The press version of the rentosertib story reads like a milestone: an experimental drug made patients roughly three years biologically younger in twelve weeks. The paper itself says something narrower. The gap between those two sentences is where this field currently lives.
The analysis, published in Nature Biotechnology on September 7, 2026, is not a new trial. It is a secondary look at serum samples from an already-completed study. Evidence level: post-hoc biomarker analysis nested inside a small, early-stage randomized controlled trial.
That parent trial was a 12-week, double-blind, placebo-controlled phase 2a study of rentosertib, an Insilico Medicine inhibitor of the kinase TNIK, in idiopathic pulmonary fibrosis. It enrolled 71 people; 55 completed; 43 consented to the proteomic substudy; one was excluded for a missing week-12 sample. Forty-two patients were analyzed.
Six proteomic clocks were run on those samples: ProtAge, ipfP3GPT, OrganAge chronological, OrganAge mortality, PAOPAC, and PAC. Four were trained to predict chronological age. Two were trained to predict mortality risk — a different question entirely.
All six clocks generally moved toward younger predicted ages in treated arms, while placebo patients showed minimal change or slight increases. The largest reductions appeared by week four, with little additional movement over the remaining eight weeks.
At week four, the four chronological clocks rated the 60 mg once-daily group about three years younger than baseline, with individual estimates ranging from 2.71 to 3.46 years. Those changes were statistically significant versus placebo. The two mortality-trained clocks did not show a comparable change at that dose.
The 30 mg twice-daily regimen was the most consistent performer. It registered across five of six clocks at week four, detected by both the chronological and mortality classes. Organ-specific mortality clocks also flagged reductions in predicted age for artery, brain, pancreas, stomach, and immune system in selected treatment groups.
Now the part that should temper every headline. All six clocks leaned on LTBP2, a fibrosis-associated protein, as a major contributor to the younger estimates. If one fibrosis protein is driving agreement across six independent models, their concordance may be measuring the disease improving, not aging slowing.
The researchers checked this directly. Changes in predicted biological age did not closely track changes in lung function. That argues against a pure fibrosis artifact, but it does not settle the question — forced vital capacity does not capture every aspect of IPF.
Two further signals cut both ways. Compared against 55,319 older adults in the UK Biobank, rentosertib-modulated proteins were enriched 1.74-fold for age-associated proteins, and the split-dose group moved opposite to normal aging trajectories. The single daily dose did not show that pattern. Senescence-associated proteins rose in placebo and fell with treatment, but no senescent cells were measured in tissue.
That is the honest state of play: suggestive, mechanistically consistent, and not proof.
That is the honest state of play: suggestive, mechanistically consistent, and not proof.
The structural problem is older than proteomics. As Cummings and Kritchevsky laid out in GeroScience in 2022, there is no consensus on what constitutes an acceptable clinical endpoint for a geroscience prevention trial, and a biomarker only becomes a surrogate once you show that treatment-induced change in the marker reliably predicts the magnitude of change in the clinical outcome. Evidence level: expert review and position paper.
Nobody has done that validation for aging clocks. Not epigenetic, not proteomic.
The epigenetic clocks went first, starting in 2013, and have the longer track record. A 2026 study in Nature Medicine built TranslAGE, a harmonized database of 51 longitudinal interventional studies, and calculated 16 prominent epigenetic clocks plus 94 other DNA methylation biomarkers across them. Evidence level: meta-analysis of existing interventional cohorts, not new trials.
Three findings from that work matter here. Clocks trained on mortality or pace of aging responded most strongly and agreed with one another. Pharmacological and lifestyle interventions drove the largest biomarker responses; pharmacological was the only category with significantly larger effect sizes. And study population mattered — several biomarkers were more responsive in disease groups than in healthy people, likely because sicker people have more room to move.
Only two intervention classes met the authors' stricter bar: TNF-reducing therapies, and two types of Mediterranean diet. Both had to modify the same biomarkers, same direction, same magnitude, and replicate in a second study.
That is what a candidate surrogate looks like in practice. It is a much higher bar than "the number went down."
Which is why the diet study from the University of Sydney deserves a careful reading. Researchers randomized 104 Australians aged 65 to 75 to one of four diets for four weeks, then calculated a biological age score from 20 biomarkers including cholesterol, insulin, and C-reactive protein. Evidence level: small short-term RCT, biomarker endpoints only, no clinical outcomes.
Three of four diet groups showed lower estimated biological age. The strongest signal came from the omnivorous high-carbohydrate group — 14 percent of energy from protein, 28 to 29 percent from fat, 53 percent from carbohydrates. The omnivorous high-fat group showed no meaningful change.
Four weeks. Twenty biomarkers. A meaningful shift. The authors themselves call it an early signal and say plainly that it is too soon to claim any effect on lifespan or age-related disease.
The lesson is not that diet does nothing. It is that clock readings move fast and for reasons that may have nothing to do with the rate of aging.
Meanwhile, one of the field's more rigorous ongoing tests is deliberately not built around a clock. At UConn Health, Jenna Bartley is leading a 20-week, blinded, multisite trial of ketone ester supplementation in older adults with early frailty. Evidence level: ongoing randomized trial, no results yet.
Its primary outcome is a composite vigor-frailty score combining physical and cognitive function — a functional endpoint, not a biological age estimate. One participant, Judy Donnelly, had a hip replacement mid-study and finished leg-pressing 203 pounds, but neither she nor the researchers know whether she received the supplement.
That design choice is telling. When the goal is healthspan, the people running the trial still want to measure healthspan.
The commercial logic pushing clocks forward is real. Regulators approve treatments for diseases, not for aging, because aging has no agreed clinical endpoint. Without approved indications, investment in geroprotective drugs has lagged. Biomarkers that show a biological signal in months instead of decades could break that deadlock, making early-phase trials faster, smaller, and cheaper.
But a pharmacodynamic marker is not a surrogate endpoint. The first tells you a drug hit its target. The second tells you patients will live better or longer because of it.
Vadim Gladyshev, a Harvard Medical School physician and coauthor on the rentosertib analysis, framed it precisely: "This study illustrates how biomarkers of aging can be incorporated into conventional clinical trials. If validated more broadly, such approaches could allow us to learn not only whether a treatment affects a particular disease, but also whether it influences some aspects of aging biology."
Validated more broadly. That phrase is carrying the entire field right now.