterisk's denote results of two-tailed unpaired Student's t-test (**P<0.01; *P<0.05). NS, not significant.
Colonization resistance is a phenomenon that occurs when members of the microbiota inhibit the establishment of environmentally acquired pathogens, thus limiting their potential to infect and cause disease. Moreover, specific members of a microbiota are more important than others in conferring colonization resistance,. Microbiota transplantation experiments in animal models have been instrumental in defining the relationship between alterations to microbiota composition in the gut and emerging diseases of man including obesity, diabetes, metabolic syndrome and inflammatory bowel disease (IBD),. Using this methodology, we have provided data to support a model whereby an individuals' MHC genotype biases IgA-mediated selection against the microbiota that results in compositional differences that impact host health by controlling microbiota-dependent colonization resistance against an enteric pathogen. Importantly, microbial transplants from resistant donors can protect highly susceptible hosts from disease. The relevance of our experimental results to humans is supported by multiple studies. Class II HLA polymorphisms in humans have been correlated with shifts in microbiota composition and diseases like ankylosing spondylitis, rheumatoid arthritis and coeliac disease. In addition, more recent deep sequencing efforts have revealed strong associations between class II HLA alleles and IgA deficiency and susceptibility to IBDs like Crohn's and ulcerative colitis. Importantly, our experiments now identify microbiota-mediated colonization resistance as a malleable resistance mechanism influenced by an individual's MHC genotype.
Encounters with highly virulent pathogens are rare and autoimmune diseases tend to manifest later in life, so health costs associated with infectious and autoimmune diseases might not be the sole driving force of selection on MHC genes. Importantly, results from our experiments provide evidence of reciprocal selection between hosts and their microbiota mediated by MHC genes; the hallmark feature of coevolution. Hosts influence the fitness of commensal microbes, and these microbes have a profound influence on host health. Perhaps, the evolution of MHC diversity reflects local co-evolutionary dynamics with commonly encountered environmental microbes whose domestication diversified the metabolic capacity of their vertebrate hosts and limited colonization by more aggressive species. MHC alleles could be maintained over time because they facilitated beneficial symbiotic interactions that outweighed potential risks associated with infectious and autoimmune diseases. In addition, perhaps, antagonistic co-evolution between hosts and rapidly evolving pathogens promoted and maintained alleles that resulted in colonization by less beneficial symbionts, which might explain individual predisposition towards microbiota-mediated diseases like IBDs, metabolic diseases or infections by enteric pathogens. We speculate that MHC gene evolution has in part been influenced by co-evolutionary forces between hosts and their resident commensal microbes.
While our study provides support for a model whereby MHC polymorphisms differentially influence antibody-mediated selection on the microbiota, there are a myriad of additional potential mechanisms by which MHC-mediated antigen presentation could influence commensal populations. MHC class II molecules present exogenously derived peptides to CD4+ helper T cells that coordinate immune responses against extracellular antigens. Recent studies have shown that multiple CD4+ TH cell subsets (TFH, Treg and TH17) directly recognize commensal antigens in the gut,. Each of these subsets are known to influence humoral immune responses against the gut microbiota, making both MHC-directed regulatory and inflammatory pathways capable of influencing microbial composition,. Recent work has also uncovered a previously unappreciated role for MHC class II presentation in innate lymphoid cells. Group 3 innate lymphoid cells utilize MHC class II presentation to anergize commensal-specific T cells to promote tolerance towards the microbiota. Finally, gut enterocytes secrete a variety of antimicrobial molecules capable of influencing microbiota composition, and these cells express MHC class II on their surface. Whether MHC class II expression by gut epithelial cells influences microbial communities is completely unknown, but warrants attention given the direct physical interaction between these cells and commensal microbes. Together, these studies highlight a central and functionally diverse role for MHC-mediated antigen presentation in modulating immune responses against commensal organisms. Future studies using conditional knockout mice will be crucial for empirically defining the process by which MHC antigen presentation sculpts microbiota architecture.
Results from our experiments suggest that MHC polymorphisms may contribute to the high degree of individuality observed in microbiota composition among humans. Understanding how MHC shapes microbiota diversity to influence colonization resistance could yield important insights into the treatment of emerging enteric infectious diseases of man. Clostridium difficile and Campylobacter jejuni are enteric bacterial pathogens that cause severe disease in susceptible people, and colonization resistance is known to play an important role in mitigating disease caused by these organisms,. In addition, microbiota transplants from uninfected donors have proven extremely effective at treating recurring C. difficile-induced colitis. Therefore, identifying how host genotype and microbiota composition synergize to determine the phenomenon of colonization resistance has clear therapeutic potential. For example, HLA-microbiota typing, coupled with microbiota transplant screens in germfree animals, might be a useful strategy for identifying the most efficacious microbiotas (that is, donor source) for transplantation therapies. In addition, such a strategy might also be a useful methodology for identifying the core ‘colonization-resistance-promoting' members of the microbiota that could be tested for their potential in future probiotic formulations. Similarly, identifying HLA alleles or haplotypes that predispose individuals to infection by these and other enteric pathogens might be a useful biomarker in preventive medicine by identifying at-risk members of the population that might benefit from a probiotic regiment of colonization-resistance-promoting species. These potential therapeutic benefits warrant further research as they might enhance personalized medicine.
Eight-week-old male and female homozygous BALB/c MHC congenic mice (hereafter denoted H2bb (C.B10-H-2b/LilMcdJ, cat#001952), H2dd (BALB/cJ, cat#000651) and H2kk (C.C3-H-2k/LilMcdJ, cat#001951)) were purchased from Jackson Laboratories and housed under SPF conditions at the University of Utah. To control for legacy effects across mouse strains, homozygote breeders were re-derived using the progeny of heterozygote crosses. Homozygote genotypes were confirmed by PAGE using linked microsatellite markers that discriminate among MHC genotypes used in these experiments. Germfree BALB/c mice were bred and reared in our own germfree facility. Age-matched male and female WT C57BL/6, MHCII−/− and B2M−/− were kindly provided by Dr Xianjian Chen. All animal use was in compliance with federal regulations and guidelines set forth by the University of Utah's Institutional Animal Care and Use Committee (Protocol#14-05009).
The LT2 strain of Salmonella enterica (serovar typhimurium) was used in all experimental infections. S. e. typhimurium was grown in LB broth overnight and animals were subsequently administered ca. 104 CFUs via oral gavage. The wasting disease caused by systemic S. e. typhimurium infection was quantified as weight lost during the course of infection. Differences in weight loss were calculated by comparing the percentages of body weight by day among treatment groups, as well as calculating the average weight lost per day from the beginning of weight loss (day 2) to the end of the experiment (day 7). To do this, the slope of the change in body weight by day was used to estimate the average loss of weight per day. Spleen weight and S. e. typhimurium loads in this organ are highly correlated and were both used to estimate systemic invasion. S. e. typhimurium CFUs were enumerated by plating spleen homogenates on MacKonkey Agar plates. No colonies are detected in the faeces or spleens of uninfected animals by this method.
