Epigenetic clocks, explained from the evidence
Your DNA keeps time. Here is what it can tell you.
Epigenetic clocks read chemical marks on DNA to estimate age across most human tissues, and across mammals. Below, what the studies show is set out as 33 claims, one per line, each with a number, a short statement, and the single paper that supports it. Filter by tag, and sort by key findings or by newest evidence.
-
1
±3.6 yearstypical error of the Horvath clock
Median error of the Horvath clock against calendar age in held-out samples (correlation 0.96), across 51 tissues and cell types.
AI draft · awaiting reviewhorvath-accuracyHorvath 2013 · Genome BiologyMethod development; 82 public array datasets -
2
+10%risk of death per extra year of GrimAge
Hazard of death for each extra year of GrimAge acceleration in about 6,900 people. GrimAge outperformed earlier clocks at predicting time to death.
AI draft · awaiting reviewgrimage-mortalityLu et al. 2019 · AgingMethod development; Framingham training set, 6,935 people in validation -
3
Up to 9 yearsgap between repeat runs of one sample
How far readings from six common clocks can differ between repeat runs of the same sample, from technical noise alone.
AI draft · awaiting reviewnoise-9yrHiggins-Chen et al. 2022 · Nature AgingMethods; technical replicates -
4
2–3% slowerpace of aging with calorie restriction
Slower pace of aging on DunedinPACE after two years of caloric restriction versus usual diet (effect size d = −0.25 at 24 months, P < 0.003), even though participants achieved about 12% restriction against a 25% target. GrimAge and PhenoAge did not change.
AI draft · awaiting reviewcalerie-paceWaziry et al. 2023 · Nature AgingSecondary analysis of a randomized trial (n = 197) -
5
+4.5%risk of death per extra year of PhenoAge
Higher risk of death from any cause for each additional year of DNAm PhenoAge, pooled across cohorts. The fastest-aging 5% had about 1.6 times the average risk.
AI draft · awaiting reviewphenoage-mortalityLevine et al. 2018 · AgingMethod development; validated in several large cohorts -
6
Noneagreed gold standard for biological age
The field has no agreed gold standard for biological age. The Biomarkers of Aging Consortium sets out shared terms and the validation steps a biomarker needs before it can be used as a trial outcome.
AI draft · awaiting reviewconsortium-frameworkMoqri et al. 2023 · CellConsensus review (Biomarkers of Aging Consortium) -
7
2.9–3.8 monthsless biological aging with omega-3 over 3 years
Slowing of biological aging over three years with 1 g a day of omega-3, on PhenoAge, GrimAge2 and DunedinPACE (standardized effects 0.16 to 0.32) in 777 adults aged 70 and over. Omega-3, vitamin D and exercise had additive benefits on PhenoAge. This was a post hoc analysis, and the authors call the effect small.
AI draft · awaiting reviewdohealth-omega3Bischoff-Ferrari et al. 2025 · Nature AgingPost hoc analysis of a randomized factorial trial (n = 777 Swiss participants) -
8
2nd generationclocks that predict death better
Clocks trained on health outcomes predict death better than clocks trained on calendar age. In a joint model, PhenoAge stayed predictive (HR 1.04 per year) while the Hannum clock added nothing and the Horvath clock was slightly inverse.
AI draft · awaiting reviewphenoage-vs-first-genLevine et al. 2018 · AgingMethod development; validated in several large cohorts -
9
13,399people used to validate GrimAge2
Blood samples from nine cohorts used to validate GrimAge2. It predicted death more strongly than GrimAge (P = 3.6×10⁻¹⁶⁷ versus 2.6×10⁻¹⁴⁴) across racial and ethnic groups.
AI draft · awaiting reviewgrimage2-mortalityLu et al. 2022 · AgingMethod development; 13,399 blood samples from nine cohorts -
10
1.5 yearsgap between repeat runs with PC clocks
Agreement between most repeat runs after the same clocks were rebuilt on principal components (PC clocks).
AI draft · awaiting reviewpc-1-5yrHiggins-Chen et al. 2022 · Nature AgingMethods; technical replicates -
11
About 0clock age of stem cells
Estimated age of embryonic and induced pluripotent stem cells. Reprogramming a cell resets its clock.
