Triplet Regimen, Biomarker Gate, Phase 3 Signal: What ILUSTRO’s Cohort 4 Actually Proves

Buried inside a Phase 2 cohort result published in Nature Medicine on March 16, 2026, is a trial design blueprint that every oncology ops team running a biomarker-enriched protocol should study before their next protocol amendment meeting. The ILUSTRO trial — Astellas’s systematic evaluation of zolbetuximab combinations in CLDN18.2-positive gastric and gastroesophageal junction (GEJ) adenocarcinoma — just delivered Cohort 4 data: first-line zolbetuximab plus mFOLFOX6 plus nivolumab in HER2-negative metastatic disease. The efficacy signal is encouraging. The design logic is the real story.

Cohort 4 enrolled patients screened and confirmed CLDN18.2-positive, a tight-junction protein expressed in roughly 38 to 40 percent of gastric adenocarcinomas according to published prevalence estimates. That biomarker gate is not incidental — it is the architectural load-bearing wall of the entire trial. Zolbetuximab, a chimeric IgG1 antibody targeting CLDN18.2, failed to demonstrate overall survival benefit in unselected populations in earlier studies precisely because the signal was diluted by receptor-negative patients. Cohort 4 does not make that mistake.

The clinical results now support advancement to a Phase 3 randomized trial, according to the Nature Medicine publication. That is a regulatory and operational pivot point sponsors cannot afford to misread.

The Biomarker Gate Is the Protocol

Old assumption: biomarker enrichment is a statistical convenience that narrows your enrollment funnel and forces you to screen more patients to hit accrual targets. New assumption: in the post-CheckMate, post-KEYNOTE era of gastric oncology, biomarker enrichment is the enrollment strategy — and the Phase 2 design exists to validate the gate, not just the drug.

Consider what Cohort 4 is actually testing. The mFOLFOX6 backbone provides the chemotherapy scaffold proven in first-line gastric disease. Nivolumab, Bristol-Myers Squibb’s PD-1 inhibitor, received FDA approval for first-line gastric and GEJ adenocarcinoma based on the CheckMate 649 trial, which enrolled 1,581 patients and showed a median overall survival of 13.8 months versus 11.6 months for chemotherapy alone in the PD-L1 combined positive score ≥5 subgroup. Adding zolbetuximab on top of that validated doublet is a deliberate escalation — but only in a population where CLDN18.2 expression gives the antibody a mechanistic rationale. Screen out that population and you are running a toxicity study masquerading as an efficacy trial.

The operational implication is direct: central biomarker confirmation, not local testing, must gate randomization. Any site initiating patients on local CLDN18.2 IHC without a validated central assay is introducing a protocol deviation that will haunt the Phase 3 data package when FDA reviewers interrogate the biomarker concordance data during NDA review.

Phase 2 as a Phase 3 Engineering Document

Here is the counterintuitive read that most clinical ops teams miss. Phase 2 cohort data from a multi-cohort trial like ILUSTRO is not primarily an efficacy signal — it is a feasibility and calibration dataset for the Phase 3 protocol. Response rates matter. But what the Cohort 4 results also establish is the operational envelope: what combination toxicity looks like at these doses, what dropout and discontinuation rates the mFOLFOX6-plus-antibody schedule produces, and whether the CLDN18.2-positive screening prevalence in real enrollment populations matches the epidemiological projections used to build the Phase 3 sample size.

FDA’s guidance on adaptive trial design, finalized in 2019, explicitly acknowledges multi-cohort platform trials as a mechanism for generating dose and population data that can inform confirmatory trial design — provided the Phase 2 inference rules are pre-specified and not mined post-hoc. The ILUSTRO design, running multiple cohorts across different combination backbones, fits that model. But the 2019 guidance also places the evidentiary burden on sponsors to demonstrate that Phase 2 results were not cherry-picked to inflate Phase 3 power assumptions. Astellas will need a clean audit trail from Cohort 4 enrollment rules to Phase 3 statistical assumptions.

Sponsors who treat Phase 2 data as a press release moment rather than a Phase 3 engineering document pay for it at End-of-Phase 2 meetings. FDA reviewers do not forget a response rate that was generated in a protocol with looser eligibility criteria than the confirmatory trial.

Who Carries the Operational Weight

Gastric and GEJ adenocarcinoma trials already carry one of the highest site burden profiles in solid tumor oncology. Patients present with malnutrition, dysphagia, and rapid performance status decline. mFOLFOX6 infusion schedules require biweekly site visits. Layer in CLDN18.2 central screening, nivolumab infusion, and zolbetuximab administration — which has a documented nausea and vomiting profile that requires antiemetic pre-medication protocols — and you have a trial that will stress any site running fewer than six concurrent oncology studies.

The sites that will hold Cohort 4’s Phase 3 successor together are academic centers with dedicated GI oncology programs and coordinators who have already managed CheckMate 649 or KEYNOTE-590 infrastructure. Community oncology networks that enrolled gastric patients on simpler doublet studies will face a ramp-up time that trial timelines rarely accommodate. Sponsors building the Phase 3 site selection strategy on ILUSTRO’s Cohort 4 enrollment map are making the right call — the sites that completed Cohort 4 have already absorbed the learning curve on combination scheduling, antiemetic management, and CLDN18.2 screening logistics.

The monitoring requirement deserves equal attention. A triplet regimen in a fragile patient population will generate serious adverse event reports at a rate that demands a risk-based monitoring plan with triggered on-site visits, not a pure central monitoring model. FDA’s guidance on risk-based monitoring, updated in 2023, supports remote and centralized monitoring as a default — but it does not support it as a substitute for on-site presence when the safety signal complexity warrants direct data verification. Zolbetuximab’s emesis-related adverse events, which drove protocol-specified antiemetic prophylaxis in prior ILUSTRO cohorts, are exactly the kind of events that require source document verification at the site level.

The Decision in Front of Clinical Ops Leaders Now

If you are building the Phase 3 protocol for this triplet regimen, your immediate task is reconciling the Cohort 4 eligibility criteria against the intended Phase 3 population before the End-of-Phase 2 meeting with FDA. CLDN18.2 expression thresholds, IHC scoring cutoffs, and the central assay vendor selected for Cohort 4 must carry forward without modification — or every deviation from that biomarker definition becomes a comparability question that FDA will raise in writing. The 2019 adaptive design guidance and FDA’s guidance on biomarker-based clinical trial enrichment strategies both establish that the confirmatory trial must honor the evidentiary basis of the Phase 2 enrichment decision. Changing the biomarker threshold between Phase 2 and Phase 3 is not a protocol update. It is a new hypothesis.

The Next Signal to Watch

The Phase 3 trial design informed by ILUSTRO Cohort 4 will almost certainly require a co-primary endpoint structure that satisfies both FDA and EMA — overall survival and progression-free survival, likely with pre-specified subgroup analyses in CLDN18.2 high-expressers versus the broader positive population. Watch for the FDA’s Type B End-of-Phase 2 meeting outcome, which will determine whether the agency accepts a single-arm comparator structure or demands a randomized controlled design with a nivolumab-plus-chemotherapy control arm. Given CheckMate 649’s approval basis, FDA will be reluctant to accept anything less than a head-to-head randomized design. That design choice will drive sample size from an estimated 400 to 600 patients in a Phase 2 range into the 700 to 900 patient territory that defines a three-to-four year enrollment timeline — and with CLDN18.2 prevalence as the screening funnel, the site network size and geographic diversity of that Phase 3 will determine whether the timeline is real or aspirational.