statistics from other independent studies 13,17,18 , we identify novel host-microbiota interactions. Furthermore, we explore the impact of potential confounding factors in modulating these genetic effects and identify potential diet-dependent host-microbiota interactions. We further assess the potential causal relationships between the gut microbiome and dietary habits, biomarkers and disease using Mendelian randomization (MR). Finally, we carry out a power analysis showing how microbiome studies, even at the current sample size, are underpowered to reveal the complex genetic architecture by which host genetics regulates the gut microbiome.
ResultsGenome-wide associations with bacterial taxa and pathways. We investigated 5.5 million common (minor allele frequency (MAF) > 0.05) genetic variants on all autosomes and the X chromosome using linear mixed models 19 to test their association with 207 taxa and 205 bacterial pathways in 7,738 individuals from the DMP cohort (Methods and Supplementary Table 1) 19 . There was no evidence for test statistic inflation (median genomic lambda 1.002 (range, 0.75-1.03) for taxa and 1.004 (range, 0.87-1.04) for pathways). We identified 37 single nucleotide polymorphism (SNP)trait associations at 24 independent loci at a genome-wide P value threshold of 5 × 10 −8 (Fig. 1 and Supplementary Table 2). Genetic variants at two loci passed the more stringent study-wide threshold of 1.89 × 10 −10 that accounts for the number of independent tests performed (Methods).The strongest signal was seen for rs182549 located in an intron of MCM6, a perfect proxy of rs4988235 (r 2 = 1, 1000 Genomes Project European populations), one of the variants known to regulate the LCT gene and responsible for lactase persistence in adults (ClinVar accession RCV000008124). The T allele of rs182549, which confers lactase persistence through a dominant model of inheritance, was found to be associated with decreased abundances of the species Bifidobacterium adolescentis (P = 7.6 × 10 −14 ) and Bifidobacterium longum (P = 3.2 × 10 −08 ), as well as decreased abundances of higher-level taxa (Supplementary Table 2 (ref. 5 )). Associations at this locus were also seen for other taxa of the same genus but at lower levels of significance (Bifidobacterium catenulatum, P = 3.9 × 10 −5 ) and for species of the Collinsella genus (Extended Data Fig. 1). The genetic association at the LCT locus has been previously described, albeit only at the genus level, in Dutch, UK and US cohorts 6,8,14 , as well as in a recent large-scale meta-analysis 13 .The second locus that passed study-wide significance consisted of genetic variants near the ABO gene. ABO encodes the BGAT protein, a histo-blood group ABO system transferase. Associations found at this locus include species Bifidobacterium bifidum (rs8176645, p = 5.5 × 10 −15 ) and Collinsella aerofaciens (rs550057, P = 2.0 × 10 −8 , r 2 = 0.59 with rs8176645 in 1000 Genomes Project Europeans) and higher-order taxa (rs550057, genus Collinsella, P = 9.3 × 10 −11 ; family Coriobacteriac...