Epigenetic Clocks in 2026: Horvath, GrimAge, DunedinPACE Compared

Epigenetic clocks have become the default surrogate endpoint for longevity trials. If you want to claim that a drug, peptide, or lifestyle intervention "reverses biological age," you almost certainly measured it with a DNA methylation clock — likely Horvath, PhenoAge, GrimAge, or DunedinPACE. In 2026, the clock landscape has matured considerably since Steve Horvath's 2013 paper, but so has the skepticism. This article is a clear-eyed 2026 update: what each major clock measures, which ones actually predict mortality, where they fail, and how they are being used in real interventional trials right now.

If you have read our foundational piece on epigenetic clocks and biological age, this article extends that primer into 2026-specific territory: second- and third-generation clocks, the pace-of-aging framework, the causal-vs-correlational debate, and the reproducibility controversies of the past three years.

What Is an Epigenetic Clock?

An epigenetic clock is a mathematical model — usually a weighted linear combination of DNA methylation levels at specific CpG sites in the genome — that produces an estimate of "biological age" from a blood, saliva, or tissue sample. The inputs are percent-methylation values at hundreds of CpG sites measured on DNA methylation arrays (originally the Illumina 27K, then 450K, then EPIC 850K, and now EPIC v2). The output is a single number in years, intended to represent how old your body looks methylation-wise, independent of calendar age.

The rationale: DNA methylation patterns drift with age in highly reproducible ways. Some CpG sites gain methylation; others lose it. These changes correlate tightly with chronological age across tissues and across individuals. If you can train a model to predict chronological age from methylation, you can then ask whether the residual — the gap between predicted and actual age — predicts disease, mortality, or response to interventions. That residual is the signal longevity researchers care about.

The Science: Generations of Clocks

First-generation clocks (chronological-age predictors).

• Horvath 2013 (Genome Biology). The foundational multi-tissue clock using 353 CpG sites, trained on >8,000 samples across 51 tissue types. Impressive cross-tissue accuracy. Predicts chronological age tightly but is a relatively weak predictor of mortality beyond age.
• Hannum 2013 (Molecular Cell). Blood-specific clock using 71 CpGs. Similar accuracy to Horvath in blood.

These clocks were revolutionary for proving that DNA methylation could "read" age, but because they were trained to predict calendar age, their residuals captured only noise plus whatever mortality signal leaked through.

Second-generation clocks (phenotype/mortality-trained).

• PhenoAge (Levine et al. 2018, Aging). Trained not on chronological age but on a composite phenotypic age derived from clinical biomarkers (albumin, creatinine, glucose, CRP, lymphocyte percent, etc.) that themselves predict mortality. PhenoAge acceleration is a stronger predictor of all-cause mortality than Horvath.
• GrimAge (Lu et al. 2019, Aging). Trained on a composite of time-to-death plus methylation-based surrogates of seven plasma proteins and smoking pack-years. GrimAge is currently the single strongest methylation-based predictor of mortality, cardiovascular events, cancer incidence, and time to first major chronic disease. An updated GrimAge2 (2022) added additional protein surrogates.

Second-generation clocks changed the field by optimizing for the thing we actually care about — longevity — rather than calendar age.

Third-generation clocks (pace-of-aging).

• DunedinPoAm (Belsky et al. 2020) and DunedinPACE (Belsky et al. 2022, eLife). Built from the Dunedin Study birth cohort. Instead of estimating a static biological age, DunedinPACE estimates your rate of aging at the moment of sampling — years of biological aging per calendar year. A healthy person might score 0.9; an unhealthy one 1.2. Pace-of-aging reframes the question from "how old are you?" to "how fast are you aging right now?" — which is more responsive to interventions, because even a brief slowdown can show up in the pace signal without waiting for the accumulated gap to change.

Next-generation work. Moqri et al. (2023, Cell) reviewed the emerging landscape of "causal" epigenetic clocks built on CpGs that appear mechanistically linked to aging rather than merely correlated. Ying et al. developed causality-enriched clocks. Several groups have published principal-component-based clocks (PC clocks) that explicitly address the noise problem of individual CpG measurement.

The Evidence: Clocks in Real Interventional Trials

Clocks are useful only if they respond to interventions. Here's where they've been tested:

Fahy et al. 2019 (Aging Cell) — TRIIM. A one-year trial of recombinant human growth hormone plus DHEA and metformin in nine men produced a roughly 2.5-year reduction in GrimAge estimates. Famous, controversial, and with a tiny sample — but it was the first human trial to show a methylation clock responding to a treatment.

Waziry et al. 2023 (Nature Aging) — CALERIE. The CALERIE trial of caloric restriction in healthy adults (originally designed to test other endpoints) was reanalyzed for epigenetic outcomes. DunedinPACE — but not Horvath or GrimAge — showed a slowing of pace-of-aging in the caloric restriction group. This was the first well-controlled RCT to demonstrate an intervention affecting a methylation clock.

Sinclair lab and partial reprogramming. In animal models, transient expression of Yamanaka factors reduces epigenetic age dramatically. Translating to humans is harder, but the underlying biology links David Sinclair's longevity research and the Yamanaka factor partial reprogramming field directly to clock-based readouts.

