Genetic and environmental control of host-gut microbiota interactions

1Studies carried out over the last decade have revealed that gut microbiota contribute to a variety of common disorders, including obesity and diabetes, colitis, atherosclerosis, rheumatoid arthritis, and cancer. The evidence for metabolic interactions is particularly strong, as a large body of data now supports the conclusion that gut microbiota influence the energy harvest from dietary components, particularly complex carbohydrates, and that metabolites such as the short-chain fatty acids produced by gut bacteria can perturb metabolic traits, including adiposity and insulin resistance. Gut microbiota communities are assembled each generation, influenced by maternal seeding, environmental factors, host genetics, and age, resulting in substantial variations in composition among individuals in human populations. Most experimental studies of host-gut microbiota interactions have employed large perturbations, such as comparisons of germ-free versus conventional mice, and the significance of common variations in gut microbiota composition for disease susceptibility is still poorly understood. Furthermore, while studies with germ-free mice have clearly implicated microbiota in clinically relevant traits, it has proven difficult to identify the responsible taxa of bacteria.

2We now report a population-based analysis of host-gut microbiota interactions in the mouse. One of the issues we explore is the role of host genetics. Although some evidence is consistent with significant heritability of gut microbiota composition, the extent to which the host controls microbiota composition under controlled environmental conditions is unclear. We also examined the role of common variations in gut microbiota in metabolic traits such as obesity and insulin resistance and mapped loci contributing to the abundance of certain microbiota. We performed our study using a resource termed the Hybrid Mouse Diversity Panel (HMDP), consisting of about 100 inbred strains of mice that have been either sequenced or subjected to high-density genotyping. The resource has several advantages for genetic analysis as compared to traditional genetic crosses. First, it allows high-resolution mapping by association rather than linkage analysis, and it has now been used for the identification of a number of novel genes underlying complex traits. Second, since the strains are permanent, the data from separate studies can be integrated, allowing the development of large, publicly available databases of physiological and molecular traits relevant to a variety of clinical disorders (systems.genetics.ucla.edu and phenome.jax.org). Third, the panel is ideal for examining gene-by-environment interactions, since it is possible to examine individuals of a particular genotype under a variety of conditions.

3We determined the composition and variability of gut microbiota in a total of 599 mice from 113 HMDP strains, of which 327 were male and 273 were female (average three mice per strain) (Supplemental Table 1). All mice in the study were bred for two or more generations in the same facility at UCLA, and each strain was maintained in separate cages (two to four mice of each strain). The mice were maintained on a chow diet (6% kcal from fat) until 8 wk of age, and then placed on a high-fat, high-sucrose (HF/HS) diet for anadditional 8 wk. We performed multiplex 16S rRNA sequencing of the V4 amplicon using the Illumina MiSeq platform. On average, 23,048 reads were obtained per sample (ranging from 6331 to 82,238). Reads were binned into individual samples based on barcode sequence, and complementary taxon-based analysis methods were used to compare 16S rRNA sequences across the cecum microbial communities. The relative abundances of phylum, class, order, family, and genus were determined for the 599 mice. We focused on abundant microbes, defined as operational taxonomic units (OTUs) with at least 0.01% relative abundance across all samples (total 439 OTUs).

4We previously showed that changing a chow to a HF/HS diet drastically changed microbiota composition across HMDP strains and that these shifts were strongly dependent on the genetic background of the mice. After HF/HS diet feeding, we identified 49 genera, where the 17 most abundant genera were present in at least 75% of the samples (n = 599). These seventeen most abundant genera accounted for 68% of reads and included members of the six phyla (Supplemental Table 2). Consistent with previous findings in both mice and humans, the most abundant phyla in the gut were Firmicutes (49.8% ± 10.9) and Bacteroidetes (41.8% ± 9.6). Compared to the chow diet, the HF/HS diet resulted in increased Firmicutes and decreased Bacteroidetes, consistent with previous studies.

5Microbiota composition varied greatly across the 110 strains of mice (Fig. 1A; Supplemental Table 2). For instance, the relative abundance of the Firmicutes across all the strains ranged between 20% and 82%. Even larger variations were observed at finer taxonomic levels; for example, a common mucus layer inhabitant in gut, Akkermansia muciniphila (A. muciniphila), varied in abundance from 0.005% to 40% across the strains of mice. In contrast to human data, we were not able to detect any members of hydrogen-consuming methanogens, although not all methanogens, which are Archaea, would be expected to amplify with bacterial 16S RNA gene primers.

