Longitudinal changes in epigenetic measures over 2 years: methodological implications

Epigenetic clocks are increasingly proposed as surrogate endpoints in aging trials, yet their short-term behavior in healthy older adults is not well characterized. We analyzed DNA methylation at baseline, year 1, and year 2 in 899 COSMOS-Blood participants (mean age 70.0; 50% women), deriving Horvath, Hannum, PhenoAge, and GrimAge clocks (original and principal component [PC] versions) and DunedinPACE. Epigenetic age acceleration was computed by regressing each clock on chronological age. Chronological age was independent of epigenetic age acceleration and DunedinPACE. PC clocks exhibited substantially smaller 2-year change variance than original clocks, indicating greater measurement stability. Linear mixed-effects models showed statistically detectable but numerically small annual epigenetic age acceleration increases for several PC clocks (e.g., PC Horvath + 0.14 year/year; PC GrimAge + 0.16 year/year), whereas DunedinPACE did not change significantly. Baseline values strongly predicted the same measure at years 1 and 2 (R² ≈ 0.71-0.88 for PC clocks). Tertile trajectories were largely stable, and first-year increases tended to be followed by second-year decreases, consistent with regression to the mean. Overall, current epigenetic measures, particularly PC clocks, appear stable on average and highly predictable over 2 years in generally healthy older adults, implying limited sensitivity to short-term change. These empirical SDs and strong baseline-follow-up correlations support ANCOVA-based analytic methods for future trials, and the study provides information for the power calculation.