Epigenetic clocks and blood cell composition: why cell mix matters
Prepared by: Research Desk
Prepared by: Research Desk. Sources are linked for verification.
A blood-based epigenetic age test does not read methylation from "blood" as a single uniform tissue. It reads methylation from DNA extracted from white blood cells. That DNA comes from several cell types, mainly neutrophils, lymphocytes and monocytes, and each type carries its own methylation pattern. When the mix changes, the bulk methylation signal changes too, and so can the clock estimate built on it.
A 2026 Genome Medicine paper by Jonkman and colleagues takes up this subject. Its title is "Blood cell composition reveals distinct biological interpretation of DNA methylation age and age acceleration" [1]. This explainer covers the measurement problem behind that question. It sets out what cell composition is, how it can move a clock, and how researchers estimate it from methylation data. It also explains why saliva and blood results are not directly interchangeable. For background on how clocks are built, read how epigenetic clocks work first.
Blood is a mixture, not a single tissue
Methylation helps define cell identity. Different white blood cell types share the same DNA sequence, but at many CpG sites their methylation levels differ. That difference is the premise underlying deconvolution methods, which estimate cell proportions from methylation data [6][7][8].
A bulk blood sample therefore reports a weighted average across cell types. Here is a simple illustration, with numbers chosen for easy arithmetic rather than taken from any study. Suppose a CpG site is 80% methylated in neutrophils and 20% in lymphocytes. A sample that is 60% neutrophils reads about 56% methylated at that site. Shift the mix to 70% neutrophils and the same site reads about 62%. No individual cell has changed its methylation.
The same logic applies within a lineage. At any site where naive and memory T cells differ, a sample with a different balance of those subsets will read differently. Multiply that effect across the dozens or hundreds of CpG sites a clock uses, and a change in cell proportions can register as a change in estimated age.
How a cell-mix shift can move a clock estimate
An epigenetic clock is, at its core, a weighted combination of methylation values at selected CpG sites. Horvath's 2013 work described DNA methylation age across many human tissues and cell types [2]. Later clocks were built with different aims:
PhenoAge was described as an epigenetic biomarker of aging for lifespan and healthspan [3].
GrimAge was reported to strongly predict lifespan and healthspan [4].
DunedinPACE was described as a DNA methylation biomarker of the pace of aging [5].
Because each clock selects CpG sites to predict a different target, each can in principle be more or less exposed to cell-type differences. Sites chosen for their link to lifespan, healthspan or pace of aging could partly reflect immune composition. If so, part of what a clock "detects" may be the cell mix itself.
That would not automatically be a flaw, because immune composition is a real biological feature. The challenge is interpretation. A higher score could reflect methylation change within cells, a change in which cells were sampled, or both. For how this ties into reliability and repeat results, read biological age test accuracy.
What the 2026 analysis focuses on
The Jonkman paper's title separates two quantities that are often blurred [1]:
DNA methylation age: the raw clock estimate in years (or a pace, for DunedinPACE).
Age acceleration: the gap between that estimate and the value expected for a person's chronological age.
The title states that blood cell composition reveals a distinct biological interpretation of DNA methylation age and of age acceleration [1].
Our inference (not a finding quoted from the paper): a careful reader should not assume that the raw estimate and the acceleration measure relate to cell mix in the same way. It is worth looking at each one on its own terms.
We have not reported which clocks were analysed, the sample size or any effect sizes. We could not verify those details against the source at the time of writing. Readers who want the specifics should go to the paper itself.
Deconvolution: estimating the cell mix from methylation
Many studies do not have direct cell counts for every sample. Instead, they estimate cell proportions from the methylation data itself, a process called deconvolution.
Reference-based approaches use methylation profiles from purified cell types as a reference. They then estimate the proportions of each type that would best reproduce the bulk sample.
Reference-free approaches do not need purified references. Houseman and colleagues described a reference-free method. They framed cell composition as something that can mediate associations, not only confound them [6].
Method choice matters. One study compared reference-free deconvolution software and produced guidelines for quantifying cell-type heterogeneity [7]. A separate systematic evaluation assessed deconvolution pipelines for sequencing-based bulk methylomes [8]. That is a different data type from the arrays many studies use. The common thread is that estimated proportions are estimates. They carry their own error, and that error passes into any adjusted result.
Intrinsic and extrinsic age acceleration
The following are general working definitions used in the field. They are not drawn from any single paper cited on this page. Once cell proportions are estimated, age acceleration can be expressed in different ways:
Intrinsic (cell-adjusted) acceleration removes the part of the clock signal explained by estimated cell proportions. It aims to capture age-related methylation change that does not depend on the cell mix.
