The Hannum clock: single-tissue blood DNA methylation age explained
Prepared by: Research Desk
Prepared by: Research Desk. Sources are linked for verification.
In 2013, two papers set the template for nearly every epigenetic age estimate in use today. One was Steve Horvath's multi-tissue clock. The other, published a few months earlier in Molecular Cell, came from Gregory Hannum and colleagues. It was built entirely from whole blood [1]. If a lab report or research paper lists "Hannum age" or "Hannum acceleration", it refers to this model. Knowing how it was built explains what that number can and cannot represent.
This explainer covers the training data, the 71 sites the model uses, why blood specificity matters, and how researchers use the clock in large cohorts today.
What the Hannum clock is
The Hannum clock is a statistical model. It takes DNA methylation levels at a fixed set of CpG sites in a blood sample and outputs an estimated age in years [1]. A CpG site is a position in the genome where a cytosine is followed by a guanine. Methylation is a chemical tag on that cytosine, and at some sites it shifts gradually with age. For a general primer on the chemistry, see how epigenetic clocks work.
The model is a "first-generation" clock. It was trained to predict chronological age, not mortality, disease or pace of change. That design choice shapes everything about how its output should be read. Our explainer on first, second and third generation epigenetic clocks places it in that wider lineage.
How it was built: 656 people, 71 sites
According to the original paper, Hannum and colleagues profiled genome-wide methylation in whole blood from 656 individuals aged 19 to 101 [1]. They used the Illumina Infinium HumanMethylation450 array, often shortened to "450K" because it measures roughly 450,000 CpG sites.
To reduce hundreds of thousands of candidate sites to a usable predictor, the team used elastic net regression. This penalized regression method shrinks the weights of uninformative variables toward zero. It tends to keep a compact set of predictors even when many of them are correlated. The final model kept 71 CpG markers [1]. Each has a weight, and the weighted sum produces the age estimate.
The paper reported that the model was highly predictive of chronological age. It also reported that the residual, the gap between predicted and actual age, could be interpreted as an "aging rate" measure. The authors examined factors associated with that rate, including sex and specific genetic variants [1]. They also tested the blood model on other tissues. Those tissues needed an adjustment to line up with chronological age, and tumour samples showed apparent age acceleration relative to normal tissue [1].
Why "single-tissue" matters
The defining contrast is with Horvath's 2013 clock. Horvath deliberately trained on many tissues and cell types, so one model could estimate age across very different samples [2]. Hannum trained on one sample type: whole blood [1].
That has practical consequences:
Strongest fit in blood. The 71 sites and their weights were tuned on blood. The model's calibration is strongest for blood samples processed on comparable arrays.
Offsets elsewhere. The original authors reported that non-blood tissues needed a correction [1]. Saliva, buccal swabs and other sample types contain different cell mixtures. A raw Hannum estimate from those samples is not directly interchangeable with a blood-based one.
Different signal from multi-tissue clocks. A multi-tissue model is pushed toward sites that age similarly everywhere. A blood-only model can pick up sites whose age association is partly specific to blood.
For the multi-tissue design in detail, see our page on the Horvath clock.
Leukocyte composition: the main caveat
Whole blood is not one cell type. It is a mixture of neutrophils, lymphocyte subsets, monocytes and other white cells, and each has its own methylation profile. Horvath's work across tissues and cell types showed how much methylation-based age estimates depend on the cells being measured [2].
The proportions of these cells are not fixed. They vary between people, shift with age, and can change with acute illness, infection, smoking or recent stress. Because the Hannum model reads methylation from the whole mixture, a shift in cell proportions can move the estimate even if nothing changed within any single cell type. This is why the research literature often reports two variants:
Extrinsic age acceleration keeps the cell-composition signal in. It partly reflects immune system aging.
Intrinsic age acceleration statistically adjusts for estimated blood cell counts. It aims to isolate change within cells.
When a study reports "Hannum acceleration", check which variant was used and whether cell counts were estimated or measured. The same blood draw can produce noticeably different acceleration values depending on that choice. Our page on biological age test accuracy discusses noise and repeat results more broadly.
