AbstractQuantifying underlying DNA methylation signatures of complex traits presents an opportunity to identify biomarkers and modes of disease intervention. Years of epigenome-wide association studies (EWAS) have shown signatures vary greatly by trait and the interpretation of signals remains difficult. This thesis explores potential explanations for this and examines the role of EWAS in understanding complex traits.
To ascertain necessary data, I led a collection of EWAS results and developed a web resource for storing and querying the 975,574 associations across 1244 EWAS. Evidence was found that results for EWAS that accounted for common biases, such as batch effects and cell composition, could partially be explained by variance and heritability of DNA methylation. Further, identified sites were enriched in promoter regions, enhancer regions and transcription factor binding sites.
Across the EWAS surveyed, DNA methylation was commonly measured in blood at roughly 450,000 sites genome-wide. I examined the predictive capacity of DNA methylation in this context and found that it captured little variance of 400 independent complex traits.
Next, commonalities between the overlap in biology highlighted by EWAS and GWAS of corresponding traits was explored and I found that the genes and genesets identified were substantially different. Trait aetiology may still be explored through EWAS, but the largely differential biology highlighted suggests the majority of EWAS results here are due to confounding and reverse causation.
Mendelian randomization (MR) analyses further suggested residual confounding as being responsible for EWAS results as marked differences were found between an EWAS meta-analysis of lung cancer and the corresponding MR analyses.
Through cataloguing published results and integrating methods and results from other fields, this thesis identifies limitations to the current EWAS study design that reveal the complexity of the role of DNA methylation on mediating the path from genotype or environment to phenotype.
|Date of Award
|11 May 2021
|Tom R Gaunt (Supervisor), Gibran Hemani (Supervisor) & Nicholas John Timpson (Supervisor)