Analysis of 3.6 million individuals yields minimal evidence of pairwise genetic interactions for height

Adult height is a highly heritable polygenic trait with heritability attributable to thousands of independent variants. Large-scale studies have been able to detect genetic variants with contributions to height in the range of approximately 1.2 millimetre per allele copy on average. Non-additive genetic interactions may, in part, account for the difference between broad-sense and narrow-sense heritability estimates. However, prior studies have failed to identify variants with non-additive effects, possibly due to the lack of statistical power. Leveraging 3.6M individuals of European genetic ancestry in the 23andMe research cohort, we performed a genome-wide analysis study (GWAS) to select 1,063 independent common SNPs associated with height (p-value < 5e-8), and then screened for evidence of non-additive effects by analysing 564,453 models including a pairwise SNP-SNP interaction term. We identified 69 pairwise models with suggestive evidence of SNP-SNP interaction (p-value < 1e-4) and, for each SNP pair, we evaluated a fully saturated model including additive, dominant, and epistatic (additive-by-additive, additive-by-dominance and dominance-by-dominance) terms. We tested for the presence of epistatic interactions by comparing models with and without epistatic terms using a likelihood ratio test. Assuming a strict Bonferroni-corrected threshold of 8.9e-8 (0.05/564,453), we found no evidence of epistatic interactions (Likelihood ratio test (LRT) p-value < 9e-07 for all models). Our analysis rules out the existence of epistatic interactions between alleles of >1% frequency with effect sizes larger than 2.42mm. Our large-scale analysis provides further evidence of the minimal contribution of non-additivity in the genetic architecture of adult human height.