Discovery of gene-environment and epistatic interactions affecting gene expression in the TwinsUK cohort via association mapping of variance and monozygotic twin discordance. A. Brown1,2,3, A. Buil2, A. Vi˝uela4, M. Davies4, K. Small4, T. Spector4, E. Dermitzakis2, R. Durbin1 1) Wellcome Trust Sanger Institute, Cambridge, United Kingdom; 2) Dep. Genetic Medicine and Development, University of Geneva, Geneva, Switzerland; 3) NORMENT, KG Jebsen Centre for Psychosis Research, Institute of Clinical Medicine, University of Oslo, Oslo, Norway; 4) Department of Twin Research and Genetic Epidemiology, King's College London, London, United Kingdom.
The identification of non-additive interactions between genetic variants (epistasis), or between genetic variants and environment (GxE), can give insight into mechanism and contribute to our understanding of the genetic architecture of traits. Here we explore epistasis and GxE affecting gene expression, considered as a set of quantitative traits measured using RNA-seq, in fat, LCLs, skin and whole blood from the TwinsUK cohort (N=~850). Because epistasis and GxE affect trait variance in a genotype dependent fashion, we used the strategy of prioritising SNPs associated with variance in expression (v-eQTL) to look for interactions. Similarly, SNPs associated with discordance of expression between monozygotic twin pairs (d-eQTL) indicate presence of GxE. We found 1198 v-eQTL in LCLs, 620 in fat, 368 in skin and 39 in blood, and 73, 211, 63 and 1 d-eQTL in those tissues. A greater proportion of v-eQTL acted as d-eQTL (21-44% depending on tissue) than vice versa (0-29%), consistent with d-eQTL as a subset of v-eQTL induced by GxE. Functional overlap with ENCODE data shows LCL v-eQTL significantly depleted in transcriptionally repressed regions (odds ratio, 0.82) and enriched in enhancers (OR 1.71); d-eQTL are enriched in promoters (OR 5.29). Skin d-eQTL are enriched in H3K36me3 regions (OR 4.02), a mark of active transcription. To find environments involved in GxE signatures, we tested all v-eQTL and d-eQTL for interactions with age, BMI and 20 expression principal components (PCs), having previously seen PCs can be highly heritable. There were three Bonferroni significant interactions between genotype and BMI affecting fat expression (p<1.94e-5). There were many interactions with PCs: 2, 10, 39 and 66 in blood, fat, skin and LCLs (p<9.70e-7). Analysis of separate dermis and epidermis data suggested some skin d-eQTL are cell specific eQTL. Finally, as v-eQTL can be induced by epistasis, we scanned the cis window for SNPs interacting with v-eQTL. Initial results found epistasis in all tissues, frequently shared across tissues. We replicated epistasis found in all 4 tissues using RNA-seq LCL data from GEUVADIS samples: for 100%, 52%, 67% and 15% of interactions in blood, fat, LCLs and skin we observed p<0.05. In summary, we detect widespread variance and discordance effects in gene expression. V-eQTL provide a way to discover replicating epistasis while d-eQTL consistently have more success at mapping GxE with phenotypes, PCs and tissue composition measures.
You may contact the first author (during and after the meeting) at