Exploring cell-type-specific cytosine modifications in somatic tissues and cancer

Thesis

Cytosine methylation (5mC) is one of the most widely studied DNA modifications and plays crucial roles in correct development, regulation of gene expression, and maintaining genome stability. 5mC landscapes vary across human cell types and can be significantly altered in cancers. Differences in methylation patterns are frequently used in studying human development and disease and can be used as prognostic biomarkers. In this thesis, I show a novel, semi-supervised method based on non-negative matrix factorisation (NMF) for the identification of highly variable, tissue-specific CpG blocks across the genome. The method is based on a TAPS β atlas representing a collection of 20 different cells and tissues. I applied the method to both heterogeneous data and FACS-sorted cells, obtaining satisfying results in cross-validation, finding that the method required minimal optimisation to be applicable to new datasets. The selected blocks cover a wide range of genomic contexts and are co-localised with tissue-specific genes, histone marks, chromatin states, and are enriched in 5hmC signals across most tissues. I present how the method can be used in the cancer context, by deconvoluting samples from patients with oesophageal adenocarcinoma (OAC) enroled in the LUD2015-005 clinical trial. The results gave insights into the cellular composition of patient samples and tumour purity and found that changes in the contribution of stomach and eosinophils between time points were associated with different survival probabilities.

Authors: Drozdz, M

• DOI: 10.5287/ora-brpd6en9z
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English