Insilico.com

A peer-reviewed paper published Monday in Nature Biotechnology has delivered a result the longevity field has been waiting years for: the first published clinical evidence that a drug can reduce predicted biological age across six independently built measurement tools — simultaneously — in human patients. The molecule is rentosertib, developed by Insilico Medicine for idiopathic pulmonary fibrosis (IPF) — and unlike rapamycin or metformin, drugs that geroscientists borrowed from other disease areas and tested against aging as an afterthought, rentosertib was designed by Insilico's AI platform for aging biology from the start with aging embedded into its target selection. The six clocks all pointed in the same direction.

The paper's lead author, Insilico founder and co-CEO Alex Zhavoronkov, is scheduled to present the findings today at the Nature conference "Redefining Healthcare in the Age of AI" at Sorbonne University in Paris.

What Six Protein Clocks Actually Measure — and Why Agreement Among Them Matters

A proteomic aging clock is a mathematical model trained on the blood protein profiles of thousands of people, using the patterns of circulating proteins to estimate how biologically old a person is — not in calendar years, but in terms of how far their body's molecular machinery has drifted from the norms of a younger body. Research has demonstrated that proteomic aging clock predicts mortality risk across diverse populations. The models are trained independently, each by a separate research group, using different training datasets, different machine-learning methods, and different targets — some optimized to predict a person's chronological age from their protein patterns, others optimized to predict mortality risk. They do not share code, features, or training data.

That independence is precisely what makes unanimous agreement across six of them statistically meaningful. The six models used in the rentosertib study — ProtAge, OrganAge (in two variants: chronological and mortality-risk-based), PAC, ipfP3GPT, and PAOPAC — were developed by separate teams at Harvard Medical School, Oxford, Peking University, and Insilico itself. Their methodologies span classical machine learning (OrganAge, PAOPAC, PAC) and deep learning (ProtAge, ipfP3GPT). If five of the six had agreed and one dissented, the dissenter could be the model artifact. When all six converge, the cross-model consensus is the scientific case.

Researchers collected blood samples from 42 of the Phase IIa trial participants at multiple time points over 12 weeks, measuring the concentrations of 2,841 proteins using the Olink Explore 3072 proximity extension assay platform. They applied all six independently developed aging clocks to those protein profiles and compared the biological age estimates in treated versus placebo patients at each time point. Across 54 treatment-versus-placebo comparisons, 21 reached statistical significance after false discovery rate adjustment.

Michael Levitt — who shared the 2013 Nobel Prize in Chemistry for developing multiscale models of complex chemical systems — offered the most direct public assessment of the finding: the cross-model consensus is what convinced him, not the size of the effect. These models share neither their features nor their training data, he noted. But he also stated plainly what the trial cannot yet prove: it cannot separate a slower aging process from a treated lung. The experiment in healthy volunteers is the one he wants to see next.

Why Designing for Aging Differs From Repurposing

Every drug that geroscientists have studied in humans for longevity effects — rapamycin, metformin, the senolytic combination dasatinib plus quercetin — was originally developed for something else. Rapamycin is an immunosuppressant. Metformin is a diabetes drug. Quercetin is a flavonoid. These compounds happen to interact with aging-relevant pathways, and researchers are studying them retrospectively for longevity potential. None of them was designed with the hallmarks of aging as the primary specification.

Rentosertib took a different path. Using its PandaOmics AI target-discovery platform, Insilico evaluated the gene TNIK — TRAF2- and NCK-interacting kinase — against the Lopez-Otin hallmarks of aging framework, which identifies 12 molecular and cellular processes that both drive and characterize biological aging. Research on TNIK hallmarks of aging assessment showed TNIK scored as a high-priority target across six of those 12 hallmarks, including cellular senescence, altered intercellular communication, deregulated nutrient sensing, loss of proteostasis, genomic instability, and mitochondrial dysfunction. It was also implicated in fibrosis signaling — making it a dual-purpose target.

The generative chemistry platform Chemistry42 then designed the small molecule specifically to inhibit TNIK. The entire journey from target identification to preclinical candidate nomination took approximately 18 months — a fraction of the typical two-and-a-half to four years for conventional drug discovery programs.

The practical consequence of this design philosophy: the anti-aging biology was embedded in the molecule from the first day, not layered on afterward. Rapamycin's geroprotective effects were discovered decades after the drug was developed. Rentosertib's were anticipated.

How Rentosertib Suppresses the Molecular Signature of Aging

Across the 12-week trial, rentosertib acted as what biologists call a senomorphic agent — a compound that suppresses the harmful secretions of senescent cells without necessarily eliminating those cells. Senescent cells — cells that have permanently stopped dividing in response to stress or damage — secrete a cocktail of inflammatory proteins known as the SASP (senescence-associated secretory phenotype), which drives chronic inflammation, tissue damage, and accelerated aging in surrounding tissue.