6In examining individual mice housed in separate cages (to avoid cage effects since mice are coprophagic), we found that microbiota compositions were much more similar within strains than between strains (P < 0.001 for unweighted and weighted UniFrac) (Fig. 1B; Supplemental Fig. 1). However, because the members of an inbred strain share a recent common ancestor, it is unclear to what extent the shared microbiota result from parent to offspring transfer of microbiota as compared to host genetic factors. The standard ways to estimate heritability, defined in an outbred population as the proportion of the phenotypic variance contributed by the genetic variance, are to examine pedigrees or compare monozygotic with di-zygotic twins. In mice, this is traditionally done using inter-crosses between strains differing in traits of interest. Because of the problem of maternal seeding, we estimated heritability by a different method, based on the proportion of phenotype variance accounted for by genetic relationships among the strains.

7All of the HMDP strains have been either sequenced or densely genotyped (http://www.jax.org), allowing us to determine their genetic relatedness. Based on this SNP-based approach (rather than a family-based approach), we were able to estimate the heritability of the abundance of the major taxa of gut microbiota (Supplemental Table 3). For the calculation, we utilized a linear mixed model and assumed additive effects (see Methods). The assumption behind the linear mixed model approach is that the covariance of the genetic component of the phenotypic data is proportional to the kinship matrix or genetic similarity matrix between the animals. In this model, each individual mouse microbiome composition (relative abundance of each taxa) is affected by a genetic random effect, which is correlated across mice by virtue of sharing some of the genetic variants affecting microbiome, and an environmental random effect, which is uncorrelated across mice.

8When maintained under controlled conditions, host genetic variation appears to explain a substantial amount of the variation in gut microbiota composition in the HMDP, up to 0.5 or more for many common taxa (Table 1; Supplemental Table 3). The range of heritabilities of microbiota was similar for phyla, families, and genera, and for males and females (Supplemental Table 3) and approached the range of heritabilities we observed for measured clinical phenotypes (Supplemental Table 4).

9Our approach for estimating the heritability of gut microbiota composition is potentially confounded by the complication of physical transmission. However, since most of the inbred strains have been separated for many decades, it seems unlikely that a particular composition would be maintained over so many generations if it was due largely to physical “seeding.” A second possible caveat in this analysis is that multiple animals from the same strain were, in some cases, housed in the same cage, and they may share similar microbiota compositions due to physical transfer rather than host genotype. We rule out this confounding factor by performing the same analysis using only one animal per strain (Supplemental Table 3). We expected lower heritability than in the complete cohort because of reduced total genetic relatedness and power. However, even in the reduced sample, while our estimates are lower than the complete data set, host genotype still accounts for a substantial fraction of the variation.

10Among the HMDP strains examined were striking differences in response to the HF/HS diet. Some strains showed as much as a sixfold increase in body fat whereas others showed no significant change, and food consumption was only modestly associated with the gain in body weight (r2 = 0.30) or the gain in fat mass (r2 = 0.04). Likewise, HOMA-IR, a measure of insulin resistance, showed over a 50-fold range among the strains. We hypothesized that the composition of the gut microbiota might contribute to this variation.

11To test the hypothesis, we performed neonatal cross-fostering experiments between two strains, A×B19 and SWR, exhibiting diverse responses after HF/HS feeding. Figure 2A shows the response of the two strains to the diet in terms of body fat increase during 8 wk of HF/HS diet feeding. The A×B19 strain gained ∼24 g fat in response to the diet while strain SWR gained ∼4 g (Fig. 2A; Supplemental Fig. 2A). We cross-fostered newborn SWR mice with A×B19 dams and observed that the gut microbiota composition of the cross-fostered pups resembled A×B19 mice rather than SWR mice, indicating efficient transfer (Fig. 2B). At 4 wk of age, we placed the pups on the HF/HS diet and monitored fat gain. In a pilot study (Supplemental Fig. 2B), both cross-fostered male and female mice initially exhibited increased weight gain as compared to SWR mice. While the male mice continued to show an increased response up to 8 wk on the diet, the female mice became similar to SWR mice after 8 wk. We then repeated the study with a larger group of cross-fostered mice (n = 8–11 per group). Again, the male cross-fostered mice showed significantly more weight gain and higher body fat composition (P < 0.01) as compared to SWR control mice (Fig. 2C,D). In addition, cross-fostered male SWR mice also showed higher levels of plasma triglyceride compared to SWR mice (Fig. 2E). The female mice, on the other hand, did not exhibit a significant increase in body fat at 8 wk of age (Supplemental Fig. 2C). After 8 wk of the HF/HS diet, the microbiota composition of the cross-fostered SWR pups moved back toward SWR microbiota composition, supporting the role of host genotype in microbiota community structure (Fig. 2F). We conclude that common variations in the composition of the gut microbiome contribute in part to the response to a HF/HS diet.

