Ethnicity was associated with marked differences in the human gut microbiota. Whereas there were no differences between ethnicities in estimated bacterial content per gram of stool (Fig. S1A), principal coordinates analysis of PhILR Euclidean distances from 16S-seq data (Table S2, n = 22 EA, 24 W subjects) revealed a subtle but significant separation between the gut microbiotas of EA and W subjects (p = 0.037, R2 = 0.037, ADONIS; Fig. 1A). Statistical significance was robust to the distance metric used (Table S5). Bacterial diversity, evaluated using three distinct metrics from our 16S-seq data, were all significantly higher in W individuals, including Faith’s phylogenetic diversity, ASV richness, and Shannon diversity (Fig. 1B). The Firmicutes and Bacteroidetes phyla were enriched in EA subjects, Verrucomicrobia were enriched in W subjects, and Actinobacteria and Proteobacteria were comparable (Fig. 1C). By contrast, analysis at the genus and ASV level did not reveal any differentially abundant groups between ethnicities, suggesting that the phylum-level trends we observed require the integration of more subtle shifts across multiple component members.
Next, we validated these results by metagenomic sequencing (Table S2, n = 21 EA, 24 W subjects). Consistent with our 16S-seq analysis, we detected a subtle but significant difference in the gut microbiomes between ethnicities based upon metagenomic species abundances (p = 0.025, R2 = 0.038, ADONIS) and gene families (p = 0.024, R2 = 0.040, ADONIS). Visualization of species within each phylum revealed marked variation in the magnitude and direction of change between ethnicities in our metagenomic (Fig. 1D,E) and 16S-seq data (Fig. S1B,C). Notably, W individuals had higher levels of Akkermansia muciniphila (Fig. 1D,E and Fig. S1B,C), which has been implicated in protection from obesity and its associated metabolic diseases [76].
Next, we used NMR-based stool metabolomics to gain insight into the potential functional consequences of ethnicity-associated differences in the human gut microbiome (Table S2, n = 10 subjects/ethnicity). Metabolite profiles were more strongly associated with ethnicity (p = 0.008, R2 = 0.128, ADONIS; Fig. 2A) than community structure (R2 = 0.033–0.048, ADONIS; Table S5) or gene abundance (p = 0.024, R2 = 0.040, ADONIS). Feature annotations revealed elevated levels of the branched chain amino acid (BCAA) valine and the short-chain fatty acids (SCFAs) acetate and propionate in EA subjects (Fig. 2B and Table S6). In contrast, proline, formate, alanine, xanthine, and hypoxanthine were found at higher levels in W subjects (Fig. 2B). To assess the statistical significance and reproducibility of these trends, we used targeted GC-MS and UPLC-MS/MS to quantify a panel of BCAAs, SCFAs, and bile acids (Table S7). Confirming our NMR data, EA subjects had significantly higher levels of stool acetate (Fig. 2C) and propionate (Fig. 2D); however, we did not detect any significant differences in BCAAs or bile acids (Fig. S2). Isobutyrate (which was not detected by NMR) was also significantly higher in EA subjects (Fig. 2E). In agreement with these metabolite levels, a targeted re-analysis of our metagenomic data revealed a significant enrichment in two SCFA-related pathways: “pyruvate fermentation to butanoate” (p = 0.023, fold-difference = 2.216) and “superpathway of Clostridium acetobutylicum acidogenic fermentation” (p = 0.023, fold-difference = 2.182).
Consistent with prior studies [22, 23, 77], we found that gut bacterial richness in W individuals was significantly associated with both BMI (Fig. 3A) and body fat percentage (Fig. 3B). Remarkably, these associations were undetectable in EA subjects (Figs. 3A,B) even when other metrics of bacterial diversity were used (Fig. S3). Re-analysis of our data separating lean and obese individuals revealed that the previously observed differences between ethnic groups were driven by lean individuals. Lean W subjects had significantly higher bacterial diversity (Fig. 3C), in addition to greater differences in both gut microbial community structure (p = 0.001, R2 = 0.096, ADONIS; Fig. 3D) and metabolite profiles (p = 0.006, R2 = 0.293, ADONIS; Fig. 3E), than did corresponding EA individuals. By contrast, obese W vs. EA individuals were not different across any of these metrics (Figs. 3C-E). Lean EA individuals were significantly enriched for the Actinobacteria and Firmicutes phyla with a trend towards increased Bacteroidetes (Fig. 3F). At the genus level, lean EA subjects had higher levels of Bacteroides, Blautia, and an unclassified Lachnospiraceae taxon (Figs. 3G,H). In contrast, the Verrucomicrobia phylum (which contains A. muciniphila) was consistently enriched in both lean and obese W subjects relative to EA individuals (Fig. 3F).