AI draft · awaiting reviewhorvath-stemcellsHorvath 2013 · Genome BiologyMethod development; 82 public array datasets -
12
185mammal species read by one clock
Mammal species, across 59 tissues, in which a single universal clock estimates age with correlation above 0.96.
AI draft · awaiting reviewuniversal-mammalLu et al. 2023 · Nature AgingMethod development; 11,754 arrays from 185 mammal species -
13
Fasterpace of aging after childhood adversity
DunedinPACE is associated with later disease, disability and death, and shows faster aging in young adults who experienced childhood adversity.
AI draft · awaiting reviewdunedinpace-adversityBelsky et al. 2022 · eLifeMethod development; Dunedin birth cohort followed from age 26 to 45 -
14
+6.6 yearsolder clock age in Down syndrome
Average epigenetic age acceleration in blood and brain from people with Down syndrome (P = 7.0×10⁻¹⁴).
AI draft · awaiting reviewdownsyndromeHorvath et al. 2015 · Aging CellCase-control; blood and brain -
15
+36 yearsolder clock age in cancer tissue
Average age acceleration in about 6,000 cancer samples from 32 datasets.
AI draft · awaiting reviewhorvath-cancerHorvath 2013 · Genome BiologyMethod development; 82 public array datasets -
16
0.96repeat-run agreement of DunedinPACE (1 = perfect)
Test-retest reliability (intraclass correlation) of DunedinPACE across 36 technical replicates on the 450K array.
AI draft · awaiting reviewdunedinpace-iccBelsky et al. 2022 · eLifeMethod development; Dunedin birth cohort followed from age 26 to 45 -
17
−2.5 yearsclock age after 1 year of treatment, 9 men, no controls
Change in Horvath clock age after 12 months of growth hormone, DHEA and metformin in 9 men, versus their own baseline (PhenoAge −3.7, Hannum −2.8, GrimAge −2.2 years). There was no control group, and several authors had financial ties to the sponsor.
AI draft · awaiting reviewtriim-reversalFahy et al. 2019 · Aging CellOpen-label pilot, 9 men, no control group -
18
Menage faster than women on the clock
Men show faster epigenetic aging than women in blood, saliva and brain. In blood, Hispanic and Tsimane people show slower intrinsic but faster extrinsic aging than people of European ancestry.
AI draft · awaiting reviewsex-ancestryHorvath et al. 2016 · Genome BiologyCross-sectional; almost 6,000 samples, seven racial and ethnic groups -
19
4% fastermethylome aging in men than women
Faster methylome aging in men than in women, in blood from 656 adults aged 19 to 101.
AI draft · awaiting reviewhannum-sexHannum et al. 2013 · Molecular CellMethod development; 656 adults aged 19 to 101 -
20
+3.3 yearsolder liver clock age per 10 BMI units
Increase in liver epigenetic age for every 10 BMI units, in 140 liver samples. Obesity did not change the clock in fat, muscle or blood.
AI draft · awaiting reviewobesity-liverHorvath et al. 2014 · PNASCross-sectional; about 1,200 tissue samples including 140 livers -
21
HR 0.37risk of death with longer DNAm telomere length (below 1 = protective)
Age-adjusted hazard ratio for death with longer DNAmTL in validation cohorts (longer telomeres are protective). The methylation estimate predicted death, heart disease and heart failure better than telomere length measured directly.
AI draft · awaiting reviewdnamtl-outperformsLu et al. 2019 · AgingMethod development; training plus 12 validation cohorts -
22
Under 6 monthserror on newborn skin cells
Error of the skin and blood clock on newborn skin cells (fibroblasts), where the original pan-tissue clock was off by more than 10 years.
AI draft · awaiting reviewskinblood-fibroblastHorvath et al. 2018 · AgingMethod development; many cell types and ex vivo cultures -
23
Detectedfaster aging in progeria cells
The skin and blood clock detected faster epigenetic aging in cells from children with Hutchinson-Gilford progeria, which the pan-tissue clock missed.
