Six proteomic clocks predicted lower biological age in a lung drug trial

What the clocks showed

Six different protein-based aging clocks looked at the same blood samples from a 42-person drug trial, and all six pointed toward a lower predicted biological age in the people receiving the drug. The placebo group barely moved.

That agreement is the reason to pay attention. Aging clocks are known for disagreeing with each other. A separate analysis of 51 human intervention studies, covered here in our report on epigenetic clocks, found that the same intervention can move one clock and leave another flat, which is why a single clock's result is weak evidence on its own. Six of them agreeing is a stronger internal signal than this field usually produces.

What it is not is evidence that anyone aged more slowly. The study is small, it ran for twelve weeks, everyone in it had a serious lung disease, and the authors are careful about all three.

What the researchers did

They went back to stored blood from a randomized, double-blind, placebo-controlled phase 2a trial of rentosertib, an experimental drug for idiopathic pulmonary fibrosis, a progressive scarring disease of the lungs. The trial ran from July 2023 to June 2024 at sites across China and had nothing to do with aging.

Forty-two people were in this analysis, mean age 67.1. They had received a placebo or one of three regimens for twelve weeks: 30 mg once a day, 30 mg twice a day, or 60 mg once a day. Blood was drawn before treatment and at weeks 2, 4 and 12. The researchers measured 2,841 proteins that passed quality checks and ran six previously published proteomic aging clocks over the results.

What a proteomic aging clock is

It is an algorithm that estimates biological age from the mix of proteins circulating in your blood.

Proteins do the body's daily work. They carry signals, run metabolism, coordinate immune responses and repair tissue, and their levels shift with age and with illness. A clock is trained by showing it protein data from many people whose ages or health outcomes are already known, so it learns which combinations tend to appear in younger or older people.

It is pattern recognition, not a stopwatch. A clock can report that someone's proteins resemble a younger reference pattern. It cannot show that the person will live longer.

When the signal appeared, and when it faded

All six clocks read lower in the treated groups than in placebo.

The paper made 54 comparisons in all: six clocks, three timepoints, three dose regimens. Twenty-one reached statistical significance, and a permutation test confirmed that 21 is well beyond what chance alone would produce.

The signal was concentrated at week 4. Of the 18 comparisons made at that timepoint, 11 showed a significantly lower shift in predicted age among treated participants. The largest single figure, a reduction of 2.71 to 3.46 years, comes from the 60 mg once daily arm at week 4, and it is the range across the four clocks trained on chronological age. In that same arm neither of the two mortality-trained clocks registered a significant change. The 30 mg twice daily arm had the broadest agreement across clocks, with nine significant comparisons.

By week 12 the number of significant comparisons had fallen, which the paper calls partial attenuation of the initial response. Comparing week 4 directly against week 12, no arm and clock combination showed a further shift in either direction, so the authors settle on plateau as the better description. They do not treat that as good news needing no explanation. They write that a rebound or plateau at later timepoints "does not necessarily imply the arrest of beneficial anti-aging changes, but may reveal a statistical artifact of context-specific protein dynamics."

This is worth stating plainly, because the figure now circulating attaches the shift of three to four years to twelve weeks of treatment. The paper's own numbers put the largest shift at week 4 and report fewer significant results by week 12.

The evidence that this was about aging

Two findings point beyond the lungs.

The drug changed the trajectories of 326 proteins, against two in the placebo group. The researchers then compared those changes against the way proteins normally shift with age in 55,319 older adults in the UK Biobank. In the twice daily arm, the treatment-induced changes ran significantly opposite to that normal aging direction. So the proteins were not merely producing a lower number inside one formula. As a group they were moving against the direction aging usually takes them.

And the dose that most improved lung function in the original trial, 60 mg once daily, was not the dose with the most consistent clock response. Change in lung function also explained little of the variation in biological age estimates, a median of 6 percent across the six clocks. If the clocks were only reading the lungs getting better, those two things should line up more closely than they do.

The evidence that it may not be

Everyone in this trial had active lung scarring and was taking a drug to treat it.

The authors identified the proteins contributing most to the biological age shift, 35 of them in all. The single most important one, and the only important feature present in all six clocks, was LTBP2, which the paper describes as a master regulator of fibrosis. Other leading contributors included further fibrotic markers and extracellular matrix proteins.

So the clocks may be substantially reading the disease improving. The paper says as much directly: proteomic clocks alone cannot separate aging effects from disease-specific effects, and "full disentanglement of the two signals is not achievable within an IPF cohort and requires validating the drug or its underlying mechanism in healthy volunteers."

The clocks also carry different weights of evidence. The four trained on chronological age tracked participants' real ages closely. The two trained on mortality risk correlated only weakly with actual age, which is expected given what they are built for, but it means any reduction they report rests on a looser relationship to age.

One thing worth knowing about the six clocks

Six clocks agreeing is stronger than one clock agreeing. It is not six independent laboratories reaching the same answer.

The six come from five published sources, since two of them are variants of the same underlying model, one trained to predict chronological age and one trained on mortality risk. Reading the paper's own method section against its author list: five of the six clocks were built by groups that include an author of this paper, and one, PAC, was not. One of the six, ipfP3GPT, was built by Insilico Medicine itself, the company that develops rentosertib. Another, PAOPAC, is described in a preprint that has not been peer reviewed.

That reading is ours rather than the paper's. The paper names its six clocks and cites the source of each in its methods, and it lists its authors on the first page; putting the two side by side is the whole of the method, and both are open for a reader to check. The paper's own competing-interests declaration covers something different and narrower: that the first author founded and runs Insilico Medicine, that seven further authors work there, and that the remaining authors declare none. It does not mention the clocks.

None of that makes the result wrong. It makes independent replication the thing that would settle it.

What would settle it

More participants, longer follow-up, and people who do not have pulmonary fibrosis. That last one is the authors' own condition, and it is the only way to find out whether the clocks were reading aging or reading a lung disease getting better.

It also means naming the aging measurements in advance rather than choosing them afterward, and comparing what the clocks say against things a person can feel or a doctor can measure: strength, fitness, cognition, frailty, later disease.

If the method holds up, disease trials that already collect blood could start answering aging questions at the same time, at very little extra cost. That would not replace long-term evidence. It could get us there sooner, and that is the argument the paper is really making.

What to take from this

The useful thing here is a research method, not a treatment. Rentosertib is experimental, it was given to people with a specific lung disease, and nothing in this paper says anything about what it would do in anyone else.

The finding to hold on to is about the clocks rather than the drug. Six of them agreed, and the protein doing most of the work in all six was a marker of the disease being treated. That is a good illustration of what these scores are: a model reading whatever is in your blood, including illness, treatment and recovery, and reporting it as an age.

If you are considering a biological age test, that is the question worth asking a clinic. What is this clock reading, what else could move it, and what would you do differently depending on the answer. Our explainer on biological age and organ age covers how the scores are built and why two of them can disagree about the same person.

Sources

• Nature Biotechnology: Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment. Zhavoronkov A, Galkin F, Chen S, et al., published 7 September 2026, open access. The 54 comparisons and the week 4 concentration are in Results, "Rentosertib-induced aging clock predictions"; the 2.71 to 3.46 year range and the mortality-clock result are in the same section; LTBP2 and the 35 contributing proteins are under "Rentosertib-induced proteomic trajectories"; the six clocks and their sources are named in Methods, "Proteomic aging clocks". The paper gives two counts for the trial's locations, 21 in Results and 22 in Methods, which is why no count appears above.
• ClinicalTrials.gov, NCT05938920. The parent phase 2a trial, which was designed to test the drug against lung fibrosis and not against aging.