Epigenetic Clocks and Drift Explained

Two people can share a birthday and nothing else biologically. One has the joints, arteries, and cell repair systems of someone 10 years younger. The other is aging faster than the calendar says. Chronological age cannot tell them apart, and that blind spot is the entire reason epigenetic clocks exist.

When a DNA methylation clock returns a biological age older than your calendar age, it means the methylation patterns across your genome have drifted further from a youthful baseline than would be expected for someone born when you were. That gap is called epigenetic age acceleration, and a positive value is the number researchers pay the most attention to because it tracks with physical functional decline, cognitive impairment, and all-cause mortality in ways chronological age alone does not capture. As for commercial tests, the reliability question has a real ceiling: different clocks were built to detect different signals, trained on different outcomes using different biomarkers, and a result that moves one clock can leave another flat entirely. No clinical trial has yet shown that reducing an epigenetic age reading through any intervention leads to fewer diseases or a longer life, so a single commercial score is better read as a probabilistic signal than as a verdict on any individual's health trajectory.

Why Chronological Age Fails as a Measure of Aging Speed

A birth date is a fixed number. It does not move, and it does not care what is happening inside a cell. Two people born on the same day in the same hospital carry an identical number on their driver's license for the rest of their lives, no matter what shape their organs are in or how well their molecular repair systems are keeping up with damage.

Stanford Medicine researchers describe biological age differently. They treat it as a measure of speed. Biological age measures how fast the aging process is moving through a person's body right now, not how old that person is.

Tony Wyss-Coray, a Stanford professor of neurology and director of the Phil and Penny Knight Initiative for Brain Resilience, has compared biological clocks to something the World Health Organization calls "intrinsic capacity": a composite of cognition, mobility, energy, and mood that doctors already use to gauge how well a person is actually functioning. Biological clocks try to do the same job at the molecular level, in a way that can be measured and repeated.

An eBioMedicine editorial from 2026 states the goal in plain terms: biological age tries to capture the molecular, physiological, and environmental changes that shape frailty and disease risk.

What epigenetic drift is

Epigenetic drift is the molecular process behind that gap. It describes the slow, cumulative loss of precision in how cells manage DNA methylation, and it is the mechanism that pulls biological age away from chronological age.

DNA methylation works by attaching a small methyl group to cytosine, usually at sites called CpG dinucleotides. These tags do not change the DNA sequence itself. They act more like switches, turning genes on or off by controlling which parts of the genome get read.

A 2026 Frontiers review lays out why that switching system breaks down as you age. The enzymes responsible for maintaining and placing these methyl tags, DNMT1, DNMT3A, and DNMT3B, lose activity or become dysregulated over time. The enzymes produce sloppier maintenance of old methylation patterns and sloppier placement of new ones.

That sloppiness moves in two directions at once. Across the genome broadly, methylation tends to decline, a pattern called global hypomethylation. At specific spots, particularly CpG islands sitting inside gene promoter regions, methylation tends to pile up instead. That buildup is not cosmetic. Hypermethylation at a promoter can silence the gene behind it, and some of the genes caught in that silencing are involved in tumor suppression and DNA repair. Meanwhile the hypomethylation happening elsewhere destabilizes regions of the genome that are supposed to stay quiet.

What causes the errors to accumulate? Oxidative stress, chronic inflammation, and the simple math of repeated cell division all play a part. The Frontiers review notes that methylation maintenance is highly accurate at any single cell division. The problem builds over thousands of divisions, as DNA damage accumulates and the copying process carries small mistakes forward each time.

One finding makes the whole concept measurable rather than theoretical. A 2025 epigenome-wide drift study found that 99% of CpG sites showing drift also showed increased variability between individuals of the same age. In other words, two 60-year-olds do not drift the same way or at the same pace. That variability is what opens a gap between the calendar and the cell, and it is what makes the gap something scientists can actually detect and quantify.

The most important word in this entire field might be "reversible." A review in MedComm describes epigenetic drift as reversible, and treats it as a central driver of organismal decline and age-related disease. If drift could only move in one direction, there would be nothing to intervene on. The fact that it can potentially be slowed or partly undone is what turns this from a diagnostic curiosity into something worth building tools around.

How scientists built clocks to detect and quantify drift

Epigenetic clocks do not measure drift directly. No instrument opens a cell and counts how much precision has been lost. Instead, clocks read the fingerprint that drift leaves behind on the methylome and infer age from that pattern. That distinction matters because a clock is an inference engine, not a drift counter, and that fact shapes how much weight any single reading deserves.

The construction process, as the Frontiers review describes it, starts with large datasets of human methylation profiles tied to known ages. Researchers scan the genome for CpG sites where methylation state changes in a predictable way as people get older. Those sites get weighted and combined into an algorithm that spits out a single number: an estimated age, based purely on methylation data.

DNA methylation makes a good raw material for this kind of model for a few practical reasons. It is chemically stable, it can be measured with standardized lab methods such as bisulfite treatment, microarrays, and next-generation sequencing, and it shows up across different tissue types. On top of that, it correlates strongly with outcomes people actually care about: functional decline, cognitive impairment, and all-cause mortality.

Once a clock produces an estimate, that number gets compared against the person's actual chronological age. The gap between the two is called epigenetic age acceleration. A positive value means the methylation-based estimate is older than the calendar says, which is the number researchers pay the most attention to.

Three generations of clocks

Clock science did not stand still after the first models came out. Each generation asked a sharper question than the one before it, moving away from simply guessing someone's birth year and toward predicting something that actually matters for their health.

