Genetic age, telomeres and epigenetic clocks: what the tests actually measure

"Genetic age" is typed into search boxes as if it were one measurement. It is not. Behind the phrase sit three different things that are often sold under similar names: the DNA sequence you inherited, the length of telomeres at the ends of chromosomes, and chemical marks on DNA that change with age. They are measured differently, they answer different questions, and when a kit promises to tell you how old you are, it most often means just one of them.

The general idea of biological age is covered in what biological age is and how it is measured, and the methods behind the number in how biological age is calculated. This article is about a narrower question: what exactly is inside the tests that carry the word "genetic", and what each of them can and cannot say.

The sequence you were born with does not tell your age

The inherited sequence a genotype test reports — the kind of test used for ancestry or for single risk variants — is set at conception. A standard saliva or blood genotype test averages that sequence over millions of cells, so at twenty and at seventy it reads essentially the same text. That is why such a test cannot tell how old someone is. Individual cells do slowly acquire their own mutations; that is not what these tests measure.

What such a test can address is a different question: how much of the variation in lifespan between people is inherited. The answer has been shrinking as the data have grown.

• Twin studies. In 2,872 same-sex Danish twin pairs born between 1870 and 1900 and followed almost to the last death, the heritability of lifespan was estimated at 0.26 in men and 0.23 in women (Herskind et al., 1996). The authors described longevity as "only moderately heritable".

• Large family trees. An analysis of population-scale online genealogies with millions of relatives (Kaplanis et al., Science, 2018) put the heritability of lifespan at about 16%.

• Correcting for who marries whom. Using Ancestry pedigrees covering hundreds of millions of people born in the 1800s and early 1900s, Ruby et al. (Genetics, 2018) found that lifespans of in-laws — relatives who share no genes — correlate almost as well as those of blood relatives. Spouses tend to resemble each other in lifespan-related factors, and that inflates classic estimates. After accounting for it, the heritability of lifespan came out well below 10%.

Genome-wide searches point the same way. The largest meta-analysis of survival to extreme ages (Deelen et al., Nature Communications, 2019; 11,262 people who reached the 90th survival percentile and 3,484 who reached the 99th, against 25,483 controls) confirmed APOE — the ε4 variant was associated with lower odds of reaching those ages, ε2 with higher odds. A further locus near GPR78 reached genome-wide significance in the discovery analysis of 90th-percentile survival; the authors reported that they could not replicate it in independently genotyped cohorts. A study of the recorded lifespans of about a million parents of genotyped participants (Timmers et al., eLife, 2019) found that the variants explaining the most variance were those that raise the risk of dementia, cardiovascular disease and lung cancer, and that the top and bottom tenths of the resulting polygenic score differed in average lifespan by about five years.

What this does not say. These are estimates for populations — mostly of European ancestry and, in the pedigree work, for people born more than a century ago. A polygenic score describes average differences between large groups; it is not a forecast for one person, and it says nothing about how fast that person is ageing now.

The sequence does change a little — but not in a way a kit can read

Strictly speaking, DNA is not completely static. Individual cells slowly accumulate their own mutations. In haematopoietic stem and multipotent progenitor cells from ten people aged 0 to 81, sequencing of 3,579 genomes from single-cell-derived colonies showed an average of about 17 new mutations per year (Mitchell et al., Nature, 2022). In the subset of those colonies sequenced on one platform (seven of the ten donors), telomere length fell by about 30 base pairs per year in adult life. Neurons, which do not divide, also accumulate mutations at a steady rate through life (Abascal et al., Nature, 2021).

This is research-grade sequencing of individual cells or single molecules. Ordinary genotyping from saliva or blood reads the inherited sequence averaged over millions of cells and does not measure this accumulation.

Telomeres: a real change with age, and a poor clock

Telomeres are repetitive caps at the ends of chromosomes. They shorten as cells divide, and very short telomeres push a cell into senescence. That makes telomere length an intuitive candidate for a "biological clock". The data are less tidy than the intuition.

How strongly it tracks age. A systematic review of leukocyte telomere length in adults (Müezzinler et al., Ageing Research Reviews, 2013; 124 cross-sectional and 5 longitudinal studies) found an inverse correlation between mean age and mean telomere length across studies of about r = −0.3 and an estimated loss of roughly 22–25 base pairs per year in cross-sectional data, 32–46 in the few longitudinal ones. A correlation of that size means that at any given age people differ widely: telomere length alone places a person on the age scale very imprecisely.

How much of it is set early. In 355 identical and 297 fraternal same-sex adult twins (individuals, not pairs) aged 19–64, followed for an average of 12 years, the heritability of telomere length at baseline was estimated at 64%, and the heritability of its rate of shortening at 28% (Hjelmborg et al., Journal of Medical Genetics, 2015). The authors conclude that whether a person has long or short telomeres is largely determined before adulthood. A single adult measurement therefore reflects a starting point as much as the ageing that followed.

