# Biological age clocks

Rating: Proven — Replicated results in human randomized trials
Type: Measure
Updated: 2026-10-14
Canonical: https://heretic.bio/articles/biological-age-clocks-measurement-and-interpretation.html

> Epigenetic clocks are regression algorithms that read age-related DNA-methylation drift and output a biological age or aging rate. Three large 2025–2026 analyses, an 18,859-person head-to-head, a 24-year longitudinal study, and a NHANES mortality analysis, confirmed GrimAge-family and DunedinPACE as genuine risk stratifiers. The caveats: clocks disagree with each other, single measurements are noisy, and no intervention has yet proven that moving a clock changes lifespan.

> DNA-methylation "clocks" are the de facto surrogate endpoints of longevity medicine — used to price DTC tests, market interventions, and adjudicate trials — yet their validation picture changed materially between Dec 2025 and Mar 2026. Large head-to-head and longitudinal studies confirmed GrimAge-family and DunedinPACE as genuinely predictive of disease and mortality, while simultaneously exposing their limits (single timepoints, assay noise, clock disagreement, zero clinical-grade standardization). **Trajectory beats snapshot; nothing about a clock reading is destiny.**

## What Clocks Actually Measure

Epigenetic clocks are regression algorithms that estimate biological age (or aging rate) from DNA methylation at hundreds of CpG sites in blood/saliva. Methylation drift is systematic with age — the clock reads the drift. Different clock generations answer **different questions** :

Generation | Clocks | Question answered | Analogy  
---|---|---|---  
1st (2013–14) | Horvath, Hannum | "How old does methylation say you are?" | Odometer (chronological fit)  
2nd (2018–19) | **GrimAge** (v1/v2), PhenoAge | "How much mortality/disease risk is in your methylation?" | Damage gauge (trained on outcomes)  
3rd (2020–22) | DunedinPoAm → **DunedinPACE** | "How fast are you aging right now?" | Speedometer (longitudinal rate)  
4th / multi-omic + retroelement (2025–26) | OMICmAge, RetroAge, DNAmEMRAge, EpiAge-R concept | "Integrate omics + EMR + retroelements into one systems readout; add resilience" | Fleet telemetry (in development)  
  
**Design notes that matter for interpretation:** \- GrimAge trains on **DNAm surrogates of plasma proteins + smoking pack-years** — it captures cumulative damage including smoking history (predictive even in never-smokers via protein surrogates). Latest version: GrimAge V2 (_Aging_ 14:9484–9549, 2022; DOI [10.18632/aging.204434](https://doi.org/10.18632/aging.204434)). Original: Lu et al., _Aging_ 11:303–27, 2019 (DOI [10.18632/aging.101684](https://doi.org/10.18632/aging.101684)). \- PhenoAge was built from clinical chemistry biomarkers, then methylation-mapped (Levine et al., _Aging_ 10:573–91, 2018; DOI [10.18632/aging.101414](https://doi.org/10.18632/aging.101414)). \- DunedinPACE was trained on **observed 12-year change in 18 organ-function biomarkers** in the Dunedin birth cohort (Belsky et al., _eLife_ 11:e73420, 2022; DOI [10.7554/eLife.73420](https://doi.org/10.7554/eLife.73420); predecessor DunedinPoAm, _eLife_ 2020; DOI [10.7554/eLife.54870](https://doi.org/10.7554/eLife.54870)). Score of 1.0 = cohort-average pace; >1 = faster aging. It has the **highest test-retest reliability** of any clock — the reason it dominates intervention trials.

## Validation Status: The Dec 2025 – Mar 2026 Trilogy

Three Nature-portfolio analyses landed in one quarter, collectively the strongest validation — and the clearest caution — to date:

  1. **Head-to-head (Generation Scotland, n=18,859):** Mavrommatis et al. compared **14 clocks against 174 incident disease outcomes** \+ all-cause mortality. 176 disease associations across 13 clocks; **GrimAge v2 and DunedinPACE showed the largest effect sizes** ; respiratory and liver outcomes were the strongest predictive domains; e.g., DunedinPACE HR 1.44 for incident diabetes (_Nat Commun_ 16, 2025; PMID [41402269](https://pubmed.ncbi.nlm.nih.gov/41402269); [PMC12708718](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12708718/)).
  2. **Longitudinal trajectories (InCHIANTI, n=699, up to 24-year follow-up):** Kuo et al. (2026) showed that adding **rate-of-change to baseline** GrimAge/GrimAgeV2/PhenoAge/DunedinPACE improved mortality prediction to C-statistics of **0.800–0.808** — measurable discrimination gains from modeling trajectories rather than snapshots.
  3. **Population mortality (NHANES, n=2,105, median 17.5-year follow-up, 998 deaths):** among all clocks examined, **GrimAge epigenetic age acceleration was the most statistically significant predictor of mortality** (p<0.0001) after standard covariate adjustment (_Geroscience_ , 2025).



