Why Does Biological Age Change Depending on the Test? Scientists Have Answers

A large new study found that five widely used biological aging clocks are each tracking something surprisingly different at the cellular level. A new gene-based approach showed stronger associations with some health outcomes, though results varied across outcomes and external datasets.

Two people born in the same year can look, feel, and function like they belong to different generations. For decades, scientists have been developing tools called “epigenetic clocks” to measure that gap more precisely. These clocks read chemical tags on DNA, patterns that shift as the body ages, and translate them into an estimate of how fast a person’s body is aging compared to their calendar years. The catch: despite being widely used in research and increasingly moving into clinical and commercial settings, nobody fully understood what these clocks were actually measuring under the hood. Now, a large study in npj Aging has begun to explain it.

Five Biological Age Clocks, Five Different Stories

At the center of this research are five aging clocks: Horvath, Hannum, PhenoAge, GrimAge, and DunedinPACE. Each was built using a different strategy. Some were trained to predict a person’s chronological age. Others were designed to predict disease risk, time until death, or the pace of physical and mental decline over time. Researchers and clinicians have used all five, often without a clear sense of which best suited a given question.

Researchers from the University of Southern California, UCLA, Indiana University, and the University of Minnesota analyzed data from 3,227 older Americans. By pairing two types of biological data collected from the same blood samples, measurements of chemical tags on DNA alongside measurements of which genes were actively switched on or off, they mapped the gene activity and biological pathways tied to each clock. These clocks, often treated as interchangeable measures of the same thing, turned out to track largely distinct biological stories.

Data came from the Health and Retirement Study, a nationally representative, ongoing study of Americans over age 50. Participants had a mean age of 70, were 58% women, and came from a range of racial and ethnic backgrounds. Blood samples were collected in 2016, and health and survival data were tracked through 2022.

When the team looked at which genes were switched on or off in people with higher scores on each clock, a sign of faster biological aging, the differences were dramatic. Horvath was linked to changes in just 49 genes. DunedinPACE was linked to changes in more than 3,200 genes. Critically, there was almost no overlap: not a single gene was shared across all five clocks.

What Each Biological Age Clock Appears to Track

Grouping individual gene signals into broader biological themes helped clarify what each clock appears to pick up on.

Horvath, the oldest of the five and designed to estimate age across many different tissue types, was linked to metabolic processes and basic cellular communication. Its signals pointed to fundamental housekeeping functions of cells rather than specific diseases.

Hannum, also an older clock but designed specifically from blood, showed stronger ties to immune system function, inflammation, and blood vessel activity, patterns consistent with how the immune system gradually breaks down and becomes more prone to chronic inflammation with age.

PhenoAge, built to predict a cluster of clinical health measures tied to mortality risk, showed associations with cellular aging, stress responses, and signs of cell division gone awry, consistent with its design to capture broad, system-wide biological stress.

GrimAge, consistently the strongest predictor of death in published research, was linked to immune surveillance pathways. Despite being associated with more gene activity changes than PhenoAge, it mapped to a narrower set of biological themes. That pattern points to a concentrated stress signal built partly from markers of smoking exposure and inflammation.

DunedinPACE, the newest of the five and designed to measure the speed of aging rather than a static biological age, showed the broadest reach of all, touching on protein metabolism, immune signaling, nervous system development, cellular energy production, and reduced activity in the pathways cells use to repair DNA damage.

Despite these differences, a few threads ran through multiple clocks. Immune system activity, particularly the kind of low-grade, chronic inflammation that tends to increase with age, showed up repeatedly. Pathways related to the body’s response to infection and internal damage were active in four of the five clocks, making immune activity one of the clearest signals shared across the clocks.

A New Gene-Based Score That May Work Better for Some Outcomes

Armed with these findings, the researchers built a new set of scores they called Transcriptomic Aging Gene Scores, or TAGS. Rather than reading chemical tags on DNA, TAGS directly measure how actively the genes linked to each aging clock are being expressed at any given moment. It is the difference between a dial set to a certain position versus actually checking whether the switch it controls is on or off.

TAGS were developed using 80% of the sample and then tested on the remaining 20%, a group of 645 individuals whose data had been set aside specifically for this purpose. Across many health outcomes measured through 2022, including deaths, frailty, difficulty with daily activities, diabetes diagnosis, heart problems, lung problems, and levels of an inflammation marker called interleukin-6, TAGS scores showed stronger associations with some outcomes than the original epigenetic clocks. For grip strength, cognitive functioning, and the anti-inflammatory marker IL-10, the associations were weaker or inconsistent.

For diabetes, heart problems, lung problems, and markers of inflammation, TAGS remained meaningful predictors even when tested directly against their parent clocks in the same statistical model, though performance varied by clock and outcome. Notably, when evaluated in the hold-out test sample, all five TAGS scores predicted all-cause mortality, while the two oldest clocks, Horvath and Hannum, did not significantly do so on their own. At the same time, for mortality and frailty specifically, GrimAge and DunedinPACE still showed distinctive contributions alongside TAGS, a reminder that the original clocks retain real value for certain outcomes even as the new scores add predictive power for others. The researchers also tested their TAGS scores in three independent external datasets, with uneven results. Associations were stronger in groups with Parkinson’s disease and early-stage cardiovascular disease, but weaker in a group with major depressive disorder.

One area where neither the original clocks nor the new TAGS showed reliable signals was psychological health. Associations with mental health outcomes were consistently weak across all analyses.

Choosing the Right Biological Age Clock for the Right Question

Because each clock tracks different biology, the researchers argue that the choice of clock genuinely matters. As an example, they suggest GrimAge, tied to immune and inflammatory activity, could fit an anti-inflammatory study better than Horvath, which appears largely disconnected from immune processes. DunedinPACE, with its broad reach across multiple body systems, might be better suited for tracking whether interventions such as exercise are slowing the overall pace of decline.

These findings arrive as aging clocks move rapidly from research laboratories into clinical settings and consumer products. In some of those settings, the five clocks are treated as roughly equivalent windows into biological age. This study makes a strong case that they are not, and that treating them as interchangeable risks drawing the wrong conclusions from the biology.

For anyone whose doctor or research team has handed them an “epigenetic age” number, the message from this work is both reassuring and complicated. Reassuring, because the clocks were tied to real, measurable differences in gene activity. Complicated, because what exactly they are measuring depends heavily on which clock was used, a detail that until now has largely been left unexplained.

Disclaimer: This article summarizes findings from a peer-reviewed study for a general audience and is for informational purposes only. It is not medical advice. Epigenetic and biological age tests cannot diagnose disease or predict an individual’s health, and results vary by which clock is used. Anyone with questions about their health or biological age results should consult a qualified healthcare provider.

---