How Hannum sits beside later clocks
Later clocks changed the training target rather than the basic method:
PhenoAge was trained against a composite of clinical blood chemistry measures linked to mortality, not against chronological age alone [3].
GrimAge combined methylation-based surrogates for plasma proteins and smoking pack-years, trained on time to death [4]. Our GrimAge explainer covers its construction.
DunedinPACE was trained on the longitudinal rate of change in multiple organ-system markers within one birth cohort. It outputs a pace rather than an age [5].
Because Hannum predicts chronological age, its acceleration residuals tend to be smaller and less tied to mortality than those of mortality-trained clocks. That is a design difference, not a flaw. The biological age test guide has a side-by-side summary of these clocks.
How researchers use the Hannum clock in cohorts
In epidemiology, the Hannum clock is rarely used alone. Studies commonly compute it alongside Horvath, PhenoAge, GrimAge and DunedinPACE. They then report how each measure relates to an outcome. Examples of this multi-clock research design in the recent literature include:
An analysis of older adults examining how biomarkers of aging biology measured across tissues relate to those measured within a tissue [6].
A study modelling DNA methylation age as a mediator between metabolic profile and cardiovascular and general aging outcomes [7].
Work on epigenetic aging in people with rheumatoid arthritis [8].
An analysis of aging measures and cancer in the U.S. Health and Retirement Study, a large observational cohort [9].
A study of epigenetic ageing and the risk of Parkinson's disease [10].
These are observational designs. They report associations, not causes. Methylation age has also appeared as a measured outcome in randomized trials. One example is the DIRECT PLUS diet trial, which assessed DNA methylation-based biological age across intervention groups [11]. In trials, the tissue-specific caveats above still apply: changes in blood cell mix during a study can shift blood-based clock values.
A note on consumer reports
Some consumer epigenetic test reports include a Hannum estimate. These tests are generally run as laboratory-developed tests within a single lab [12]. CLIA certification relates to laboratory quality standards for testing [13]. It does not validate what an aging clock means. Our page on whether biological age tests are FDA approved explains the regulatory background.
Further reading
Books on epigenetics and aging science vary widely in how current and how well sourced they are. A few neutral criteria are useful when choosing one. Check the publication date, since titles written before 2019 predate GrimAge and DunedinPACE. Check whether the author cites primary papers. Look for a clear distinction between age-prediction clocks and mortality-trained clocks. Finally, decide on the format and technical depth you want, from narrative introductions to statistics-heavy textbooks.
Look for books that explain elastic net regression, blood cell composition adjustment and array platforms in plain terms. These are the concepts that make clock outputs interpretable.
Frequently asked questions
How many CpG sites does the Hannum clock use?
The published model uses 71 CpG sites, selected by elastic net regression from 450K array data [1]. Each site carries a weight, and the weighted sum gives the age estimate.
Is the Hannum clock the same as the Horvath clock?
No. Both appeared in 2013, but Hannum trained on whole blood only, while Horvath trained across many tissues and cell types [1][2]. They use different CpG sites and can give different estimates for the same sample.
Can the Hannum clock be applied to saliva?
It was developed in blood. The original authors reported that other tissues required an adjustment to match chronological age [1]. Estimates from non-blood samples are therefore not directly interchangeable with blood-based values.
Why do intrinsic and extrinsic Hannum acceleration differ?
Extrinsic acceleration keeps the effect of shifting white blood cell proportions. Intrinsic acceleration adjusts for estimated cell counts. Because blood composition varies, the two can diverge noticeably for the same person.
Related on this site
The Horvath clock: the 2013 multi-tissue epigenetic age estimator
GrimAge explained: how the mortality-trained epigenetic clock works
First, second and third generation epigenetic clocks, explained
Biological age tests: how epigenetic clocks work and what they mean
To learn how pages are sourced and how the site is funded, read about this site.
Last updated: 23 October 2026.
Not medical advice. This page reports published research and regulatory records for information only.
Research-only notice: Content on this site is provided for research and educational purposes only and is not intended for human use. It is not medical advice.
Translation notice: Some content on this site may have been machine-translated using Google Translate and may contain errors.