The first generation, built by Horvath and Hannum in 2013, aimed at one target: predict chronological age as accurately as possible from methylation data. Stanford Medicine researchers note that these clocks were designed to reproduce the number already printed on a driver's license. The real discovery came as a side effect. When someone's predicted age diverged from their actual age, that gap turned out to carry information. The clocks worked like an odometer built to match a known reading, and the mismatches between the readings were where the interesting signal lived.

The second generation shifted the training target from time to health. PhenoAge is trained on clinical and phenotypic markers of aging rather than chronological age alone. The Frontiers review lists what goes into it: DNA methylation data combined with chronological age and nine clinical biomarkers, including albumin, creatinine, glucose, C-reactive protein, lymphocyte percentage, mean cell volume, red blood cell distribution width, alkaline phosphatase, and white blood cell count. Because CRP and glucose sit inside that architecture, the same metabolic and inflammatory markers a doctor might already track on a basic blood panel are built directly into the age estimate itself.

GrimAge, also second generation, took a different route to a similar goal. It was trained on time-to-death data using a two-stage process: first building DNA methylation surrogates that stand in for plasma proteins and smoking history, then training those surrogates against actual mortality outcomes. David Rehkopf, a Stanford professor of epidemiology and population health, has described these models as more useful because they focus more directly on capturing the biology of aging that predicts disease.

Third generation clocks, led by DunedinPACE, changed the question again. Instead of estimating how much biological age someone has accumulated, DunedinPACE estimates how fast they are aging right now. The odometer-versus-speedometer comparison fits well here. First-generation clocks tried to read total mileage. DunedinPACE reads current speed. The Frontiers review notes it showed strong correlations with morbidity, disability, and mortality, along with predictive value for functional decline, cognitive aging, and early-life adversity.

That rate measurement means two people can have the exact same biological age today and still be headed in very different directions, if one of them has a DunedinPACE score showing faster current drift than the other.

What drives faster drift

Drift is not simply a function of years passed. It responds to identifiable exposures and behaviors, and that fact is what makes epigenetic age a readout of choices and circumstances rather than pure inherited fate.

An eBioMedicine editorial points to two recent studies showing that modifiable cardiovascular behaviors shape both how much biological aging someone has accumulated and how fast it continues to move. A BMC Medicine study found that favorable cardiovascular health tracked with lower epigenetic age acceleration, and the pattern split by sex. In men, avoiding nicotine and keeping glucose regulated were the stronger predictors. In women, physical activity, glucose regulation, and a healthy BMI carried more weight.

At the molecular level, the Frontiers review names oxidative stress, chronic inflammation, and environmental exposures, including air pollution and radiation, as contributors to faster methylation dysregulation. These are not abstract risk factors sitting somewhere outside the model. Because CRP and glucose are literally built into PhenoAge's formula, inflammation and metabolic dysfunction are not just associated with faster aging. They are direct inputs into the number the clock produces.

That raises a fair question: if the inputs are partly behavioral, the clock can be bent back only to a degree. Knowing what accelerates drift tells researchers what to study and what to watch. It does not yet tell anyone a guaranteed way to reverse it, and the next section explains why that line matters.

What does it mean when your biological age reads older than your chronological age?

Across a range of study populations and outcomes, epigenetic age acceleration tracks with disease risk and mortality in ways that chronological age alone does not capture. The Frontiers review documents a strong correlation between methylation-based age estimates and outcomes including physical functional decline, cognitive impairment, and all-cause mortality.

DunedinPACE adds a layer on top of that. Beyond predicting functional decline, cognitive aging, and the effects of early-life adversity, its rate-of-aging measure predicted mortality risk even after accounting for a person's baseline epigenetic age. That detail matters. It means the direction and speed of change over time carries information that a single baseline number misses. A one-time reading is a photograph. A trajectory is a trend line, and the trend line turned out to predict outcomes on its own.

The clocks also connect to specific disease pathways. A systematic review and meta-analysis synthesized 13 studies on accelerated biological aging and stroke risk. The association was stronger for first-ever stroke than for recurrent stroke, which hints that these clocks might be picking up something about vascular vulnerability that standard risk factors like blood pressure and cholesterol do not fully explain on their own.

None of this proves causation. The associations are consistent and have been replicated across different clocks and cohorts, but an association between faster drift and worse outcomes is not the same as proof that the drift itself causes those outcomes. Stanford researchers frame healthspan, the years spent in good health, as the more useful measure than lifespan alone. Biological clocks are built to track that healthspan trajectory. Treating any single score as a certain prediction of a specific disease outcome goes further than the current evidence supports.

Whether lifestyle and intervention can move the clock

Lifestyle changes can shift epigenetic age readings in controlled studies. The evidence for that claim is uneven, and no trial so far has shown that moving a clock's number buys someone more healthy years.

One of the most cited studies in this space comes from Fitzgerald and colleagues, documented in the eBioMedicine literature. The trial tested a multi-modal intervention combining diet changes, exercise, sleep guidance, relaxation practices, and supplemental probiotics and phytonutrients, all delivered together rather than as separate interventions.

The results did not land the same way across every clock used to measure them. DunedinPACE showed no change in that trial, even as other measures shifted. That mismatch matters for how anyone should read a single test result. Different clocks were built to detect different signals, trained on different outcomes, using different biomarkers. A result that moves one clock and leaves another flat is a reminder that these instruments are not interchangeable.

The Frontiers review names inconsistent longitudinal performance as one of the open problems still facing this field. Readings do not always track in a stable, predictable way over repeated measurements in the same person.

No clinical trial has yet shown that reducing someone's epigenetic age through intervention leads to fewer diseases or a longer life. The biological logic connecting the two is reasonable, and the associational evidence supporting it is substantial. Proof that lowering the number changes what actually happens to a person's health has not arrived yet. That is the gap between what the science suggests and what it has demonstrated, the one to keep in view before treating any clock reading as a verdict.