Abstract
Both host genetics and environmental factors determine human immune-re-
lated phenotypes and disease, including asthma, allergies, and rheumatic
disorders. Although many known environmental exposures contribute to
immune disease risk and severity, the extent to which the human microbiome
is responsible for variation in these phenotypes remains largely unknown.
Additionally, whether gene-by-environment interactions with the microbiome
influence these phenotypes remains uncharacterized, as well as which specific
genetic variants and microbes play roles in this process. To address these
gaps, we examine both the proportion of variation explained by host genetics
(PVE-G, or heritability) and gut microbiome composition (PVE-M) in a unified
framework for ~30 immune-related phenotypes using the large TwinsUK
cohort, for which microbiome, genetic, and phenotypic data are available for
2,500 individuals. First, we examine the relative contributions of PVE-M and
PVE-G for each trait using a linear mixed model framework, by incorporating
random effects that account for the extent of sharing of both genotypes (genet-
ic relatedness matrix) and microbes (beta diversity matrix) among individuals.
We find that PVE-M accounts for up to 44.8% of non-genetic variation in
traits such as body mass index (BMI). Second, we explore whether explicitly
including gut microbiome composition in genome-wide association study
(GWAS) models improves power to detect genetic variants associated with
immune phenotypes. To do so, we compare linear mixed models both with or
without microbiome random effects, applied to 1.3 million genotyped variants
for each phenotype. Including the microbiome in the model does not increase
power to detect genetic variants associated with certain phenotypes (including
height, asthma, and eczema), but does for other phenotypes (BMI, white blood
cell counts, and level of vitamin D in blood). For example, in the case of BMI,
including the microbiome term pushes the significance of the top associated
SNP to surpass the genome-wide Bonferroni threshold (rs1036286, P = 8.6 x
10-8) as compared to the model without the microbiome term (P = 1.8 x 10-8).
Using these data, we have conducted one of the first GWA studies to explicitly
model the microbiome, demonstrating that the microbiome is capable of
explaining a large proportion of variation due to non-genetic effects and can
improve power to detect genetic variants associated with phenotypes.