**Consensus reading:** 2nd/3rd-generation clocks are real risk stratifiers — comparable to or better than conventional risk factors for several outcomes. They are **not** deterministic organ age, they disagree with each other (partially non-overlapping variance), and single readings carry assay noise that marketing omits.

### The frontier: multi-omic and resilience

  * **OMICmAge** integrates multi-omics with electronic medical records into a single biological-age readout; outperformed prior methylation-only clocks on precision and 5-/10-year mortality discrimination in head-to-head analysis (_Nat Aging_ , 2026; DOI [10.1038/s43587-026-01073-7](https://doi.org/10.1038/s43587-026-01073-7)).
  * **EpiAge-R framework** (Yamada, _Int J Mol Sci_ 27:1908, 2026; PMID [41752045](https://pubmed.ncbi.nlm.nih.gov/41752045), [PMC12940517](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12940517/)): clocks are odometers or speedometers but measure **no resilience** — the capacity to recover homeostasis after stress. Two people with identical GrimAge and DunedinPACE can diverge sharply after surgery or infection.
  * **AmiAge — amino-acid biological-age clock** (_Nat Commun_ 17:6733, 2026; DOI [10.1038/s41467-026-73371-y](https://doi.org/10.1038/s41467-026-73371-y)): Random Forest regressor trained on plasma concentrations of **18 amino acids** (n = 11,000 in-house + 270,000+ public samples, ages 1–89). The **AmiAge Gap** correlates with frailty index, telomere length, age-related disease incidence, and mortality. Distilled to an **8-amino-acid clinical panel** : alanine, glutamine, glycine, histidine, leucine, phenylalanine, tyrosine, valine. Notably taurine is one of the 18 input features but drops out of the distilled panel — consistent with the de Cabo 2025 + Marcangeli 2025 finding that taurine alone is **not** a primary aging biomarker; its signal is redundant with BCAAs at the population level. Implication: **panel >> single amino acid** for biological-age tracking; commercial metabolomics-grade amino acid panels are the natural extension of methylation clocks for users who can't access venipuncture-based assays. See [Taurine - The Aging Amino Acid](taurine-the-aging-amino-acid.html) for the full biomarker debate.



## What Has and Hasn't Moved Clocks (Honest Intervention Ledger)

Intervention | Clock result | Read  
---|---|---  
GLP-1 agonist (semaglutide, 32 wk Ph2b RCT, [NCT04019197](https://clinicaltrials.gov/study/NCT04019197), n=84 PWH) | DunedinPACE **−9%** (p=0.01); PhenoAge −4.9 yr (p=0.004); PCGrimAge −3.1 yr (p=0.007); GrimAgeV2 −2.3 yr (p=0.009); **OMICmAge −2.2 yr (p=0.009)** ; **RetroAge −2.2 yr (p=0.030)** | **Best RCT clock data to date** — 5/5 clocks significant across 1st–4th gen. Liver-fat responders drove the signal — see [GLP-1 Peptides (Semaglutide, Tirzepatide, Retatrutide)](glp-1-peptides-semaglutide-tirzepatide-retatrutide.html). Corley MJ et al., _Nat Commun_ 17:6606, 2026. PMID [42156721](https://pubmed.ncbi.nlm.nih.gov/42156721/) · [PMC13381876](https://pmc.ncbi.nlm.nih.gov/articles/PMC13381876/).  
Rapamycin (PEARL, TruAge clocks, n=24) | **Null** (mixed, non-significant) | Absence of evidence ≠ evidence of absence; assay + n issues — see [Rapamycin mTOR Inhibition - PEARL Trial Evidence](rapamycin-mtor-inhibition-pearl-trial-evidence-and-clinical-translation.html)  
Therapeutic plasma exchange (first human RCT 2025) | Mixed across 36 clocks | Signal, not proof — see [Therapeutic Plasma Exchange (TPE)](therapeutic-plasma-exchange-tpe.html)  
Caloric restriction (CALERIE) | DunedinPACE slowed vs. control | Slow aging rate with sustained CR  
Taurine ([NCT06613542](https://clinicaltrials.gov/study/NCT06613542)) | Pending — first taurine RCT with clock endpoints | See [Taurine - The Aging Amino Acid](taurine-the-aging-amino-acid.html)  
Most supplements | No credible clock data | Marketing claims unsupported  
  
⚠️ **Surrogate-endpoint caveat:** no intervention has yet shown that changing a clock score changes lifespan or healthspan. Clocks are the best available intermediate endpoint, not proof of slowed aging.