AI draft · awaiting reviewskinblood-progeriaHorvath et al. 2018 · AgingMethod development; many cell types and ex vivo cultures -
24
353 sitesHorvath clock, 51 tissues
The Horvath pan-tissue clock estimates calendar age from 353 CpG sites in 51 tissues and cell types. It is a first-generation clock, best for development, tissue comparisons and cell reprogramming.
AI draft · awaiting reviewclock-horvathHorvath 2013 · Genome BiologyMethod development; 82 public array datasets -
25
1,030 sitesGrimAge, trained on time to death
GrimAge is trained on time to death, through methylation stand-ins for 7 plasma proteins and smoking pack-years. It is a second-generation blood clock for mortality and healthspan risk.
AI draft · awaiting reviewclock-grimageLu et al. 2019 · AgingMethod development; Framingham training set, 6,935 people in validation -
26
513 sitesPhenoAge, trained on clinical blood tests
DNAm PhenoAge is trained on a phenotypic age built from routine clinical blood tests. It is a second-generation blood clock for disease and mortality risk.
AI draft · awaiting reviewclock-phenoageLevine et al. 2018 · AgingMethod development; validated in several large cohorts -
27
173 sitesDunedinPACE, measures pace of aging
DunedinPACE estimates the pace of aging from decline in 19 organ-system indicators measured at ages 26, 32, 38 and 45. It is built to track change, for example in trials.
AI draft · awaiting reviewclock-dunedinpaceBelsky et al. 2022 · eLifeMethod development; Dunedin birth cohort followed from age 26 to 45 -
28
71 sitesHannum clock, blood
The Hannum clock estimates calendar age from 71 CpG sites in whole blood. It is a first-generation clock for blood-based age estimates in adults.
AI draft · awaiting reviewclock-hannumHannum et al. 2013 · Molecular CellMethod development; 656 adults aged 19 to 101 -
29
6 clocksrebuilt to cut measurement noise
PC clocks are six established clocks rebuilt on principal components to cut technical noise. They are suited to repeat measures in the same person.
AI draft · awaiting reviewclock-pcclocksHiggins-Chen et al. 2022 · Nature AgingMethods; technical replicates -
30
391 sitesskin and blood clock
The skin and blood clock estimates age from 391 CpG sites in skin, blood, saliva, cheek cells and cultured cells. It suits saliva and cheek-swab tests, cell culture and progeria studies.
AI draft · awaiting reviewclock-skinbloodHorvath et al. 2018 · AgingMethod development; many cell types and ex vivo cultures -
31
+2 proteinsGrimAge2 adds CRP and HbA1c
GrimAge2 adds methylation stand-ins for C-reactive protein and HbA1c to GrimAge, on the same 1,030 CpG sites. It predicts mortality and disease across ancestries.
AI draft · awaiting reviewclock-grimage2Lu et al. 2022 · AgingMethod development; 13,399 blood samples from nine cohorts -
32
59 tissuesuniversal mammal clocks
The universal pan-mammalian clocks are trained on log age and on age relative to each species' lifespan or maturity, using conserved sites on a mammalian array. They suit cross-species and animal studies.
AI draft · awaiting reviewclock-universalLu et al. 2023 · Nature AgingMethod development; 11,754 arrays from 185 mammal species -
33
140 sitesDNAmTL, estimates telomere length
DNAmTL estimates leukocyte telomere length from 140 CpG sites in blood. It also predicts death and heart disease.
AI draft · awaiting reviewclock-dnamtlLu et al. 2019 · AgingMethod development; training plus 12 validation cohorts
Glossary
- DNA methylation
- A chemical tag (a methyl group) on DNA, mostly at CpG sites, that helps switch genes on or off. Its pattern changes with age in predictable ways.
- CpG site
- A place in the genome where a C is followed by a G. Clocks read the methylation level at a few hundred to a few thousand of these sites.
- Epigenetic clock
- A formula that turns methylation levels at chosen CpG sites into an age estimate or an aging rate.
- Chronological age
- Age in years since birth.
- Biological age
- An estimate of how old the body behaves. Each clock defines it differently, and there is no agreed gold standard.
- Age acceleration
- The gap between clock age and calendar age, after adjusting for calendar age. Positive means older than expected.
- First-generation clock
- Trained to predict calendar age, for example the Horvath and Hannum clocks.
