Bayesian Genome-wide TWAS method to leverage both cis- and trans- eQTL information through summary statistics

Abstract

Transcriptome-wide association studies (TWAS) have been widely used to integrate gene expression and genetic data for studying complex traits. Due to the computation burden, existing TWAS methods neglect distant trans-expression quantitative trait loci (eQTL) that are known to explain a significant proportion of the variation for most expression quantitative traits.To leverage both cis-and trans-eQTL information for TWAS, we propose a novel TWAS approach based on Bayesian variable selection regression model, which not only accounts for both cis-and trans-SNPs of the target gene but also enables efficient computation by using summary statistics of standard eQTL analyses and a scalable EM-MCMC algorithm. Simulation studies illustrate that our Bayesian approach achieved higher TWAS power compared to existing methods. By application studies, we identified gene ZC3H12B whose GReX is associated with both Alzheimer's dementia (AD) (p-value=2.15) and a global measure of AD pathology (p-value=2.438, which is completely driven by trans-eQTL. We also identified gene KCTD12 whose GReX is associated with ߚ -amyloid load (p-value=7.63which is driven by both cis-and trans-eQTL. Particularly, four of the top driven trans-eQTL of ZC3H12B are located in gene APOC1 (<12KB away from the well-known risk gene APOE of AD) and are also known GWAS signals of AD and blood lipids. Free software for implementing our proposed Bayesian TWAS approach is available on Github.

Novel Variance-Component TWAS method for studying complex human diseases with applications to Alzheimer’s dementia

“…Next, we applied VC-TWAS to the stage1 summary-level GWAS data of AD from IGAP [3], which has a much larger sample size (∼54K with 17,008 AD cases and 37,154 controls). We considered filtered cis-eQTL DPR weights as used in the above applications with individual-level GWAS data, as well as the SNP weights of both cis- and trans-eQTL generated by the BGW-TWAS method [28]. LD covariance matrices from ROS/MAP individual-level GWAS data were used for implementing VC-TWAS with summary-level GWAS data.…”

“…TIGAR 8 provides a more flexible approach to nonparametrically estimate cis-eQTL effect sizes by a Bayesian DPR method [13] (Text S2). Additionally, we also considered modeling gene expression using both cis- and trans-eQTL effect sizes estimated by the recently proposed Bayesian genome-wide TWAS (BGW-TWAS) method [28].…”

“…Additionally, we also considered modeling gene expression using both cis-and trans-eQTL effect sizes estimated by the recently proposed Bayesian genome-wide TWAS (BGW-TWAS) method [28].…”

“…To provide a complementary list of significant genes by considering both cis-and trans-eQTL, we conducted VC-TWAS with the IGAP summary data using the cis-and trans-eQTL weights generated by the BGW-TWAS method [28]. We detected total 71 significant genes with FDR < 0.05 (Fig 4; Table S4; Fig S11B), among which 6 genes were identified by GWAS [42,[50][51][52][53]] and 22 genes were shown to be related with AD or other neurological diseases by previous studies and (Table 3).…”

“…By considering only cis-eQTL effect sizes, VC-TWAS identified both novel and known risk genes for AD within 2MB of the well-known major risk gene APOE of AD, including the known risk gene TOMM40 and APOE . Considering both cis- and trans-eQTL effect sizes estimated by the Bayesian Genome-wide TWAS (BGW-TWAS) method [28], VC-TWAS detected 71 risk genes for AD which complemented existing TWAS results using only cis-eQTL SNP data. After describing the results, we provide a brief discussion summarizing our findings and describing implementation of VC-TWAS into our previously developed software tool TIGAR [8] for public use.…”