Leveraging the correlation structure in DNA methylation data to identify stable and persistent regulatory networks in the human methylome

Abstract

DNA methylation (DNAm) is an epigenetic modification that influences genetic function, which can be altered by environmental and genetic factors. Relationships between DNAm sites are important because genome biology functions as a system, and as such it is unlikely that DNAm sites act in isolation. Identifying key relationships between DNAm sites may uncover systems or pathways which regulate, or are regulated by, DNAm. This could lead to
a better understanding of regulatory mechanisms involved in disease susceptibility, which may lead to new therapeutic targets being identified.
Previous studies have shown that DNAm sites in close proximity often have correlated DNAm states. However the drivers of this correlation have so far only been established for selected DNAm sites. This thesis provides a comprehensive description of correlation structure across the entire 450k array in cis and trans, and how this is preserved in the same individuals over time, as well as across datasets and ethnicities.
I illustrate that across the genome (as measured by the 450k), cis correlation structure is consistent and replicable, both across cohorts and across ethnicities. I show that genetic influences on DNAm correlation structure are present, but that they do not seem to reflect LD structure. I show that highly correlated trans DNAm sites are enriched for active transcription start sites, promotors, and transcription regulation. Network visualisation shows that these highly correlated trans sites are interconnected.
I finally create DNAm networks using WGCNA, to ascertain whether correlated DNAm sites form pathways that associate with biological functions and phenotypes related to development. I find that DNAm modules are strongly preserved over time, between datasets, and between ethnicities.
This work shows that correlations between DNAm sites are replicable, stable, and biologically meaningful, and can be leveraged to gain novel insights about genome function.

Date of Award	24 Mar 2020
Original language	English
Awarding Institution	University of Bristol
Supervisor	Tom R Gaunt (Supervisor), Josine L Min (Supervisor) & Nicholas John Timpson (Supervisor)