Extrinsic acceleration deliberately keeps, or emphasises, the immune-composition signal. The reasoning is that immune aging is part of what should be measured.
Neither version is the "true" one. If the research question concerns methylation change inside cells, unadjusted cell shifts act as a confounder. If the question concerns immune aging, adjusting them away removes the signal of interest. When a consumer report gives a single acceleration number, it is worth knowing which approach, if any, produced it. Our guide to first, second and third generation epigenetic clocks covers how acceleration measures developed.
Saliva and blood are not interchangeable
As general background (not drawn from the sources cited on this page), saliva is usually described as containing a mix of white blood cells and cells shed from the lining of the mouth, and that balance is not fixed. Blood DNA comes mainly from leukocytes. So a saliva sample and a blood sample are likely to start from different cell mixtures.
Horvath's multi-tissue work showed that methylation age can be estimated across many tissues and cell types [2]. Also as general background, rather than a claim drawn from the sources cited here, several widely used clocks are generally described as having been developed mainly in blood-derived DNA. Applying a clock developed in one tissue to another is an extrapolation unless the lab has validated it in that tissue.
Two consequences follow. First, a saliva result and a blood result for the same person are not directly comparable. Second, a change between two tests run on different sample types is not, by itself, evidence of a change in aging.
Timing of sampling
A 2024 study reported that epigenetic age estimates oscillate during the day [9]. Whatever the full explanation, that finding supports keeping sample timing consistent, along with sample type, when two results are lined up against each other.
Questions to put to a test provider
The questions below address cell composition directly and are worth asking before ordering a test:
Which tissue is sampled: venous blood, a finger-prick blood spot or saliva?
Which clock is reported, and was it trained or validated in that tissue?
Is cell composition estimated, and are the estimated proportions shown in the report?
Is the acceleration figure cell-adjusted, and if so, by which deconvolution method?
What is the lab's regulatory status? Some consumer tests are offered as laboratory-developed tests, a category described by the FDA [10]. In September 2025 the FDA published a notice implementing the vacatur of its earlier rule on laboratory-developed tests [11]. CLIA certification concerns laboratory testing quality standards [12]. It is not a statement that a score means what marketing implies. Our page on whether biological age tests are FDA approved explains both in detail.
The broader biological age test guide sets these questions alongside clock design and interpretation.
Further reading: choosing a book on epigenetics and aging
For readers who want depth beyond single papers, a well-chosen book can explain methylation biology, immune aging and statistics in one place. Neutral criteria for picking one:
Publication date. Books written before 2019 predate GrimAge and DunedinPACE, and few discuss deconvolution in detail.
Author background. Look for authors who publish in epigenetics, immunology or biostatistics and who cite primary studies.
Method coverage. A useful book explains how arrays work, how clocks are trained and why cell composition matters, not just headline numbers.
Tone. Measured explanation is more useful than promises about extending life.
Format and price. Print, ebook and audiobook editions vary; figures and tables tend to read better in print or ebook.
No book replaces the primary literature, which moves faster than publishing schedules. The sources listed at the end of this page link directly to the original studies and records.
Frequently asked questions
Does a change in cell composition mean a clock result is wrong?
Not necessarily. Immune composition is a real biological feature, and some clocks may partly capture it. The challenge is interpretation, because a shift could reflect the cell mix, methylation change within cells, or both.
What is the difference between intrinsic and extrinsic age acceleration?
These are general working terms. Intrinsic acceleration adjusts for estimated cell proportions to focus on within-cell methylation change, while extrinsic acceleration keeps the immune-composition signal. Which one fits depends on the research question.
Can a saliva result be lined up against an earlier blood result?
Not directly. Saliva and blood are likely to contain different cell mixtures, and as general background, several widely used clocks are described as having been developed mainly in blood-derived DNA. Results are most interpretable when the same tissue, lab method and clock are used each time.
How do labs know the cell mix without counting cells?
They estimate it from the methylation data using deconvolution. This works either with reference profiles from purified cell types or with reference-free methods [6][7]. The results are estimates with their own error, not direct counts.
Related on this site
Biological age tests: how epigenetic clocks work and what they mean
Are biological age tests FDA approved? How aging tests are regulated
How epigenetic clocks work: DNA methylation and age estimation
GrimAge explained: how the mortality-trained epigenetic clock works
Every page here follows the same sourcing approach; read how this site works for details.
Last updated: 22 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.