GSK3915393 met futility in IPF: a transglutaminase 2 inhibitor with literature-only human support

What was tried

GlaxoSmithKline ran TRANSFORM, a Phase 2 randomized, quadruple-masked, placebo-controlled study of GSK3915393 in idiopathic pulmonary fibrosis (NCT06317285). Participants received 80 mg orally twice daily or placebo for 26 weeks. The study opened on 4 April 2024 and reached primary completion on 1 October 2025 with 158 randomized, 106 to drug and 52 to placebo.

The cohort was older than most: mean age 72.1 years on drug and 72.4 on placebo, 127 of 158 male. Entry required a guideline-consistent IPF diagnosis within five years, centrally read computed tomography showing usual interstitial pneumonia or probable usual interstitial pneumonia, forced vital capacity at or above 45 percent of predicted, and diffusing capacity at or above 25 percent. Background antifibrotics were permitted at a stable dose.

The registered reason for stopping is unambiguous. The sponsor states the study met futility criteria at a pre-planned interim analysis, showing no clinical efficacy.

The biological hypothesis

GSK3915393 is a small molecule inhibitor of transglutaminase 2, the product of TGM2. The enzyme catalyzes calcium-dependent crosslinking between glutamine and lysine residues, and the resulting isopeptide bonds stabilize extracellular matrix against proteolytic turnover. If crosslinking converts provisional matrix into durable scar, blocking the crosslinker should keep new matrix reversible.

The preclinical case is real. Transglutaminase 2 is elevated in fibrotic lung, Tgm2 knockout mice resist bleomycin-induced fibrosis with preserved lung function, and selective inhibitors have been proposed for IPF.

The human case is thinner, and the asset history shows it. GSK3915393 came out of a transglutaminase 2 program aimed at celiac disease, where the enzyme is the autoantigen. Reporting from February 2023 describes GSK ending celiac development while keeping the asset for other indications. IPF became the other indication. Open Targets scores the TGM2 to idiopathic pulmonary fibrosis association at 0.079, with literature at 0.652 as the only contributing evidence type, no genetic association datatype, and zero drugs or clinical candidates registered against TGM2. In gnomAD the gene is loss-of-function tolerant, observed-to-expected 0.72 with a 90 percent interval of 0.58 to 0.90, so reduced transglutaminase 2 in humans does not declare a protective lung phenotype.

What actually happened

At Week 4 the posterior median treatment difference in absolute change from baseline forced vital capacity was 85.5 mL favoring GSK3915393, 95 percent highest posterior density interval 28.7 to 142.0. At Week 8 it was 76.0 mL, interval 24.2 to 126.7. Both exclude zero. By Week 12 it was 38.3 mL, interval -15.3 to 95.4, and by Week 18 it was 24.4 mL, interval -50.4 to 97.5.

At the Week 26 primary endpoint the direction had reversed. Change from baseline was -110.3 mL on drug, interval -156.0 to -62.9, against -79.0 mL on placebo, interval -125.5 to -10.2, a posterior median difference of -33.0 mL, interval -112.7 to 39.5, from 53 evaluable on drug and 24 on placebo. A responder analysis agreed. The proportion with a relative FVC decline of 5 percent or less was 0.60 on drug, interval 0.46 to 0.73, against 0.79 on placebo, interval 0.62 to 0.93, odds ratio 0.40, interval 0.06 to 1.00.

Safety was not the stated reason for stopping, but the imbalance is on the record. Serious adverse events occurred in 13 of 106 on drug and 1 of 52 on placebo, with one death on drug and none on placebo. Pneumonia accounted for 3.

Failure mechanism, best guess

The most economical reading is that the early forced vital capacity separation was not disease modification and was never going to become disease modification.

Two features support that. The trajectory is monotonic in the wrong direction, falling from 85.5 mL to 76.0, 38.3, 24.4, and finally -33.0 mL, while a real antifibrotic effect should widen with exposure. The shape also fits an early non-structural influence on the measurement, superimposed on a placebo arm whose Week 4 decline of -71.3 mL was steeper than its later trajectory implied.

A second contributor is target validity. Crosslinking inhibition may act on a matrix compartment already stabilized by the time a patient with established disease enrolls. The enzyme would then be a legitimate biochemical target and a poor clinical one, because the drug can only touch matrix laid down after dosing starts. Nothing in the posted results separates that from plain absence of target engagement, because no pharmacodynamic crosslinking measure appears among the outcomes and the molecule-specific secondaries are pharmacokinetic only.

