Epigenetic clocks

Epigenetic clocks measure biological age by reading chemical tags on your DNA, called methylation marks, at hundreds of specific sites across the genome. The premise is exciting: instead of waiting decades to see whether a habit or intervention extends your life, you get a number today that might tell you how fast you're aging. Whether that number means what you hope it means is a harder question.

Four generations, different purposes

The original Horvath clock, published in 2013, tracks 353 methylation sites and was trained to predict chronological age [1]. It's accurate for that purpose and not much else. The second-generation PhenoAge clock, developed by Morgan Levine in 2018, was instead trained on clinical phenotype and mortality data. This makes it a stronger signal for health outcomes [2]. GrimAge, from Lu in 2019, incorporated plasma protein surrogates and smoking history and currently holds the best record for predicting all-cause mortality [3]. The most recent tool, DunedinPACE, works differently. Rather than estimating a static biological age, it quantifies the pace at which organ systems are declining over time, based on 20 years of tracked data in a cohort study [4]. For measuring whether something is changing, DunedinPACE is the most reliable of the four.

Interventions do move these clocks

Three well-designed studies found measurable effects. CALERIE, a two-year caloric restriction trial in humans, slowed DunedinPACE [5]. DO-HEALTH, a large European trial testing 1 g daily omega-3, slowed PhenoAge, GrimAge2, and DunedinPACE [6]. In mice, Sinclair's OSK reprogramming work reset clock readings entirely [7]. These are meaningful findings. They also share a critical limitation: clock movement is a surrogate outcome, not a direct measure of lifespan. The open question of whether a younger clock reading translates to a longer or healthier life remains unanswered.

The reliability floor is lower than most people expect

First-generation clocks carry technical noise that can swing readings by up to nine years between replicates of the exact same sample [8]. A single consumer test, run once, sits inside that noise band. It can't reliably tell you whether you're biologically 42 or 51. Repeat testing using the same platform, under the same conditions, narrows the window, but the variance is a genuine constraint on individual interpretation.

What the inventors themselves say

Neither Horvath nor Levine oversells what these clocks can do. Levine has noted there is "still debate whether it's causal... causal versus correlative" [9]. Horvath has asked directly: "we don't have many interventions that reverse the epigenetic aging rate, but do we know that reversing it helps?" [10]. These are the people who built the tools. Their caution deserves attention.

Peter Attia's position is similar: he treats epigenetic clocks as a research instrument, not something to steer clinical or personal decisions by [11].

Where the evidence is debated

The central dispute is whether these clocks measure a cause of aging or correlate with processes that cause aging. If methylation patterns are downstream noise, then reversing them would be cosmetic, like painting over rust. If they're upstream regulators, as the information theory of aging framework proposes, resetting them could slow deterioration.

Current evidence doesn't resolve this. The intervention studies [6, 5, 7] show that clocks move in response to things that appear to slow aging by other measures, which is consistent with a causal role. Correlation between two things that both respond to the same underlying biology is also exactly what a non-causal surrogate would produce. Horvath and Levine both acknowledge this gap [10, 9]. Until trials can show that patients whose clocks slow live longer or stay healthier in hard endpoint terms, clock readings remain a promising signal without confirmed meaning for individuals.

Functional measures, grip strength, balance, cardiovascular fitness, and sit-to-rise score, have stronger direct mortality validation than any consumer DNAm test. Clocks are a scientific frontier. Treat them that way.

References

1. Horvath 2013, Genome Biol — pan-tissue DNAm clock · PubMed 24138928
2. Levine et al. 2018, Aging — DNAm PhenoAge · PubMed 29676998
3. Lu et al. 2019, Aging — DNAm GrimAge · PubMed 30669119
4. Belsky et al. 2022, eLife — DunedinPACE · PubMed 35029144
5. Kraus et al. 2019, Lancet D&E (CALERIE) — 2yr caloric restriction · PubMed 31303390
6. DO-HEALTH 2025, Nature Aging — omega-3 & DNAm aging clocks · PubMed 39730768
7. Lu et al. 2020, Nature — OSK reprogramming restores vision (mouse) · PubMed 33268865
8. Higgins-Chen et al. 2022, Nat Aging — clock reliability (PC-clocks) · PubMed 36277076
9. raw/foundmyfitness/fmf-levine-phenoage-age-acceleration.txt
10. FoundMyFitness — Steve Horvath: epigenetic aging, PhenoAge & GrimAge
11. Attia #290 — liquid biopsies & the role of epigenetics in aging