Flow cytometry was performed on an LSR Fortessa instrument (BD Biosciences). Positive cell populations were defined by the use of single stain and/or appropriate isotype controls in all experiments. Representative flow cytometry plots defining cell populations that differed significantly among genotypes are provided in relevant figures. Cells were harvested from tissues as previously described. For surface antibody staining, isolated cells were plated at 500,000 cells per well in a 96-well plate and washed twice using sterile HBSS buffer (CellGro) supplemented with 2.5% FBS (HyClone). Cells were then suspended in 100 μl of buffer containing fluorescently conjugated antibodies (1/250 (v/v) dilution) and placed in the dark at 4 °C for 20 min. Cells were washed twice with buffer to remove unbound antibody, and then analysed. For intracellular staining, cells were surface stained as described, and then fixed overnight. The next day, cells were washed to remove fixative and then stained in 50 μl of 1 × permeabilization buffer (eBiosciences) with fluorescently conjugated antibodies (1/50 (v/v) dilution) in the dark at 4 °C for 30 min. Cells were washed twice with buffer and then analysed. For summary of markers used to distinguish cell subsets and a full list of antibodies, working concentrations, and catalogue numbers used in all flow cytometry experiments refer to Supplementary Table 2.
Microbiota stocks were derived from each MHC congenic genotype by scraping luminal contents from the gut (duodenum to rectum) of donor animals (n=2–3). Contents were diluted in sterile 1 × HBSS to create a 50:50 (v/v) mix. Mixtures were homogenized, vortexed for 20 s, and then spun at 400 × g for 10 min to separate course materials from the microbial suspension. Suspensions were then divided into 500 μl single-use aliquots, flash frozen in liquid nitrogen, and immediately stored at −80 °C until use. For colonization, age-matched germfree BALB/c mice were orally gavaged with 100 μl of a thawed aliquot of appropriate microbiota stock for 7 days. Colonization dynamics typical of this method are shown in Supplementary Fig. 5b.
For analysis of faecal microbial communities, fresh faecal pellets were collected from 8- to 12-week-old animals and then stored at −80 °C until further use. For analysis of mucosally associated communities, the entire gut (duodenum to anus) was collected from an animal. Luminal contents were gently removed with forceps, and then tissues were rinsed in sterile 1 × HBSS. The luminal face of the gut epithelium was scraped using sterile forceps and immediately stored at −80 °C until further use. Three independent sequencing experiments were performed utilizing two different sequencing technologies; Illumina MiSeq and pyrosequencing. Bacterial genomic DNA was extracted using MO BIO PowerSoil DNA Isolation Kit (MO BIO Laboratories) for all experiments. Cage effects (for example, maternal effects, strain isolation effects, stochastic drift and so on) are unavoidable confounding variables in mouse microbiota studies that cannot be fully controlled for. As detailed below for each experiment, we have endeavoured to minimize its contribution to our overall findings by (1) sampling across multiple cages when possible, (2) deriving animals from heterozygous crosses in an attempt to equalize microbiota exposure across genotypes and (3) by performing three independent tests of the hypothesis that MHC antigen presentation sculpts microbiota composition using animals from different mouse facilities, representing different sexes and different genetic models, and by characterizing the effect of MHC genotype between two sites in the gut. Pyrosequencing experiment #1 (reported in Fig. 2a–d): Male animals from a long-term breeding colony of C57BL/6, B2M−/−, and MHCII−/− animals maintained in the Pathology Department at the University of Utah were used to compare faecal communities in Fig. 2a–d. Animals were sampled across multiple cages per genotype. Sequencing runs for these samples were conducted at Baylor College of Medicine in the lab of our collaborator Dr Joseph Petrosino using a previously described pyrosequencing approach targeting the V3–V5 region of the bacterial 16S rRNA gene. The 16S rRNA gene libraries were generated by the Center for Metagenomics and Microbiome Research at Baylor College of Medicine, using the V3–V5 (357F/926R) primer in accordance with standard Human Microbiome Project protocols . The 16S rRNA libraries were sequenced by the Human Genome Sequencing Center at Baylor College of Medicine using a Roche 454 GS FLX+ instrument (Roche, Indianapolis, IN) operated with titanium chemistry. 16S rRNA gene sequences were assigned into OTUs at a similarity cutoff value of 97% using the UPARSE pipeline in QIIME and the SILVA Database. Before analysis all samples were first rarified to a depth of 3,382 sequences per sample. Sequences corresponding to data shown in Fig. 2a–d, Supplementary Fig. 4a, Supplementary Fig. 4b and Supplementary Fig. 4e have been deposited at the NCBI SRA under accession number PRJNA296810. Illumina MiSeq experiment #1 (reported in Fig. 2e–h): Faecal pellets were collected from female MHC homozygote and heterozygote animals that were members of a long-term breeding colony maintained in the Pathology Department at the University of Utah. These animals served as the source animals for all other experiments described in this manuscript. The experiments described in Figs 2e–h and and4a4a represents the results of a single experiment where five animals from each of the three different MHC genotypes had their microbiotas sequenced. In the case of the H2dd (green dots) and H2kk (red dots) animals, the five animals from each genotype were both litter- and cagemates (that is, for each genotype, five female animals from the same litter were housed in the same cage and used for analysis). The H2bb animals (blue dots) represent five female animals derived from three different mothers that were housed in three separate cages. This design provides an internal control for the possibility that drift or maternal effects significantly influences our results by demonstrating that H2bb animals still significantly cluster by genotype when compared with the other two H2 congenic cohorts. Re-deriving homozygote congenic breeder pairs from heterozygote crosses, and using their progeny for microbiota sequencing analysis, minimizes any potentially confounding legacy effect due to the independent husbandry of these strains for several decades (the strain isolation effect). Also, five heterozygous H2bd animals depicted in Fig. 4a were derived from a single breeder pair (H2dd × H2bb cross) that exposes animals to both H2bb and H2dd microbiota (Supplementary Fig. 7). From these animals, we obtained paired-end 300 nucleotide Illumina MiSeq reads from 16S rRNA gene variable regions 3 and 4, then processed and overlapped them using mothur and QIIME as previously described. Open reference 97% similarity OTU picking against the Greengenes 13_8 reference database was employed. Before analysis all samples were rarified to a depth of 16,000 sequences. Sequences corresponding to data shown in Figs 2e–h and and4a,4a, Supplementary Fig. 2c and Supplementary Fig. 6a–d have been deposited at the NCBI SRA under accession number SRP062960. Pyrosequencing experiment #2 (reported in Fig. 2i–l): Faecal and mucosal samples were collected from separately housed male MHC congenic BALB/c animals from a long-term breeding colony housed in the Biology Department at the University of Utah to compare faecal and mucosal communities. Animals were sampled across multiple cages per genotype. Sequencing runs for these samples were conducted at Baylor College of Medicine in the lab of our collaborator Dr Joseph Petrosino. The 16S rDNA V4 region was amplified by PCR and sequenced in the MiSeq platform (Illumina) using a 2 × 250 nucleotide paired-end protocol yielding pair-end reads that overlap almost completely. The read pairs were demultiplexed based on the unique molecular barcodes, and reads were merged using USEARCH v7.0.1001 (ref. ). 16S rRNA gene sequences were assigned into OTUs at a similarity cutoff value of 97% using the UPARSE pipeline in QIIME and the SILVA Database. Abundances were recovered by mapping the demultiplexed reads to the UPARSE OTUs. Before comparison of faecal and mucosal communities in male BALB/c animals, all samples were rarified to a depth of 560 sequences. For independent analysis of the effect of host genotype on faecal and mucosal communities, samples were rarified to a depth of 4,700 and 560 sequences, respectively. Sequences corresponding to data shown in Fig. 2i–l, Supplementary Fig. 4d and Supplementary Fig. 4e have been deposited at the NCBI SRA under accession number PRJNA296810.
