Prostate cancer (PrCa) is a clinically and molecularly heterogeneous disease. Recent studies have focused on the genomic and transcriptomic landscapes1. Since the comprehensive methylation landscape of PrCa has not been defined, we generated a dataset of whole-genome bisulfite sequencing (WGBS) on 110 prostate tumors, 24 matched lymph node metastases; and 8 normal prostate tissues. The tumors also have matched WGS; RNA-seq and clinicopathologic history.
Bulk tumor methylation landscapes include multiple layers of alterations due to distinct biological processes: A) tumor microenvironment (TME); B) global epigenomic remodeling; and C) cis-regulatory methylation events driving transcription2. We adopt a systematic strategy that uses multi-omic modalities to delineate these regulatory layers.
A. Incorporating reference methylation data of specific immune cell subtypes enabled a finer resolution of the immune composition of the tumors. We demonstrate that although the total immune composition was not predictive of biochemical recurrence in PrCa, the presence of immune cell subtypes, CD14 monotypes and CD4 T cells were indeed prognostic. Next, we applied a novel algorithm to peel away the confounding effects of TME in the observed methylation signal from bulk tumor tissues.
B. We revealed two epigenomic remodeling signatures, i) a replication-related methylation loss signature; and ii) a methylation gain signature. They were responsible for pervasive and stochastic methylation changes at many promoters and enhancers in PrCa. What comes first, genomic or epigenomic alterations? The methylation gain signature was linked with key cancer driver mutations (KLF5, RB1) suggesting that genomic drivers can alter the epigenome. Conversely, higher levels of replication-loss methylation signature correlated with lower mutational burden at CpG sites, hinting at the contrary.
C. We also found a subset of enhancers with de-novo methylation loss at transcription factor binding sites (TFBS) in PrCa. These enhancers were enriched for binding (ChIP-seq) of key prostate master TFs including AR, FOXA1 and HOXB13, and linked with downstream gene targets involved in cell cycle and androgen response. We leveraged read-level methylation data to deliver pseudo single-cell resolution enabling insights into epiclonal dynamics. These identified enhancers demonstrated deterministic dynamics indicative of methylation loss at TFBS under ‘selection forces’ facilitating downstream gene aberration in PrCa.