Genome-wide DNA methylation profiling and identification of potential pan-cancer and tumor-specific biomarkers

Abstract

DNA methylation alterations have already been linked to cancer, and their usefulness for therapy and diagnosis has encouraged research into the human epigenome. Several biomarker studies have focused on identifying cancer types individually, yet common cancer and multicancer markers are still underexplored. We used The Cancer Genome Atlas (TCGA) to investigate genome-wide methylation profiles of 14 different cancer types and developed a three-step computational approach to select candidate biomarker CpG sites. In total, 1991 pan-cancer and between 75 and 1803 cancer-specific differentially methylated CpG sites were discovered. Differentially methylated blocks and regions were also discovered for the first time on such a large scale. Through a three-step computational approach, a combination of four pan-cancer CpG markers was identified from these sites and externally validated (AUC = 0.90), maintaining comparable performance across tumor stages. Additionally, 20 tumor-specific CpG markers were identified and made up the final type-specific prediction model, which could accurately differentiate tumor types (AUC = 0.87–0.99). Our study highlights the power of the methylome as a rich source of cancer biomarkers, and the signatures we identified provide a new resource for understanding cancer mechanisms on the wider genomic scale with strong applicability in the context of new minimally invasive cancer detection assays.