12To identify which bacteria contribute to obesity and metabolic phenotypes, we sought to identify potential relationships between metabolic traits and the gut microbiota. Altogether, we identified many correlations, including some novel and some known relationships (Supplemental Table 5). Several of these appear to be significant, based on a false discovery rate of <0.01.

13Two taxa from the family Lachnospiracea, Roseburia spp. and Ruminococcus gnavus, were positively associated with obesity and metabolic traits including body fat increase on a HF/HS diet, insulin levels, and HOMA-IR (P < 0.001). The same traits were also associated with an unknown species of Lactobacillus (Supplemental Table 5). Our data are consistent with a recent study, showing that increased abundance of Roseburia spp. in obese subjects is positively correlated with body mass index and inflammation, and Lactobacillus reuteri has been previously linked to increased obesity in humans. A. muciniphila was inversely correlated with body fat (r = −0.15; P = 9.02 × 10−4) and insulin levels (r = −0.20; P = 4.57 × 10−6) (Supplemental Table 5). A. muciniphila is a mucin-degrading, gram-negative anaerobe residing in intestinal mucus layers that has been associated with obesity and insulin resistance in humans and mice.

14To test causality of the relationship, we administrated live or heat-killed A. muciniphila to obesity-prone A×B19 male mice (Supplemental Fig. 3A). Ten-week-old male A×B19 mice were treated five times per week with A. muciniphila by oral gavage at a dose of 1.44 × 109 cfu/0.2 mL (HF/HS-Akk), while control mice were treated with an oral gavage of an equivalent volume of heat-killed A. muciniphila (HF/HS). After 1 wk of gavage, all mice were put on HF/HS diet for four additional weeks. After 5 wk of gavage, we observed that mice given A. muciniphila showed significantly improved metabolic parameters. Figure 3A shows that body weight and total body fat, including all fat depots examined, were significantly reduced in A. muciniphila-treated mice. Plasma lipid levels showed substantial decreases in total cholesterol and triglycerides (Fig. 3B). Most striking were the effects on insulin resistance, with dramatically decreased levels of both glucose and insulin (Fig. 3B). Our data are consistent with the correlations observed among the HMDP as well as recent findings.

15In addition to metabolic changes, the administration of A. muciniphila altered the gut microbiota composition (Fig. 3C). Thus, using both chow and HF/HS diets, A. muciniphila-treated mice clustered separately from mice that received heat-inactivated bacteria (Supplemental Fig. 3A,B). This difference was evident at the phylum level, showing significant shifts between two dominant phyla, Bacteoridetes and Firmicutes (Supplemental Fig. 3C). Surprisingly, we did not observe significant differences in the total abundance of A. muciniphila, perhaps because DNA from heat-killed bacteria was also present.

16Next, we aimed to obtain evidence for specific interactions between gut microbiota and host genetics. Rather than using linkage analysis, as is traditional in mouse genetics, we employed association across the HMDP strains, since the resolution of mapping is one or two orders improved. Such association analysis has now been used to identify novel genes which were subsequently validated in a number of cases, but given the structure of the inbred mouse population, there is some potential for long-range linkage disequilibrium. The proportion of each common taxon was treated as an individual trait, and association analyses were performed with 198,431 informative SNPs spaced throughout the mouse genome using a mixed-model algorithm that corrects for population structure. The threshold for genome-wide significance was based on simulation and permutations as previously described. Altogether, seven genome-wide significant loci (P < 4 × 10−6) were found to be associated with common genera (Fig. 4; Supplemental Table 6). Loci ranged from 800 kb to 3 Mb in size and in most cases contained several genes within a linkage disequilibrium block. The majority of these significant associations were detected with members of the classes Clostridia (Lachospiraceae, Ruminococcacea, and Bacilli), and most exhibited similar associations in both sexes (Supplemental Table 6). In order to test whether the GWAS results were inflated by the effect that multiple animals from the same strain were housed in the same cage, we performed GWAS using only one animal per strain. Even with the reduced sample size, we were able to detect GWAS associations in the same regions (Supplemental Table 6), albeit with reduced significance. We have carried out expression profiling of adipose and liver of the HMDP strains when maintained on HF/HS diets and used the data to identify cis expression quantitative trait loci (eQTL). These provide a useful means of prioritizing candidate genes at the relevant loci since they provide evidence of functional variation. The significant cis-eQTLs at each of the GWAS loci are shown in Supplemental Figure 5 and Supplemental Table 7. Here, we focus on those genera that show strong correlation with clinical traits, as discussed above; additional loci are described in detail in the Supplemental Material.

