Epigenetic clocks and inflammaging: pitfalls caused by ignoring cell-type heterogeneity

Abstract

Epigenetic clocks are machine learning predictors of chronological age that have been consistently shown to be also informative of biological age measures, such as all-cause mortality. A recent study has argued against the use of machine learning methods for building epigenetic clocks, on grounds that these predictors do not enrich for features that correlate with disease. This study further argues that standard epigenetic clocks can not measure inflammaging, an age-related trait, proposing an ad-hoc feature rectification strategy and an inflammation clock that predicts age-acceleration in inflammatory diseases. Here we debunk this study, by demonstrating that their inflammation clock only captures the known increase in the granulocyte to lymphocyte ratio associated with inflammatory diseases like rheumatoid arthritis, and that it fails to capture the shift from naive to mature lymphocytes, which is a key feature of inflammaging. As such, their clock is subsumed and outperformed by an ordinary cell-type deconvolution algorithm. Our analysis underscores the critical need for epigenetic clock and epigenome-wide association studies to always estimate cell-type fractions in the tissue of consideration, and to adjust for its variation, in order to disentangle effects due to changing cell-type composition, from DNA methylation changes that happen in specific cell-types.