Epigenetic biomarkers show promise for longevity—but which clock moves which needle?

Led by Yale University researchers in collaboration with colleagues from Harvard University, UC-San Diego and epigenetic testing lab TruDiagnostics, the study curated TranslAGE, a harmonized database of longitudinal human intervention studies with blood-based DNA methylation data. These are patterns of chemical tags on DNA that have been associated in research with mortality risk, cardiovascular disease and other age-related outcomes that are then plugged into an epigenetic clock to estimate biological age.

The team calculated 16 widely used epigenetic clocks alongside 94 additional DNA methylation biomarkers to assess whether these tools respond consistently to interventions intended to influence aging biology.

“The central finding is that epigenetic aging biomarkers are not all measuring the same thing, and they do not respond equally to interventions,” said Raghav Seghal, faculty at Yale University and lead author on the study.

“Some biomarkers showed measurable change over time in response to interventions, while others were largely unchanged. This is important because the field has often treated ‘epigenetic age’ as a single construct. Our study shows that choice of biomarker matters greatly when evaluating whether an intervention may be influencing biology relevant to aging.”

Surrogate biomarkers for the leap of faith

A main issue with aging research is that assessing the impact of any sort of intervention on healthspan or lifespan outcomes requires decades-long trials or a certain leap of faith.

The concept of a DNA-methylation-based age predictor using saliva emerged from research in the early 2010, leading UCLA geneticist Steve Horvath to introduce the first epigenetic clock in 2013 that could estimate age across tissues. Since then, subsequent generations of epigenetic clocks increasingly sought to measure health, aging and mortality risk.

Although associations do not confirm that methylation itself leads to a longer, healthier life, epigenetic clocks are increasingly being used in exploratory studies and consumer-facing longevity testing in the healthy aging space to narrow the leap from theory to evidence.

“Aging biomarkers can potentially allow researchers to rapidly monitor the impact of an aging intervention without the need for decade-spanning trials,” the Yale research team wrote. “However, before the use of aging biomarkers, such as epigenetic clocks, as surrogate endpoints, their responsiveness to interventions that target aging must be tested.”

Rather than relying on any single trial or clock, they asked a broader question: Which measures are sensitive to change, under what conditions, and how consistently do those changes hold up across studies?

“That is a necessary step toward making biomarker-based intervention research more rigorous,” Sehgal said.

The Yale study found that newer, reliability-optimized epigenetic biomarkers—especially DunedinPACE, PCGrimAge, GrimAgeV2, PCPhenoAge and SystemsAge—responded more consistently to longevity interventions than older clocks trained mainly to predict chronological age, suggesting they may be better suited for future clinical trials.

DunedinPACE appeared broadly responsive across lifestyle studies, while GrimAgeV2 and PCGrimAge were particularly sensitive to drug interventions.

The study highlighted that future trials should match biomarker selection to the intervention, study population and biological mechanism. In this context, more “explainable” markers tied to systems such as inflammation, metabolism, kidney, lung and musculoskeletal function could help clarify which parts of aging biology are actually changing and reduce the risk of overclaiming from isolated clock changes.

“There is real value in these tools, but the field needs to be disciplined about interpretation,” Sehgal said. ”A biomarker can help us rapidly prioritize interventions for further study; it cannot, on its own, establish that an intervention slows aging. The goal is to reduce the gap between molecular measurements and clinically meaningful health.”

Interventions with the strongest signals, and without

Across the database, the research team observed quantifiable biomarker changes in several intervention categories, including dietary approaches, lifestyle changes, smoking cessation and pharmacologic interventions.

The Mediterranean diet and anti-TNF therapies that specifically target inflammation appeared among the more reproducible examples, with changes observed across related biomarkers and, in some cases, replicated across separate studies. Supplement and medical procedure categories showed less consistent effects.

For supplement-related interventions, the researchers included trials with with available longitudinal DNA-methylation data, including studies of vitamin D, omega-3 fatty acids and multinutrient formulations.

Within the past two years, post-hoc analysis of the three-year DO-Health trial tested vitamin D, omega-3 fatty acids and exercise, while the two-year COSMOS trial assessed a multivitamin-multimineral and a cocoa extract. Both studies found statistically significant but modest effects on newer-generation clocks.

