What actually moves an epigenetic clock, and what does not

By Marcus Renshaw ·

If you have bought a biological age test, the obvious next question is what to do about the result. The market has an answer ready, and it is usually a supplement. The literature has an answer too, and it is less flattering: in 2026 two separate reviews scored dozens of interventions against next-generation epigenetic clocks, and the things with the loudest longevity marketing are concentrated in the column marked no detectable effect. This is the companion question to what the clocks actually measure. Not what the number means, but what moves it.

Across two 2026 reviews, exercise, a plant-rich diet, caloric restriction, omega-3, semaglutide, a multivitamin and pitavastatin all lowered next-generation epigenetic clock readings in human studies. Nicotinamide riboside, rapamycin and senolytics showed no detectable effect. Plasmapheresis made clocks worse. Effect sizes are small and the endpoint is a surrogate.

What do we actually know about moving a clock?

More than you would guess from how the field talks, and the reason is one piece of unglamorous work. A group at Yale, with co-authors from the testing company TruDiagnostic, built a harmonized database of 51 longitudinal interventional studies called TranslAGE and recalculated a consistent set of 16 epigenetic clocks, plus 94 other DNA methylation biomarkers, for every one of them. That paper circulated as a preprint from 2024 and was published in Nature Medicine in August 2026.

Harmonization is the whole contribution. Before it, every study used whichever clock its authors preferred, so “this intervention moved GrimAge” and “that one did not move Horvath” were not comparable statements. Recalculating all 16 clocks on all 51 studies makes them comparable for the first time.

Three findings from it shape how to read everything below. Clocks trained to predict mortality or pace of aging respond most strongly across interventions, and agree with one another, while clocks trained to predict chronological age respond less. Pharmacological and lifestyle interventions produce the strongest responses. And the characteristics of the study population and the duration of the study are themselves major determinants of whether anything shows up, which means a null result in a short study of healthy young people is weak evidence of nothing happening.

Which interventions have lowered epigenetic age in humans?

The second 2026 review went at the question from the other direction. Rather than harmonizing clocks, it catalogued outcomes: 41 human studies reporting the effect of an intervention on at least one next-generation clock, sorted into what went down, what did nothing, and what went up.

CategoryWhat the review placed there
Decreased epigenetic ageExercise, a plant-rich diet, caloric restriction, omega-3 fatty acids, a multivitamin-multimineral supplement, the GLP-1 agonist semaglutide, the statin pitavastatin, ketamine, umbilical cord plasma
No detectable effectNicotinamide riboside, rapamycin, senolytics, and several others
Accelerated epigenetic agingPlasmapheresis, among other therapeutics

Read that table as a list of categories, not a ranking. Membership in the first column does not mean equal evidence: exercise and a plant-rich diet are supported by far more human data than ketamine or umbilical cord plasma, and lumping them together is the table’s limitation rather than a claim. What the table does establish is direction, and where the burden of proof currently sits.

The awkward pattern is hard to miss. The top column is mostly free or cheap and mostly unglamorous. The middle column contains three of the most heavily marketed interventions in consumer longevity.

What does it mean that rapamycin and NR showed nothing?

Less than the enthusiasts fear and less than the skeptics hope. “No detectable effect” is a weaker claim than “does not work”, and it is weaker in a specific way: these are clock readings, not outcomes. An intervention could plausibly extend healthy life without moving a methylation pattern, and could move a methylation pattern without extending anything.

But it is not nothing either, and for rapamycin in particular it lines up with everything else. The best human trial, PEARL, missed its primary endpoint, and the fastest instruments available do not show it slowing aging. When the cheap screening tool and the expensive trial agree, that is worth more than either alone.

For nicotinamide riboside and senolytics the finding lands harder, because clock movement is close to the only human evidence those categories have been able to point at. Take that away and the case rests on mechanism and mouse data.

How big are these effects, actually?

Small enough that the size matters more than the direction, and there is one trial clean enough to show you exactly how small.

A prespecified ancillary study inside the COSMOS trial randomized 958 older adults to a daily supermarket multivitamin or placebo for two years and measured five clocks. The multivitamin slowed two of them: the between group difference in yearly change was -0.113 years for PCGrimAge (95% CI -0.205 to -0.020, P = 0.017) and -0.214 years for PCPhenoAge (-0.410 to -0.019, P = 0.032), where negative means the clock rose more slowly on the multivitamin. The effect was larger in participants whose biological aging was already running fast. Cocoa extract, tested in the same trial on the same people, moved none of the five.

So a statistically significant result from a well designed two year randomized trial amounts to roughly a tenth of a year per year on a biological age estimate. The authors say so themselves: the effects are small, and whether they are clinically relevant is undetermined. That is the honest scale of the best evidence in this field, and it is the number to hold in mind whenever someone shows you a dramatic before and after.

One guard on reading the two reviews together. You cannot line up effect sizes across these studies as if they came from one experiment. “A multivitamin beat rapamycin” is a comparison between different trials in different populations over different durations using different clocks, and the harmonization paper’s own finding is that population and duration drive responsiveness. The defensible statement is about direction and about which claims have any human clock evidence at all, not about rank order.

Who is telling you this?

Both reviews come from people with a commercial stake in epigenetic testing, and this is worth stating plainly because it cuts in different directions in each case.

