People with "old" brain astrocytes developed Alzheimer's at 12.6 times the rate of those with "young" astrocytes. In carriers of two copies of the APOE4 allele, the gap reportedly widened to nearly 40-fold, per the Wyss-Coray and Topol review.
That number, from Tony Wyss-Coray and Eric Topol's Nature Medicine review, is read from plasma proteins in a tube of blood. It is the kind of finding that makes the case that biological-age measurement has genuinely arrived.
And in the same review, the authors write that molecular clocks carry "multiple caveats and a lack of readiness for use in medical practice."
Both statements are true, and the distance between them is the entire problem with the field right now.
Both statements are true, and the distance between them is the entire problem with the field right now.
Start with why the astrocyte finding is so striking. It reportedly comes from cell-type clocks that assessed over 40 cell types across more than 60,000 individuals using more than 7,000 proteins, per the review. About one in four participants showed accelerated aging in at least one cell type.
The astrocyte signal reportedly predicted Alzheimer's risk independent of APOE4 status, per the review. That matters because it separates people who carry the risk allele but are not actually accelerating from those who are—something we could not do before.
This is not an isolated trick. Organ clocks built from the same proteomic approach have repeatedly tied a fast-aging organ to disease indexed to that organ: heart age to heart failure, brain age to dementia. In the UK's 1946 birth cohort, all born the same week, organ ages spread across roughly a decade at chronological age 63.
Two systems stand out as gatekeepers. When brain and immune clocks show normal-paced aging, survival over 15 to 17 years approaches 100 percent. The measurement, in other words, is real, reproducible across cohorts, and mechanistically suggestive.
So the clocks can see something. The uncomfortable question is whether what they see is anything a clinician can act on—and here the evidence turns.
The Berlin Aging Study II ran the cleanest test available: sixteen validated markers of biological aging, measured in the same 1,083 people, followed for an average of 7.4 years. Epigenetic clocks, a proteomic clock, telomere length, SkinAge, BioAge, Brain Age, psychological aging—all in one cohort, judged against real outcomes like frailty, mobility, cognition, and morbidity.
Of those sixteen, only two showed strong, consistent associations with age-related outcomes both cross-sectionally and longitudinally: the Allostatic Load Index and DunedinPACE.
Those two were not marginal. Added individually to a prediction model, each reportedly improved accuracy for incident metabolic syndrome, high cardiovascular risk, and incident frailty by up to 24 percentage points, per the Berlin Aging Study II.
But read the result the other way. Fourteen of sixteen widely used aging markers did not clear the bar—and the two survivors include Allostatic Load, which is built from ordinary lab values, not exotic omics.
That is the tension the glossy proteomic headlines obscure. The most fashionable clocks are not automatically the most predictive, and in a head-to-head, most of them underperform.
An npj Aging perspective puts the challenge even more bluntly, asking in its title whether we actually need aging clocks at all. Its argument is definitional: biological age is an abstract, non-measurable quantity that "exists only as the output of the algorithm claiming to measure it."
Every new clock, in other words, defines its own biological age. There is no ground truth to check against, which is why the authors insist that any clock justify its advantage over established alternatives—expert risk scores, direct outcome predictors, large health models—before anyone treats it as actionable.
This is not clock-bashing. It is the demand that a clock beat the cardiovascular risk calculator a physician already has, rather than merely correlate with the same outcomes at greater expense.
And correlation is precisely where the field is stuck. As the lifespan.io summary of the Wyss-Coray–Topol review notes, aging clocks "do not establish causality," were built largely on population-level data, and have not been established for clinical use. There is no standardization across the field and no regulatory approval for the products already being sold.
Topol himself is candid about the commercial mess. A simple epigenetic test can be run for ten or twenty dollars, he told Scientific American, while some companies charge two hundred—and the "lack of standards from one company to the next is disconcerting."
Which brings the real problem into focus. The gap is not a shortage of clocks. It is the missing evidentiary step: proof that moving a clock changes what happens to a patient.
Consider the semaglutide result published in Nature Communications. In a post hoc analysis of an 84-person trial in HIV-associated lipohypertrophy, semaglutide reportedly slowed multiple epigenetic clocks—PhenoAge, PCGrimAge, GrimAge V2, OMICmAge, and DunedinPACE, which ran 9 percent slower than placebo, per Nature Communications.
Set aside whether semaglutide works as a longevity drug; that is a separate argument. The instructive detail is structural: epigenetic aging was not pre-specified, the sample was small, the cohort HIV-specific, and follow-up ran only 32 weeks. The authors say plainly that "prospective trials are needed."
This is the shape of nearly all clock-and-intervention evidence: a clock moved, measured after the fact, with no demonstration that the person got healthier as a result. A slower DunedinPACE is a changed number. It is not yet a prevented disease.
Two pieces of infrastructure actually close this gap, and neither is a new clock.
The first is standardization at the bench. A platform called qCAGEs—Quantitative Cellular AGing Evaluation system—offers a dual-parameter, high-throughput way to screen for drug-induced cellular senescence, explicitly built to move past the unreliability of relying on any single senescence marker. That is the unglamorous plumbing a field needs before its readouts mean the same thing across labs.
The second, and the one that matters most for the causal question, is a properly designed randomized trial. FAXAge is that trial.
FAXAge is a 52-week randomized controlled study in healthy adults aged 65 and older, comparing time-restricted feeding, exercise, and their combination. It uses DNA methylation clocks as biomarkers alongside cognitive, physical, metabolic, and molecular outcomes.
The decisive feature is not the 52 weeks. It is the planned 10-year follow-up—the link between clock movement measured early and actual health outcomes measured later.
That is the experiment the field has been skipping. If a lifestyle intervention slows a methylation clock at one year, and the people whose clocks slowed most are demonstrably healthier a decade on, then the clock has earned its status as a surrogate endpoint. Trials could then read out in months instead of decades.
If the clock moves but the outcomes do not follow, then the number was an expensive correlation, and the whole surrogate-endpoint premise collapses.
Both the Wyss-Coray–Topol review and the npj Aging critique name this as the prize: biological age as a surrogate endpoint that lets geroprotective trials avoid decade-long lifespan observations. Everyone agrees on the goal. What has been missing is the design that can actually prove or disprove it.
So the honest status report is this. Measurement has matured to the point of reading a near-40-fold Alzheimer's risk from plasma proteins—a real, reproducible advance in risk stratification.
But risk stratification is not intervention. Until a trial built like FAXAge shows that bending a clock bends a person's trajectory, biological age remains a powerful way to flag who is in trouble, dressed up as a target for fixing it.
The clocks tell time. Whether resetting them buys anyone more of it is still, in 2026, an open question—and, for the first time, a testable one.