Colon-specific epigenetic clocks from minimal features reveal disease-driven aging

Abstract Epigenetic clocks estimate chronological and biological age from DNA methylation patterns, but conventional models typically train on hundreds of thousands of CpG sites and large training cohorts. We previously demonstrated that tissue-unique methylation sites change in a predictable manner upon aging and disease. Here, we demonstrate that clocks built from tissue-unique methylation sites enable accurate age prediction in the human colon using a compact feature set and limited training data. We trained a machine learning model on healthy colon tissue, identifying CpG sites that capture both chronological age and anatomical location (proximal vs. distal). This clock maintains high predictive performance (r = 0.978; MAE 3.9 years) while using an order of magnitude fewer sites and samples than traditional approaches. Applying the model to tissues from individuals with HIV infection, inflammatory bowel disease (IBD), and colonic polyps reveals consistent patterns of accelerated aging, while aspirin treatment is associated with partial deceleration. Our findings establish tissue-unique CpGs as a powerful basis for efficient, interpretable clocks and offer new insights into how chronic inflammation and neoplasia shape the aging landscape of the colon.