https://www.thelancet.com/journals/lanhl/article/PIIS2666-7568(26)00019-X/fulltext
chatGPT(5.5paid):
Paper
Shan et al., “Physical activity and biological age measured by DNA methylation clocks: a systematic review and meta-analysis,” Lancet Healthy Longevity, 2026.
Summary
This paper asks whether people who are more physically active show a lower “biological age” when measured using DNA methylation clocks. The authors searched six databases from 2011 to June 2025 and included 44 studies covering 145,465 participants. These studies used multiple epigenetic clocks, including first-generation clocks such as Horvath and Hannum, second-generation clocks such as PhenoAge and GrimAge, and third-generation clocks such as DunedinPoAm and DunedinPACE.
The overall finding is that higher physical activity generally correlates with lower DNAm age / lower epigenetic age acceleration, but many individual study results were not statistically significant. Only seven cross-sectional studies had sufficiently compatible data for the formal meta-analysis. In that meta-analysis, one standard deviation higher MET-minutes per week was associated with:
| Clock | Association with higher physical activity |
|---|---|
| Horvath EAA | 0.03 SD lower biological age acceleration |
| GrimAge EAA | 0.09 SD lower biological age acceleration |
| Hannum EAA | Not statistically significant |
| PhenoAge EAA | Not statistically significant |
So the clearest signal is for GrimAge, with a smaller and less robust signal for Horvath.
The authors conclude that physical activity is associated with lower biological age, but that the evidence is mostly observational and cross-sectional, so it cannot prove that exercise itself slows biological ageing. They call for longitudinal studies and randomised trials using standardised, objectively measured physical activity.
Novelty
The main novelty is that this appears to be the first systematic review and meta-analysis specifically assessing physical activity against multiple generations of DNA methylation clocks, rather than focusing on one clock or one cohort. The authors explicitly compare evidence across first-, second-, and third-generation clocks.
A second useful contribution is the attempt to harmonise physical activity exposure using MET-minutes per week, allowing a pooled estimate across studies where possible. This is valuable because the underlying literature uses very different exercise measures: step counts, accelerometers, self-report questionnaires, occupational activity, frequency measures, and MET-based measures.
A third point of novelty is the separation of results by clock type. The paper shows that the association is not uniform across clocks: GrimAge appears more responsive or more strongly associated with activity than Hannum or PhenoAge, while Horvath shows a smaller signal. That matters because these clocks measure partly different things: chronological-age-like methylation drift, mortality-associated methylation patterns, inflammatory/physiological risk, or pace of ageing.
Critique
The paper is useful, but its conclusion should be read as association, not causation.
The biggest limitation is that the meta-analysis is based on only seven cross-sectional studies, despite the full review including 44 studies. Cross-sectional data cannot distinguish whether physical activity lowers biological age, or whether biologically younger people are more capable of being physically active. The authors acknowledge this directly.
The second problem is measurement heterogeneity. Physical activity was measured in many different ways: self-report, accelerometry, steps, occupational activity, MET estimates, and intervention assignment. Self-reported exercise is particularly vulnerable to recall bias and social desirability bias. This makes the pooled effect less clean than it may appear.
Third, the effect sizes are small. A 0.03 SD lower Horvath EAA per SD higher physical activity is statistically significant but modest. The GrimAge result is more meaningful, but still observational. It may partly reflect the fact that GrimAge captures smoking, inflammation, plasma-protein-like signals, cardiometabolic risk, and mortality-related biology, all of which are correlated with an active lifestyle.
Fourth, residual confounding is a major issue. More physically active people often differ in diet, BMI, smoking, alcohol use, socioeconomic status, comorbidity burden, medication use, sleep, and frailty. The authors used minimally adjusted models to improve comparability, but that also leaves a real possibility of unmeasured confounding.
Fifth, the paper does not adequately resolve dose-response or exercise type. It cannot say whether resistance training, aerobic training, mixed activity, high-intensity activity, or simply avoiding sedentary behaviour is most important. Only one included study explicitly distinguished aerobic, resistance, and mixed physical activity, making modality-specific meta-analysis impossible.
Sixth, the possibility of non-linear effects is important. The authors note that some studies suggest U-shaped or J-shaped associations, meaning very high activity or elite athletic training might not simply produce “younger” DNAm profiles in a linear way. Standard meta-analysis of linear coefficients may therefore miss the true shape of the relationship.
Seventh, generalisability is limited. Many DNAm clocks were developed and validated mainly in people of European ancestry, and the authors note that their performance may be weaker in non-European populations.
Bottom line
This is a well-conducted and useful synthesis showing that physical activity is associated with lower epigenetic age, especially by GrimAge and, more weakly, Horvath. However, it does not prove that exercise reverses or slows biological ageing as measured by methylation clocks. The strongest interpretation is that physical activity tracks a healthier biological state, with GrimAge probably capturing part of that signal.
The most important next study would be a large randomised trial comparing aerobic, resistance, and combined exercise, with objective activity monitoring, repeated DNAm sampling, cell-composition adjustment, and multiple clocks including GrimAge2, DunedinPACE, DNAmFitAge, immune-age, and organ/system-specific clocks.