CoChat
Sign Up
Cited by 31Epigenetics and DNA MethylationGenetics, Aging, and Longevity in Model OrganismsTelomeres, Telomerase, and Senescence

The DOI matches its Crossref record; 63 of the 63 references it cites were found in a registry or an index; 4 of the 19 sources its claims cite were checked for support.

Responsiveness of epigenetic aging biomarkers to longevity interventions in humans

Open access
Sehgal, Raghav, Borrus, Daniel, Armstrong, Jenel F., Gonzalez, John, Kasamoto, Jessica, Markov, Yaroslav, Priyanka, Ahana, Smith, Ryan, Carreras-Gallo, Natàlia, Lasky-Su, Jessica, Dwaraka, Varun B., Corley, Michael J., Higgins-Chen, Albert·2026·Nature Medicine
DOI 10.1038/s41591-026-04562-9PMC PMC13577923
Open this paper in CoChat — free accountOpen PDF

Abstract

Aging biomarkers can potentially allow researchers to rapidly monitor the impact of an aging intervention without the need for decade-spanning trials. However, before the use of aging biomarkers, such as epigenetic clocks, as surrogate endpoints, their responsiveness to interventions that target aging must be tested. Here we curate TranslAGE, a harmonized database of 51 public and private longitudinal interventional studies, and calculate a consistent set of 16 prominent epigenetic clocks for each study, along with 94 other DNA methylation (DNAm) biomarkers that can help explain the changes observed for each clock. Using this database, we discover patterns of responsiveness across a variety of interventions and DNAm biomarkers. For example, clocks trained to predict mortality or pace of aging show the strongest responses across all interventions and show consistent results with one another; pharmacological and lifestyle interventions drive the strongest responses from DNAm biomarkers; and the characteristics of the study population and study duration are key factors in determining the responsiveness of DNAm biomarkers to an intervention. Moreover, clocks with multiple subscores (that is ‘explainable clocks’) provide specificity and greater mechanistic insight into the responsiveness of interventions than single-score clocks. These findings can help to design future clinical trials by guiding the choice of interventions and of specific subsets of DNAm biomarkers to minimize multiple testing, study duration, study population and sample size, with the eventual aim of uncovering DNAm biomarkers that can be used as surrogate aging endpoints.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

The paper curates TranslAGE, a harmonized database of 51 longitudinal human longevity intervention studies (3,128 samples), and calculates 16 epigenetic clocks plus 94 other DNAm biomarkers to test responsiveness. It finds that reliable second-generation clocks, especially DunedinPACE and PCGrimAge, are most responsive, with pharmacological and lifestyle interventions and disease populations showing stronger effects.

Key findings

  • Across 51 interventions, 19 significantly decreased epigenetic age across the 16 biomarkers (13 after multiple-testing correction), 5 significantly increased it (3 after correction), and 26 had no significant effects. Pharmacological interventions had a mean effect size of −0.09307 (t = 4.948) and lifestyle interventions −0.0393 (t = 7.579); pharmacological effects were larger than lifestyle (P = 0.0063), supplements (P = 0.0009) and medical procedures (P = 0.0517).
  • Reliable generation 2+ biomarkers were most responsive. DunedinPACE decreased significantly in 16 interventions and increased in only 1, with pooled mean effect size −0.0891 (P = 0.0012). Other significant decreases were GrimAgeV2 (−0.0814, P = 0.0075), PCPhenoAge (−0.08876, P = 0.003), PCGrimAge (−0.07843, P = 0.0003) and SystemsAge (−0.05859, P = 0.0043). Only PCHorvath1 among generation 1 clocks significantly decreased (mean −0.06674, P = 0.029), and PhenoAge decreased (mean −0.06735, P = 0.0408) but was not significant after multiple-testing correction.
  • Disease populations showed larger biomarker responses than healthy populations: PCPhenoAge mean −0.32 (P = 0.0057), PCGrimAge −0.22 (P = 0.0022) and SystemsAge −0.213 (P = 0.0105), all significant after correction. In healthy groups, only DunedinPACE (−0.074, P = 0.0143) and PCGrimAge (−0.043, P = 0.0191) were responsive, neither significant after correction; DunedinPACE responded similarly in both populations (mean difference 0.07, P = 0.25).
  • Explainable GenX biomarkers provided mechanistic detail: 39 biomarkers changed significantly across all interventions, including PCGrimAge components PCPACKYRS (−0.05824, t = 4.636), PCCystatinC (−0.1404, t = 3.630) and PCTIMP1 (−0.08526, t = 3.666); SystemsAge components kidney (−0.1021, t = 3.753), lung (−0.08812, t = 4.615) and musculoskeletal (−0.07132, t = 3.729). Horvath1 correlated with its stochastic component (R = 0.50), whereas PhenoAge did not (R = −0.06); DamAge and AdaptAge changes were strongly inversely related (R = −0.89).

