Bemarituzumab's FGFR2b benefit lived in gastric cancer, so the solid-tumor basket was cut

What was tried

Bemarituzumab is an afucosylated humanized IgG1 antibody against the IIIb isoform of fibroblast growth factor receptor 2, the FGFR2b protein encoded by FGFR2 (Wainberg et al., Lancet Oncology, 2022). Amgen ran NCT05325866 as a Phase 1b/2 open-label basket study of bemarituzumab monotherapy in advanced solid tumors selected for FGFR2b overexpression, spanning multiple tumor types. The study started in September 2022 and enrolled 260 patients before termination, with an actual primary completion date of 28 January 2026 and a posted reason of Sponsor Decision (ClinicalTrials.gov, NCT05325866). The basket ran alongside the company's gastric-focused program, where bemarituzumab was paired with chemotherapy.

The biological hypothesis

FGFR2b overexpression and FGFR2 amplification drive a subset of epithelial cancers, most clearly gastric and gastro-oesophageal adenocarcinoma. In Open Targets, FGFR2 associates with gastric carcinoma at 0.65, supported by genetic literature evidence at 0.87 and somatic mutation evidence at 0.46, and with the broader neoplasm class at 0.61 (Open Targets Platform, FGFR2, ENSG00000066468). An afucosylated antibody adds enhanced antibody-dependent cellular cytotoxicity on top of ligand blockade. The basket hypothesis extended a tumor-agnostic logic. If FGFR2b protein is present on the cell surface, blocking it should help regardless of tissue of origin.

What actually happened

The gastric data validated the target. In the randomized Phase 2 FIGHT trial (NCT03694522), bemarituzumab plus mFOLFOX6 produced median progression-free survival of 9.5 months versus 7.4 months on placebo plus chemotherapy, with a hazard ratio of 0.68 (95 percent CI 0.44 to 1.04, p = 0.073) across 155 FGFR2b-selected patients (Wainberg et al., Lancet Oncology, 2022). In patients with FGFR2b expression in at least 10 percent of tumor cells, the progression-free survival hazard ratio fell to 0.43 and the overall survival hazard ratio to 0.52 at final analysis (Wainberg et al., Gastric Cancer, 2024). The basket study in other tumor types ran as monotherapy without a comparator arm, enrolled 260 patients across its tumor cohorts, and did not produce a comparable public signal before it was closed by sponsor decision (NCT05325866). The benefit concentrated where the biology was strongest, in the FGFR2b-high gastric setting that the FIGHT data had already defined.

Failure mechanism, best guess

Two forces explain the basket termination, and they reinforce each other. First, the efficacy gradient was steep with respect to FGFR2b expression. The clean separation appeared in high expressers in gastric cancer, and a tumor-agnostic basket dilutes that signal by enrolling tumors where FGFR2b is present but not the dominant driver. Second, the antibody carries an on-target safety cost. In FIGHT, all-grade corneal adverse events occurred in 67 percent of bemarituzumab patients versus 10 percent on placebo, and grade 3 corneal events in 24 percent (Wainberg et al., Lancet Oncology, 2022), consistent with FGFR2b expression in the corneal epithelium. A drug with a real ocular toxicity needs a strong efficacy return to clear its benefit-risk bar, and that return was demonstrated in gastric cancer but not across the basket. The rational move was to concentrate on the indication where FGFR2b biology pays for the toxicity, the Phase 3 gastric program, and to stop spending on the rest.

How to prevent this next time

Two quantitative tools would have framed the basket decision before 260 patients were enrolled. The first is biomarker enrichment with a prespecified expression threshold. The FIGHT data show the effect size moving from a hazard ratio of 0.68 in the broadly selected population to 0.43 in the at-least-10-percent subgroup, which argues for setting basket entry at the expression level where benefit is demonstrable rather than at any detectable FGFR2b. The second is a Bayesian predictive-probability gate evaluated per tumor cohort. The predictive probability of success is

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

evaluated separately for each tumor type, with a prior anchored on the gastric effect size and updated as each cohort accrues. A cohort whose posterior probability of clearing a clinically meaningful response stays low after a planned interim should close, releasing patients and budget for the cohorts that clear it. A third lever, translational validation sequencing, would have required early pharmacodynamic or ex vivo evidence that FGFR2b blockade alters tumor biology in a given tissue before that tissue earned a full expansion cohort. The single highest leverage change would have been running the basket as a set of independently gated Bayesian cohorts keyed to an expression threshold, so each tumor type had to earn continuation on its own posterior rather than riding a shared tumor-agnostic assumption.

What this means for similar programs

FGFR2b is a validated target with an approval-grade signal in gastric cancer, so the end of the basket is not a verdict on the mechanism. It is a reminder that tumor-agnostic logic only holds when the target is the dominant driver across tissues, and that an on-target toxicity raises the efficacy bar every cohort must clear. Programs built on surface-protein selection should treat expression level as a continuous variable that sets the benefit-risk line, and should be willing to narrow to the indications that clear it. A basket that treats every FGFR2b-positive tumor as equivalent spends its largest cost, patient exposure to a known ocular toxicity, on cohorts least likely to return benefit.

Open questions

Will Amgen release the basket's tumor-by-tumor response data, which would show exactly where FGFR2b blockade did and did not translate. Does the corneal toxicity scale with FGFR2b expression in a way that narrows the viable population further. And can next-generation FGFR2b agents, including bispecifics or antibody-drug conjugates, widen the therapeutic window enough to revive a tumor-agnostic strategy.

Sources

1. • ClinicalTrials.gov, NCT05325866, full record (Phase 1b/2 FGFR2b basket, enrollment 260, terminated, primary completion 28 January 2026, reason Sponsor Decision). https://clinicaltrials.gov/study/NCT05325866 - Wainberg ZA, et al. Bemarituzumab in patients with FGFR2b-selected gastric or gastro-oesophageal junction adenocarcinoma (FIGHT). Lancet Oncology, 2022 (PFS 9.5 vs 7.4 months, HR 0.68; corneal AEs 67 percent vs 10 percent, grade 3 24 percent). https://- Wainberg ZA, et al. Bemarituzumab final analysis of the randomized phase 2 FIGHT trial. Gastric Cancer, 2024 (FGFR2b at least 10 percent: PFS HR 0.43, OS HR 0.52). https://- ClinicalTrials.gov, NCT05052801, FORTITUDE-101 Phase 3 bemarituzumab plus chemotherapy in FGFR2b-overexpressing gastric cancer (active gastric program). https://clinicaltrials.gov/study/NCT05052801 - Open Targets Platform, FGFR2 (ENSG00000066468), gastric carcinoma 0.65 and neoplasm 0.61 associations. https://platform.opentargets.org/target/ENSG00000066468 - ClinicalTrials.gov, NCT03694522, FIGHT Phase 2 source record. https://clinicaltrials.gov/study/NCT03694522.