Peyer's patches of age-matched MHC congenic animals were collected, placed in 500 μl of Qiazol RNA stabilization solution, and frozen at −80 °C until further processing. After phenol-chloroform extraction of RNA in the aqueous phase, we further cleaned the RNA using Qiagen RNeasy columns. Following DNase digestion and inactivation, the cleaned RNA from each sample was used as template for separate cDNA synthesis reactions for each Ig repertoire (IgA, IgD, IgG and IgM) using gene-specific primers (Supplementary Table 3) designed to target the CH1 region of the heavy chain constant region in BALB/c mice. Primers were designed based on alignments of IMGT reference sequences and degenerate nucleotides were introduced to account for multiple alleles when present or, in the case of IgG, to cover multiple subclasses. Each gene-specific cDNA was then split evenly and used as template among triplicate 25 μl PCR reactions using the same gene-specific primer used in cDNA synthesis as a reverse primer along with a mix of 17 forward primers (mixed based on abundance of their targets) previously identified as targeting BALB/c heavy chain variable (VH) regions. We appended partial Illumina adaptors to the 5′-end of these VH primers (Supplementary Table 3) to serve as targets for primers in a second PCR. PCR using high-fidelity Phusion HotStart II polymerase was performed as follows: 98 °C initial denaturation for 2 min; 26 cycles of 98 °C for 20 s, 60 °C anneal for 20 s, 72 °C extension for 20 s; final single extension at 72 °C for 2 min. Triplicate PCR reactions were then combined and cleaned using Qiagen MinElute columns and eluted with 27 μl. 10 μl of the eluate was used as template for a second 50 μl PCR designed to add the rest of the Illumina adaptors to amplified products with the same reaction conditions as for the first PCR, except with only 16 cycles. Reverse primers targeting the constant heavy region for the second PCR were internal to the set used in the cDNA synthesis and first PCR, and had full Illumina adaptors with eight-nucleotide indexes appended to their 5′-end. Forward primers in the second PCR target the partial Illumina adaptors added to the VH region during the first PCR and add the rest of the Illumina adaptors, as well as another eight-nucleotide index sequence (Supplementary Table 3). Final indexed reactions were subsequently cleaned with Qiagen MinElute columns, mixed evenly and submitted for sequencing on the Illumina MiSeq with paired-end 300 nucleotide reads. Sequences were initially processed to create a single, primer-trimmed, high-quality contig from overlapping paired-end sequences using the mothur command make.contigs and subsequently screened for sequences that did not have at least 20 nucleotides overlapping. To reduce the number of sequences analysed, we then used the OTU picking method within QIIME to cluster identical sequences and create biom table of counts of Ig sequences by samples for each repertoire. A representative sequence for each cluster of identical sequences was then submitted for analysis by IMGT/HighV-QUEST. The resulting tables of Ig sequence characteristics (for example, V-region identity, CDR3 sequences and nucleotide substitution rates reported in Fig. 1) were then parsed and added to biom tables as observation metadata to facilitate downstream analyses. For all analyses each repertoire was rarefied to an even depth of sequences per sample.
Percentages of Ig-bound bacteria was measured in the faeces of animals. Briefly, fresh faecal pellets were collected and homogenized in 500 μl of sterile 1 × HBSS buffer. Faecal suspensions were spun and 400 × g for 5 min to precipitate course materials, and supernatants containing suspended bacteria were placed in new 1.5 ml Eppendorf tubes. Tubes were spun for 10 min at 8,000 × g to pellet bacteria and supernatants containing unbound antibodies were discarded. To eliminate unbound antibodies further, bacterial pellets were washed by re-suspending them in 1 ml of sterile 1 × HBSS, spinning at 8,000 × g for 5 min, and then discarding supernatants. Two washes were performed. An antibody against mouse IgA (Southern Biotech, PE-conjugated rat anti-mouse IgA, cat#1165-09L) was used to stain for IgA-bound bacteria. Antibody was diluted 1/500 in sterile 1 × HBSS containing 10% FBS as a blocking agent. Antibody stains were pipetted at 100 μl volumes into a round-bottomed 96-well plate. 5 μl of faecal bacteria suspensions were added to appropriate wells, mixed by pipetting, and then incubated in the dark at 4 °C for 30 min. Antibody stains were then removed by spinning plates at 4,000 r.p.m. and flicking off stain. Bacteria were washed twice by suspending in 200 μl of sterile 1 × HBSS, spinning at 4,000 r.p.m., and flicking off supernatants. Bacteria were then suspended in 250 μl of sterile 1 × HBSS containing 5 μl of 1 × SYBR green I stain (Molecular Probes). Bacteria were incubated in the dark at 4 °C for 20 min before enumeration on a flow cytometer. RAG1−/− faecal bacteria samples were included in all experiments as negative controls.
The concentration of IgA in faecal pellets were quantified using an IgA-specific ELISA kit (eBioscience: Mouse IgA Ready-SET-go kit (cat#88-90450-88)) following kit protocols. All concentration estimates are standardized by faecal weight and depicted as concentration per gram of faeces.
All quantitative PCR reactions were conducted in 12.5 μl volumes using the SYBR green Master Mix (Roche). Quantitative PCR experiments were conducted on a Lightcycler LC480 instrument (Roche). Template quantity and quality was assessed using a Nanodrop spectrophotometer. Abundance estimates are standardized to the concentration of input DNA per reaction and are represented as copies per nanogram of faecal DNA. Template extraction for quantification of faecal bacteria loads: DNA was extracted from fresh faecal pellets using the PowerFecal DNA Isolation Kit (Mo Bio) following kit instructions. Bacterial loads were quantified using previously validated bacterial group-specific 16S primers (Supplementary Table 3).