Methods

  • The authors curated 51 public and private longitudinal interventional studies with pre- and post-intervention blood DNAm data (total 3,128 samples), standardized study metadata, and calculated more than 110 DNAm biomarkers using MethylCIPHER 2.0, including 16 prominent epigenetic clocks and 94 other biomarkers.
  • Biomarkers were age-residualized within each dataset, standardized using the Health and Retirement Study reference standard deviation, and analyzed with paired t-tests for pre–post changes, one-sample t-tests for deviations from zero effect size, and univariate linear regressions for study-level covariates. Multiple-testing correction used a PCA-based effective number of independent tests.
  • Interventions were categorized as lifestyle, pharmacological, supplements or medical procedures; the database also included 4 control datasets and 9 pro-aging event datasets. Analyses were conducted in GraphPad Prism 9.4.0. The secondary analysis was approved by the Yale IRB (study number 2000040453).

Limitations

  • Stated: the study pooled heterogeneous interventional studies differing in participant age ranges, intervention modalities, durations and designs, limiting biological specificity and precluding causal interpretation; the intent was descriptive. It also notes variability in study quality, sample size and intervention relevance, lack of harmonized preprocessing and batch correction across studies, and caution that generation 2+ clocks may reflect intermediate health or inflammatory states rather than direct aging biology.
  • Stated: no empirically defined minimal clinically important difference exists for DNAm clocks, a key barrier to surrogate endpoint use. The authors also note that standardized effect sizes may appear modest, but for widely used clocks with epigenetic age deviation SD of ~4–6 years, an effect size of ~0.2 corresponds to an approximate 1-year shift.
  • Evident from the text: many reported effects are based on nominal P < 0.05, and several key findings are not significant after multiple-testing correction; the database also includes privately held TruDiagnostic datasets alongside public repositories, which may affect representativeness.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41591-026-04562-9

Cites

63 of 63 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

4 of 19 cited sources checked · 4 supported · 0 flagged

Show all 57 claims

  • Probably not a substantial claim. 5
  • It may state two separate claims as one. 30
  • The quote only partly states this claim. 1
36 of 36

  • The paper relies on this without support.

    It is unlikely that multiple clocks and multiple studies would repeatedly yield the same false positives.

    “Although false positives for a given clock/intervention combination are likely to occur, it is unlikely that multiple clocks and multiple studies would repeatedly yield the same false positives.”

  • The paper relies on this without support.

    All interventions in the dataset are targeted at ameliorating pathological aging potentially by reducing biological age.

    “We reasoned that because all interventions in our dataset are targeted at ameliorating pathological aging potentially by reducing biological age, then if the DNAm biomarkers are indeed responsive they should show a significant decrease when all effect sizes were pooled together”

  • The paper relies on this without support.

    If DNAm biomarkers are responsive, they should show a significant decrease when all effect sizes are pooled together.

    “if the DNAm biomarkers are indeed responsive they should show a significant decrease when all effect sizes were pooled together”

  • The paper relies on this without support.

    Comparatively little is known about biomarker responsiveness to interventions.

    “whereas comparatively little is known about biomarker responsiveness to interventions”

  • The paper relies on this without support.

    Fragmented literature and inconsistent clock panels make it difficult to compare studies, identify consistently responsive biomarkers or determine whether discordant findings reflect biological differences, methodological variation, inconsistent biomarker testing or publication bias.

