Bayesian Genome-wide TWAS Method to Leverage both cis- and trans-eQTL Information through Summary Statistics

Abstract

Transcriptome-wide association studies (TWASs) have been widely used to integrate gene expression and genetic data for studying complex traits. Due to the computational burden, existing TWAS methods do not assess distant trans-expression quantitative trait loci (eQTL) that are known to explain important expression variation for most genes. We propose a Bayesian genome-wide TWAS (BGW-TWAS) method that leverages both cis-and trans-eQTL information for a TWAS. Our BGW-TWAS method is based on Bayesian variable selection regression, which not only accounts for cis-and trans-eQTL of the target gene but also enables efficient computation by using summary statistics from standard eQTL analyses. Our simulation studies illustrated that BGW-TWASs achieved higher power compared to existing TWAS methods that do not assess trans-eQTL information. We further applied BWG-TWAS to individual-level GWAS data (N ¼ $3.3K), which identified significant associations between the genetically regulated gene expression (GReX) of ZC3H12B and Alzheimer dementia (AD) (p value ¼ 5.42 3 10 À13 ), neurofibrillary tangle density (p value ¼ 1.89 3 10 À6 ), and global measure of AD pathology (p value ¼ 9.59 3 10 À7 ). These associations for ZC3H12B were completely driven by trans-eQTL. Additionally, the GReX of KCTD12 was found to be significantly associated with b-amyloid (p value ¼ 3.44 3 10 À8 ) which was driven by both cis-and trans-eQTL. Four of the top driven trans-eQTL of ZC3H12B are located within APOC1, a known major risk gene of AD and blood lipids. Additionally, by applying BGW-TWAS with summary-level GWAS data of AD (N ¼ $54K), we identified 13 significant genes including known GWAS risk genes HLA-DRB1 and APOC1, as well as ZC3H12B.

MOSTWAS: Multi-Omic Strategies for Transcriptome-Wide Association Studies

“…We considered two scenarios for each set of simulation parameters: (1) an ideal case where the leveraged association between the distal-SNP and gene of interest exists in both the reference and imputation panel, and (2) a “null” case where the leveraged association between the distal-SNP and the gene of interest exists in the reference panel but does not contribute to phenotype heritability in the imputation panel. We ran 1,000 simulations for every unique set of simulation parameters in both simulation scenarios and computed TWAS power for models trained using local-only modeling (traditional FUSION models), Bayesian Genome-Wide TWAS (BGW-TWAS), a concurrent TWAS method that includes distal variants [ 31 ], and MOSTWAS.…”

“…We also compared MOSTWAS external predictive performance with BGW-TWAS, which employs Bayesian variable selection to train predictive models using both local- and distal-eQTLs [ 31 ]. Here, as BGW-TWAS models were trained on the entire ROS/MAP dataset, we extracted expression data from the PsychENCODE project for 646 samples from dorsolateral prefrontal cortex tissue and corresponding individual-level genotype data; these 646 samples are derived from patients in the control group, with no symptoms of schizophrenia or bipolar disorder [ 38 , 39 ].…”

“…We then imputed expression using both MOSTWAS and BGW-TWAS models trained using ROS/MAP and assessed predictive performance via McNemar’s adjusted R 2 . Here, we note that BGW-TWAS uses a different inclusion criteria than MOSTWAS, where a BGW-TWAS model is trained if at least one eQTL with non-zero effect and sufficiently large posterior causal probability is detected without heritability estimation or cross-validation performance [ 31 ]. Possibly due to MOSTWAS’s more conservative inclusion criteria, BGW-TWAS imputes a larger number of genes than MOSTWAS (11,990 versus 4,931, respectively).…”

“…Three of the 4 loci also showed significant associations with distal variants, above and beyond the association with local variants, at FDR-adjusted P <0.05 ( S2 Table ). We also assessed TWAS associations for Alzheimer’s disease risk using MOSTWAS for the 11 autosomal risk genes identified by Luningham et al using BGW-TWAS [ 31 ]. Due to more conservative inclusion criteria in MOSTWAS, we only trained significant models for 5 of these genes ( AEP1 , BTN3A2 , GPX1 , APOC1 , and HLA-DRB1 ).…”

“…We compared MOSTWAS to a contemporary TWAS method that incorporates distal-eQTLs, BGW-TWAS [ 31 ]. BGW-TWAS can impute a larger number of gene models than MOSTWAS, but among genes both BGW-TWAS and MOSTWAS were able to impute in external PsychENCODE and TCGA-BRCA data [ 38 , 39 ], MOSTWAS generally showed larger predictive adjusted R 2 .…”