The DNA methylation landscape of metastatic prostate cancer: from characterization to liquid biopsy applications

Epigenetic alterations are observed in virtually all cancer types, yet there is limited understanding of their role in tumorigenesis and evolution. The role of DNA methylation has been particularly elusive in this context. While this epigenetic mark has been extensively profiled in healthy and cancerous samples, our ability to understand its relationship with underlying biological processes is still limited. Moreover, recent advancements in the profiling of cell-free DNA in circulation have sparked renowned attention toward tissue-specific and cancer-specific DNA methylation patterns. In this thesis, I present results to improve and refine the computational characterization of DNA methylation in cancer, focusing on metastatic castration-resistant prostate cancer. The first contribution is the development and performance assessment of Rockermeth, a computational methodology to leverage large-scale DNA methylation profiling data to nominate robust differentially methylated regions (DMRs). Rocker-meth can retrieve biologically relevant DNA methylation changes, as demonstrated by extensive integrative analyses with gene expression, chromatin states, and genomic annotations. The second contribution is the generation of a map of DNA methylation changes across prostate cancer progression. The application of Rockermeth and other tailored methodologies can be used to trace the critical evolutionary steps of this disease, from the healthy tissue to the most lethal metastatic AR-independent counterpart. The main result is the evidence of the ability of DNA methylation to capture a snapshot of the active transcription factors in each state of the disease, offering orthogonal information compared to standard genomic sequencing. The third contribution is the design and development of NEMO, a tailored liquid biopsy sequencing panel approach to allow non-invasive neuroendocrine castration-resistant prostate cancer detection in patients with metastatic disease. Based on previous results and the comprehensive analysis of multiple datasets, I designed a set of informative genomic regions to estimate disease burden and evidence of neuroendocrine transdifferentiation. The actual implementation of the NEMO panel produced a scalable and cost-effective strategy, which has been extensively benchmarked using both in silico and in vitro approaches. The application of NEMO to patient-derived cfDNA samples demonstrated accurate tumor content estimation and robust detection of neuroendocrine disease, making it a promising instance of liquid biopsy for CRPC.