Abstract
Machine learning models based on DNA methylation can be used to predict the age of biological samples, but their interpretability is limited due to the lack of causal inferences. Here, we lever-aged large-scale genetic data and performed epigenome-wide Mendelian Randomization to identify CpG sites causal to aging-related traits. We show that neither the existing epigenetic clocks nor DNA methylation changes are enriched in causal CpG sites. Causal CpGs include similar numbers of sites that contribute to aging and protect against it, yet their combined contribution negatively affects age-related traits. We developed a framework for integrating causal knowledge into epigenetic clock models and constructed DamAge and AdaptAge that measure age-related damaging and adaptive changes, respectively. DamAge acceleration is associated with various adverse conditions (e.g., mortality risk), whereas AdaptAge acceleration is related to beneficial adaptations. Only DamAge is reversed upon cell reprogramming. Our results offer a comprehensive map of CpG sites causal to lifespan and healthspan, allowing to build causal biomarkers of aging and rejuvenation and assess longevity interventions, age reversal, and aging-accelerating events.
Competing Interest Statement
The authors have declared no competing interest.