    “Fragmented literature and the calculation of inconsistent clock panels make it difficult to compare studies, identify consistently responsive biomarkers or determine whether discordant findings reflect biological differences, methodological variation, inconsistent biomarker testing or publication bias.”

  • The paper relies on this without support.

    Investigators designing new clinical trials have limited evidence to guide biomarker selection.

    “Consequently, investigators designing new clinical trials have limited evidence to guide biomarker selection.”

  • The authors state this as an assumption.

    The data are normally distributed for the one-sample t-tests.

    “The normality of the data was assumed, and Prism automatically performed the test using the standard t-test formula.”

  • The paper relies on this without support.

    Interventions of similar categories frequently elicit correlated physiological responses.

    “interventions of similar categories (for example, diets, exercise regimens and pharmacological agents) frequently elicit correlated physiological responses.”

  • The paper relies on this without support.

    The standard deviation of epigenetic age deviation is typically on the order of ~4–6 years for widely used clocks such as Horvath, Hannum, PhenoAge and GrimAge.

    “where the standard deviation of epigenetic age deviation is typically on the order of ~4–6 years”
9 of 9

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Intervention outcome counts total 50, not 51

    Minor

    The reported groups contain 19 + 5 + 26 = 50 interventions, but the analysis says it covers all 51 interventions. One intervention is unaccounted for.

    Reported 51 · recomputed 50

  • One listed intervention is later described as two

    Minor

    The category of interventions that did not decrease any general generation 2 reliable clock lists only gastric bypass. The following paragraph refers to two interventions in that category.

  • Biomarker totals disagree

    Minor

    The stated panel of 16 clocks and 94 other biomarkers totals 110, whereas the pipeline says it calculated more than 110 biomarkers for each dataset.

3 of 3

Citations

Cites

Flagged first, then by citations

Show all 60 references

Research works printed but not linked: 2

  1. Schork, N. J., Beaulieu-Jones, B., Liang, W., Smalley, S. & Goetz, L. H. Does modulation of an epigenetic clock define a geroprotector? Adv. Geriatr. Med. Res. 4, e220002 (2022).
  2. Parashar, S. Role of DNA Methylation in Skeletal Metastasis and Bone Remodeling: Diagnostic, Prognostic and Therapeutic Implications. PhD thesis, McGill University (2016).

Cited by

Sorted by citations

See all 31

Related

See all

Metadata

Authors
Sehgal, R., Borrus, D., Armstrong, J. F., Gonzalez, J., Kasamoto, J., Markov, Y., Priyanka, A., Smith, R., Carreras-Gallo, N., Lasky-Su, J., Dwaraka, V. B., Corley, M. J., & Higgins-Chen, A.
Affiliations
  • Yale University (US)
  • Brigham and Women's Hospital (US)
  • University of California San Diego (US)
ORCID
Venue
Nature Medicine
Publisher
Springer Science and Business Media LLC
Published
2026-08-21
Volume
32
Issue
9
Pages
3477-3490
ISSN
  • 1078-8956 (ISSN-L)
  • 1546-170X
DOI
10.1038/s41591-026-04562-9
PMC
PMC13577923
OpenAlex
W7203885907
Semantic Scholar
ff8c0662ad3ca2fb7a1d6e01619862b24938c53a
Cited by
31
References
60
Citations by year
2026: 16
FWCI
23.64
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Medicine (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed Central (submittedVersion)
Funders
  • Yale University
  • University of Pennsylvania
  • University of Washington
  • Johns Hopkins University
  • National Institutes of Health
  • National Institute on Aging (RC2AG036495, RC4AG039029, R01AG065403, U24AG041689, U01AG009740)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Epigenetics and DNA Methylation
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Aging › Genetics, Aging, and Longevity in Model Organisms
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Cell Biology › Telomeres, Telomerase, and Senescence
Keywords
epigenetic clocks, DNA methylation, aging biomarkers, surrogate endpoints, aging interventions, pharmacological interventions, lifestyle intervention, senolytics, TNF-alpha inhibitors, mortality prediction, pace of aging

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review