Checkpoint Inhibitors Are Failing Recurrent GBM. This Phase 1 Trial Design May Explain Why They Keep Trying.

Pull up the Nature Medicine phase 1 trial of relatlimab with or without nivolumab in recurrent glioblastoma and read past the abstract. The trial enrolled 46 patients, 23 per cohort, in a disease where median survival after recurrence is measured in months. That sample size tells you something the investigators would not say outright: the field has abandoned the idea that a clean efficacy signal is coming from a small cohort in glioblastoma. The goal here was safety, signal, and protocol architecture. The goal was to figure out how to run a trial that a larger adaptive study can actually build on.

Three developments in CNS immunotherapy over the past eighteen months have been converging on a single uncomfortable conclusion, one that the industry has not quite named yet: the barrier to immunotherapy success in glioblastoma is not the drug target. It is the trial design ecosystem that surrounds it. Call it the CNS Protocol Deficit.

The Signal Everyone Has Been Ignoring

The history of anti-PD-1 in recurrent glioblastoma is a graveyard of promising mechanisms and underpowered conclusions. In KEYNOTE-028, pembrolizumab monotherapy in PD-L1-positive recurrent GBM produced an overall response rate of 8% among 26 enrolled patients, with two partial responses lasting 8.3 and 22.8 months after a median follow-up of 14 months. Two responses. The oncology community looked at that number and largely moved on. That was a mistake.

The mistake was not interpreting 8% as failure. The mistake was interpreting it as a complete picture. Glioblastoma’s immunosuppressive microenvironment does not just suppress T cells. It exhausts them through multiple parallel pathways simultaneously. LAG-3, a co-inhibitory receptor expressed on tumor-infiltrating lymphocytes, operates independently of the PD-1 axis. Blocking one pathway while the other remains active is like closing one lane of a four-lane highway and calling it traffic management. The relatlimab plus nivolumab hypothesis was mechanistically sound before the trial started. What was missing was a protocol architecture capable of detecting a signal in a population where the noise floor is catastrophically high.

The relatlimab trial addressed that directly. By running parallel cohorts of 23 patients each, one receiving relatlimab monotherapy and one receiving the combination, the investigators created an internal comparator without requiring a separate placebo arm that no recurrent GBM patient would consent to in 2026. That structural choice matters more than it appears.

Why the Blood-Brain Barrier Argument Has Always Been Incomplete

The standard objection to checkpoint inhibitors in primary brain tumors goes like this: antibodies cannot cross the blood-brain barrier in meaningful quantities, so systemic immunotherapy cannot reach the tumor. The data behind this objection is real. Approximately 0.1% of circulating antibodies cross the BBB, according to a September 2025 MDPI analysis of CNS checkpoint inhibitor pharmacokinetics. That fraction sounds disqualifying.

It is not, and the relatlimab trial design implicitly acknowledges why. The mechanism of CNS checkpoint blockade does not require the antibody to reach the tumor in bulk. It requires the antibody to activate peripheral and infiltrating T cells that then traffic into the CNS, a process that occurs at the BBB margin rather than through it. The CheckMate 143 phase 1 study, which evaluated nivolumab alone and nivolumab plus ipilimumab in recurrent GBM, documented a toxicity profile that actually confirmed T cell CNS infiltration, because immune-related adverse events at grade 3 or higher appeared at rates requiring protocol-level safety management. You do not get CNS irAEs from a drug that never reaches the tumor microenvironment.

The relatlimab trial built on this by requiring systematic neurological adverse event monitoring at intervals that earlier GBM immunotherapy trials treated as optional. That operational discipline, pre-specified imaging windows, neurocognitive assessments tied to adverse event grading, creates a dataset that downstream adaptive trials can use to set safety stopping rules with actual evidence behind them rather than extrapolations from non-CNS oncology trials.

What Sponsors Need to Build Next

Here is the counterintuitive implication that most sponsors reading this trial will miss. The conventional assumption is that a phase 1 safety signal in recurrent GBM justifies a direct move to a randomized phase 2, probably 100 to 150 patients, powered for progression-free survival. That path is available. It is also likely to produce a fourth or fifth inconclusive result in a disease where the recurrent GBM therapeutic market was valued at USD 3.5 billion in 2026, projected to reach USD 5.37 billion by 2033, yet systemic checkpoint therapy has not delivered a meaningful survival benefit to date.

The more defensible path is adaptive, and it requires FDA alignment before the phase 2 IND goes in. The FDA’s February 2024 draft guidance on Data Monitoring Committees in Clinical Trials specifically addresses the integration of DMCs into adaptive designs, acknowledging that pre-specified interim analyses with adaptive dose or cohort modifications require DMC charters written with considerably more specificity than those used in conventional oncology trials. In a CNS indication where progression is difficult to distinguish from pseudoprogression on standard MRI, the DMC’s role as the arbiter of whether a patient has progressed or responded becomes a protocol design question, not just a safety question.

For CROs managing GBM studies, the relatlimab trial’s parallel-cohort architecture also signals something operationally important. Site selection criteria for CNS immunotherapy trials increasingly require neuro-oncology teams with experience reading immunotherapy response on modified RANO criteria, a competency that is not evenly distributed across academic medical centers. Sponsors who assume that any NCI-designated cancer center can serve as a GBM immunotherapy site are building recruitment models on phantom capacity.

Technology vendors should pay attention to the irAE monitoring architecture embedded in this trial design. The systematic neurological assessment windows, tied to imaging and clinical evaluation at protocol-specified intervals, represent exactly the kind of structured data capture that eCOA platforms have been promising for CNS indications for years without delivering at the protocol-specification level. The relatlimab trial shows what the data collection requirements actually look like in practice. Building eCOA instruments that map to modified RANO response criteria and neurocognitive adverse event grading scales is not a feature request for the next product cycle. Sponsors designing phase 2 adaptive GBM trials in 2027 will require it at the RFP stage.

The recurrent GBM checkpoint inhibitor field has produced ten years of negative or inconclusive trials, and the temptation is to read each new phase 1 as another step in a long march toward eventual failure. But the relatlimab trial’s actual contribution is the protocol blueprint it leaves behind, the parallel cohort structure, the CNS-specific irAE monitoring cadence, the internal comparator design that sidesteps placebo consent barriers. Sponsors who mine the methods section rather than the response rate table will find the signal that actually matters. The next trial that fails in recurrent GBM will fail because it ignored that blueprint, not because the LAG-3 target was wrong.

References

1. Nature Medicine — “Anti-LAG-3 with or without anti-PD-1 in recurrent glioblastoma: a phase 1 trial”
2. PubMed — KEYNOTE-028: Pembrolizumab monotherapy in PD-L1-positive recurrent glioblastoma, ORR 8% in 26 patients
3. MDPI — “Immune Checkpoint Inhibition in Patients with Brain Metastases from Non-Small-Cell Lung Cancer: Emerging Mechanisms and Personalized Clinical Strategies” (September 2025): ~0.1% of circulating antibodies cross the BBB
4. MDPI — CheckMate 143 phase 1: nivolumab alone or combined with ipilimumab in recurrent GBM, CNS irAE toxicity profile
5. Advarra — FDA February 2024 Draft Guidance on Data Monitoring Committees in Clinical Trials
6. Coherent Market Insights — Recurrent GBM therapeutic market valued at USD 3.5 billion in 2026, projected USD 5.37 billion by 2033