Rentosertib suppressed specific SASP-related proteins — including EREG, ESM1, IGFBP4, MMP10, MMP13, and SPP1 — while downregulating two major growth-factor signaling pathways linked to accelerated aging, RTK–PI3K and RAS–ERK. It did this through TNIK inhibition — a mechanism distinct from the established senomorphic agents (rapamycin blocks mTOR; JAK inhibitors block the JAK-STAT pathway). TNIK inhibition is a pharmacologically new entry point into the same biological process.

To validate that the drug was reversing general aging-protein trajectories and not merely treating lung disease, the research team compared the proteins altered by rentosertib against the protein expression patterns of 55,319 participants in the UK Biobank — one of the largest and most comprehensive health datasets in the world. The proteins rentosertib changed were enriched among those that normally shift with aging in the general population — and they moved in the opposite direction of those typical age-related trajectories. That is the proteomic signature of a drug that is doing something to the biology of aging specifically, not merely treating a lung disease whose proteomic footprint happens to overlap with the aging-clock features.

Ludger Goeminne, a Research Fellow in Medicine at Harvard Medical School and a co-author of the paper, noted that the pathway analysis confirmed rentosertib's biological impact extends beyond fibrosis.

What Does It Mean That the Geroprotective Signal Was Strongest at a Different Dose?

The most scientifically significant detail in the paper may be a dose-response divergence that the headline number does not capture. In the Phase IIa trial, the strongest improvement in forced vital capacity — the lung-function measure that tracks disease progression in IPF — appeared in the 60 mg once-daily dose group, where patients gained a mean of 98.4 mL of lung capacity against a mean decline of 20.3 mL in the placebo group. But the strongest and most consistent biological age reversal signal appeared in participants receiving 30 mg twice daily — a different dose schedule, different total exposure pattern, strongest clock signal at Week 4. The Phase IIa TNIK inhibitor trial data showed this dose-response divergence clearly.

If the aging-clock reversal were simply a downstream consequence of treating IPF — if the clocks were measuring "this person's lungs are less diseased" rather than "this person's biology is younger" — the aging and lung-function signals would track each other at the same dose. They don't. The divergence is the evidence that rentosertib's geroprotective activity operates through at least partially independent mechanisms from its anti-fibrotic activity.

Prof. Jing-Dong Jackie Han of the Peking-Tsinghua Center for Life Sciences, whose team developed the PAOPAC clock used in the study, said the cross-model consistency demonstrates that rentosertib's effect on aging-related proteomic signals is not a model-specific artifact, but rather a highly robust biological phenomenon.

What the Study Cannot Prove — and What Comes Next

The researchers are explicit about the limits of what the data establishes. Proteomic clocks measure patterns of protein expression that correlate with aging in population studies; they are mathematical models of aging, not direct measures of the biological mechanisms that cause it. A clock shift in an IPF cohort cannot fully disentangle the effects of treating a serious disease from the effects of genuine systemic geroprotection. The 42-participant sample is small, the 12-week duration is short, and all participants had a disease that itself alters the serum proteome in ways that partially overlap with aging signatures.

Independent commentators reached similar conclusions. Analysis from remio.ai noted that the study found a younger biomarkers not younger patients pattern — a coordinated shift in blood proteins associated with younger predicted age, but not proof that rentosertib reversed human aging. The study also acknowledged limitations including the modest sample size, the 12-week observation window, and reliance on computational approaches without complementary omics modalities, as detailed in biological age reversal limitations covered by Unite.AI.

The next required experiment, as Levitt explicitly named it, is healthy volunteers. A trial in people without IPF — people whose serum proteome is not already shaped by fibrosis — would directly test whether the biological age signal survives in a population where the disease confound does not exist.

Does Rentosertib Treat Aging or Just Treat Lung Disease Better?

This is the study's central unresolved question, and the researchers say so plainly. Because all 42 participants in the proteomic analysis had IPF, the clock reversal could reflect the drug treating a disease that shares proteomic overlap with aging, rather than the drug acting on aging itself. The dose-response divergence — strongest clock signal at a different dose than strongest lung benefit — is the main evidence that the effects are at least partially separable. But definitive proof requires a trial in people without IPF, which the study explicitly calls for as the critical next experiment.