17For Roseburia spp., we identified significant associations spanning 2.6 Mb on Chromosome 15 (Fig. 4A; Supplemental Table 6). The same region showed a significant association with subcutaneous fat mass on a HF/HS diet (P < 10−7) (Supplemental Fig. 4A), a clinical trait that is also positively correlated with the abundance of Roseburia spp. (r = 0.25, P = 3.9 × 10−10) (Table 2). Global gene expression in epididymal adipose tissue and liver showed a significant cis-eQTL between the peak SNP (rs31843241) and transcript levels of the Kif21a, Lrrk2, and Irak4 genes (Table 2; Fig. 5A; Supplemental Fig. 5; Supplemental Table 7). The expression of Irak4, a gene involved in the initiation of the innate immune response, was correlated with the abundance of Roseburia spp. and HOMA-IR, suggesting a causal relationship (Fig. 5B,C).

18Ruminococcus gnavus exhibited genome-wide significant association to a locus on Chromosome 19 (Fig. 4B; Supplemental Table 6). The peak SNP (rs30796836, P = 8.37 × 10−7) has a significant cis-eQTL with the transcript levels of the Osbp (oxysterol binding protein) gene in adipose tissue of HF/HS-fed mice (females: P = 1.51 × 10−12; males: P = 2.54 × 10−10), and the abundance of Ruminococcus gnavus negatively correlated with expression of Osbp (r = −0.36; P = 0.00014) (Table 2; Supplemental Fig. 5; Supplemental Table 7).

19Finally, we detected significant associations for the abundance of A. muciniphila on Chromosomes 7 (rs33129247; P = 2.59 × 10−6) and 2 (rs27323290, P = 6.67 × 10−6) (Figs. 4C, 6A; Supplemental Table 6). The peak SNP on Chromosome 7 (rs33129247) was also associated with triglyceride levels (P = 6.47 × 10−9) and gonadal fat (P = 7.44 × 10−7) (Table 2; Fig. 6B; Supplemental Fig. 4B). Strong candidates for this locus are Igf1r and Nr2f2 genes, since both have been shown to play a role in glucose and insulin regulation. The Chromosome 2 locus contains Ctnnbl1, a gene implicated in obesity and two interesting candidates, bactericidal/permeability-increasing protein (Bpi) and lipopolysaccharide binding protein (Lpb). Ctnnbl1 also showed a significant association with food intake (P = 1.17 × 10−9) and total weight after 8 wk on the HF/HS diet (P = 5.8 × 10−8) (Table 2). The Lbp and Ctnnbl1 genes both have significant cis-eQTLs in adipose and liver (Table 2; Fig. 6C; Supplemental Fig. 5; Supplemental Table 7) and are associated with body fat percentage increase and insulin levels (Fig. 6D,E).

20We previously showed that inbred strains of mice differ dramatically in their response to a high-fat, high-sucrose diet. Based on the large body of evidence indicating that gut microbiota can influence metabolic traits, we hypothesized that the dietary response was dictated in part by differences in gut microbiota. We showed that different inbred strains differ strikingly in the composition of gut microbiota and provided evidence that the variation is determined in part by the host genetic background. Consistent with our hypothesis, we showed that cross-fostering between two strains of mice affected dietary response to the high-fat, high-sucrose diet. By correlating microbiota composition with dietary response among the HMDP inbred strains, we were able to identify several candidate microbiota influencing dietary response. We chose one of these, A. muciniphila, to examine using gavage with the cultured microbe and observed striking effects on weight gain, adiposity, plasma lipids, and insulin resistance. Finally, to help identify novel host-microbiota interactions, we have mapped loci controlling certain microbiota taxa. We discuss each of these findings in turn below.