- Second-generation clock
- Trained to predict health outcomes or death, for example PhenoAge and GrimAge.
- Pace-of-aging clock
- Estimates how fast someone is aging now, rather than how old they are, for example DunedinPACE.
- Test-retest reliability
- How closely two measurements of the same sample agree. Often reported as an intraclass correlation (ICC), where 1 is perfect agreement.
- Principal-component (PC) clock
- A clock rebuilt on summary patterns across many CpG sites, which makes it less sensitive to noise at any one site.
- Methylation array
- The lab chip most clocks are built on, such as Illumina's 450K and EPIC arrays. Results can shift between array versions and labs.
- Hazard ratio
- How much the risk of an event, such as death, changes per unit of a measure. 1.10 means 10% higher risk per unit.
- Pluripotent stem cell
- A cell that can become any cell type. Embryonic and reprogrammed (iPS) cells read close to age zero on the Horvath clock.
All 17 sources
- Bischoff-Ferrari HA, Gängler S, Wieczorek M, et al. Individual and additive effects of vitamin D, omega-3 and exercise on DNA methylation clocks of biological aging in older adults from the DO-HEALTH trial. Nature Aging 2025. Post hoc analysis of a randomized factorial trial (n = 777 Swiss participants).
- Lu AT, Fei Z, Haghani A, et al. Universal DNA methylation age across mammalian tissues. Nature Aging 2023. Method development; 11,754 arrays from 185 mammal species.
- Moqri M, Herzog C, Poganik JR, et al. Biomarkers of aging for the identification and evaluation of longevity interventions. Cell 2023. Consensus review (Biomarkers of Aging Consortium).
- Waziry R, Ryan CP, Corcoran DL, et al. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nature Aging 2023. Secondary analysis of a randomized trial (n = 197).
- Belsky DW, Caspi A, Corcoran DL, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife 2022. Method development; Dunedin birth cohort followed from age 26 to 45.
- Higgins-Chen AT, Thrush KL, Wang Y, et al. A computational solution for bolstering reliability of epigenetic clocks. Nature Aging 2022. Methods; technical replicates.
- Lu AT, Binder AM, Zhang J, et al. DNA methylation GrimAge version 2. Aging 2022. Method development; 13,399 blood samples from nine cohorts.
- Fahy GM, Brooke RT, Watson JP, et al. Reversal of epigenetic aging and immunosenescent trends in humans. Aging Cell 2019. Open-label pilot, 9 men, no control group.
- Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging 2019. Method development; Framingham training set, 6,935 people in validation.
- Lu AT, Seeboth A, Tsai PC, et al. DNA methylation-based estimator of telomere length. Aging 2019. Method development; training plus 12 validation cohorts.
- Horvath S, Oshima J, Martin GM, et al. Epigenetic clock for skin and blood cells applied to Hutchinson Gilford Progeria Syndrome and ex vivo studies. Aging 2018. Method development; many cell types and ex vivo cultures.
- Levine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging 2018. Method development; validated in several large cohorts.
- Horvath S, Gurven M, Levine ME, et al. An epigenetic clock analysis of race/ethnicity, sex, and coronary heart disease. Genome Biology 2016. Cross-sectional; almost 6,000 samples, seven racial and ethnic groups.
- Horvath S, Garagnani P, Bacalini MG, et al. Accelerated epigenetic aging in Down syndrome. Aging Cell 2015. Case-control; blood and brain.
- Horvath S, Erhart W, Brosch M, et al. Obesity accelerates epigenetic aging of human liver. PNAS 2014. Cross-sectional; about 1,200 tissue samples including 140 livers.
- Hannum G, Guinney J, Zhao L, et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Molecular Cell 2013. Method development; 656 adults aged 19 to 101.
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biology 2013. Method development; 82 public array datasets.
About this draft
An AI co-scientist read the papers listed above and wrote each claim. Every claim cites exactly one paper, and its number was checked against that paper. Claims stay marked “AI draft” until the expert reviewer accepts, corrects or rejects them.
“Key findings” puts the results the field relies on most at the top, in an order the draft proposes and the reviewer can change. “Newest” sorts by the publication year of each claim’s supporting paper.
This site is for education and is not medical advice. A biological age result is not a diagnosis.