Statistics were carried out using JMP9.0 (SAS), Prism 6.0 (Graphpad) and R software. permutational analysis of variance was used for hypothesis testing of significance between groups shown in PcoA plots. A multivariate analysis of variance was used in Fig. 3g to test for a significant difference in the effect of colonization with different microbiotas on the abundance of immune cell parameters (a sum model was employed).
A two-tailed unpaired Student's t-test was used for all other pairwise statistical comparisons unless otherwise noted. A Welch's correction was applied when a Levine's test revealed unequal variance in otherwise normally distributed data sets. A nonparametric Mann–Whitney U-test was used on highly skewed data sets (that is, Salmonella CFU data). Error bars in all figures represent ±s.d. with the exception of Figures depicting trends in weight loss where error bars represent ±s.e.m. Before data analysis outliers were identified and excluded using the ROUT method in PRISM 6.0. A minimum sample size of five mice were used for all experiments. Sample size, #replicates, statistical test and estimates of dispersion are reported in all figure captions.
How to cite this article: Kubinak, J. L. et al. MHC variation sculpts individualized microbial communities that control susceptibility to enteric infection. Nat. Commun. 6:8642 doi: 10.1038/ncomms9642 (2015).
Supplementary Figures 1-7 and Supplementary Tables 1-6
We would like to thank members of the Round and O'Connell labs for their critical review of the manuscript. We thank James Marvin for assistance with flow sorting. The flow cytometry core is supported by the National Center for Research Resources of the National Institutes of Health under award number 1S10RR026802-01. Some of the germfree mice used in this publication were provided from UNC's Gnotobiotic Facility which is supported by grants 5-P39-DK034987 AND 5-P40-OD010995. J.L.K. and C.P. are supported by a T32 fellowship in microbial pathogenesis (AI-055434). R.S. was supported by a T32 genetics training grant (GM007464). R.M.O. is supported by the NIH New Innovator Award DP2GM111099-01, the NHLBI R00HL102228-05, an American Cancer Society Research Grant and a Kimmel Scholar Award. Support for this project comes from the Edward Mallinckrodt Jr. Foundation, Pew Scholars Program, NSF CAREER award (IOS-1253278), Packard Fellowship in Science and Engineering and NIAID K22 (AI95375) and NIAID (AI107090, AI109122) and an NIH Directors Innovator award DP2AT008746-01 to J.L.R.
Author contributions J.L.K. and J.L.R. designed experiments. J.L.K. performed all experiments. J.L.K. analysed all data with assistance from W.Z.S. during analysis of microbiota and Ig sequencing data. N.J.A. and J.F.P. performed pyrosequencing experiments and assisted in microbiota community analysis. R.S. and C.P. assisted in immune phenotyping experiments, and R.S. assisted during experimental Salmonella infections. J.L.K. and J.L.R. wrote the first draft of manuscript. W.Z.S., R.S., C.P., N.J.A, J.F.P, T.C., L.G., W.K.P, P.E.J., R.M.O. assisted in manuscript preparation. L.M. assisted in mouse genotyping. J.L.R. supervised the project.
All animals are populated by microbes, and, contrary to popular belief, most microbes appear highly beneficial to their hosts. They are critical in animal nutrition and immune defense, and they can serve as important catalysts for the effective development and functioning of host tissues. It also is becoming increasingly clear that they can contribute to host behavior. It has been hypothesized that one way they do so is by producing the components of chemical signals that animals use to communicate. We tested and confirmed first predictions of this hypothesis in hyenas, demonstrating that the bacterial and odor profiles of hyena scent secretions covaried and that both profiles varied with characteristics of hyenas known to be communicated through their chemical signals.
All animals harbor beneficial microbes. One way these microbes can benefit their animal hosts is by increasing the diversity and efficacy of communication signals available to the hosts. The fermentation hypothesis for mammalian chemical communication posits that bacteria in the scent glands of mammals generate odorous metabolites used by their hosts for communication and that variation in host chemical signals is a product of underlying variation in the bacterial communities inhabiting the scent glands. An effective test of this hypothesis would require accurate surveys of the bacterial communities in mammals’ scent glands and complementary data on the odorant profiles of scent secretions—both of which have been historically lacking. Here we use next-generation sequencing to survey deeply the bacterial communities in the scent glands of wild spotted and striped hyenas. We show that these communities are dominated by fermentative bacteria and that the structures of these communities covary with the volatile fatty acid profiles of scent secretions in both hyena species. The bacterial and volatile fatty acid profiles of secretions differ between spotted and striped hyenas, and both profiles vary with sex and reproductive state among spotted hyenas within a single social group. Our results strongly support the fermentation hypothesis for chemical communication, suggesting that symbiotic bacteria underlie species-specific odors in both spotted and striped hyenas and further underlie sex and reproductive state-specific odors among spotted hyenas. We anticipate that the fermentation hypothesis for chemical communication will prove broadly applicable among scent-marking mammals as others use the technical and analytical approaches used here.
Every animal is populated by communities of microbes that can profoundly affect its biology, often in beneficial ways (, ). Indeed, symbiotic microbes are critical contributors to animal nutrition and immune health, and they serve as important catalysts for the effective development and functioning of animal tissues and neural circuitry (–). It also is becoming apparent that symbiotic microbes can extend host behavioral phenotypes in beneficial ways, including facilitating their feeding, antipredator, reproductive, and communicative behaviors (6, ).
An effective communication system is a critical component of an animal’s behavioral repertoire, and one way in which symbiotic microbes might contribute to their hosts’ behavioral phenotypes is by increasing the diversity and/or efficacy of the signals available to them (6, ). Most animals communicate to some extent via chemical means, and mammals in particular often rely on odorous secretions from integumental scent glands to signal conspecifics (–10). These glands occupy myriad locations on mammals’ bodies and are typically warm, moist, nutrient-rich, and largely anaerobic. As such, they are conducive to the proliferation of symbiotic, particularly fermentative, bacteria (10). The fermentation hypothesis for mammalian chemical communication posits that as bacteria ferment or otherwise metabolize the nutrient-rich substrates in these glands, they generate odorous metabolites that subsequently are used by their hosts to communicate with conspecifics (6, 10–). The hypothesis further suggests that variation in chemical signals among mammals with specialized scent glands results largely from an underlying variation in the odor-producing bacterial communities within these glands. If this hypothesis is true, then (i) mammalian scent gland secretions should contain fermentative, odor-producing bacteria, (ii) the bacterial and odor profiles of secretions should covary, and (iii) these profiles should vary with the host characteristics being signaled, such as species identity, group membership, sex, or reproductive state (6).