Senolytic and GlyNAC exploratory signals. Small, mostly exploratory analyses have reported directional improvements in various clocks after senolytic or amino-acid interventions. Most are underpowered.

Consumer interventional signals. Programs like Bryan Johnson's Blueprint and various precision-medicine clinics have self-reported DunedinPACE and GrimAge improvements on intensive lifestyle protocols — a signal consistent with the CALERIE findings, though selection and expectation effects matter.

The emerging consensus: DunedinPACE responds to interventions most rapidly; GrimAge responds to longer or larger interventions; first-generation Horvath barely budges because it was never optimized for that.

Current Clinical Status: Who Sells Clocks and Who Uses Them

The epigenetic clock ecosystem in 2026 includes:

• TruDiagnostic (TruAge) — the most widely used consumer and research clock service, reporting Horvath, PhenoAge, GrimAge, DunedinPACE, and proprietary derivatives. Works with clinicians and several large longevity studies.
• Elysium Health (Index) — saliva-based consumer clock using a proprietary blend.
• Thorne Biological Age — home-collection consumer kits.
• MyDNAge — one of the earlier consumer clocks, Horvath-derived.
• Clock Foundation / Horvath lab — academic reference clocks, GrimAge licensing, ongoing clock development.
• FOXO Technologies and others — specialized applications (life insurance underwriting, wellness).

On the trial side, NIH-funded studies, longevity-focused biotechs, and large observational cohorts (UK Biobank, MESA, Framingham) now routinely include methylation clock assays. Clocks are increasingly accepted as exploratory secondary endpoints in intervention trials, though the FDA has not accepted any clock as a validated surrogate endpoint for drug approval.

Connection to Gene Editing & Peptides

Clocks matter for gene editing and peptide therapeutics because they are how we'll measure whether rejuvenation actually worked.

• Reprogramming readouts. Any clinical attempt at partial reprogramming with Yamanaka factors will be judged in part by its effect on methylation clocks. This is already how preclinical reprogramming studies are scored.
• Gene therapy endpoints. Companies exploring longevity-oriented gene therapies (follistatin, klotho, FOXO3) will look to clocks as supporting evidence.
• Peptide trials. As peptides for longevity enter more rigorous trials, DunedinPACE in particular is the clock most likely to pick up modest intervention signals.
• Editing aging genes directly. Hypothetically, base editing could alter loci that influence methylation patterns — though this remains speculative and is not on anyone's near-term roadmap.

The bigger point: without reliable biomarkers of biological age, longevity trials would have to wait decades for hard endpoints. Clocks, whatever their flaws, make the field empirically tractable.

Limitations

The honest 2026 picture includes several caveats longevity enthusiasts often gloss over:

1. Test-retest noise. Individual CpG methylation measurements are noisy. For some clocks, test-retest variability is several years on a single sample taken twice. PC-based clocks and DunedinPACE improve this, but noise remains a concern for individual-level decisions.
2. Tissue specificity. Blood is convenient but not the same as liver, brain, or muscle. Tissue-specific methylation patterns differ, and blood clocks may miss organ-specific aging.
3. Ethnic calibration. Many clocks were trained on predominantly European ancestry cohorts. Performance in non-European populations can differ systematically.
4. Diet and cell composition artifacts. Short-term dietary changes can alter methylation and immune cell proportions in blood, producing transient "age" changes that don't reflect real aging.
5. Correlation vs causation. A shift in a clock estimate is not the same as a shift in biological aging. Chen et al. and other methylation researchers have argued that some clock "reversal" reports reflect model artifacts rather than rejuvenation.
6. Commercial incentives. Clock-selling companies have financial interest in visible clock movements, creating subtle pressures in how results are reported.
7. Within-person change is small relative to between-person variation. The absolute clock changes produced by real interventions are often smaller than baseline variability between individuals.

FAQ

Which clock is the most accurate?

"Accuracy" depends on what you mean. For predicting chronological age: Horvath. For predicting mortality: GrimAge (and GrimAge2). For detecting intervention effects quickly: DunedinPACE. Most rigorous researchers report several clocks simultaneously.

Should I buy a consumer epigenetic age test?

It depends on your purpose. For individual health decisions, the noise and interpretation uncertainty limit usefulness. For tracking changes over years on a protocol, serial testing with the same provider can be informative — but expect noise.

Has any drug been approved based on an epigenetic clock?

No. The FDA has not accepted any clock as a validated surrogate endpoint for approval. They remain exploratory endpoints in trials.

How much does biological age actually move with intervention?

Well-controlled trials typically show sub-year to low-single-digit shifts in clock estimates over months to years. Claims of "eight years reversed" usually come from open-label or observational studies using non-validated clocks.

Is DunedinPACE better than GrimAge?

They measure different things. DunedinPACE measures rate of aging; GrimAge measures accumulated age-related risk. Both are useful; neither replaces the other.

Are epigenetic clocks the same as telomere testing?

No. Telomere length is a different biomarker covered in our telomeres and aging article. Telomeres and methylation clocks correlate modestly but measure different biology.

Further Learning

• Epigenetic Clocks and Biological Age — the foundational primer this article updates
• Hallmarks of Aging Explained — where epigenetic alterations fit in the framework
• Telomeres and Aging — a complementary biological age biomarker