Genetic interactions drive heterogeneity in causal variant effect sizes for gene expression and complex traits

Abstract

Despite the growing number of genome-wide association studies (GWAS) for complex traits, it remains unclear whether effect sizes of causal genetic variants differ between populations. In principle, effect sizes of causal variants could differ between populations due to gene-by-gene or gene-by-environment interactions. However, comparing causal variant effect sizes is challenging: it is difficult to know which variants are causal, and comparisons of variant effect sizes are confounded by differences in linkage disequilibrium (LD) structure between ancestries. Here, we develop a method to assess causal variant effect size differences that overcomes these limitations. Specifically, we leverage the fact that segments of European ancestry shared between European-American and admixed African-American individuals have similar LD structure, allowing for unbiased comparisons of variant effect sizes in European ancestry segments. We apply our method to two types of traits: gene expression and low-density lipoprotein cholesterol (LDL-C). We find that causal variant effect sizes for gene expression are significantly different between European-Americans and African-Americans; for LDL-C, we observe a similar point estimate although this is not significant, likely due to lower statistical power. Cross-population differences in variant effect sizes highlight the role of genetic interactions in trait architecture and will contribute to the poor portability of polygenic scores across populations, reinforcing the importance of conducting GWAS on individuals of diverse ancestries and environments.

Gene expression in African Americans and Latinos reveals ancestry-specific patterns of genetic architecture

“…All four grandparents of each study participant also identified as Latino for GALA II or African American for SAGE. The median age of the participants varied from 13.2 (PR) to 16.0 (AA) years old. About 50% of the participants were female and 45% (MX) to 62% (PR) had physician-diagnosed asthma.…”

“…We and others have shown that gene expression prediction models trained in predominantly European ancestry reference datasets, such as the Genotype-Tissue Expression (GTEx) project 2 , have substantially lower accuracy to predict gene expression levels when applied to populations of non-European ancestry 3,12,13 . The importance of having ancestry-matched training datasets for prediction accuracy is also reflected by the limited cross-population portability of other multi-SNP prediction models, such as polygenic risk scores (PRS) [14][15][16] . Therefore, the limited diversity in genetic association studies and reference datasets is a major obstacle for applying existing integrative genomic studies to non-European populations.…”

Population differentiation of polygenic score predictions under stabilizing selection

“…Indeed the prediction accuracy of polygenic scores for some traits was recently shown to be quite variable across different groups within an ancestry, suggesting that GxE and environmental variation are quite prevalent for some traits [ 45 ]. In addition, associated variants on the same haplotype were found to have differing effects in European–Americans and admixed African–Americans, suggesting that genetic or environmental interactions modify additive effect sizes across groups [ 47 ]. Thus, we caution that a better understanding of the portability of polygenic scores across populations also requires a stronger understanding of the causes of variation in prediction accuracy within populations.…”

“…For one, the associated loci are usually not the causal loci underlying trait variation; instead they tag the effects of linked causal sites. The interpretation of the effect size of an associated variant can be tricky because of: (i) linkage disequilibrium (LD), whereby it absorbs the effects at correlated causal sites [ 27 – 31 ]; (ii) population stratification, whereby it absorbs the effects of covarying environments [ 32 – 36 ]; and (iii) gene-by-environment (GxE) or gene-by-gene (GxG) interactions, whereby its estimate is averaged over the interacting environmental contexts or genetic backgrounds in the sample [ 28 , 37 – 47 ]. As all of these factors can and will vary across populations, the effect sizes of alleles will differ among them and so polygenic scores will have lower prediction accuracy, i.e.…”

Transcriptome-Wide Gene-Gene Interaction Association Study Elucidates Pathways and Functional Enrichment of Complex Traits

“…Often, interactions are only tested between loci that are significant in single-locus GWAS or phenotypic variance test effects, or are based on hypothesized pathways or networks. While such methods improve feasibility by reducing the number of tests, they constrain the ability to detect novel epistatic effects or new pathways and networks involved in complex traits 8 , and in some cases do not indicate whether the interactor effect is an environment or a second gene 28,29 . Similarly, if a strong interaction between two loci exists, the main effects estimated in a single-variant GWAS could be muted 7 , reducing the likelihood of identifying such interactions in two-stage approaches.…”