Why the Framework May Matter More Than the Finding

The study's authors, led by Zhavoronkov, argue that the trial's most lasting contribution may be the methodology it demonstrates rather than the specific result. The traditional path to identifying a geroprotective drug has been to wait for a drug approved for a different disease, study it for years in aging models, and eventually run a dedicated longevity trial in healthy volunteers — a process that takes decades. The rentosertib trial demonstrates that aging biomarkers can be embedded prospectively, from the start, as exploratory endpoints within a standard disease-focused Phase IIa trial — adding geroprotective assessment at minimal incremental cost and no delay to the primary clinical program.

The study lays out a stepwise framework: first embed aging and senescence biomarkers as exploratory endpoints in disease trials; then, as the data matures, pursue formal qualification of those biomarkers under the FDA Biomarker Qualification Program and the FDA-NIH BEST (Biomarkers, EndpointS, and other Tools) framework, which provides a regulatory pathway for qualifying aging biomarkers as composite clinical endpoints. The BEST framework does not yet recognize proteomic aging clocks as qualified surrogate endpoints — that qualification would require much larger studies. But the methodology for generating that evidence base is now published and open-sourced.

All proteomic data from the Phase IIa trial have been deposited publicly with the China National Center for Bioinformation under accession OMIX008341, and the Insilico open-source analysis pipeline is available as a Python library on GitHub.

Rentosertib has now entered a Phase III trial in IPF, enrolling approximately 320 patients across 47 centers in China — a larger population and longer observation period that may strengthen the biological age signal alongside the primary lung-function endpoints.

On the business side, Insilico reported revenues of approximately $106 million in the first half of 2026, a 287% year-over-year increase, and achieved its first profitable half-year since listing on the Hong Kong Stock Exchange in December 2025 — with an adjusted net profit exceeding $51 million. The company's cumulative contract value across major partnerships — including collaborations with Eli Lilly, Takeda, Servier, and SK Biopharmaceuticals — has reached approximately $11 billion since 2021.

For the broader longevity field, the most significant number in the study is not the financial one. It is this: six independent aging clocks, built by six independent research groups, using methods ranging from classical machine learning to deep learning, trained on targets ranging from chronological age to mortality risk — all looking at the same 42 patients, and all pointing in the same direction.

Frequently Asked Questions

What is a proteomic aging clock and how does it differ from an epigenetic clock?

Both are computational models designed to estimate how biologically old a person is — independent of their calendar age. Epigenetic clocks (like Horvath's clock, developed in 2013) measure patterns of DNA methylation to estimate biological age. Proteomic aging clocks measure patterns of circulating proteins in blood. Proteins are more proximal to biological function: they reflect what the body is actively doing — inflammatory signaling, tissue remodeling, metabolic regulation — rather than the epigenetic marks governing gene expression. Proteomic platforms like Olink Explore 3072 can now measure nearly 3,000 proteins simultaneously from a small blood sample, enabling the construction of high-dimensional statistical models of aging that are more directly interpretable in terms of known biological mechanisms.

Why does it matter that six different clocks were used instead of just one?

Because each clock was built independently — different training datasets, different machine-learning methods, different target outcomes (some predicting chronological age, others predicting mortality risk). A single clock might show a drug-induced shift because the clock happens to be sensitive to proteins altered by that specific drug or disease. When six clocks trained by different groups with different methods all agree, the likelihood that the signal is a model-specific artifact drops dramatically. The cross-model convergence is the primary statistical argument for robustness in this study.

Can rentosertib actually reverse aging — and when will we know?

Not proven yet. The study found a coordinated shift in blood proteins associated with younger predicted biological age in treated IPF patients. It cannot prove the drug reversed the aging process itself, because all study participants had IPF — a disease that alters the very protein patterns the clocks are trained on. The critical next test, which Nobel laureate Michael Levitt explicitly named, is a trial in healthy volunteers without IPF. If the biological age signal persists in people without the disease confound, the case for systemic geroprotective activity becomes substantially stronger. The ongoing Phase III trial in IPF (320 patients, 47 centers in China) is designed to assess lung function, not aging per se — but its larger sample may provide additional proteomic data over a longer time horizon.

What makes rentosertib different from other longevity drugs like rapamycin or metformin?

Rapamycin, metformin, and the senolytic combination dasatinib plus quercetin are repurposed drugs: they were developed for other indications (organ transplant, diabetes, and cancer, respectively) and discovered through subsequent research to interact with aging-relevant pathways. None was designed with aging biology as the primary specification. Rentosertib was. Insilico's AI platform scored TNIK — the drug's molecular target — against the 12 hallmarks of aging framework, found it relevant to six hallmarks, and designed a molecule specifically to inhibit it. That makes rentosertib, if its geroprotective effects are confirmed in further studies, the first drug created from scratch specifically for the biology of aging to produce a clinical signal in humans.