How to prevent this next time

No patient-level data were released, so a formal re-analysis is not possible. The qualitative levers are the honest ones.

Base-rate adjustment first. A pair carrying a 0.079 Open Targets association with literature as its only evidence type is a thin foundation for a 26-week Phase 2 with a spirometric primary endpoint. Programs should name, before Phase 2, what human evidence beyond animal models connects target to disease, and say so when there is none.

Biomarker enrichment second. Transglutaminase 2 makes a measurable product, and a validated assay for isopeptide crosslink burden in plasma or lavage would have made this trial interpretable either way. A null result with confirmed crosslink suppression rules the mechanism out of IPF. A null result without suppression indicts the molecule instead. Without either, the program spent 158 participants and learned neither.

Red-team third, aimed at the interim. The gap between the Week 8 and Week 26 estimates is 109.0 mL, well outside the Week 8 interval of 24.2 to 126.7. Teams reading those early numbers owed themselves a pre-committed decay pattern that would falsify them. Forced vital capacity works as a surrogate because it tracks mortality over long horizons, and a four-week movement in it is not evidence about scar.

The single highest leverage change would have been requiring a validated pharmacodynamic readout of transglutaminase 2 crosslinking activity as a gate before the efficacy study opened, so that a null result would have separated a wrong target from an inadequate molecule.

What this means for similar programs

Matrix crosslinking as a class deserves the caution this result earns it. The idea has appealing chemistry and repeated preclinical support, and it has now produced a clean human null in the indication with the largest fibrotic burden. Teams working on lysyl oxidase enzymes, other transglutaminases, or matrix stabilization should assume the established-scar problem applies to them, and should test in earlier disease.

The celiac precedent is worth separating from the fibrosis one. A randomized trial of a transglutaminase 2 inhibitor in celiac disease supports the enzyme as a tractable human target. Tractability travels. Disease relevance does not.

The Claidex Mechanism Risk Score for TGM2 now stands at 39, yellow band, driven mainly by genetic deficit at 13.81 of a possible 15. Saturation is 2.72 because almost nobody else works here. Weak human evidence plus an empty field is a pattern to recognize rather than mistake for white space.

Open questions

Was crosslinking activity suppressed in lung tissue at 80 mg twice daily, and is TRANSFORM pharmacodynamic data available through the sponsor's data-sharing route?

Does the serious adverse event imbalance of 13 of 106 against 1 of 52 reflect drug effect, the enrolled population, or chance at this sample size?

Would earlier disease, for example interstitial lung abnormalities without honeycombing, respond differently?

Sources

1. • ClinicalTrials.gov, NCT06317285, TRANSFORM, protocol and posted results. https://clinicaltrials.gov/study/NCT06317285 - GSK Study Register, trial 220929, TRANSFORM-IPF record and termination status. https://www.gsk-studyregister.com/trials/220929 - Open Targets Platform, TGM2 (ENSG00000198959) and idiopathic pulmonary fibrosis (EFO_0000768): association score, evidence types, tractability, genetic constraint, drug candidate count. https://platform.opentargets.org/target/ENSG00000198959 - Pulmonary Fibrosis Foundation pipeline entry describing GSK3915393 as a transglutaminase 2 targeting small molecule. https://www.pulmonaryfibrosis.org/patients-caregivers/medical-and-support-resources/clinical-trials-education-center/pipeline/drug/idiopathic-pulmonary-fibrosis/gsk3915393 - Fierce Biotech, 1 February 2023, GSK ending GSK3915393 development in celiac disease while retaining the asset. https://www.fiercebiotech.com/biotech/gsk-drops-celiac-disease-program-acquired-sitari-buyout-and-delivers-blow-bid-avert - Olsen KC et al. Transglutaminase 2 and its role in pulmonary fibrosis. Am J Respir Crit Care Med, 2011. https://- Transglutaminase 2 knockout mice are protected from bleomycin-induced lung fibrosis with preserved lung function. Physiological Reports, 2024. https://- Transglutaminase 2: a novel therapeutic target for idiopathic pulmonary fibrosis using selective small molecule inhibitors. Amino Acids, 2021. https://- Schuppan D et al. A Randomized Trial of a Transglutaminase 2 Inhibitor for Celiac Disease. N Engl J Med, 2021. https://- Decline in forced vital capacity as a surrogate for mortality in patients with pulmonary fibrosis. Respirology, 2023. https://- openFDA FAERS drug event endpoint queried for GSK3915393, no matching reports returned. https://api.fda.gov/drug/event.json.