A summary of the related research paper published:
Six Clocks, One Drug, and the Problem of Telling Aging from Disease
Insilico Medicine reanalyzed serum samples from its 12-week phase 2a trial of rentosertib, an AI-designed TNIK inhibitor for idiopathic pulmonary fibrosis, running six independently published proteomic aging clocks across 42 patients at four timepoints. All six clocks pointed the same direction: treated patients looked biologically younger than placebo, with the strongest and most consistent signal at week 4 in the 30 mg twice-daily arm, and reductions of roughly 2.6 to 6.2 predicted years depending on clock and regimen. The effect plateaued or partly reversed by week 12. Critically, the clock changes did not track the drug’s actual lung-function benefit: the arm with the best breathing outcome had the weakest aging signal, and change in forced vital capacity explained a median of only 6 percent of the variance in biological age shift. Supporting analyses showed suppression of senescence-associated proteins, downregulation of growth factor and IGF-binding-protein signaling, and reversal of age-associated protein trajectories benchmarked against 55,319 UK Biobank samples. The paper is a methods demonstration for dual-purpose trial design rather than evidence that rentosertib is a geroprotector.
A drug built to treat scarred lungs may have done something else at the same time, and a group of researchers has just shown how you might catch it.
Rentosertib is an inhibitor of TNIK, a kinase that Insilico Medicine’s AI platform flagged as a target sitting at the intersection of fibrosis and aging biology. In 2024 the company finished a 71-patient phase 2a trial in idiopathic pulmonary fibrosis, a disease that kills most patients within five years of diagnosis. The headline result was modest but real: the highest dose improved lung capacity by about 98 millilitres over 12 weeks while placebo patients declined.
What the company did next is the interesting part. It went back to the frozen blood, measured close to 2,900 proteins in each sample, and ran the results through six different aging clocks. These are statistical models trained on large population datasets to guess a person’s age, or their risk of dying, from the proteins circulating in their blood. If a drug makes people look younger to a clock, that is at least a hint the drug is touching aging itself rather than only the disease.
All six clocks agreed. Patients on rentosertib registered lower biological ages than placebo patients, with the clearest effect after four weeks. The regimen that produced the most consistent signal across every clock was 30 milligrams twice a day. That result is awkward, because a different regimen, 60 milligrams once a day, produced the better breathing outcome. Same total daily dose, different schedules, and the aging readout and the lung readout came apart.
That dissociation is the paper’s real argument. If the clocks were simply detecting sicker or less sick lungs, the arm with the best respiratory result should have looked the youngest. It did not. When the team regressed biological age change on lung function change, respiratory improvement accounted for almost none of the variation.
Two other lines of evidence point the same way. Proteins associated with cellular senescence, the state where damaged cells stop dividing and start leaking inflammatory signals, rose in placebo patients and fell in every treated group. And when the team compared how treatment moved each protein against how normal aging moves that same protein in the UK Biobank, the twice-daily regimen was reversing the aging direction.
The caveats are heavy. Nine to eleven people per arm. Twelve weeks. Everyone had advanced lung fibrosis, which distorts exactly the protein families the clocks weigh most. The signal peaked and then faded. And the study was designed, funded and analyzed by the company that owns the drug.
There is also a wrinkle the authors flag themselves. The single protein contributing most to the aging score across all six clocks was LTBP2, a regulator of the fibrosis pathway. The clocks are, in part, reading the disease they were supposed to see past.
The authors are careful about all of this. Their claim is procedural rather than pharmacological: aging endpoints can be bolted onto disease trials cheaply, using blood already being drawn, and regulators and companies should start doing it. Rapamycin and metformin both took decades to move from disease indication to serious geroprotector candidacy. Measuring aging biology from the first trial onward would compress that timeline considerably.
Actionable Insights
Nothing here is directly usable. Rentosertib is an investigational drug available only in trials, tested in people with a fatal lung disease.
Three take-home points hold up.
First, be skeptical of any single biological age number. The six clocks in this study disagreed substantially with one another. The four clocks trained to predict chronological age tracked real age reasonably well, correlations of 0.70 to 0.84, with typical errors under four years. The two trained on mortality risk barely tracked age at all, correlations of 0.16 to 0.23, with typical errors above 11 years. If your consumer test reports a single number without telling you which family of model it uses, that number is close to uninterpretable.
Second, the reported effect is smaller than it sounds. A reduction of 2.7 to 3.5 predicted years is roughly one unit of the clocks’ own measurement noise. Standardized effect sizes, read from the paper’s figures, cluster around 1.0 to 1.6 in Cohen’s d terms. That is large on paper, but with nine to eleven people per group, effects of that size are routinely inflated and often shrink by half in larger replications.
Third, dosing schedule mattered more than total dose. Two arms took the same 60 milligrams per day and produced different protein signatures and different aging readouts.
Context and Source
-
Open Access Paper: Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment.
-
Institutions: Insilico Medicine AI Limited (Abu Dhabi, UAE), Insilico Medicine Shanghai and Insilico Medicine US (Cambridge, MA), with academic co-authors from Peking University, Westlake University, Peking Union Medical College Hospital, RWTH Aachen University (Germany), Massachusetts General Hospital, the Broad Institute of MIT and Harvard, Stanford University, the University of Washington, and Brigham and Women’s Hospital / Harvard Medical School.
-
Countries: United Arab Emirates, China, United States, Germany. The underlying clinical trial was run entirely in China.
-
Journal: Nature Biotechnology. Published online 07 September 2026
-
Impact evaluation: The impact score of this journal is 44.5 (2025 Journal Impact Factor, publisher-reported), evaluated against a typical high-end range of 0 to 60+ for top general science and biomedical journals, therefore this is an Elite impact journal.