The clinical trial for a lung fibrosis drug was designed not only to confirm the drug's safety and effect on lung function, but also to incorporate multiple aging-biology analyses from the outset. When participants' blood samples were run through six different statistical models for assessing aging, all of them pointed toward a lower predicted age in the treatment groups. So far, this seems straightforward. But these six clocks were never designed with the same purpose or built using the same methods in the first place. Behind the headline that "the clocks agreed" lies a deeper question: what are aging clocks actually counting?
Blood from 42 people, run through six clocks with different goals and methods
The Phase IIa trial "GENESIS-IPF," conducted at 21 sites in China, administered rentosertib (development code ISM001-055), a TNIK inhibitor developed by Insilico Medicine, to 71 patients with idiopathic pulmonary fibrosis (IPF). Participants were split into four groups—30mg once daily, 30mg twice daily, 60mg once daily, and placebo—and followed for 12 weeks. The primary endpoint was safety and tolerability, which the trial achieved. From the outset, the study was designed to incorporate aging-biology analysis using the serum proteome.
Of the participants, 43 consented to additional blood protein testing over time. One was excluded due to missing 12-week data, leaving 42 participants (drawn from all four groups) for analysis. Using the Olink Explore 3072 platform, researchers measured 2,841 different proteins and ran the results through six "proteomic aging clocks." The reference population consisted of data from 55,319 people in the UK Biobank.
All six clocks pointed toward a lower predicted age. The peak effect appeared at week 4 in the 30mg-twice-daily group, showing a reduction equivalent to 3–4 years, with one clock recording a reduction of up to 6 years. The change from week 4 to week 12 was not statistically significant, and the paper characterizes this less as a "weakening" effect and more as a plateau.
What about the actual lung function outcomes? Forced vital capacity (FVC) increased by an average of 98.4 mL at 12 weeks in the 60mg-once-daily group, while the placebo group decreased by 20.3 mL. However, the 30mg-once-daily group decreased by 27.0 mL, while the 30mg-twice-daily group increased by 19.7 mL—meaning there wasn't a clean dose-dependent relationship across all four groups, only a general "trend toward increase." Moreover, the dose that showed the greatest improvement in lung function (60mg once daily) did not match the dose that showed the strongest aging signal (30mg twice daily).
What exactly are aging clocks counting?
When people hear "aging clock," they tend to imagine a single device that directly measures the phenomenon of aging within the body. In reality, that's not what it is. An aging clock is nothing more than a regression model: researchers gather data from a large number of people of known ages, identify patterns that statistically correlate with age or mortality risk, and then work backward from those patterns. What counts as an "aging signal" varies, and so does the content of the clock itself.
The six clocks used in this study were built independently by five different research groups: Argentieri, Kuo, Han, Galkin, and Goeminne. Looking closely at the breakdown, though, two of the six clocks were actually built by the same Goeminne group—one version predicts chronological age, the other predicts mortality risk. Of the remaining four, only one was trained to predict mortality risk; the other three were built to predict chronological age. The methods also differ, split between classical machine learning and deep learning.
In other words, "six clocks agreeing" doesn't mean six replicate experiments measuring the same thing arrived at the same answer. Rather, it means several models with different goals and different construction methods happened to point in the same direction.
Looking at the broader taxonomy, aging clocks fall into even larger categories. There are "proteomic clocks," like the ones used here, which measure blood proteins via Olink panels, and "epigenetic clocks," which look at DNA methylation patterns—the chemical modifications that influence how genes are expressed. Methylation arrays can cover roughly one million sites. In Japan, since 2024, the epigenomics company Rhelixa has developed an epigenetic clock called "EpiClock®" calibrated for Japanese populations, which is offered as a testing service at aesthetic clinics in Ginza and Kyoto.
| Category | Proteomic Clocks (the 6 used here) | Epigenetic Clocks (e.g., EpiClock®) |
|---|---|---|
| What's measured | Blood protein patterns | DNA methylation patterns |
| Scale of measurement | ~2,841 proteins | ~1 million CpG sites |
| Prediction target | Chronological age or mortality risk (varies by clock) | Mainly chronological age (some models adjust for mortality-related indicators) |
The "~1 million sites" figure in the table refers to the general scale of methylation array technology—epigenetic clocks were not actually measured in this rentosertib trial. Nor have proteomic clocks and epigenetic clocks been directly compared and validated against each other. Just because proteomic clock predictions decreased doesn't mean the same people would appear younger on an epigenetic clock as well. It's more accurate to think of aging clocks not as a single ruler measuring one underlying reality called aging, but as multiple tools—each counting a different signal, depending on the category and the specific clock.
This can't simply be explained away as "the disease got better"
Since proteomic clocks count proteins that correlate with age or mortality risk, if factors other than aging drive those same proteins, the clock can easily be thrown off. IPF is a disease accompanied by systemic inflammation, and inflammatory states tend to show up in blood protein patterns. If a drug calms lung inflammation, the same protein groups that aging clocks reference could shift as a result. The research team itself cautions that this data alone cannot confirm that aging itself was slowed, and that additional validation in non-IPF populations is needed.
That said, this explanation alone doesn't fully account for everything. TNIK, the drug's target, was chosen from the very start of the AI-driven discovery process by Insilico's drug discovery platform precisely because it was identified as a molecule involved in both multiple hallmarks of aging and the fibrotic cascade of IPF. The connection to aging isn't a discovery made after the fact—it was intended from the target-selection stage.
Furthermore, the fact that the dose showing the greatest improvement in lung function doesn't match the dose showing the strongest aging signal doesn't fit neatly into a simple narrative of "the numbers moved only because the disease got better." Based on the current data, it's impossible to disentangle the effects of reduced inflammation from a potentially independent effect mediated through the target molecule.
Whether to call it an anti-aging drug depends on understanding the clocks themselves
Whether rentosertib is a promising IPF treatment and whether it's a drug that slows aging are two separate questions. The former has already advanced to Phase III trials. The latter rests on an analysis of a small group of just 42 people, and only one of the six clocks showed a reduction of up to 6 years. Whether the effect persists also remains unclear, given that the 12-week data point looks close to a plateau.
Still, there's something to be learned from this case. When you encounter the phrase "aging clock," it's worth checking how many types of clocks are being referenced, which biomolecules each one counts, and whether its purpose is to predict chronological age or mortality risk. Studies that measure the same people using both proteomic and epigenetic clocks to see whether the results align, or trials that test the same drug in healthy participants without disease, would offer more clues as to whether this is a "byproduct of the disease improving" or an "independent rejuvenating effect." Next time you come across the phrase "biological age," it's worth pausing to check which clock produced that number—and what it was actually designed to predict.