## Practical Measurement Guide

**Which test:** \- **Rate question ("am I aging fast?")** → DunedinPACE-based panel (e.g., TruDiagnostic, Tally Health DTC offerings). \- **Risk question ("accumulated damage?")** → GrimAge V2 / PCGrimAge. \- Organ-specific methylation clocks (brain, heart, kidney ages) exist; treat as exploratory.

**Interpretation rules:** 1\. **Trajectory > snapshot.** Retest at 6–12-month intervals; one reading is noise-prone. 2\. **Compare like assays only.** Different labs/platforms are not interchangeable; stay with one provider for longitudinal tracking. 3\. **Effect sizes in trials are small** (single-digit % pace changes). A DTC report claiming years of reversal from a supplement protocol is marketing. 4\. **Anchor to functional markers.** Pair with VO2max, DEXA, ApoB, HbA1c, hs-CRP (see [Overview - Longevity Science](longevity-science-overview.html) biomarker table). 5\. **No regulatory clearance:** as of 2026 no clock is an FDA-cleared in-vitro diagnostic; results are wellness-grade.

## Sources

1. Lu AT, et al. DNA methylation GrimAge version 2. Aging 14(23):9484–9549, 2022. DOI 10.18632/aging.204434 . (DOI 10.18632/aging.204434)
2. Lu AT, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging 11(2):303–27, 2019. DOI 10.18632/aging.101684 . (DOI 10.18632/aging.101684)
3. Levine ME, et al. An epigenetic biomarker of aging for lifespan and healthspan (PhenoAge). Aging 10(4):573–91, 2018. DOI 10.18632/aging.101414 . (DOI 10.18632/aging.101414)
4. Belsky DW, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife 11:e73420, 2022. DOI 10.7554/eLife.73420 . (DOI 10.7554/eLife.73420)
5. Belsky DW, et al. DunedinPoAm, quantification of the pace of biological aging. eLife 2020. DOI 10.7554/eLife.54870 . (DOI 10.7554/eLife.54870)
6. Mavrommatis C, et al. An unbiased comparison of 14 epigenetic clocks in relation to 174 incident disease outcomes. Nat Commun 16, 2025. PMID 41402269 . (PubMed 41402269)
7. PMC12708718 (PMC12708718)
8. OMICmAge quantifies biological age by integrating multi-omics with EMRs. Nat Aging , 2026. DOI 10.1038/s43587-026-01073-7 . (DOI 10.1038/s43587-026-01073-7)
9. Yamada H. Epigenetic Clocks, Resilience, and Multi-Omics Ageing: the EpiAge-R Conceptual Framework. Int J Mol Sci 27(4):1908, 2026. PMID 41752045 . (PubMed 41752045)
10. PMC12940517 (PMC12940517)
11. AmiAge — amino-acid biological age clock. Nat Commun 17:6733, 2026. DOI 10.1038/s41467-026-73371-y . Random Forest on 18 plasma amino acids (n>281,000); distilled 8-AA clinical panel (alanine, glutamine, glycine, histidine, leucine, phenylalanine, tyrosine, valine); taurine in full panel but drops out of distilled 8 — consistent with de Cabo 2025 + Marcangeli 2025 single-marker null. (DOI 10.1038/s41467-026-73371-y)
12. Corley MJ, Dwaraka VB, Pang APS, et al. "Semaglutide slows epigenetic aging in a randomized trial of HIV-associated lipohypertrophy." Nat Commun 17:6606, 2026. PMID 42156721 · PMC13381876 · DOI 10.1038/s41467-026-72861-3 . Post-hoc analysis of NCT04019197 (32-wk, n=84 PWH; semaglutide n=45, placebo n=39). Significant slowing across PhenoAge −4.9y (p=0.004), PCGrimAge −3.1y (p=0.007), GrimAge V2 −2.3y (p=0.009), OMICmAge −2.2y (p=0.009), RetroAge −2.2y (p=0.030), DunedinPACE −9% (p=0.01) . First GLP-1 epigenetic-clock RCT; HIV-specific cohort limits generalizability; epigenetic endpoints were post-hoc. See GLP-1 Peptides (Semaglutide, Tirzepatide, Retatrutide) . (NCT04019197, PubMed 42156721, PMC13381876, DOI 10.1038/s41467-026-72861-3)
13. NCT06613542 (NCT06613542)
14. Marcangeli V, et al. "Experimental Evidence Against Taurine Deficiency as a Driver of Aging in Humans." Aging Cell 24(10):e70191, 2025. PMID 41061678 · PMCID PMC12507425 . Independent UQAM/McGill refutation of taurine as aging biomarker; n=137 men, ages 20–93, with DXA + pQCT + mitochondrial respiration endpoints. See Taurine - The Aging Amino Acid . (PubMed 41061678, PMC12507425)