Next, we sought to understand the potential drivers of differences in the gut microbiome between ethnic groups in lean individuals within the IDEO cohort, focusing on birth location, time spent in the USA, dietary intake, and host metabolic phenotypes. Although everyone in the cohort was recruited from the San Francisco Bay Area, birth location varied widely (Fig. S4). There was no significant difference in the proportion of subjects born in the USA between ethnicities (75% W, 54.5% EA; p = 0.15, Pearson’s χ2 test). There was also no significant difference in the geographical distance between birth location and San Francisco [W median 2,318 (2.2-6,906) miles; EA median 1,986 (2.2-6,906) miles; p = 0.69, Wilcoxon rank-sum test) or the amount of time spent in the San Francisco Bay Area at the time of sampling [W median 270 (8.00-741) months; EA median 282.5 (8.50–777) years; p = 0.42, Wilcoxon rank-sum test). While obese subjects were markedly distinct from lean individuals of both ethnicities with regard to measured metabolic and laboratory parameters (Table S1), there were no statistically significant differences between ethnic groups after separating lean and obese individuals (Table S1).
Surprisingly, we did not detect any significant differences in either short- (Table S8) or long-term (Table S9) dietary intake between ethnicities. Consistent with this, Procrustes analysis did not reveal any significant associations between dietary intake and gut microbial community structure: procrustes p = 0.452 (DHQIII) and p = 0.445 (ASA24) relative to PhILR transformed 16S-seq ASV data. The Spearman Mantel statistic was also non-significant [r = 0.09511, p = 0.094 (DHQIII) and r = 0.02953, p = 0.313 (ASA24)], relative to PhILR transformed 16S-seq ASV data. Despite the lack of a strong overall shift in the gut microbiota, we were able to identify 11 ASVs associated with dietary intake in lean W individuals (Fig. S5). In contrast, there were no significant ASV-level associations in lean EA subjects.
Given the marked variation in the gut microbiome at the continental scale [1–3], we hypothesized that the observed differences in lean EA and W individuals may be influenced by a participant’s current address at the time of sampling. Consistent with this hypothesis, we found clear trends in ethnic group composition across ZIP codes in the IDEO cohort (Figs. 4A,B) that were mirrored by the 2018 US census data (Pearson r = 0.52, p = 0.026 for neighborhoods with greater than 50% white subjects; Fig. 4D). Obese individuals from both ethnicities and lean W subjects tended to live closer to the center of San Francisco relative to lean EA subjects (Fig. 4C). Distance between current ZIP code and the center of San Francisco was associated with both gut microbial diversity (Fig. 4E) and community structure (Fig. 4F). These analyses were robust to the central point used, as shown using the Bay Bridge as the central reference point (Fig. S6).
Taken together, our results support the hypothesis that there are stable ethnicity-associated signatures within the gut microbiota of lean EA vs. W individuals that are independent of diet. To experimentally test this hypothesis, we transplanted the gut microbiota of a representative lean W and lean EA individual into germ-free C57BL/6J mice (Fig. 5A and Table S4). Despite maintaining the genetically identical recipient mice on the same autoclaved low-fat, high plant-polysaccharide (LFPP) diet, we detected significant differences in gut microbial community structure (Fig. 5B), bacterial richness (Fig. 5C), and taxonomic abundance (Figs. 5D-E) between the two ethnicity-specific recipient groups, reflecting the differences we saw when initially analyzing the stool samples of the human participants themselves. A replication experiment using an independent pair of donors revealed similar trends, though they did not reach statistical significance (Fig. S7). To assess the reproducibility of these findings across multiple donors and in the context of a distinctive dietary pressure, we next fed 20 germ-free mice a high-fat, high-sugar (HFHS) diet for 4 weeks prior to colonization with microbiota from a W vs. EA donor, and then maintained the mice on this diet following colonization (per group n = 10 mice, 5 donors; per donor n = 2 mice, 1 cage; Fig. 5F). Remarkably, this experiment replicated our original findings on the LFPP diet, including altered gut microbial community structure (Fig. 5G), increased richness in mice receiving W donor microbiota (Fig. 5H), and higher levels of Bacteroides in mice receiving EA donor gut microbiota (Figs. 5I-J).
Moreover, mice transplanted with gut microbiomes of EA and W individuals displayed significant differences in body composition. Mice that received W donor microbiota and were fed the LFPP diet gained more body weight (Fig. 6A) and increased their adiposity (Fig. 6B), in conjunction with a reduction in lean mass (Fig. 6C), relative to mice that received the EA donor microbiota. These overall trends were mirrored in the human microbiota recipient mice that were fed the HFHS diet (Figs. 6E-G), though they did not reach statistical significance due to marked variations between donors independent of ethnicity (Fig. S8). There were no significant differences in glucose tolerance between mice receiving stool transplants from donors of different ethnicity in either experiment (Figs. 6D,H).