While neither DO-Health or the COSMOS trials appear in the TranslAGE dataset due to data availability constraints, it did consider a selection of smaller, shorter-term studies testing other nutrient and ingredient combinations, reporting that they produced more mixed results. In one case, for example, data for the same intervention showed changes in the same biomarkers but in inconsistent directions.

“The appropriate conclusion is not that supplements do nothing, nor that a particular supplement is an anti-aging treatment,” Sehgal said. “The available trials are often small, short and heterogeneous, and different biomarkers can give different answers. At present, the evidence does not justify treating a change in one epigenetic clock as proof that a supplement will extend human healthspan or lifespan.”

Population health status also influenced responses. Several biomarkers, including PCPhenoAge, PCGrimAge and SystemsAge, showed larger decreases in disease populations, possibly reflecting greater biological disruption at baseline and more room for improvement. DunedinPace showed a more consistent response across health groups, suggesting potential utility changes in both the healthy and the diseased.

“The clearest signals were generally seen in interventions that produced substantial physiological change or were studied in populations with existing disease,” Sehgal said. “This does not mean that an intervention has been proven to slow aging or extend lifespan; it means that some biomarkers registered a measurable biological response.”

Capturing more structured, biologically meaningful change

The study highlighted that consistency across both biomarkers and sufficiently-powered studies are critical to distinguish genuine biological response from isolated statistical noise.

“The next step is to move beyond asking whether a biomarker changes and determine whether that change predicts meaningful improvements in people’s lives: lower disease risk, better physical and cognitive function, preserved independence and ultimately longer survival,” Sehgal said. “That requires larger randomized trials with repeated biomarker measurements and clinical outcomes collected together.”

Commenting on the study findings, David Furman, director of the AI and Bioinformatics Core at the Buck Institute for Research on Aging and head of the Stanford 1000 Immunomes Project, noted that the paper establishes that certain clocks move in a consistent statistically robust direction in response to intervention.

“That’s a necessary precondition for a surrogate endpoint, but the paper is explicit that it isn’t sufficient; they state plainly that whether short-term biomarker change corresponds to long-term gains in healthspan or lifespan remains unresolved,” he said. “There is no direct evidence linking change in an epigenetic clock to future outcomes.”

He and his team are working to advance the epigenetic clock model by factoring in real-world intrinsic capacity. Instead of using a molecular proxy and inferring a functional or clinical outcome, intrinsic capacity captures real-world functional ability across multiple measure of functional capacity that contribute to the risk of mortality and disability.

Last June, Furman and colleagues from the Buck Institute and IHU Toulouse published a paper in Nature Aging introducing the IC Clock to bridge molecular readouts of aging and clinical assessments of intrinsic capacity. It is trained on the clinical evaluation of cognition, locomotion, psychological well-being, sensory abilities and vitality of 1,014 individuals between the ages of 20 and 102 in the INSPIRE-T cohort.

“We’re testing whether IC trajectories track and predict the same hard endpoints these methylation clocks are being validated against but without the added inferential distance of ‘does moving a DNAm site actually move biology that matters’,” he explained.

The IC clock work is complemented by the ARPA-H PROSPR/THRIVE program, which integrates wearable, clinical and multi-omic data across some 20 large aging cohorts to test whether IC trajectories provide a more direct link between biological aging and meaningful health outcomes.

Regarding the responsiveness of pharmacological and structured lifestyle interventions identified in the Yale study, Furman said the data is consistent with what his team has seen in its real-world-data work linking foods, supplements and exercise to organ-level clocks (OrganAge and DiseaseAge).

“Supplements showing the weakest effect in this dataset tracks with what we generally see: Individual nutraceuticals rarely move composite biological age markers reliably at the population level,” he added.

“If I were to rank what currently has the most quantifiable, reproducible impact: (1) targeted anti-inflammatory pharmacology in appropriate populations, (2) sustained dietary pattern change (Mediterranean-style, not short courses), (3) exercise/lifestyle programs with measured functional outcomes and (4) supplements, real but currently the weakest and most heterogeneous signal.”

Source: Nature Medicine. doi: 10.1038/s41591-026-04562-9. “Responsiveness of epigenetic aging biomarkers to longevity interventions in humans”. Authors: Sehgal et al.