An AI Lung Drug Shifted Biological Age Clocks in Patients

Rentosertib, an AI-designed lung drug, cut predicted biological age by 2.71 to 3.46 years on four protein clocks in a 12-week fibrosis trial. The reading comes from blood, not from extra birthdays, and it comes from people who already have a scarring lung disease.

Insilico Medicine published the analysis on September 7, 2026, in Nature Biotechnology, then restated it as a 3-to-4-year reversal. The underlying study is narrower, and the company’s own next trial still scores breathing.

Six Protein Clocks Read Younger Blood in a Fibrosis Trial

Rentosertib is an oral inhibitor of TNIK, a kinase Insilico’s software flagged as a fibrosis and aging target, then built a molecule against with its Chemistry42 generator. The first job was idiopathic pulmonary fibrosis, an age-tied lung disease with a median survival of 2 to 4 years after diagnosis in the United States, where incidence sits between 10 and 60 cases per 100,000 people and rises about tenfold after 65.

The aging work is a second pass on GENESIS-IPF (NCT05938920), a randomized, double-blind, placebo-controlled phase 2a study run in China in 2023 and 2024. Of 128 people screened, 71 were randomized to 30 mg once daily, 30 mg twice daily, 60 mg once daily, or placebo. Forty-three of them consented to extra serum screening; one lacked an end-of-trial sample, leaving 42 Asian patients, mean age 67.1 years, with blood drawn at baseline and at weeks 2, 4, and 12.

Researchers ran those sera, 2,841 proteins on an Olink Explore 3072 panel, through six proteomic clocks on the same samples: ProtAge, OrganAge in chronological and mortality versions, PAC, ipfP3GPT, and PAOPAC. All six pointed younger in treated arms. Placebo moved little or slightly older. First author Alex Zhavoronkov, Insilico’s founder and co-CEO, presented the paper on September 8 at a Nature meeting at Sorbonne University in Paris.

The Dose That Helped Lungs Was Not the Clock Favorite

The original trial’s best published lung signal sat in a different arm from the broadest clock signal. At 12 weeks, the 60 mg once-daily group posted a mean FVC change of +98.4 ml (95% CI 10.9 to 185.9) against -20.3 ml (95% CI -116.1 to 75.6) on placebo. Forced vital capacity is the amount of air a person can blow out, and it is the number IPF trials live or die on.

On the clocks, that same 60 mg arm at week 4 moved all four chronological models by -2.71 to -3.46 years, with no significant shift on either mortality-trained clock. The 30 mg twice-daily arm, which is the same total daily milligrams split in two, produced the most consistent clock pattern, with nine significant tests out of the comparisons run and movement on both chronological and mortality models. Insilico later described a peak effect at week 4 in that twice-daily group.

THE DOSE SPLIT IN THE AGING SET

The aging paper treats that mismatch as a hint that the clock shift is not only the lung getting better. It also says the two signals cannot be fully pulled apart in people who have IPF. By week 12 the significant tests thinned; the authors call the shape a plateau rather than a rebound, because no arm-clock pair then shifted further.

What the Six Clocks Measure

These models estimate age or death risk from protein patterns in blood. They are not a stopwatch on cells. The four chronological clocks (ProtAge, OrganAge chrono, ipfP3GPT, PAOPAC) tracked calendar age in this cohort with Spearman correlations of 0.70 to 0.84 and, after a simple offset correction, errors below 4 years. The two mortality clocks (PAC and OrganAge mortality) correlated only 0.16 to 0.23 with calendar age, which the authors expected in a sick group, because those models are built to see disease burden as extra age.

Across 54 arm-clock-timepoint tests, 21 met a false-discovery cutoff of Q below 0.10, far above a permutation null of 0.15. Eleven of those 21 clustered at week 4. Chronological clocks agreed with one another (r 0.78 to 0.89) more than they agreed with mortality clocks (r 0.25 to 0.61).

