On 7 September 2026, Insilico Medicine and an international team reported in Nature Biotechnology that rentosertib (ISM001-055) — an AI-designed small-molecule TNIK inhibitor developed for idiopathic pulmonary fibrosis (IPF) — shifted blood proteomic aging clocks younger in a Phase IIa cohort. Across 42 patients with longitudinal Olink data covering 2,841 proteins, all six independently built clocks pointed the same way relative to placebo. The peak reported signal was roughly 3–4 years of predicted biological-age reversal at week 4 on 30 mg twice daily, with up to 6 years on one clock. This is an exploratory biomarker read on people with fibrotic lung disease, not an approved longevity drug and not a healthy-volunteer aging trial. Nobel chemist Michael Levitt, quoted in Insilico’s release, put the hedge plainly: the study cannot yet separate slower aging from a treated lung.
First author Alex Zhavoronkov (Insilico) is presenting the work on 8 September 2026 at the Nature conference Redefining Healthcare in the Age of AI at Sorbonne University in Paris. Collaborators include groups at Harvard Medical School, Stanford, the Broad Institute, RWTH Aachen, Peking University, and Westlake University.
Why it matters
Longevity research usually waits for drugs approved for something else — metformin, rapamycin, GLP-1s — then hunts for aging signals after the fact. Rentosertib was built the other way around: Insilico’s AI platforms nominated TNIK as a dual-purpose target tied to fibrosis and to multiple hallmarks of aging, then generated the chemistry. The clinical program’s primary job is still IPF. What is new is that the same Phase IIa serum samples were run through six proteomic clocks that do not share features or training data, and they agreed.
For a stranger, the stake is simple. If disease trials can host credible aging endpoints without waiting decades for lifespan studies, geroprotective candidates can surface years earlier — and regulators have a paper trail to judge. The counter-stake is just as simple: IPF patients are sick, lung function and systemic proteomes move together, and a younger clock score is not extra healthy years. The authors and outside commenters flag that gap. The next decisive experiment is healthy volunteers, not another press chart.
Key numbers
| Metric | Value | Source |
|---|---|---|
| Paper | Nature Biotechnology, announced 7 Sep 2026 | DOI 10.1038/s41587-026-03286-y; Insilico release |
| Drug / target | Rentosertib (ISM001-055) / TNIK | Insilico; prior Nat Biotechnol / Nat Med |
| Trial backdrop | Phase IIa IPF, NCT05938920 | Nat Med 2025; this analysis |
| Proteomic cohort | n=42 with longitudinal Olink | Insilico release |
| Proteins measured | 2,841 | Insilico release |
| Clocks tested | 6 (ProtAge, OrganAge chrono + mortality, PAC, ipfP3GPT, PAOPAC) | Insilico release |
| Direction | All six → lower predicted biological age vs placebo | Insilico release |
| Peak age-clock signal | ~3–4 years younger at week 4, 30 mg BID; up to 6 on one clock | Insilico release |
| Prior FVC (same program, earlier paper) | 60 mg QD: mean FVC +98.4 mL vs placebo −20.3 mL (−62.3 excl. outlier) | Nat Med 31:2602–2610 (2025) |
| External comparison | Trajectories vs 55,319 UK Biobank profiles | Insilico release |
| Data deposit | CNCB OMIX008341 | Insilico release |
Do not treat clock-years as calendar years of life extension. Do not invent per-clock p-values the press release does not give.
How they showed it
The drug’s path. Insilico’s PandaOmics / Pharma.AI stack flagged TNIK across aging-hallmark assessments; Chemistry42 designed rentosertib. Target-to-preclinical-candidate took about 18 months, with the discovery chemistry package in Nature Biotechnology in 2024. Phase IIa IPF results in Nature Medicine (June 2025) met the primary safety endpoint and showed a dose-dependent lung-function trend — including the FVC numbers above for the 60 mg once-daily arm.
Why proteomics were sitting there. The Phase IIa protocol prospectively collected longitudinal serum proteomes for exploratory biomarkers. Those samples, now deposited at CNCB under OMIX008341, became the substrate for the aging-clock paper.
Six clocks, one cohort. Researchers applied ProtAge, OrganAge (chronological and mortality variants), PAC, ipfP3GPT, and PAOPAC to the same 42-patient longitudinal series. Methods range from classical machine learning to deep learning; training targets mix chronological age and mortality risk. Despite that diversity, every clock trended toward younger predicted age on treatment versus placebo. The loudest point estimate in the company summary is week 4 at 30 mg BID.
Not just a happier lung. Insilico reports that the dose with the strongest lung-function gain was not the dose with the strongest age-clock signal — evidence, they argue, that the proteomic shift is not merely a downstream echo of better FVC. Pathway reads frame rentosertib as partly senomorphic, dialing down senescence-linked proteins (including EREG, ESM1, IGFBP4, ITGA2, MMP10, MMP13, SPP1) and growth-factor routes such as RTK–PI3K and RAS–ERK. Those mechanistic claims are hypothesis-generating from the same exploratory dataset.
Open materials. Pipeline code is released as an open-source Python library; proteomic data sit at OMIX008341 so others can rerun the clocks.
What this is not
- Not an approved anti-aging medicine. Rentosertib’s clinical path is IPF; Phase III for fibrosis is underway in China. No regulator has cleared it to reverse aging.
- Not a healthy-volunteer longevity trial. Everyone in this analysis had idiopathic pulmonary fibrosis. Treating a lethal lung disease can move blood proteins that aging clocks also watch.
- Not proof of extra lifespan or healthspan. Clock-years are model outputs. Mortality and function over years were not the endpoints here.
- Not a claim that all six clocks moved by the same amount. The headline agreement is directional; the “up to 6 years” figure is one clock at peak, not a universal effect size.
- Not a substitute for the earlier efficacy paper. FVC gains cited above come from the 2025 Nature Medicine Phase IIa report, not from inventing new spirometry in this aging analysis.
What to watch
- Healthy-volunteer or non-fibrosis cohorts that ask whether the clock shift survives when the lung is not the disease under treatment.
- Whether FDA / BEST-style biomarker qualification ever accepts multi-clock proteomic panels as dual-purpose endpoints — the framework the paper pitches.
- Phase III IPF outcomes for rentosertib: if fibrosis benefit fails, the aging story does not rescue the medicine.
- Independent re-analysis of OMIX008341 with the open pipeline — same direction, or clock-specific noise?
- Whether dose separation between lung benefit and clock benefit holds in larger samples.
Six different blood-aging models, built by different labs, looked at the same 42 fibrosis patients on an AI-designed TNIK pill and came back younger on treatment than on placebo. The punchiest public number is about 3–4 years at week 4 on 30 mg twice daily. That is a serious signal for how to run dual-purpose trials. It is still a signal inside diseased lungs, read with exploratory clocks — not a fountain-of-youth approval. The honest next step is the experiment Levitt asked for: healthy people, same rulers, same drug.
Sources
- Zhavoronkov, A. et al. Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment. Nat Biotechnol (2026). DOI: 10.1038/s41587-026-03286-y
- Insilico Medicine. Nature Biotechnology | Rentosertib proteomic aging clocks (7 Sep 2026). https://insilico.com/news/rnt0709261-rentosertib-proteomic-aging-clocks
- Xu, Z. et al. A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial. Nat Med 31, 2602–2610 (2025). DOI: 10.1038/s41591-025-03743-2
