Abstract
PURPOSE: Cancers present significant DNA methylation changes, which arise in a stochastic manner, marked by extensive epigenetic variation, indicative of high epigenetic instability. We aimed to evaluate the utility of epigenetic instability for cell-free DNA (cfDNA)-based cancer detection. EXPERIMENTAL DESIGN: Through analysis of cancer DNA methylation datasets (n = 2,084), we identified a set of 269 CpG island regions that robustly captures this instability in a cancer-specific manner. We developed metrics to measure this epigenetic instability, termed the epigenetic instability index (EII), for cancer screening via cfDNA methylation. RESULTS: Machine learning classifiers using the EII of these 269 regions efficiently identified breast and lung cancers from cfDNA, differentiating even stage IA lung adenocarcinoma with ∼81% sensitivity and early-stage breast cancer with ∼68% sensitivity, both at 95% specificity. CONCLUSIONS: Our studies demonstrate that quantifying epigenetic instability is a novel, capable approach to distinguishing cancer from normal cases using cfDNA, performing better than standard approaches using absolute methylation changes. The epigenetic instability-based approaches for cancer detection developed here, along with their validation in independent datasets, support further development and the potential for future clinical application of these strategies in cancer screening.