THE SIX MODELS ON THE SAME SERA

• ProtAge: A deep-learning clock trained to predict chronological age from proteins.
• OrganAge chrono: A classical model of calendar age, with organ-level variants.
• ipfP3GPT: A deep-learning chronological clock, one of the six applied to every consenting patient.
• PAOPAC: A classical chronological clock, grouped with the other age-trained models.
• PAC: A mortality-trained clock that tracks death risk rather than birthday age.
• OrganAge mortality: The death-risk twin of OrganAge, including an artery submodel that moved -6.95 to -16.57 years in treated arms, an exploratory organ reading the authors kept separate from the six-clock headline.

Company copy called the six clocks independent groups. The author list includes Insilico scientists beside labs at Harvard Medical School, Peking University, and others, so the set is methodologically mixed, not a panel of six outside vendors. Linear mixed models found 326 proteins on a different path from placebo, against two proteins in the placebo arm. The twice-daily group carried 142 unique protein changes, the widest proteomic footprint, matching its clock record.

Authors Cannot Split a Younger Clock From a Treated Lung

The abstract is blunt: proteomic clocks alone cannot deconvolute aging- and disease-specific effects. Full separation, the authors write, is not achievable inside an IPF cohort and needs the drug or its mechanism in healthy volunteers. Pathway work found senescence-linked proteins moving down, including a SenMayo signature, plus shifts in growth-factor signaling, which they offer as indirect support rather than proof of systemic rejuvenation.

Michael Levitt, the 2013 Nobel laureate in chemistry, put the same limit in the company’s release, and he did not dress it up.

Six proteomic clocks from six independent groups, applied to the same 42 patients, all reported a younger biological age in the treated arms. What convinces me is not the size of the effect but the agreement, because these models share neither their features nor their training data. This trial cannot yet separate slower aging from a treated lung, and the authors say so plainly. The experiment in healthy volunteers is the one I want to see next.

Michael Levitt, PhD, 2013 Nobel Prize laureate in Chemistry, Insilico Medicine release

Zhavoronkov has said that if biological age is lower, a person is likely to die later, while warning that the result is early and that long-term effects in larger groups are unknown. Ludger Goeminne, a research fellow in medicine at Harvard Medical School and a co-author, said organ-specific clocks moved and that pathway analysis showed an impact beyond fibrosis reduction. That is not a license. No medicines regulator treats a proteomic age score as a surrogate endpoint that can carry a drug.

WHAT WE KNOW

• The clock direction: All six models predicted lower biological age in treated arms versus placebo over 12 weeks.
• The week-4 range: Four chronological clocks in the 60 mg once-daily arm read -2.71 to -3.46 years.
• The lung number: The same 60 mg arm gained a mean 98.4 ml of FVC at 12 weeks in the parent trial.

WHAT IS UNCONFIRMED

• True geroprotection: Whether the clocks moved because aging slowed, because a fibrotic lung got quieter, or both.
• People without IPF: Whether the same protein shift appears in healthy volunteers.
• Felt youth: Whether anyone in the trial felt younger or more energetic; the aging paper does not report that.

Insilico’s own launch post went further than the figures, saying that after 12 weeks all six clocks read patients 3 to 4 years younger. The paper’s peak is week 4, the 2.71-to-3.46-year band is the 60 mg chronological set, and the twice-daily arm is the one with the widest agreement. That gap between the post and the plots is the story the clocks actually support.

🧬 Rentosertib just turned back biological age in people.

⏱️ 6 aging clocks. 6 independent groups.
After 12 weeks, all 6 clocks read them 3 to 4 years younger.

🚀 The first AI-designed drug. The first clinical demonstration of 3–4 year biological age reversal.

The first… pic.twitter.com/2ZV6x1iF1R

— Insilico Medicine (@InSilicoMeds) September 7, 2026

Sixteen Patients Left the 12-Week Study Early

GENESIS-IPF’s primary endpoint was safety: the share of patients with at least one treatment-emergent adverse event. Those rates were 72.2% on 30 mg once daily (13 of 18), 83.3% on 30 mg twice daily (15 of 18), 83.3% on 60 mg once daily (15 of 18), and 70.6% on placebo (12 of 17). Treatment-related serious events were low and similar across arms. The events that most often stopped treatment were liver toxicity or diarrhea.