How reproducible the measurement is. In an international study where ten laboratories measured the same blinded DNA samples by three techniques, absolute results from different laboratories differed so widely that they could not be compared directly, even though the ranking of samples agreed well. Between-laboratory coefficients of variation averaged about 10% for Southern blotting and STELA and more than 20% for qPCR (Martin-Ruiz et al., International Journal of Epidemiology, 2015). The authors concluded that reference ranges cannot be shared between laboratories. qPCR — the method most often used at scale in biomarker studies — is also the noisiest between laboratories.

What it is associated with. In 64,637 adults from two Copenhagen cohorts, people in the shortest tenth of telomere length had higher all-cause mortality than those in the longest tenth, with a hazard ratio of 1.40 after adjustment (Rode et al., JNCI, 2015). That is an observational association. In the same study, genetically determined short telomeres were associated with lower cancer mortality and were not associated with all-cause mortality.

Longer is not simply better. A Mendelian randomisation study — which uses inherited variants as a natural experiment to get closer to cause and effect — combined data from 103 genome-wide studies (Telomeres Mendelian Randomization Collaboration, JAMA Oncology, 2017). Genetically longer telomeres were associated with a higher risk of several cancers, including glioma, lung adenocarcinoma and melanoma, and with a lower risk of coronary heart disease and abdominal aortic aneurysm. A later analysis of 472,174 UK Biobank participants (Codd et al., Nature Genetics, 2021) likewise found that genetically determined telomere length was linked to diseases in both directions.

So a telomere result does not come with a direction that is good for everything. We did not find a professional society guideline that recommends measuring telomere length in healthy people to assess ageing.

"Genetic age" in a kit is usually epigenetic age

When a commercial test promises to reveal your "genetic" or "DNA" age, what it most often measures is not the sequence and not telomeres but DNA methylation — chemical marks at particular sites in the genome that shift predictably with age. A statistical model, an epigenetic clock, converts that pattern into years. The clocks that defined the field were published in 2013: Horvath's, trained across many tissues, and Hannum's, trained on blood. How these clocks differ, and why they disagree with each other, is covered in how biological age is calculated and what the gap between biological and chronological age means.

Two properties matter for someone holding a result.

Technical noise can be as large as the effect people hope to see. In repeated measurements of the same samples, six widely used clocks gave estimates that differed by up to nine years between technical replicates (Higgins-Chen et al., Nature Aging, 2022). Versions of the same clocks rebuilt from principal components brought most replicates within 1.5 years. Whether a particular commercial test uses such a version is rarely stated.

There is no agreed standard yet. A 2023 consensus paper from the Biomarkers of Aging Consortium (Moqri et al., Cell, 2023) opens with the statement that the field currently lacks standards and consensus on what makes an ageing biomarker reliable, and proposes a framework for validating them before clinical use.

How to tell what a test actually measures

Before interpreting any "genetic age", it is worth getting answers to four questions. They are not about trust in a brand; they are about what the number can mean at all.

1. What is measured? The inherited sequence (genotype), telomere length, or DNA methylation. These are three different things, and only the last two are used to track age.

2. In what sample? Saliva, a cheek swab and blood contain different mixes of cells, and most clocks were trained almost exclusively on blood. When 83 people aged 9–70 gave five sample types at once, clock estimates from cheek and saliva samples differed significantly from blood estimates in the same person — by close to 30 years on average for some clocks (Apsley et al., Aging Cell, 2025). Correlations between oral and blood estimates remained low after controlling for cellular proportions. A result from one sample type is not interchangeable with another.

3. Which model or method, and is it published? A named, peer-reviewed clock or assay can be looked up. An undisclosed proprietary algorithm cannot be interpreted.

4. What is its repeatability? If the same sample measured twice can differ by years, a change of a few years between two of your own tests may be noise.

What these tests do not say

• None of them is a diagnosis. They do not detect a disease and do not tell anyone what to treat.

• None of them predicts one person's lifespan. Heritability, polygenic scores, telomere associations and clock associations are all established in groups.

• A shorter or "older" result does not show that something is going wrong. For telomeres, genetically longer is associated with higher risk of some cancers; for clocks, the difference may be within measurement error.

• Results from different companies, methods or sample types are not comparable.

What to do with a result

If you already have a number, it is most useful as part of a conversation with a doctor that includes the details above: what was measured, in what sample, by which method. The measures that change clinical decisions — blood pressure, lipids, glucose, kidney function, cardiorespiratory fitness and others — are measured with established methods and have reference ranges, and they are worth tracking over time with the same method. That is what the Lonevi record is built for: seeing how your own values move from test to test, rather than a single figure detached from its history.

This material is informational and does not replace consultation with a doctor.