21Experimental studies have shown that the host genetic background can influence gut microbiota composition. For example, mice with mutations affecting inflammatory signaling or diabetes differ in microbiota composition from their wild-type littermates. However, the importance of common genetic variants in contributing to the composition of the gut microbiota is unclear. While twin pairs and related individuals share gut microbiota to a greater extent than unrelated individuals, early studies did not find a statistically significant difference in gut microbiota sharing in monozygotic (MZ) and dizygotic (DZ) twins. However a recent study with 416 twin pairs demonstrated that MZ twins have greater overall microbial community similarities than DZ twin pairs and identified several microbial taxa with relative abundances that differ depending on host genetics. Studies using genetic crosses of mice, where the environment can be controlled, also suggest that host genetics can significantly alter gut microbiota composition.

22Heritability represents the fraction of variation that is attributable to genetic variation and is a relative value that depends on the environment and the degree to which the population varies. Traditionally, heritability has been estimated using pedigrees in outbred populations or by comparing monozygotic versus di-zygotic twins. In mice, heritability is generally estimated by analyzing genetic crosses, but for studies of gut microbiota, this is confounded by the fact that there is physical transmission of “seed” microbiota from generation to generation. To circumvent this problem, we used a SNP-based approach to determine relatedness rather than a family-based approach. All of the inbred strains constituting the HMDP are separated from one another by many generations and thus are unlikely to share microbiota as a result of physical transmission. Our results indicate a high degree of heritability of the major groups of microbiota in mice, ranging from ∼0.3 to >0.5, although we note certain caveats in our approach (see Results). This high heritability presumably results from the fine tuning of a symbiotic relationship that has co-evolved for millions of years. Among the likely contributing factors are differences in immunoglobulin and antibacterial molecules secreted into the gut lumen, differences in the mucosal gut structure, and differences in bile acid metabolism.

23Previous studies have shown that certain large effect mutations affecting inflammatory signaling or metabolic traits can significantly affect microbiota composition and that, in some cases, these can be transmitted by transplantation of gut microbiota from such mice. Here, we have examined whether common variations in gut microbiota are also causally involved in metabolic traits. To test this possibility, we chose two inbred strains, A×B19 and SWR, that differ strikingly in the response to a HF/HS diet for cross-fostering studies. When male SWR mice cross-fostered by A×B19 dams were subjected to the HF/HS diet, they gained ∼8% in total body fat as compared to ∼2% for SWR, while A×B19 gained ∼25% (Fig. 2C). Thus, while the majority of the response was dictated by the host genetic background, the gut microbiota did contribute significantly. Human gut microbiota exhibit more diversity and quantitative variation than what we have observed among common mouse strains, suggesting that a significant fraction of variance in obesity and insulin resistance in human populations is due to microbiota composition. Conceivably, this could explain some of the “missing heritability” observed in GWAS. We used correlation analysis to identify candidate microbiota contributing to the response to the HF/HS diet. Several genera were found to be strongly associated with traits such as body fat, plasma lipids, and insulin resistance, including some, such as Roseburia spp. and A. muciniphila, that have previously been implicated in metabolic traits. We chose A. muciniphila, a mucin-degrading, gram-negative anaerobe, to examine its potential response to our dietary challenge. We administered by oral gavage either live or heat-killed bacteria to strain A×B19 male mice for 1 wk and then began the HF/HS challenge, continuing to administer the live or heat-killed bacteria for a total of 5 wk. Significant differences were observed between the groups in body fat gain, plasma lipid levels, and insulin resistance. We noted significant correlations between the abundances of certain taxa across the panel of strains (data not shown), providing information about microbiota community interactions.

24Finally, we used the HMDP to perform high-resolution mapping of loci contributing to microbiota abundance. Using association analysis, we identified seven significant loci for five out of 17 common genus level taxa. Most of the loci were observed in both males and females, supporting the conclusion that they are true positives. The loci contain a number of strong candidate genes based on the literature, functional variants, and correlations with clinical and molecular traits (see Supplemental Material). The Chromosome 15 locus for Roseburia spp. contains the Irak4 gene, which is involved in signaling innate immune responses from Toll-like receptors. Mice deficient in Irak4 expression are more susceptible to viral and bacterial infections, and Irak4 has previously been associated with gut microbiota composition in a subset of BxD RI strains. The A. muciniphila locus in Chromosome 2 contains two closely linked genes, Bpi and Lbp. Both bind to bacterial lipopolysaccharide (LPS) and elicit immune responses by presenting LPS to CD14 and TLR4 and signaling the acute-phase immunological response. Bpi acts as an endogenous antibiotic protein with potent killing activity against gram-negative bacteria.