Effective testing of these predictions of the fermentation hypothesis requires accurate surveys of the bacterial communities in the scent gland secretions of mammals as well as complementary data on the odorant profiles of these secretions. Historically, technical limitations of cultivation-based surveys and, to a lesser extent, molecular fingerprinting surveys, of symbiotic bacteria have impeded our ability to test these predictions effectively because these approaches often underestimate the actual diversity in bacterial communities (6, ). As a consequence, evaluations of the hypothesis typically have concluded that the bacterial diversity in integumental scent gland secretions is insufficient to underlie the observed diversity of chemical signals (6, ). In a recent study, we used next-generation sequencing to thoroughly survey the bacterial communities in the scent gland secretions of adult female spotted hyenas, Crocuta crocuta (). That study revealed more types of bacteria than the 15 previous surveys of specialized mammalian scent glands combined and demonstrated that most of these bacteria were members of fermentative, odor-producing clades. It also revealed that the bacterial communities in scent secretions varied among hyena social groups, suggesting that their diversity was sufficient to explain social group-specific odors in spotted hyenas (16). Although that study afforded support for the fermentation hypothesis, it did not include complementary data on the odor profiles of hyena scent gland secretions—data needed to evaluate the hypothesis effectively—and the scope of host traits considered was limited. Here we concurrently analyze the bacterial and odor profiles of scent gland secretions collected from wild spotted and striped hyenas, Hyaena hyaena, in Kenya (Fig. S1) to determine whether the two profiles covary in each species and to ascertain the extent to which the two profiles vary with hyena species, sex, and, in the spotted hyena, female reproductive state.
The lifestyles of spotted and striped hyenas differ greatly. Spotted hyenas—found throughout sub-Saharan Africa—live in large, hierarchically structured groups, called clans, that typically contain 40–80 individuals (17). Clans include multiple breeding males and multiple overlapping generations of females, and adult members cooperatively maintain and defend their group’s territory against neighboring clans (18). To mediate the complex social relationships within and among clans, spotted hyenas use a rich repertoire of tactile, visual, vocal, and chemical signaling behaviors (19, 20). In contrast, striped hyenas—found in North, West, and East Africa—live in small groups containing one or two reproductively mature females and one or more adult males (21). Although the home ranges of group members overlap considerably, striped hyenas usually rest, travel, and forage alone; therefore they seldom interact directly with groupmates (21). Little is known of striped hyena communicative behavior, especially in natural populations, but striped hyenas appear to have a very modest vocal signaling repertoire, with no long-distance vocalizations (22). Therefore, among striped hyenas, chemical signaling likely serves a prominent role in territorial behavior and potentially in reproduction as well.
Despite their very different lifestyles, spotted and striped hyenas both commonly exhibit a conspicuous chemical signaling behavior called “pasting,” a form of scent marking in which a hyena deposits an odorous secretion, called “paste,” from its subcaudal scent pouch on a grass stalk (20, 22). The major volatile constituents in paste are volatile fatty acids (VFAs), esters, hydrocarbons, alcohols, and aldehydes (23, 24). Previous investigations have shown that the odors of spotted hyena pastes vary with individual identity, group membership, sex, and, potentially, female reproductive state (16, 24, ). Effects of striped hyena traits on paste odors have not yet been investigated.
This study of mammalian scent marking marries data from in-depth, next-generation bacterial surveys with targeted odor analyses of scent secretions from natural populations. We show that the bacterial communities in hyena pastes are dominated by fermentative bacteria and that the structures of these communities covary with the VFA profiles of pastes. Furthermore, we show that the bacterial and VFA profiles of paste differ between spotted and striped hyenas and that, among spotted hyenas in the same social group, both profiles vary with hyena sex and reproductive state. As such, this study illustrates that the diversity of symbiotic bacterial communities in paste appears sufficient to underlie chemical signaling of host traits in hyenas and affords strong empirical support for the fermentation hypothesis for chemical communication.
Scanning electron micrographs confirmed that the scent glands of spotted and striped hyenas were inhabited by symbiotic (i.e., resident) microbes (Fig. S2). Subsequent 16S rRNA gene surveys using an operational taxonomic unit (OTU) definition of 97% homologous nucleotide base similarity revealed that the bacterial communities in spotted and striped hyena pastes were markedly different but that each was dominated by fermentative anaerobes.
The structure of bacterial communities in adult spotted hyena pastes differed markedly from those in the pastes of adult striped hyenas (Fig. 1 A and B and Dataset S1 A and C) sampled either in Laikipia [analyses of similarity (ANOSIM), R = 1.0, P = 0.0001] or Shompole (R = 1.0, P = 0.0001). There was much greater variation in the structure of bacterial communities among the pastes of spotted than striped hyenas [permutational tests of multivariate dispersions (PERMDISP); Masai Mara National Reserve (MMNR) (0.387 ± 0.092) vs. Laikipia (0.131 ± 0.023), P = 0.0001; MMNR vs. Shompole (0.144 ± 0.030), P = 0.0001; Fig. 1A)]. The paste bacterial communities of spotted hyenas also were more OTU-rich (Chao1 index; tunequalvar. = 2.55, P = 0.018; Table S1). Last, membership in paste bacterial communities differed between the two hyena species (Jaccard index, ANOSIM; MMNR vs. Laikipia: R = 1.0, P = 0.0001; MMNR vs. Shompole: R = 1.0, P = 0.0001; Fig. 1B and Table S2) (, ). The great majority of bacteria (>95% of sequences) in the pastes of both hyena species were members of the order Clostridiales—fermentative anaerobes—in the phylum Firmicutes (28). Less than 1% of Clostridiales sequences in striped hyena pastes were assigned to previously characterized genera (Table S2). In contrast, the Clostridiales in spotted hyena pastes were primarily members of the genera Anaerococcus, Clostridium, Fastidiosipila, Finegoldia, Murdochiella, Peptoniphilus, and Tissierella. Spotted hyena pastes also consistently contained fermentative genera outside the phylum Firmicutes, such as Corynebacterium, Propionibacterium (both Actinobacteria), Porphyromonas (Bacteroidetes), and Fusobacterium (Fusobacteria) (28–33). At an OTU level, only 11 of 461 OTUs were shared between the pastes of adult spotted hyenas and those of adult striped hyenas from at least one population. Two were prominent (i.e., among the top 15 based on average sequence abundance) members of bacterial communities in striped hyena paste (OTUs 0003 and 0180), but none were prominent in spotted hyena pastes.
Differences in the bacterial (OTU) and VFA profiles of the pastes of adult spotted and striped hyenas. (A) A nonmetric multidimensional scaling (nMDS) plot showing a difference in structure (Bray–Curtis index) between the paste bacterial communities of adult spotted hyenas from the MMNR and adult striped hyenas from Laikipia and Shompole. (B) A Clearcut cladogram () of the prominent (i.e., top 15 based on average abundance) OTUs in the pastes of hyenas from the three populations and an accompanying heat map reflecting the mean abundances (out of 1,600 sequences) of these OTUs in pastes. These data were log-transformed before plotting (values in parentheses). Order- and genus-level classifications of OTUs, as determined by the Ribosomal Database Project’s Classifier (), are noted also. (C) An nMDS plot showing a difference in structure between the paste VFA profiles of spotted and striped hyenas. (D) A heat map of the mean percent abundances of VFAs in the pastes of hyenas from the three populations. Sample sizes were 19, 13, and 9 for MMNR, Laikipia, and Shompole, respectively.
