Large-scale association analyses identify host factors influencing human gut microbiome composition

Abstract

To study the effect of host genetics on gut microbiome composition, the MiBioGen consortium curated and analyzed genome-wide genotypes and 16S fecal microbiome data from 18,340 individuals (24 cohorts). Microbial composition showed high variability across cohorts: only 9 of 410 genera were detected in more than 95% of samples. A genome-wide association study of host genetic variation regarding microbial taxa identified 31 loci affecting the microbiome at a genome-wide significant (P < 5 × 10^-8) threshold. One locus, the lactase (LCT) gene locus, reached study-wide significance (genome-wide association study signal: P = 1.28 × 10^-20), and it showed an age-dependent association with Bifidobacterium abundance. Other associations were suggestive (1.95 × 10^-10 < P < 5 × 10^-8) but enriched for taxa showing high heritability and for genes expressed in the intestine and brain. A phenome-wide association study and Mendelian randomization identified enrichment of microbiome trait loci in the metabolic, nutrition and environment domains and suggested the microbiome might have causal effects in ulcerative colitis and rheumatoid arthritis.

Figure 1.. Diversity of microbiome composition across the MiBioGen cohorts.

(a) Sample size, ethnicity, genotyping array and 16S rRNA gene profiling method. The SHIP/SHIP-TREND and GEM_v12/GEM_v24/GEM_ICHIP subcohorts are combined in SHIP and GEM, respectively (Online Methods; see Supplementary Note for cohort abbreviations). This merge resulted in the total of 21 cohorts depicted in the figure. (b)* Total richness (number of genera with mean abundance over 0.1%, i.e. 10 reads out of 10,000 rarefied reads) by number of cohorts investigated. (c)* Number of core genera (genera present in >95% of samples from each cohort) by number of cohorts investigated. (d) Pearson correlation of cohort sample size with total number of genera. Confidence band represents the standard error of the regression line. (e)* Unweighted mean relative abundance of core genera across the entire MiBioGen dataset. (f)* Per-sample richness across the 21 cohorts. Asterisks indicate cohorts that differ significantly from all the others (pairwise Wilcoxon rank-sum test; FDR<0.05). (g) Diversity (Shannon index) across the 21 cohorts, with the DanFund and PNP cohorts presenting higher and lower diversity in relation to the other cohorts (pairwise Wilcoxon rank sum test; FDR<0.05). (*) For all boxplots, the central line, box and whiskers represent the median, IQR and 1.5 times the IQR.

Figure 2.. Heritability of microbiome taxa and its concordance with mbQTL mapping.

(a) Microbial taxa that showed significant heritability in the TwinsUK cohort (ACE model, nominal P<0.05, no adjustment for multiple comparison). Taxa with at least one genome-wide significant (GWS) mbQTL hit are marked red. Only taxa present in more than 10% of pairs (>17 MZ pairs, >41 DZ pairs) are shown. Circles and diamonds represent heritability value. Error bars represent 95% CI. (b) Correlation of monozygotic ICC between TwinsUK and NTR cohort. Only taxa with significant heritability (ACE model P<0.05) that are present in both TwinsUK and NTR are shown. Red and blue dots indicate bacterial taxa with/without GWS mbQTLs (P<5×10⁻⁸), respectively. Segments represent 95% CI. (c) Correlation between heritability significance (−log₁₀P_H2 TwinsUK) and the number of loci associated with microbial taxon at relaxed threshold (P_mbQTL<1×10⁻⁵). Taxa with at least one GWS-associated locus are marked red. Error bars represent 95% confidence intervals.

Figure 3.. Manhattan plot of the mbTL mapping meta-analysis results.

MbQTLs are indicated by letters. MbBTLs are indicated by numbers. For mbQTLs, the Spearman correlation test (two-sided) was used to identify loci that affect the covariate-adjusted abundance of bacterial taxa, excluding samples with zero abundance. For mbQTLs, p-values (two-sided) were calculated by logistic regression. Horizontal lines define nominal genome-wide significance (P=5×10⁻⁸, red) and suggestive genome-wide (P=1×10₋₅, blue) thresholds.

Figure 4.. Association of the LCT locus (rs182549) with the genus Bifidobacterium.

(a) Forest plot of effect sizes of rs182549 and abundance of Bifidobacterium. Effect sizes and 95% CI are defined as circles and error bars. Effect sizes were calculated from Spearman correlation p-values (Online Methods). (b) Meta-regression of the association of mean cohort age and mbQTL effect size. Confidence bands represent the standard error of the meta-regression line. (c) Meta-regression analysis of the effect of linear, squared and cubic terms of age on mbQTL effect size. Confidence bands represent the standard error of the meta-regression line. (d) Age-dependence of mbQTL effect size in the GEM cohort. Blue boxes include samples in the age range 6–16 years old. Red boxes include samples with age ≥17 years. The C/C (rs182549) genotype is a proxy of the NC_000002.11:g.136608646=(rs4988235) allele, which is associated to functional recessive hypolactasia. The central line, box and whiskers represent the median, IQR and 1.5 times the IQR, respectively. See Supplementary Note for cohort abbreviations.

Figure 5.. Phenome-wide association study (PheWAS) domain enrichment analysis.

The analysis covered top-SNPs from 30 mbTLs and 20 phenotype domains. Three thresholds for multiple testing were used: 0.05, 8.3×10⁻⁵ (Bonferroni adjustment for number of phenotypes and genotypes studied) and 5×10⁻⁸ (an arbitrary genome-wide significance threshold). Only categories with at least one significant enrichment signal are shown.

Figure 6.. Mendelian randomization (MR) analysis.

The X-axes show the SNP-exposure effect and the Y-axes show the SNP-outcome effect (SEs denoted as segments). (a) MR analysis of class Actinobacteria (exposure) and ulcerative colitis (outcome). (b) MR analysis of genus Bifidobacterium (exposure) and ulcerative colitis (outcome). (c) MR analysis of family Oxalobacteraceae (exposure) and rheumatoid arthritis (outcome).