Insilico Medicine has reported a provocative new analysis of its AI-designed idiopathic pulmonary fibrosis drug rentosertib, showing that six independently developed proteomic aging clocks consistently shifted toward younger predicted biological ages among treated patients compared with placebo. The peer-reviewed study, published in Nature Biotechnology on September 7, analyzed serum samples from 42 participants in the earlier randomized Phase 2a IPF trial and measured 2,841 circulating proteins. Across the six aging models, 21 of 54 treatment-versus-placebo comparisons reached the study’s statistical significance threshold, with the clearest signal at week four and in patients receiving 30 mg twice daily.
The finding is unusually interesting because rentosertib was not originally tested in Phase 2a as a longevity medicine. It is an oral small-molecule inhibitor of TRAF2- and NCK-interacting kinase, or TNIK, discovered and designed using Insilico’s artificial-intelligence drug-discovery platform for idiopathic pulmonary fibrosis, a progressive scarring disease strongly associated with aging biology. The underlying Phase 2a trial had already produced a lung-function signal at 60 mg once daily, where forced vital capacity increased by a mean 98.4 mL at 12 weeks compared with a 20.3 mL decline under placebo. Yet the strongest biological-age signal in the new proteomic analysis came from 30 mg twice daily rather than the dose producing the largest FVC change, raising the possibility that the aging-clock effect is not simply a mathematical reflection of improved lung function.
What exactly did the six biological aging clocks measure in the rentosertib study?
Proteomic aging clocks do not measure chronological age in the ordinary sense. Instead, they analyze patterns across proteins circulating in blood and use statistical or machine-learning models trained to estimate whether a person’s molecular profile resembles that of someone biologically older or younger. The six models used in the rentosertib study included ProtAge, OrganAge variants, PAC, ipfP3GPT and PAOPAC, which were developed independently and differ in their training data, architecture and biological targets.
Researchers applied these clocks longitudinally to serum samples collected from 42 patients who participated in the 12-week Phase 2a IPF study. Across all six models, the placebo group showed relatively little movement or slight increases in predicted biological age, whereas treatment groups moved directionally toward younger predicted profiles. Of 54 treatment-versus-placebo comparisons across doses and time points, 21 met the study’s false-discovery-rate threshold of Q below 0.10. Eleven of those significant differences occurred at week four, suggesting that the proteomic effect appeared relatively early and then became less pronounced or plateaued.
The 30 mg twice-daily regimen generated nine significant comparisons, the greatest number among the three dosing schedules. The 60 mg once-daily group produced seven and the 30 mg once-daily group five. Insilico summarized the peak 30 mg twice-daily effect as roughly three to four years younger predicted biological age across several clocks, with one model showing an approximately six-year shift. Those numbers describe changes in mathematical biomarker predictions, however, rather than six literal years being removed from a patient’s biological lifespan.
Does this mean rentosertib actually reversed aging in people?
No. That is the most important limitation of the story. The Nature Biotechnology authors explicitly state that proteomic clocks alone cannot distinguish an aging-specific effect from changes caused by treatment of the underlying disease. IPF itself alters inflammatory, metabolic and tissue-remodeling proteins, so making a patient’s disease biology healthier could potentially make an aging model predict a younger age even if the drug has not changed fundamental human aging or extended healthy lifespan.
The study also involved only 42 patients, lasted 12 weeks and was not designed to measure mortality, frailty, cardiovascular events, dementia, cancer incidence or other outcomes associated with aging. No patient became chronologically younger, and there is no evidence from this analysis that rentosertib extends life. Even Insilico’s own release includes commentary acknowledging that the trial cannot cleanly separate slowing of aging from successful treatment of a diseased lung.
What makes the result scientifically worthwhile is therefore not a claim that an anti-aging drug has been proven. It is the consistency across six independently built proteomic clocks and the possibility that clinical trials for age-related diseases can incorporate aging biomarkers prospectively. If several unrelated models all detect similar changes, researchers have a stronger reason to investigate whether a genuine systemic biological effect exists.
Why is the difference between the 30 mg BID and 60 mg once-daily results so interesting?
The original Phase 2a study enrolled 71 patients with IPF and randomized them to placebo, 30 mg rentosertib once daily, 30 mg twice daily or 60 mg once daily for 12 weeks. The greatest FVC improvement occurred in the 60 mg once-daily group, where mean change reached positive 98.4 mL. The 30 mg twice-daily group showed a much smaller mean FVC increase of 19.7 mL.
The aging-clock results moved differently. Thirty milligrams twice daily produced the most consistent younger-age signal even though total daily dose was identical to 60 mg once daily. The Nature Biotechnology authors suggest that maintaining more stable drug exposure throughout the day might be relevant to pathways associated with aging biology, while a larger single peak may be more important for the pulmonary FVC response. That remains a hypothesis rather than a demonstrated pharmacological explanation.
The divergence is useful because it weakens one simplistic explanation: that patients merely looked younger on the protein clocks because their lungs functioned better. If lung-function improvement were the only driver, researchers might have expected the 60 mg once-daily arm to dominate both analyses. Instead, the dose-response patterns diverged, suggesting that TNIK inhibition may influence several biological processes with different exposure requirements.