Importantly, microbial biogeography alone cannot explain the observed differences in the membership of the paste bacterial communities in adult spotted and striped hyenas. First, although they were not present in the pastes of adult striped hyenas, 13 of the 15 top OTUs in the pastes of MMNR spotted hyenas were present in the pastes of at least one juvenile striped hyena from Laikipia or Shompole, indicating that these OTUs were not restricted geographically to the MMNR. Second, the bacterial communities in the pastes of two spotted hyenas sampled serendipitously in Shompole clustered with bacterial communities from MMNR spotted hyenas rather than with those from sympatric striped hyenas (Fig. S3).
In addition to containing disparate bacterial communities, the pastes of adult spotted and striped hyenas had markedly differently VFA profiles (ANOSIM; MMNR vs. Laikipia: R = 0.770, P = 0.0001; MMNR vs. Shompole: R = 0.727, P = 0.0001; Fig. 1 C and D). The VFA profiles of spotted hyena pastes differed from those of striped hyenas in Laikipia and Shompole in nearly identical ways (Dataset S1 B and C), again indicating that biogeography is not a primary factor in species differences. The pastes of spotted hyenas had higher percentages of acetic, propanoic, butanoic, pentanoic, and hexanoic acids than those of striped hyenas, whereas the pastes of striped hyenas had much higher percentages of isopentanoic acid (Fig. 1D and Dataset S1 B and C). As with bacterial profiles, there was much greater variation in the VFA profiles of the pastes of spotted than striped hyenas (PERMDISP; MMNR (0.268 ± 0.106) vs. Laikipia (0.114 ± 0.029): P = 0.0001; MMNR vs. Shompole (0.096 ± 0.060): P = 0.0001; Fig. 1C).
The bacterial and VFA profiles of paste covaried among spotted hyenas from the general MMNR population (Mantel test, R = 0.437, P = 0.0002), and the Talek clan (R = 0.565, P = 0.0001). There also was strong covariance between the OTU–VFA correlation matrices of MMNR and Talek pastes (correlation r, Mantel test, R = 0.721, P = 0.0003; analysis of 10 shared prominent OTUs; Fig. S4), indicating that the abundance of specific OTUs correlated with the relative abundance of specific VFAs in similar ways at population and clan levels.
At the population level, there was a tendency for the bacterial profiles of paste to differ between male and female spotted hyenas (ANOSIM, R = 0.113, P = 0.0792), but there was not a consistent effect of sex on the VFA profiles of paste (R = −0.011, P = 0.4427). Still, the bacterial and VFA profiles of paste did covary among both MMNR males (Mantel test, R = 0.393, P = 0.0397) and females (R = 0.400, P = 0.0231).
Among members of the Talek clan, there were pronounced effects of both sex and female reproductive state on the bacterial profiles of paste (Fig. 2 A and B, Table 1, and Dataset S1 D and F). The bacterial communities in the pastes of Talek males, lactating females, and pregnant females contained similar numbers of OTUs (Chao1 index, ANOVA, F2,12 = 2.033, P = 0.1600; Table S1). However, bacterial communities in the pastes of males and lactating females were more even than those associated with pregnant females (Simpson index, ANOVA, F2,18 = 8.876, P = 0.0021; male vs. pregnant female: Tukey’s test, Q = 5.90, P = 0.0017; lactating female vs. pregnant female: Q = 3.67, P = 0.046; Table S1), largely because the bacterial communities in the pastes of pregnant females were dominated by members of OTU 0001 (Dataset S1 D and F). Overall, there were 343 OTUs in the pastes of Talek hyenas; 120 OTUs were exclusive to Talek males, 54 were exclusive to lactating females, and 46 were exclusive to pregnant females. Few exclusive OTUs were widespread among their respective host class, and those that were widespread were minor members of their communities. In general, Talek males, lactating females, and pregnant females shared the prominent members of their paste bacterial communities, but the relative abundances of these members varied with host sex and reproductive state (Fig. 2B, Table 1, and Dataset S1 D and F).
Variation in the bacterial (OTU) and VFA profiles of the pastes of immigrant male, lactating female, and pregnant female spotted hyenas in the Talek clan. (A) An nMDS plot showing variation in the structure (Bray–Curtis index) of paste bacterial communities among Talek clan members. (B) A heat map of the mean abundances (out of 1,600 sequences) of the prominent (i.e., top 15 based on average abundance) OTUs in the pastes of Talek hyenas. These data were log-transformed before plotting (values in parentheses). (C) An nMDS plot showing variation in the structure of paste VFA profiles in Talek hyenas. (D) A heat map of the mean percent abundances of VFAs in the pastes of Talek hyenas. Seven hyenas were sampled from each reproductive class.
Nonparametric multivariate analyses of variance confirming that the bacterial (OTU) and VFA profiles of the pastes of Talek immigrant males, lactating females, and pregnant females vary
Composition of paste OTU profiles (Jaccard index) | |
Global effect | F = 2.468, P = 0.0001 |
Male vs.lactating female | F = 2.485, P = 0.0011 |
Male vs.pregnant female | F = 3.070, P = 0.0020 |
Lactating vs. pregnant female | F = 1.757, P = 0.0177 |
Structure of paste OTU profiles (Bray–Curtis index) | |
Global effect | F = 4.181, P = 0.0001 |
Male vs.lactating female | F = 3.789, P = 0.0006 |
Male vs.pregnant female | F = 6.023, P = 0.0007 |
Lactating vs. pregnant female | F = 2.411, P = 0.0398 |
Structure of paste VFA profiles (Bray–Curtis index) | |
Global effect | F = 16.23, P = 0.0001 |
Male vs.lactating female | F = 5.779, P = 0.0116 |
Male vs.pregnant female | F = 43.81, P = 0.0008 |
Lactating vs. pregnant female | F = 9.209, P = 0.0069 |
Seven hyenas were sampled from each reproductive class.
The VFA profiles of paste also varied among Talek males, lactating females, and pregnant females (Fig. 2 C and D, Table 1, and Dataset S1 E and F). The pastes of females, especially pregnant ones, had higher percentages of pentanoic, hexanoic, and heptanoic acids, whereas the pastes of Talek males had higher percentages of acetic, propanoic, isobutanoic, and butanoic acids. Notably, the structures of the bacterial and VFA profiles of paste covaried among Talek’s lactating (Mantel test, R = 0.7641, P = 0.0001) and pregnant (R = 0.3768, P = 0.0468) females, and tended to covary among Talek’s males (R = 0.3077, P = 0.0766).