25Our data constitute a resource for the further dissection of mechanistic host-gut microbiota interactions. We have identified a number of highly significant associations between gut microbiota and clinical traits, and the loci reported here provide a means of identifying novel host factors controlling gut microbiota abundances.

26All mice were obtained from The Jackson Laboratory and were bred at UCLA for two or more generations for use in this study. Briefly, until 8 wk of age mice were maintained on a chow diet (Ralson Purina Company) and then placed on a high-fat, high-sucrose diet (Research Diets D12266B) for an additional 8 wk. Samples were obtained from the cecum of 599 mice from 113 strains, with an average of six mice per strain (327 males and 297 females) (Supplemental Table 1). Mice from different strains and genders were housed in separate cages but in the same room throughout the study. Cecum and fecal samples were snap-frozen with liquid nitrogen and stored at −80°C. The animal protocol for the study was approved by the Institutional Care and Use Committee (IACUC) at the University of California, Los Angeles.

27Microbial DNA was extracted using the PowerSoil DNA Isolation Kit (MO BIO Laboratories). Amplification and sequencing of the V4 hypervariable region of the 16S rRNA gene was performed using the validated, region-specific bacterial primers 515F and 806R according to previously described methods optimized for the Illumina MiSeq platform. The reverse amplification primer contained a 12-bp Golay error-correcting barcode sequence, and amplicons were generated in triplicate using 5 Prime Hot MasterMix (Fischer Scientific). The PCR conditions consisted of an initial denaturation step of 94°C for 3 min; 35 cycles of 94°C for 45 sec, 50°C for 30 sec, and 72°C for 90 sec, followed by 72°C for 5 min. Replicate amplicons were quantified with a Quant-iT PicoGreen dsDNA Assay kit (Life Technologies), pooled (200 ng from 96 samples), and purified using with the UltraClean PCR Clean-up kit (MO BIO Laboratories). High-throughput sequencing analysis of bacterial rRNA genes was performed on the purified, pooled sample using the Illumina MiSeq platform (Illumina).

28De-multiplexing 16S rRNA gene sequences, quality control, and operational taxonomic unit binning were performed using the open source pipeline Quantitative Insights Into Microbial Ecology (QIIME) version 1.7.0. The total number of sequencing reads was 13,805,813 (an average of 23,048 reads per sample) with an average length of 153-bp reads. Sequences were binned into OTUs based on 97% identity using UCLUST against the Greengenes reference database. Each sample's sequences were rarefied to 7000 reads to reduce the effect of sequencing depth. Seven samples were omitted from further analysis due to insufficient sequence coverage, yielding 592 samples.

29Microbial composition at each taxonomic level was defined using the summarize_taxa function in QIIME. Prior to genome-wide association analysis, taxa at any taxon present in fewer than 75% of samples was discarded, yielding a total of 43 common taxa from different taxonomic levels (three phyla, five classes, six orders, 12 families, and 17 genus level taxa). The relative abundance of each taxon was calculated by dividing the sequences pertaining to a specific taxon by the total number of bacterial sequences for that sample.

30Heritability was estimated using a linear mixed with the EMMAX software model. In this approach, the phenotypes (in this case, the relative abundances) are assumed to be generated by genetic and environmental components. The assumption behind the linear mixed model approach is that the covariance of the genetic component of the phenotypic data is proportional to the kinship or genetic similarity matrix between the animals. The analysis provides estimates of σg2 and σe2, the variances corresponding to the genetic and environmental component, respectively. The heritability is then the fraction of the variance accounted for by the genetics or and is computed for each relative abundance. We note that the kinship matrix must be standardized for these estimates to be consistent with the classical definition of heritability. A standardized kinship matrix has a mean along the diagonal of 1 and a sum of 0.

