Search

The incidence of renal cell carcinoma (RCC) and small renal masses (SRMs), defined as < 4cm in diameter, is increasing dramatically. SRMs encompass a variety of potential diagnoses, including benign and malignant tumours, the most common of which is clear cell RCC (ccRCC). Current methods are unable to confidently distinguish pathological subtypes of SRMs, meaning patients with benign tumours are undergoing unnecessary invasive surgery. In addition, there are difficulties risk stratifying patients with ccRCC and predicting outcomes. Genomic alterations, such as mutational analysis, may have a role in RCC diagnosis and prognostication, but are unlikely to be sufficient alone due to low recurrence rates and significant intra-tumoral heterogeneity (ITH), which limit detection. Changes in DNA methylation are abundant and often early events in tumorigenesis, which may overcome these challenges as potential tumour markers both in tissue and liquid biopsies. To address the aforementioned diagnostic challenge, I characterised DNA methylation and gene expression in tissue from patients with benign and malignant renal tumours to elucidate similarities and differences between tumour subtypes. Subsequently, DNA methylation data were combined on over 1200 tissue samples and these were used to train and test MethylBoostER (Methylation and XGBoost for Evaluation of Renal tumours), a machine learning model to predict common pathological subtypes of renal tumours. MethylBoostER was externally validated on four independent publicly available datasets (N=518), demonstrating a high accuracy (receiver operating characteristic area under the curve; AUC >0.90). MethylBoostER predicted consistent classification of multi-region samples from the same patient in 90% of individuals, suggesting ITH does not limit model applicability in a biopsy setting. Subsequently, I undertook a systematic evaluation of methylation heterogeneity in ccRCC, exploring associations with clinical/prognostic parameters and highlighting implications for biomarker selection. I evaluated multi-region tissue samples (N=135) from ccRCC patients (N=18) and assessed heterogeneity between patients, within a patient and within a sample. Inter-patient heterogeneity dominated over intra-tumoural heterogeneity. My analysis represents the first evaluation of epipolymorphism, a measure of methylation heterogeneity within a sample, in ccRCC. Significant differential epipolymorphism was noted in ccRCC versus normal kidney at the promoter region of genes known to be implicated in kidney cancer and this finding was externally validated in an independent cohort (N=71). Although changes in epipolymorphism are believed to be a stochastic process, my results suggest that disordered methylation may accumulate in functionally relevant loci which are known to contribute to ccRCC tumorigenesis. Circulating tumour DNA (ctDNA) represents a promising target for non-invasive liquid biopsy in both diagnostic and prognostic applications. Mutational analyses of ctDNA have produced disappointing detection rates in ccRCC, possibly hampered by low ctDNA levels and high mutational ITH. I therefore performed targeted methylation analysis of ctDNA using a novel method- Nimbus (Non-destructive Integration of Methylation to Boost Underlying Signals). Targeted analysis of hypomethylated regions in plasma ctDNA distinguished ccRCC from cancer-free controls with an AUC of 0.96 and produced superior detection rates compared to mutational analysis (93% vs 50%). My results suggest that tumour signal may be enriched in post-biopsy fluid (proximal sample) compared to plasma (distal sample), a strategy that could be useful in patients with SRMs to complement the current diagnostic pathway and overcome low concentrations of plasma ctDNA. In summary, I comprehensively characterise DNA methylation in tissue and liquid samples derived from patients with renal tumours. In the future, DNA methylation analysis of renal tumour biopsy tissue and/or liquid biopsy samples could enable improved diagnosis of patients with SRMs as well as facilitating prognostic stratification.