GENESIS-IPF: the Phase 2a trial in China

GENESIS-IPF was a 12-week trial in 71 people with idiopathic pulmonary fibrosis (IPF) at hospitals across China, and it is the only trial in which rentosertib has been tested in people with the disease and the results published in full. Its main purpose was to check safety, not to prove that the drug works, and it was too small to do the latter.

What the trial was designed to do

The trial ran from mid-2023 to mid-2024 and was registered as NCT05938920. Sources disagree about how many hospitals took part: the Nature Medicine paper says 21 sites, Insilico Medicine’s press release says 22, and one registry field has been reported as 29.

Of 128 people screened, 71 were randomised into four groups: 30 mg once daily (18 people), 30 mg twice daily (18), 60 mg once daily (18), and placebo (17). Everyone took capsules for 12 weeks, and carried on with whatever IPF treatment they were already taking, so rentosertib was tested on top of existing care rather than instead of it.

The primary endpoint, the single question the trial was built to answer, was the percentage of people who had at least one side effect while on treatment. Lung function was a secondary measure. The protocol contains no sample size calculation, so the trial was never sized to detect a difference in lung function between groups. Any efficacy finding from it is exploratory.

Who took part

Everyone in the trial was Chinese; the paper records the ethnicity of all 71 participants as Asian. The average age was 66.7 years and 90% were men (64 of 71). UK IPF populations are also mostly older men, but a UK cohort would not be a single ethnicity.

Two imbalances between the groups matter when reading the results.

The placebo group started off with worse lung function. Forced vital capacity (FVC), a measure of how much air you can breathe out, is usually expressed as a percentage of what would be expected for someone of the same age, sex and height. Averaged across the trial this was 73.7%, but the placebo group was at 66.5% and the 60 mg once-daily group at 78.5%. A sicker placebo group tends to decline faster, which flatters the comparison.

Use of existing antifibrotic medicines was also uneven. Just over half the participants (37 of 71) were taking one: 22 on nintedanib and 15 on pirfenidone. Nintedanib was being taken by half the 60 mg once-daily group but only 17.6% of the placebo group, and nintedanib itself can raise liver enzymes.

What happened to lung function

Expressed as percentage points of predicted FVC, the changes were −0.57 (placebo), −0.67 (30 mg once daily), +0.65 (30 mg twice daily) and +3.05 (60 mg once daily). No p-values were reported for FVC. About 70% of the breathing tests were done on central devices and 30% on local hospital equipment.

How to read those numbers

Only one group, 60 mg once daily, has a confidence interval that does not include zero (10.9 to 185.9 mL). A confidence interval is the range of values consistent with the data; when it crosses zero, the result is compatible with no effect at all. For the other two dose groups it does cross zero. That interval is also very wide, running from about 11 mL to about 186 mL, which is what 18 people over 12 weeks buys you.

The 30 mg once-daily group was numerically slightly worse than placebo (−27.0 mL against −20.3 mL). A genuine dose response would normally show an orderly progression across doses, and this one does not. The authors themselves describe the finding as “a trend toward an increase in FVC”, not as a demonstration of benefit.

The subgroup not taking other antifibrotics

Among people in the 60 mg once-daily group who were not taking nintedanib or pirfenidone, the reported FVC change was +187.8 mL (68.6 to 306.9). This figure appears widely in coverage, and Insilico has cited it.

It comes from roughly nine people, it was not the trial’s pre-specified comparison, and subgroup analyses of small trials routinely produce large effects that disappear in larger studies. It has also been misreported: at least one conference summary compared this subgroup figure with the whole-cohort placebo figure, which is not a valid comparison.

Other things that were measured

Changes in DLCO, a measure of how well oxygen passes from the lungs into the blood, were small and similar across all four groups. So were changes in FEV1 and in the six-minute walk test.

The Leicester Cough Questionnaire, a measure of how much cough affects daily life, was the one other measure to show a difference: p=0.0495 for the 60 mg group, corresponding to roughly a two-point improvement. Only the p-value was reported. A p-value of 0.0495 sits a hair below the conventional 0.05 threshold, and in a trial measuring many secondary outcomes without correction for multiple comparisons, a result that marginal should not be relied on.