31Body composition, food intake, and blood and plasma assays were as previously described. Briefly, mice were measured for total body fat mass and lean mass using nuclear magnetic resonance (NMR) using the Bruker minispec with software from Eco Medical Systems. Blood was collected from mice following fasting for 4–5 h, and plasma was isolated by centrifugation in Microtainer tubes with EDTA (Becton Dickinson). Plasma glucose levels were measured using a Beckman Glucose Analyzer 2 (Beckman Instruments). Plasma total cholesterol, HDL cholesterol, free cholesterol, triglycerides, and free fatty acid concentrations were determined by enzymatic assays employing colorimetric endpoints as described previously. Insulin levels and HOMA-IR were determined as previously described.

32Within 24 h of birth, the pups from SWR females were removed from birthing cages and placed with A×B19/PgnJ mothers. Pups were weaned on postnatal day 21, and at 8 wk of age mice were placed on a high-fat/high-sucrose diet (Research Diets D12266B) for an additional 8 wk. Controls from both strains were fostered with different mothers from the same strain.

33A. muciniphila (ATCC BAA-835) was grown in a Columbia broth medium supplemented with 0.05 % hog gastric Mucin type III (Sigma) under anaerobic conditions. Cells were harvested in late logarithmic phase by centrifugation at 6000 rpm in a tabletop Fisher centrifuge at room temperature and resuspended in 0.05 volume sterile anaerobic PBS containing 25% glycerol to a concentration of 7.2 × 1010 per mL prior to storage at −80°C. For gavage, anaerobic cell suspensions were diluted 10-fold in PBS. The sterile anaerobic PBS (pH 7) was supplemented with 0.05% cysteine HCl, degassed with N2, sealed in serum bottles with butyl rubber stoppers under anaerobic conditions provided by a gas phase of 1.8 atm N2/CO2 (80:20, vol/vol).

34A. muciniphila was administrated to 10-wk-old A×B19 male mice (n = 5 per group), housed in groups of 2–3 mice per cage, by oral gavage at a dose 1.44 × 109 cfu/0.2 mL suspended in sterile anaerobic PBS (HF-Akk). Treatment was 5 d per week for 5 wk. Control groups were treated with an equivalent volume of heat-inactivated A. muciniphila. After the first week of A. muciniphila treatment, all mice were placed on the HF/HS diet for an additional 4 wk.

35Association analyses of taxa were performed using the Factored Spectrally Transformed Linear Mixed Models (FaST-LMM) algorithm adjusting for population structure and using gender as a covariate. To achieve a normal distribution, the sequence counts for each taxonomic bin were Arcsine-transformed. A total of 198,431 informative SNPs (minor allele frequency > 5%; missing genotype rate < 10%) spaced throughout the genome were used. For genome-wide significance, we used a P value threshold < 4 × 10−6, based on permutation and simulation and which roughly corresponds to a Bonferroni correction. In some cases, multiple animals from the same strain were housed together, and in order to rule out the possibility that GWAS results were inflated because strains shared similar microbiota compositions due to physical transfer, we also performed GWAS choosing a single sample from each strain at random (n = 113). Linkage disequilibrium (http://pngu.mgh.harvard.edu/~purcell/plink/) boundaries were determined by calculating SNP correlations and visualizing r2 > 0.8 in Haploview.

36To help identify candidate genes at loci associated with taxa abundances, we carried out global expression analysis of epididymal adipose and liver tissue in male and female mice (16 wk old) fed a HF/HS diet as described. Isolated RNA (two mice per strain) was analyzed for global gene expression using Affymetrix HT_MG-430A arrays and filtered as described. Microarray data are available in the Genome Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE64770. The loci controlling transcript levels were mapped with FaST-LMM and are referred to as expression quantitative trait loci. Loci are defined as cis if the peak SNP mapped within 1 Mb of gene position (P value threshold < 1.4 × 10−3).

37All correlations involving bacterial relative abundance were performed using biweight midcorrelation, which is robust to outliers. The statistical cutoff of P = 0.1 after false discovery rate (FDR) correction for multiple comparisons was used to define statistical significance for correlations. Statistical analyses were performed using GraphPad Prism. Data are expressed as mean ± SEM, and significance was set at a two-tailed P value < 0.05.

3816S rRNA sequencing data generated for this study have been submitted to the NCBI Sequence Read Archive (SRA; http://www.ncbi.nlm.nih.gov/sra/) under accession number SRP059760. The summary tables for both genders are posted on our website (systems.genetics.ucla.edu/data/hmdp2) with the link called “Download high-fat microbiota vs. clinical trait correlation table” and can also be found in Supplemental Material (Supplemental Table 8).