Abstract
BACKGROUND: Profiles of DNA methylation of many tissues relevant in human disease have been obtained from microarrays and are publicly available. These can be used to generate maps of chromatin compartmentalization, demarcating open and closed chromatin across the genome. Additionally, large sets of genome-wide transcription factor binding profiles have been made available thanks to ChIP-seq technology. METHODS: We have identified genomic regions with altered chromatin compartmentalization in prostate adenocarcinoma tissue relative to normal prostate tissue, using DNA methylation microarray data from The Cancer Genome Atlas. DNA binding profiles from the Encyclopedia of DNA Elements (ENCODE) ChIP-seq studies have been systematically screened to find transcription factors with inferred DNA binding sites located in discordantly open/closed chromatin in malignant tissue (compared with non-cancer control tissue). We have combined this with tests for corresponding up-/downregulation of the transcription factors' putative target genes to obtain an integrated measure of cancer-specific regulatory activity to identify likely transcriptional drivers of prostate cancer. RESULTS: Generally, we find that the degree to which transcription factors preferentially bind regions of chromatin that become more accessible during prostate carcinogenesis is significantly associated to the level of systematic upregulation of their targets, at the level of gene expression. Our approach has yielded 11 transcription factors that show strong cancer-specific transcriptional activation of targets, including the novel candidates KAT2A and TRIM28, alongside established drivers of prostate cancer MYC, ETS1, GABP and YY1. CONCLUSIONS: This approach to integrated epigenetic and transcriptional profiling using publicly available data represents a cheap and powerful technique for identifying potential drivers of human disease. In our application to prostate adenocarcinoma data, the fact that well-known drivers are amongst the top candidates suggests that the discovery of novel candidate drivers may unlock pathways to future medicines. Data download instructions and code to reproduce this work are available at GitHub under 'edcurry/PRAD-compartments'.