An ancestry-based approach for detecting interactions

Abstract

Background: Epistasis and gene-environment interactions are known to contribute significantly to variation of complex phenotypes in model organisms. However, their identification in human association studies remains challenging for myriad reasons. In the case of epistatic interactions, the large number of potential interacting sets of genes presents computational, multiple hypothesis correction, and other statistical power issues. In the case of gene-environment interactions, the lack of consistently measured environmental covariates in most disease studies precludes searching for interactions and creates difficulties for replicating studies.

Results: In this work, we develop a new statistical approach to address these issues that leverages genetic ancestry, defined as the proportion of ancestry derived from each ancestral population (e.g., the fraction of European/African ancestry in African Americans), in admixed populations. We applied our method to gene expression and methylation data from African American and Latino admixed individuals, respectively, identifying nine interactions that were significant at P<5×10-8. We show that two of the interactions in methylation data replicate, and the remaining six are significantly enriched for low P-values (P<1.8×10-6).

Conclusion: We show that genetic ancestry can be a useful proxy for unknown and unmeasured covariates in the search for interaction effects. These results have important implications for our understanding of the genetic architecture of complex traits.

Keywords: admixture; gene-environment interaction; gene-gene interactions.

Figure 1.

Examples of How Genetic Ancestry Can Be A Proxy for Interacting Covariates. (a) Model of how genetic ancestry θ can be correlated with various environmental exposures, some of which affect a phenotype. (b) Example of how the correlation between the probability of an AA genotype (bars 2–4) and values of θ (bar 1) increase with higher levels of SNP allele frequency differentiation. In this plot p₁ and p₂ denote the allele frequency of allele A in ancestral populations 1 and 2 respectively. (c) Example of how effect sizes at a tag-SNP may differ due to differential LD on distinct ancestral backgrounds (here, EUR and AFR).

Figure 2.

Power Plots for Pairwise Interaction Simulations. Power of testing G × θ (a) versus testing pairwise SNPs directly (b) as a function of the difference in the ancestral allele frequencies at a differentiated SNP.

Figure 3.

Power Plots for Multi-way Pairwise Interaction Simulations Power of testing G × θ as a function of the difference in the ancestral allele frequencies for multiple interacting SNPs.

Figure 4.

Power Plots for G × E Interaction Simulations. Power of testing G × θ as a function of the correlation between an environmental covariate and genetic ancestry.