There is a long standing controversy regarding the contribution of epistasis (here defined as interactions between alleles of two or more genetic loci) to complex traits in humans. Most recently, epistasis has been suggested as a possible explanation for the relatively small proportion of a traits narrow sense heritability (h2 = additive genetic variance), explained by single nucleotide polymorphism (SNP) associations. This suggestion centres on an over-estimation of the total h2 inferred indirectly from population (namely twin) data, due to over-simplified or incorrect model specification (e.g., non- or under-estimation of non-additive genetic variance), or perhaps other underlying models which cannot be distinguished from an additive model.
To empirically investigate the occurrence of epistasis in the general population, 1000 Genomes data was utilised to perform allelic interaction tests for autosomal gene-associated SNPs genotyped in 365 individuals of European ancestry (85 CEPH, 93 Finnish, 89 British and 98 Toscani). Only SNPs polymorphic in each of the four subsets were utilised to allow post-hoc analysis of significant epistatic loci (i.e., assess consistency across populations and potential effect of stratification). To capture the ~90% of SNPs likely to influence expression, gene boundaries of known human protein-coding and non-protein-coding (RefSeq) genes were expanded by 15 kb. Genes with overlapping boundaries were merged, resulting in 7,336 non-overlapping regions encompassing 22,504 unique genes and 4,904,974 (M) SNPs. Spectral decomposition of matrices of pairwise linkage disequilibrium (LD) between SNPs for each of the 7,336 non-overlapping gene regions estimated the total effective number of independent SNPs (Meff) to be 1,965,694. Thus indicating the total M(M - 1)/2 = 12,029,382,517,851 (SNP-SNP) interaction tests to be equivalent to approximately Meff(Meff - 1)/2 = 1,931,975,467,971 (Meff-Meff) independent interaction tests and an appropriate genome-wide significance threshold to be 0.05/Meff(Meff - 1)/2 = 2.59 × 10-14.
Results of these analyses, which include the surprising observation of millions of significant interactions will be presented.