The harmonization paper’s declared interests go further than consulting. Three co-authors are employees of TruDiagnostic, which sells epigenetic age tests, and both the first and the senior author take consulting fees from it. More to the point, those same two are named co-inventors of a patented epigenetic clock of their own. A paper establishing that clocks are responsive to interventions, and ranking which clocks respond best, is good for anyone who sells or owns one. That is not an accusation of bad faith, and the work is careful. It is a reason to read the framing, and especially the ranking, with the ownership in view.

The catalogue review is written by an employee of Tally Health and by one of its founders, and Tally Health also sells epigenetic age tests. Here the incentive points the other way: their review’s most quotable finding is that three popular longevity interventions do nothing measurable, which is not a commercially convenient thing to publish. A finding against interest deserves more weight, not less.

And the COSMOS trial is the cleanest of the three on this dimension, being an academic randomized trial of a supermarket multivitamin with no clock vendor in the author list.

None of this means the work is wrong. It means the field’s evidence base is largely generated by its vendors, which is a reason to prefer the results that embarrass them.

So is a clock worth tracking?

Conditionally, and the conditions do most of the work.

If you track one, the harmonization paper argues for using a clock trained on mortality or pace of aging rather than one trained on chronological age, since those are the ones that respond and agree. Space your retests far enough apart that a real change can outrun the measurement noise, which for these clocks means a year rather than a quarter, and treat the trend across several tests as the signal rather than any single reading. The reliability problem with single clock readings has not gone away just because the responsiveness question got answered.

The self-experimenter version is a budget question. A clock test costs real money, moves by tenths of a year under the best studied interventions, and reports a surrogate rather than an outcome. The same money spent on ApoB and Lp(a) buys you numbers that feed directly into a treatment decision with actual outcome trials behind it. That is not an argument against curiosity. It is an argument about ordering, and the same ordering the rest of this site applies: the markers that change what you do come before the markers that tell you a story about yourself.

It is also worth noticing what the top column of that table is: exercise, diet, caloric restriction, omega-3. The interventions with the most consistent effect on the clocks are the ones you would be doing anyway, and none of them requires a test to justify. If the clock is what finally makes somebody train and eat differently, it has earned its price. If it is a scoreboard for a supplement stack, the two reviews above suggest the scoreboard is mostly reading the training and the diet.

The obvious missing study is a trial that moves a clock and then follows people long enough to see whether anything happened to them. Nobody has run it, which means every claim in this area, including the encouraging ones, is still a claim about a marker.

FAQ

What lowers epigenetic age?

In human studies catalogued by a 2026 review of 41 trials, exercise, a plant-rich diet, caloric restriction, omega-3 fatty acids, a multivitamin-multimineral supplement, semaglutide, pitavastatin, ketamine and umbilical cord plasma were all associated with decreased next-generation clock readings. The evidence behind those is very uneven, with exercise and diet far better supported than the exotic entries, and all of it measures a marker rather than an outcome.

Does rapamycin slow epigenetic aging?

Not detectably. A 2026 review of 41 human studies placed rapamycin among the interventions with no detectable effect on next-generation epigenetic clocks, alongside nicotinamide riboside and senolytics. That is a weaker statement than “it does not work”, since clocks are surrogate markers, but it lines up with the best human rapamycin trial missing its primary endpoint.

Does NAD or nicotinamide riboside affect epigenetic age?

Nicotinamide riboside showed no detectable effect on next-generation clocks in that review. That matters more for NR than for some other interventions, because clock movement is close to the only human evidence the category has been able to point to.

Does a multivitamin really slow aging?

A well designed trial found a small effect on a surrogate marker, which is not the same claim. In the COSMOS ancillary study, 958 older adults took a daily multivitamin or placebo for two years, and two of five epigenetic clocks rose more slowly on the multivitamin, by roughly a tenth to a fifth of a year per year. Cocoa extract in the same trial did nothing. The authors state the effects are small and of undetermined clinical relevance.

Which epigenetic clock responds best to interventions?

Clocks trained to predict mortality or pace of aging, according to the harmonized analysis of 51 interventional studies across 16 clocks. Those also agreed with one another, while clocks trained to predict chronological age responded less. Clocks that break into multiple subscores gave more mechanistic detail than single score clocks.

Can anything make your epigenetic age worse?

Yes. The same review found plasmapheresis, among other therapeutics, accelerated epigenetic aging measures. That is a useful reminder that these instruments are not a one way scoreboard that only rewards intervention.

Are these studies trustworthy given who funded them?

Read them knowing the field’s evidence is largely produced by its vendors. The harmonization paper includes TruDiagnostic employees and a consulting senior author, and the catalogue review was written by a Tally Health employee and founder. Both companies sell epigenetic age tests. The catalogue review’s central finding is unflattering to popular products, which makes it a finding against interest and worth more weight. The COSMOS multivitamin trial had no test vendor among its authors.

Should I spend money on a clock test or on bloodwork?

Bloodwork first, on the evidence. Clock readings move by tenths of a year under the best studied interventions and report a surrogate, while markers like ApoB and Lp(a) feed into treatment decisions that have real outcome trials behind them. Track a clock if the curiosity is worth the cost to you, but order it after the markers that change what you actually do.

Sources