WHO FINISHED THE 12 WEEKS

• Dropouts: 16 of 71 patients, or 22.5%, stopped before week 12.
• Completers: 55 patients, 77%, finished the placebo-controlled period.
• High-dose arms: Only 12 of 18 (67%) finished in both the 30 mg twice-daily and 60 mg once-daily groups, versus 16 of 18 (89%) on 30 mg once daily and 15 of 17 (88%) on placebo.
• Aging subset: The clock analysis used the 42 people with complete serum timepoints, including 9 on 60 mg, not the full randomized 71.

The clock-favorite twice-daily arm was also a high-dropout arm in the parent study. A permutation check still held after the authors pulled six people with higher-grade adverse events, so they do not chalk the clock shift up to those cases alone. Approved IPF drugs, nintedanib and pirfenidone, slow decline; they have not been shown to reverse the course of the disease. That is the bar rentosertib has to clear on FVC, not on a protein score.

Phase 3 Scores Breath Over 52 Weeks, Not Birthdays

Insilico announced on July 7, 2026, that it had started a 52-week Phase 3 study in 320 patients (NCT07687459, also CTR20262475). The U.S. registry listed the trial as not yet recruiting in that July update, with an estimated start of August 30, 2026, and an estimated completion of October 30, 2029, across 47 sites in China. The design is randomized, double-blind, and placebo-controlled, with once-daily rentosertib over 52 weeks.

Zuojun Xu of Peking Union Medical College Hospital is leading principal investigator, with Nanshan Zhong of the Chinese Academy of Engineering and Chang Chen, president of Shanghai Pulmonary Hospital, as co-leads. Carol Satler, Insilico’s senior vice president for clinical development, non-oncology, said the Phase 3 is meant to test whether the Phase 2a safety profile and lung-function signal become a meaningful benefit in IPF. The aging clocks are not the registrational endpoint.

FROM TARGET TO A YEAR-LONG LUNG TRIAL

1. February 2023: The U.S. FDA grants rentosertib (then ISM001-055) orphan designation for IPF.
2. March 2024: Nature Biotechnology publishes the discovery path, AI target pick to first-in-human, after about 18 months to a preclinical candidate.
3. June 2025: Nature Medicine publishes GENESIS-IPF: 71 patients, 12 weeks, FVC +98.4 ml at 60 mg once daily versus -20.3 ml on placebo.
4. July 7, 2026: Insilico announces Phase 3 initiation for 320 patients over 52 weeks, still an IPF lung trial.
5. September 7, 2026: Nature Biotechnology publishes the six-clock reanalysis of 42 of those patients.

Feng Ren, co-CEO and chief scientific officer, whose surname the drug carries, has described IPF as a clear clinical example of an age-related disease where fibrosis, inflammation, matrix remodeling, and senescence meet. That is the company’s dual-purpose pitch: treat a lethal indication regulators already recognize, and harvest aging biomarkers along the way. The Phase 3 still has to win on lungs.

Healthy Volunteers Are Still the Missing Experiment

The authors themselves write that the next clean test is the drug, or its mechanism, in people who do not have IPF. Until that dataset exists, a younger protein clock in scarred lungs remains a treated-disease reading with an aging overlay, not a lifespan claim. Insilico has also cleared a first-in-human path in China for an inhaled form, a separate formulation bet that still points at the lung.

Proteomic data from the phase 2a study sit in the China National Center for Bioinformation under accession OMIX008341, and the team released analysis code as an open-source Python library. That makes the clock work checkable. It does not make the clocks a substitute for a year of FVC, hospitalizations, and survival in 320 people.

Rentosertib remains investigational and has no regulatory approval. The six clocks agreed that treated blood looked younger than placebo blood. The trial that can tell whether that is aging, or a quieter lung, has not been run.

Disclaimer: This article is news reporting and analysis of a published clinical reanalysis and company statements; it is informational only. It is not medical advice, a treatment recommendation, or an assessment of whether any reader should seek, start, or stop a drug for idiopathic pulmonary fibrosis, aging, or any other condition. Readers with lung disease or questions about experimental therapies should consult a qualified physician, ideally a pulmonologist, before making health decisions. Figures, trial statuses, and regulatory positions reflect the papers and registries cited and may change as Phase 3 proceeds and as further analyses appear.