Variance quantitative trait loci

: development of novel software and methodology to facilitate discovery, analysis and sharing

Abstract

Genetic epidemiological studies have largely focused on SNP mean effects, but variance effects may also exist that can indicate the presence of SNP interaction effects. Identification of these effects may be useful for improving understanding of disease mechanisms, prediction of disease outcomes, and in combination with other data may provide opportunities for developments in precision medicine. This thesis aims to develop methodology and software to identify and analyse variance loci applied to serum biomarker concentration. To achieve these aims, a regression-based Brown-Forsythe variance test was evaluated and implemented in C++ and R which enables adjustment of covariates and provides an unbiased variance effect estimate for normally distributed traits (Chapter 4). This model was subsequently applied in variance genome-wide association studies (vGWAS) of 30 serum biomarkers in UK Biobank identifying 468 variance loci of 209 million SNPs tested. These loci were investigated to detect 82 gene-environment and six gene-gene interactions including three novel epistatic effects (Chapter 5). The utility of these vGWAS summary statistics in detecting violation of Mendelian randomization homogeneity assumptions was explored through a series of simulation studies. This approach was subsequently applied to investigate the impact of homogeneity violation of low-density lipoprotein, urate and glucose on cardiovascular disease, gout, and type 2 diabetes, respectively. There was no strong evidence of difference in causal estimate after removing instruments associated with exposure variance. These findings are consistent with the main analysis targeting the population average causal effect (Chapter 6). To facilitate sharing and future analyses of vGWAS summary statistics, an efficient and robust storage format was developed using the variant call format that can be used for any GWAS analysis along with Python packages, web-interface and data processing pipeline which are widely used and embedded within the MRC-IEU OpenGWAS infrastructure (Chapter 7).

Date of Award	6 Dec 2022
Original language	English
Awarding Institution	University of Bristol
Supervisor	Tom R Gaunt (Supervisor), Kate M Tilling (Supervisor), Louise A C Millard (Supervisor) & George Davey Smith (Supervisor)