Safety: what the trial found

Safety was the point of the trial, and the safety findings are more substantial than the lung function ones.

Side effects judged related to treatment rose steadily with dose, from 29.4% on placebo to 77.8% on 60 mg once daily. Serious or severe events also rose, reaching 38.9% in the 60 mg group against 17.6% on placebo.

The commonest events were low blood potassium (3, 5, 4 and 2 people across the 30 mg once daily, 30 mg twice daily, 60 mg once daily and placebo groups), diarrhoea (2, 3, 5 and 0), upper respiratory tract infection (1, 4, 1 and 1) and raised uric acid (2, 1, 1 and 0).

The liver signal

This is the most important safety finding. Sixteen of the 71 participants (22.5%) stopped early, 12 of them because of a side effect, and 7 of those 12 stopped because of liver injury or abnormal liver function.

By group, the rate of stopping for liver problems was 0% on 30 mg once daily, 22.2% on 30 mg twice daily and about 17% on 60 mg once daily. Raised ALT, a liver enzyme, was recorded in 1, 1, 6 and 1 people across the 30 mg once daily, 30 mg twice daily, 60 mg once daily and placebo groups; raised bilirubin in 0, 1, 4 and 2.

Two things temper this. Four of the seven were also taking nintedanib, which is itself associated with raised liver enzymes, and none was taking pirfenidone. And in every case the liver abnormality resolved once the drug was stopped. Brian Buntz, writing in Drug Discovery & Development in June 2025, was among the first journalists to draw attention to these discontinuations; the South China Morning Post was one of few outlets to report the liver findings alongside the lung function figures.

Acute exacerbations

An acute exacerbation is a sudden, severe worsening of IPF, and it is a serious event that often leads to hospital admission. There were 3 (16.7%) in the 60 mg once-daily group against 1 (5.9%) on placebo. All four people were admitted to hospital, for an average of 23.3 days.

The same group also recorded two cases of pneumonia, one of interstitial lung disease and one of respiratory failure. The authors suggest the drug may alter immune function in a way that increases susceptibility to infection, but this is a hypothesis, not a finding, and the numbers are far too small to settle it.

Deaths and withdrawals

One person died, from heart failure, in the 30 mg twice-daily group. The investigators judged the death unrelated to treatment.

The overall withdrawal rate of 22.5% is high for a 12-week trial, and 22 participants had no week-12 FVC measurement, so their values were estimated statistically. When nearly a third of the key measurements in a group are imputed, the resulting averages carry more uncertainty than the confidence intervals alone suggest.

Blood biomarkers

The researchers measured 2,841 proteins in participants’ blood. Levels of several proteins associated with scarring, including COL1A1, FAP, FN1 and MMP10, fell with increasing dose and over time, and the falls were inversely correlated with FVC change. The anti-inflammatory protein IL-10 rose. The most strongly downregulated pathway was extracellular matrix organisation, the process by which scar tissue is laid down. The data are deposited as OMIX008341. These are exploratory findings and are not a substitute for clinical outcomes.

What the authors say the trial cannot show

The paper’s own limitations section lists the following: the number of participants was small; all were recruited in China and were of a single ethnicity; follow-up was only 12 weeks; 22.5% of participants withdrew; and the sponsor was involved in the study. The authors say the next steps needed are drug-interaction studies, trials lasting at least 12 months, and studies in globally diverse populations.

What others have said

Marinka Zitnik of Harvard University wrote an independent News & Views commentary in Nature Medicine in July 2025, describing the trial as marking “a turning point”. Her comment is about the method by which the drug was found rather than about clinical benefit. She has no identified financial relationship with Insilico.

Qureight, a medical imaging company based in Cambridge, reported a deep-learning analysis of the trial’s CT scans in July 2025. Its analysis found no imbalance in disease severity on imaging at baseline and concluded that the Chinese cohort was broadly representative of global IPF populations, which if correct would weaken one criticism of the trial. The underlying abstract has no traceable identifier, Qureight’s own press page no longer resolves, and there is no evidence that Qureight is involved in the Phase 3 trial.

The UK connection

The only direct UK involvement in this trial was data processing, carried out by Fortrea Clinical Pharmacology Services in Leeds. No UK hospital took part and nobody in the UK was enrolled.