The bacterial and VFA profiles of striped hyena pastes covaried (Mantel test, R = 0.7697, P = 0.0001), and this covariance was evident among pastes from both Laikipia (R = 0.7176, P = 0.0028) and Shompole (R = 0.7952, P = 0.0002). Furthermore, the OTU–VFA correlation matrices for Laikipia and Shompole covaried (correlation r, Mantel test, R = 0.2341, P = 0.0461; Fig. S4), indicating that the abundance of specific prominent OTUs correlated with the percent abundance of specific VFAs in similar ways in the pastes of striped hyenas from the two different populations.
The structure of paste bacterial communities differed between the Laikipia and Shompole populations (ANOSIM, R = 0.5716, P = 0.0001; Fig. 3 A and B and Dataset S1 G and H). Variation in bacterial community structure was very low within each population, especially among adults (Fig. 3A and Dataset S1H). Membership of paste bacterial communities also differed between the Laikipia and Shompole populations (Jaccard index, ANOSIM, R = 0.5747, P = 0.0001). There were 443 OTUs in the pastes of striped hyenas; 59 were found exclusively in the Laikipia population, and 314 were found exclusively in the Shompole population. However, very few of the exclusive OTUs (2/59, 4/314) were found in more than half of the pastes from hyenas in the respective populations. In general, striped hyenas in Laikipia and Shompole shared the prominent members of their paste bacterial communities, but the relative abundances of these members differed between the two populations (Fig. 3B and Dataset S1 G and H). As a potential consequence, there was a tendency for paste VFA profiles to differ between the Laikipia and Shompole populations as well (ANOSIM, R = 0.1091, P = 0.0508; Fig. 3 C and D, Dataset S1 G and H).
Differences in the bacterial (OTU) and VFA profiles of the pastes of striped hyenas in Laikipia and Shompole. (A) An nMDS plot showing a difference between the structure (Bray–Curtis index) of paste bacterial communities in the Laikipia and Shompole populations. Lighter shading denotes samples obtained from juveniles. (B) A heat map of the mean abundances (out of 1,600 sequences) of the prominent (i.e., top 15 based on average abundance) OTUs in the pastes of striped hyenas. These data were log-transformed before plotting (values in parentheses). (C) An nMDS plot of the structures of paste VFA profiles in Laikipia and Shompole. (D) A heat map of the mean percent abundances of VFAs in the pastes of striped hyenas from the two populations. Twenty striped hyenas were sampled from Laikipia (8 male/12 female), and 13 (six male/ seven female) were sampled from Shompole.
Controlled for source population, there was a modest effect of sex on the structure of bacterial communities in the pastes of striped hyenas (two-way ANOSIM; population: R = 0.5563, P = 0.0001; sex: R = 0.1170, P = 0.0382) but not on paste VFA profiles (population: R = 0.0942, P = 0.0933; sex: R = 0.0281, P = 0.2712). However, when controlled for source population, the bacterial and VFA profiles of pastes did covary among male and female striped hyenas (partial Mantel test; male: R = 0.8446, P = 0.0001; female: R = 0.7397, P = 0.0002).
We tested first predictions of the fermentation hypothesis for chemical communication in hyenas and showed that (i) hyena pastes contained fermentative bacteria, (ii) the bacterial and VFA profiles of hyena pastes covaried, and (iii) these profiles differed between hyena species and, within a spotted hyena clan, they varied with sex and reproductive state.
The bacteria in hyena pastes belong to clades of fermentative bacteria whose metabolisms yield varying amounts of acetic, propanoic, isobutanoic, butanoic, isopentanoic, isohexanoic, and hexanoic acids (28–33). These differences suggest that variation in the structure of paste bacterial communities should result in variation in the structure of paste VFA profiles. Here we found that the structures of the bacterial and VFA profiles of hyena pastes strongly covaried. Prior studies using microbial fingerprinting surveys revealed that the bacterial and odor profiles of human axillae covaried among people who adhered to presampling guidelines on bathing and deodorant use (34) and that the bacterial and odor profiles of the urine-marks of laboratory mouse strains, Mus domesticus, were partially correlated (). The current study focused on wild populations of mammals, and showed strong, consistent covariance between the bacterial and odor profiles of mammalian scent gland secretions. The study also demonstrated that the two profiles covaried not only across but also within species, sex, and, in the spotted hyena, reproductive classes.
There were robust effects of species identity on the bacterial and VFA profiles of hyena paste. Spotted and striped hyena pastes contained each of the VFAs studied here, but they were present in markedly different proportions. The two hyenas are sympatric in areas of West and East Africa, including Kenya (36), so this finding is consistent with it being advantageous for sympatric taxa to distinguish readily between the homologous signals of conspecifics and heterospecifics (). Interestingly, the structures of spotted and striped hyena paste VFA profiles also differed in their degree of intraspecific variation, with variation among spotted hyena paste VFA profiles far exceeding the variation among striped hyena pastes. Given that spotted hyenas are highly social and striped hyenas are largely solitary, this finding is consistent with the social complexity hypothesis for animal signaling, which posits that frequent interactions in various contexts with many different individuals results in the evolution of more complex signaling systems (, ).
At the clan level, there was a pronounced effect of sex on both the bacterial and VFA profiles of spotted hyena paste. A prior study of a captive colony of spotted hyenas showed that they discriminated readily between the pastes of males and females in the colony, indicating that paste odor profiles, on a local scale, were sex specific (). A different, population-level study of spotted hyenas in the Serengeti National Park, Tanzania, did not find evidence that sex affected paste odor profiles (24). We also did not find a consistent effect of sex on the bacterial or VFA profiles of spotted hyena paste at the population level. The bacterial and odor profiles of spotted hyena pastes are generally clan specific (, 16). The purported mechanism for clan-specific paste odors is that clan members develop more homogeneous paste bacterial communities than the general population through cross-infection fostered by consistently overmarking the same pasting sites (, , 16, 40). Indeed, in this study, the structures of bacterial communities in the pastes of spotted hyena males and lactating females were more variable in the general MMNR population than in the Talek clan [PERMDISP; MMNR (0.387 ± 0.092) vs. Talek (0.292 ± 0.059), P = 0.0017]. Collectively, these data suggest that social interactions at the clan level (e.g., sex-specific patterns in overmarking) may confound population-level analyses. Among striped hyenas, there was an effect of sex on the bacterial but not the VFA profiles of paste. This finding suggests that, in the striped hyena, sex may not be communicated through paste or that it is communicated via volatiles other than those studied here.
The reproductive state of female spotted hyenas substantially affected the structures of paste bacterial and VFA profiles. Pregnancy can dramatically alter mammalian oral, vaginal, and gut bacterial communities (–). Here we show that pregnancy can alter the microbiota of specialized signaling organs as well. During pregnancy, spotted hyena females have elevated levels of testosterone and estrogen (, ). It is well established that steroid hormones are present in mammalian sebaceous and apocrine glands (i.e., the machinery of scent glands) and that they affect gland morphology, production, and chemistry (6, –). It appears they also may affect the structures of bacterial communities in these glands in ways that could signal the reproductive state of female hosts effectively.
