Fig. 2: Associations of host genetic variants with oral microbiome composition overlap risk loci for dentures use.

a, Converting relative abundances of M microbial species (left) into M orthogonal PCs (middle) allows combining chi-squared statistics for a given genetic variant (one per PC) into a single chi-squared test statistic with M degrees of freedom (right). b, Genome-wide associations with oral microbiome composition in SPARK (top, n = 12,519) and dentures use in UKB (bottom, n = 418,039). Nonsense (red squares), missense (green triangles) and multi-allelic copy number variants (CNVs) (blue diamonds) are highlighted. c, Associations of variants at the FUT2 locus with relative species abundance for the five microbial species with the strongest associations (left five plots); colour indicates effect direction (plots with red points correspond to species which are more abundant in people with functional FUT2 (that is, secretors); blue, less abundant) and colour saturation indicates linkage disequilibrium with rs601338 (FUT2 W154X). Association strengths from the combined test for association with oral microbiome composition show much greater statistical power (rightmost plot). d, Effect sizes (in s.d. units) on relative abundance of microbial species for individuals heterozygous for functional FUT2 (light-filled circles) and for homozygotes (dark-filled circles) relative to those with no functional FUT2 (empty circles). For each effect direction, the ten most significantly associated species are shown. P values are from a recessive model of FUT2 W154X genotype. Error bars, 95% CIs. e, Microbial taxa whose abundance associated with FUT2 genotype (FDR < 0.1) shown on the phylogenetic tree of 439 species (red, taxa whose relative abundances increased with functional FUT2; blue, decreased). Two significantly associated phyla (Firmicutes and Actinobacteria; P = 1.2 × 10⁻⁴ and 4.0 × 10⁻⁵, respectively) are highlighted with yellow sectors. At the species level (outermost circle), dot sizes increase with statistical significance. P values were computed using one-sided chi-squared test (top half, b), two-sided linear regression (bottom half, b) or two-sided linear mixed models (c,d).