What did the comparison with more than 55,000 UK Biobank participants add?
Researchers went beyond running aging clocks and compared rentosertib-induced protein changes with ordinary age-associated protein trajectories observed in 55,319 older adults from UK Biobank. Of 2,832 proteins available in both datasets, 758 showed significant age-related changes in the population data. Rentosertib-modulated proteins were enriched for these age-associated proteins by about 1.74-fold compared with background expectation.
The 30 mg twice-daily regimen showed a statistically significant negative correlation with normal aging-associated protein changes, with a Spearman correlation coefficient of minus 0.30 and a p-value below 0.01. In other words, proteins that ordinarily tended to move one way with advancing age were more likely to move in the opposite direction under that rentosertib regimen. The 60 mg once-daily arm did not show the same significant correlation despite having generated the stronger FVC signal.
Researchers also identified changes in pathways associated with cellular senescence, metabolism and growth-factor signaling. These mechanistic findings are exploratory, but they provide a biological framework for why TNIK inhibition might influence systemic age-associated protein patterns rather than affecting only fibrotic tissue.
How was rentosertib originally discovered using artificial intelligence?
Rentosertib, previously known as ISM001-055 or INS018_055, is a small-molecule inhibitor of TNIK. Insilico used its AI-powered target-discovery system to identify TNIK as relevant to fibrosis and several biological hallmarks associated with aging, then used generative chemistry tools to design drug candidates capable of inhibiting that target. The program subsequently moved through preclinical development and human testing.
That development pathway has made rentosertib a flagship example for the emerging AI-drug-discovery industry. Artificial intelligence can generate targets and molecular designs rapidly, but the field ultimately succeeds only if those molecules survive the same toxicology, manufacturing, randomized clinical trials and regulatory review required for conventionally discovered drugs.
Rentosertib has progressed unusually far for an AI-originated molecule. Following Phase 2a, Insilico initiated Phase 3 clinical development in China for IPF, moving the program from an interesting demonstration of generative chemistry toward a trial intended to determine whether the drug produces reproducible clinical benefit in a much larger population.
What did the original Phase 2a trial say about rentosertib safety?
The earlier randomized study was primarily designed to assess safety and tolerability. Treatment-emergent adverse events occurred in 72.2% of patients taking 30 mg once daily, 83.3% taking 30 mg twice daily, 83.3% taking 60 mg once daily and 70.6% receiving placebo. Treatment-related serious adverse events were uncommon, although liver toxicity and diarrhea were among events leading to discontinuation.
There were also three acute IPF exacerbations in the 60 mg once-daily group compared with one in placebo during the short study, and all three patients in the high-dose group were hospitalized. The small numbers make it impossible to determine whether that imbalance represents chance, background disease or a treatment-related issue, but it is an example of why a promising FVC number from 71 patients cannot replace larger safety experience.
Phase 3 will consequently need to determine not just whether rentosertib preserves lung function but whether the benefit-risk profile remains acceptable when substantially more patients are exposed for longer periods.
Could aging clocks eventually become endpoints in ordinary drug trials?
That may be the most durable implication of the Nature Biotechnology paper. Clinical trials normally evaluate one disease at a time, even though many chronic diseases share biological mechanisms associated with aging. Researchers developing medicines for pulmonary fibrosis, cardiovascular disease, neurodegeneration or metabolic disease may therefore be altering biological pathways relevant to aging without measuring those effects systematically.
Adding proteomic or other aging clocks to existing trials could allow researchers to identify those secondary effects without constructing an enormous longevity study from the beginning. A medicine showing consistent aging-associated biomarker changes could then be investigated specifically in healthier older populations or in trials designed around functional aging endpoints.
The danger is overinterpretation. Aging clocks remain predictive models rather than validated surrogate endpoints proving longer life or freedom from age-related disease. A drug should not receive an “anti-aging” label because a blood test predicts a younger biological age after several weeks of treatment.
Rentosertib illustrates both sides of that opportunity. Six independent models moved in the same direction, and the protein-level comparison with UK Biobank adds mechanistic depth. Yet the analysis involves just 42 patients with a serious age-related lung disease, making it impossible to determine whether the result represents genuine geroprotection.
What should we watch next for rentosertib?
The most important program remains Phase 3 IPF development, not a longevity trial. If rentosertib cannot demonstrate meaningful and safe antifibrotic efficacy in the disease for which it is being developed, its aging-clock profile becomes considerably less clinically relevant. Conversely, successful Phase 3 results could create a large clinical dataset in which biological-aging measurements might be explored prospectively.
A separate study in healthier individuals would ultimately be necessary to test whether the proteomic effects persist independently of IPF. Longer treatment would also be required to determine whether the early week-four signal is durable or whether the clocks simply settle into a temporary pharmacodynamic equilibrium.
For now, the most accurate conclusion is also the most interesting one. An AI-designed drug developed for pulmonary fibrosis caused six independent protein-based models to predict younger biological profiles in treated patients, and the strongest effect did not occur at the dose producing the best lung-function response. That is enough to justify serious aging research. It is not enough to say that researchers have reversed human aging.