This study provides strong empirical support for the fermentation hypothesis for chemical communication in hyenas. However, two further predictions need to be tested. First, if symbiotic bacteria are the source of paste VFAs, then, if provided with appropriate growth conditions, paste cultivars should produce, to varying degrees, the VFAs studied here. Alternatively, their genomes should contain genes coding for the fermentation pathways leading to the production of these VFAs. Second, if the variation we found in paste VFA profiles has signaling relevance, then hyenas should discriminate among synthetic mixtures of these volatiles representing samples from this study. Importantly, we explicitly tested the fermentation hypothesis for chemical communication because it was proposed nearly 40 y ago to explain the scent marking systems of mammals (11, ), and many of the volatile components of hyena paste are known products of bacterial fermentation (23, 24). However, extrapolated to a general symbiotic hypothesis for animal chemical communication, the fermentation hypothesis accommodates other odorous microbial metabolites (e.g., longer-chain fatty acids and their esters), symbiotic microbes (e.g., fungi and archaea), signaled host characteristics (e.g., genotype, health, and social status), and animal classes. In fact, the explanatory potential of this hypothesis is limited only by the capacity of hosts’ social and physiological circumstances to alter the structure of their symbiotic microbial communities in ways that consequently affect hosts’ odor profiles in signaling-relevant ways (6). Therefore, evaluating the potential of the symbiotic hypothesis for animal chemical communication will be a critical step in elucidating the contributions of symbiotic microbes to animal behavior.
Pastes were collected directly from the subcaudal scent pouches of anesthetized (SI Methods) spotted hyenas in the MMNR (1994–2008) and striped hyenas in the Laikipia District (2001–2003) and at the Shompole and Olkirimatian Group Ranches (2007–2009; hereafter, “Shompole”), Kenya (Fig. S1). Paste samples were placed in sterile cryogenic vials, stored in liquid nitrogen, and transported to Michigan State University, where they remained frozen at −80 °C until their bacterial and VFA profiles were determined (Table S3).
The 40 sampled spotted hyenas resided in the north-central region of the MMNR. They represented the general MMNR population (nine males/10 females from >10 clans) as well as a single, intensively studied clan (seven males/seven lactating females/seven pregnant females from the Talek clan). Only one individual, CNLF428, appeared in our prior study (). Females from the general MMNR population were lactating when sampled. Talek lactating females did not give birth in the 150 d after they were sampled and therefore were, to the best of our knowledge, not pregnant [110 d mean gestation time (19)]. Talek pregnancies were confirmed via ultrasound imaging of fetuses and/or by the female giving birth within 80 d of being sampled. Juvenile spotted hyenas were not included in this study because they do not consistently produce appreciable amounts of paste (40). Paste samples were serendipitously obtained from two spotted hyenas in Shompole. These samples were large enough for us to characterize their bacterial but not their VFA profiles, so they were included only in a single supplemental analysis (Fig. S3). The 33 sampled striped hyenas represented the general populations of the north-central Laikipia District (eight males/12 females) and Shompole (six males/seven females). They included both adults (22) and juveniles (11) and constituted all the individual striped hyenas for which paste samples were sufficiently large for bacterial and VFA analysis. Reproductive data were not available for adult female striped hyenas.
DNA was extracted from paste sample aliquots (∼0.05 g) using a MO BIO UltraClean fecal DNA kit. Bacterial 16S rRNA genes in extractions were PCR amplified (SI Methods) using two broadly conserved, degenerate primers targeting the V6–V4 variable regions of the 16S gene (1046R: 5′–CGACRRCCATGCANCACCT–3′; 518F: 5′–CCAGCAGCYGCGGTAAN–3′). Nucleotide sequencing was performed on 454 GS FLX Titanium and GS Junior instruments at the Marine Biological Laboratory in Woods Hole, MA, and at Michigan State University. Postsequencing, 454 run files were processed using mothur software (v. 1.27.0; SI Methods) (). Sequences were binned into OTUs based on a 97% sequence similarity. Each paste bacterial community then was iteratively subsampled 15 times to the depth of the least-represented sample (1,600 sequences), and the mean abundances of individual OTUs per sample were calculated and rounded to the nearest whole number. The final data set contained 865 OTUs (471 singletons). Representative sequences of these OTUs and associated metadata are available in GenBank (accession nos. KC705471-KC706325) and Dataset S2.
Branched and linear VFAs were extracted from aliquots (0.025 g) of paste using methyl tert-butyl ether as solvent. Specifically, we targeted acetic, propanoic, isobutanoic, butanoic, isopentanoic, pentanoic, isohexanoic, hexanoic, and heptanoic acids. Samples were analyzed using an Agilent Technologies’ 6890N/5973 inert GC/MS system equipped with a 30-m DB-wax column (250-µm inner diameter × 0.25-µm film thickness). One microliter of the sample was injected using an Agilent 7683 auto-injector. Two blanks consisting of the solvent mixture spiked with 5 µL 85% (wt/vol) formic acid were included between each sample to prevent any carry-over. The percent peak abundances of VFAs for each sample were determined using QuanLynx software (Micromass). Detailed protocols for VFA extraction and GC/MS analyses, including the use of internal and external standards, are provided in SI Methods.
Before analyses, OTU abundance data were log10 (x +1) transformed to temper the contributions of highly prominent OTUs to quantitative similarity index calculations (). Quantitative (i.e., structural) similarities of the OTU or VFA profiles of paste samples were characterized using the Bray–Curtis similarity index (51). Unless otherwise noted, multivariate statistical tests were based on the Bray–Curtis index. Qualitative (i.e., compositional) similarities of the OTU profiles of pastes were characterized using the Jaccard similarity index (51). A detailed discussion of graphical and statistical data analysis is provided in SI Methods.
Our research, described in Animal Research Protocol Institutional Animal Care and Use Committee (IACUC) 05/11–110-00, was approved most recently on June 15, 2012 by the IACUC at Michigan State University and complies with Kenyan Law.
We thank the Kenyan Ministry of Education, Science and Technology, the Kenyan Wildlife Service, the Narok County Council, the Senior Warden of the Masai Mara National Reserve, the Loisaba, Mpala, and Kisima Ranches (Laikipia), and the Shompole and Olkirimatian Group Ranches for permissions and support of this research. We thank Christine Drea (Duke University), Hilary Morrison (Marine Biological Laboratory), and Dan Jones, Scott Smith, Abby Vanderberg, Kylie Farrell, and Alex Schmidt (Michigan State University) for contributing their technical expertise. This study was funded by National Science Foundation and BEACON Center for the Study of Evolution in Action Grants IOS0920505, IOS0819437, IOS0618022, IOS1121474, and DBI0939454.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Data deposition: The sequences reported in this paper have been deposited in the GenBank database (accession nos. KC705471–KC706325).
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1306477110/-/DCSupplemental.