Genome-wide meta-analysis of joint tests for genetic and gene-environment interaction effects

Abstract

Background: There is growing interest in the study of gene-environment interactions in the context of genome-wide association studies (GWASs). These studies will likely require meta-analytic approaches to have sufficient power.

Methods: We describe an approach for meta-analysis of a joint test for genetic main effects and gene-environment interaction effects. Using simulation studies based on a meta-analysis of five studies (total n = 10,161), we compare the power of this test to the meta-analysis of marginal test of genetic association and the meta-analysis of standard 1 d.f. interaction tests across a broad range of genetic main effects and gene-environment interaction effects.

Results: We show that the joint meta-analysis is valid and can be more powerful than classical meta-analytic approaches, with a potential gain of power over 50% compared to the marginal test. The standard interaction test had less than 1% power in almost all the situations we considered. We also show that regardless of the test used, sample sizes far exceeding those of a typical individual GWAS will be needed to reliably detect genes with subtle gene-environment interaction patterns.

Conclusion: The joint meta-analysis is an attractive approach to discover markers which may have been missed by initial GWASs focusing on marginal marker-trait associations.

Fig. 1

Quantile-quantile plots for a single realization of Y. Trait values have been simulated using the model Y = b_G1G1 + b_G2 + b_EE + ε, where b_G1 = 0.12, b_G2 = 0.05, b_E = 0.05, and b_G2E = 0.1. The genetic inflation factors λ_GC of the marginal test, the standard interaction test and the joint test are equal to 1.019, 1.001, and 1.007 (respectively) for a, and 1.019, 1.001, 1.007 for b.

Fig. 2

Power of the marginal, joint and interaction tests as more studies are included in the analysis. HPFS: Health Professional Follow-up Study, NHS: Nurse Health Study, CHD: coronary heart disease, BRCA: breast cancer, T2D: type 2 diabetes. Phenotypic data (100,000 realizations of Y) were simulated using the model Y = b_G1G1 + b_G2G2 + b_EE + b_G2E G2 × E, where b_G1 = 0.12, b_G2 = 0.05, b_E = 0.05, and b_G2E = 0.1.

Fig. 3

Power comparison of the marginal tests of G1 and G2, the standard G2-E interaction test and the joint G2-E test. Phenotypic data (1,000 realizations of Y for each point) were simulated using the model Y = b_G1G1 + b_G2G2 + b_EE + b_G2EG2.E + ε, were b_G1, b_E and σ were fixed and respectively equal to 0.12, 0.05 and 1, while we varied b_G2 and b_G2E across a grid defined by b_G2 ∊ {0.01, 0.04, 0.07, 0.1}, and b_G2E ∊ {−0.15 to 0.15 by 0.2}.