Analysis of epistasis in human complex traits

Thesis

Thousands of genetic mutations have been associated with many human complex traits and diseases, improving our understanding of the biological mechanisms underlying these phenotypes.

The great majority of genetic association studies have focused exclusively on the direct effects of single mutations, ignoring possible interactions (epistasis). However, since genes operate within complex networks, interactions are expected to exist. The modelling of epistasis could further biological understanding, but the detection of such effects is complicated by a vast search space.

In this thesis, we present a new statistical method to detect genetic interactions affecting quantitative traits in large-scale datasets. Our approach is based on testing for an interaction between a variant and a polygenic score (PGS) comprising a group of other mutations. We develop a new computational algorithm for PGS construction, and show through simulations that this method is robust to false-positives while retaining statistical power.

We apply our approach to 97 quantitative traits in the UK Biobank (UKB) and find 144 independent interactions with the PGS for 52 different traits, including important variants known to affect disease risk at the APOE, FTO and LDLR genes, for example.

We also develop a test to identify, for each variant interacting with the PGS, the variants driving that interaction. This recovers previously-known interactions and identifies several novel signals, primarily for biomarker traits. An example is a large network of genes (including ABO, ASGR1, FUT2, FUT6, PIGC and TREH) affecting alkaline phosphatase levels, or an interaction between IL33 and ALOX15 impacting eosinophil count, potentially implicated in asthma.

Lastly, we extend our analysis to a new dataset of imputed variation at HLA genes in the UKB and find, among others, a new interaction for glycated haemoglobin involving HLA-DQA1*03:01, an allele previously associated with diabetes.

Our results demonstrate the potential for detecting epistatic effects in presently-available genomic datasets. This can allow the uncovering of key 'core' genes modulating the impacts of other regions in the genome, as well as the identification of subgroups of interacting variants of likely functional relevance.

Authors: Fonseca Ferreira, LA

• DOI: 10.5287/ora-7r7rzdpop
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: GWAS; epistasis; quantitative traits; complex traits; genetic interactions; UK Biobank; polygenic scores
• Subjects: Genetics; Epistasis (Genetics)