Favezelimab in Hodgkin lymphoma: the LAG-3 trial that was closed by a colorectal cancer readout

What was tried

Merck Sharp & Dohme opened MK-4280-003 (NCT03598608) on 17 October 2018 to test favezelimab, an anti-LAG-3 monoclonal antibody (ChEMBL CHEMBL4297784, maximum phase 3), together with pembrolizumab in haematological malignancies. Part A was an open-label dose-finding stage across three favezelimab dose levels, with co-primary endpoints of dose-limiting toxicity rate in cycle 1, overall adverse event rate, and discontinuation due to adverse events. Part B expanded the recommended Phase 2 dose into classical Hodgkin lymphoma, diffuse large B-cell lymphoma and indolent non-Hodgkin lymphoma cohorts, and added a 1:1 randomised comparison of favezelimab monotherapy against pembrolizumab monotherapy in Hodgkin lymphoma. Objective response rate was secondary, and prior anti-LAG-3 exposure was an exclusion criterion.

The study enrolled 137 participants, ran for more than seven years, and closed with a completion date of 28 January 2026. The registry gives the stopping reason in two words: business reason. No results have been posted.

The biological hypothesis

LAG3 (Ensembl ENSG00000089692) encodes an inhibitory receptor on activated and exhausted T cells. It binds MHC class II and fibrinogen-like protein 1, and suppresses CD4 T cell activation partly by trans-endocytosis of MHC class II from antigen-presenting cells (doi:10.1016/j.celrep.2024.114655). Upregulation of LAG-3 marks a functionally exhausted T cell subpopulation and tracks with disease progression (doi:10.4049/jimmunol.1402176). The therapeutic logic is co-blockade: PD-1 inhibition alone leaves a second brake engaged. A cryo-electron microscopy structure confirmed that favezelimab binds the LAG-3 D1 domain (doi:10.1016/j.str.2023.07.013).

Classical Hodgkin lymphoma was a defensible place to test this. It is the tumour type where PD-1 blockade performs best, and LAG-3 expression in Hodgkin lymphoma has been associated with survival differences (doi:10.3389/fonc.2022.957208). The Open Targets association score for LAG3 and Hodgkin lymphoma was 0.11076, from literature (0.367) and clinical (0.164) evidence with no genetic association contribution. For diffuse large B-cell lymphoma the score was 0.1341, with a literature component of 0.903 and genetic association of zero. LAG3 scores highest against melanoma at 0.5916, the indication where relatlimab reached approval.

The class was and remains busy. Open Targets records 6 clinical programmes against LAG3: relatlimab at approval, fianlimab, eftilagimod alfa and favezelimab at Phase 3, tebotelimab at Phase 2/3, and ieramilimab at Phase 2.

What actually happened

The trial itself never reported a negative result. What happened sat one level up, in the programme.

On 25 September 2024 Merck announced that KEYFORM-007, a randomised Phase 3 study of a favezelimab and pembrolizumab fixed-dose combination against regorafenib or trifluridine-tipiracil in 441 people with previously treated PD-L1-positive microsatellite stable metastatic colorectal cancer, did not meet its primary endpoint of overall survival. The safety profile was described as consistent with prior studies, with no new signals.

On 16 December 2024 Merck discontinued KEYFORM-008, the Phase 3 trial of the same fixed-dose combination against physician's choice chemotherapy in anti-PD-1 relapsed or refractory classical Hodgkin lymphoma. Merck attributed the decision to a review of data across the favezelimab programme and stated that it was not based on safety concerns about the combination. Participants already on therapy could continue. KEYFORM-008 was described as the only Phase 3 study in the programme for which results were not available.

The Hodgkin lymphoma data that had motivated KEYFORM-008 were published afterwards. In an unanchored cross-trial comparison, 27 participants from MK-4280-003 with anti-PD-1-refractory classical Hodgkin lymphoma had an objective response rate of 37 percent (95 percent CI 15 to 51) against 2 percent (95 percent CI 0 to 6) among 81 KEYNOTE-087 participants who received pembrolizumab beyond progression. A reduction of at least 50 percent in target lesion size occurred in 13 of 27 (48 percent) versus 4 of 81 (5 percent), and the mean change was minus 49 percent versus minus 0.4 percent (doi:10.1182/bloodadvances.2024014654). That comparison is not randomised evidence, and the authors framed it as an estimate of favezelimab contribution.

MK-4280-003 closed thirteen months after the KEYFORM-008 decision.

Failure mechanism, best guess

The archetype is strategic reprioritisation. This was not an efficacy failure of the trial that was terminated, and the sponsor stated it was not a safety decision. The mechanism of loss was sequencing.

The programme let its largest and most commercially attractive indication read out first. Microsatellite stable colorectal cancer has the weakest prior for checkpoint co-blockade of any setting favezelimab entered, because these tumours are immunologically cold and have resisted PD-1 blockade as a class. The Open Targets evidence for LAG3 in colorectal cancer (0.353) is carried by clinical trial records rather than by genetics, so it did not represent independent biological support. When the colorectal readout failed, it removed the commercial case for the fixed-dose combination, and the Hodgkin lymphoma question was retired with it. The 37 percent response rate in anti-PD-1-refractory disease was never tested against a randomised control.

How to prevent this next time

There is no endpoint-level dataset from MK-4280-003 to model, so the levers here are portfolio-design levers rather than statistical ones.

Sequence confirmatory trials by prior strength, not by market size. The indication with the highest probability of a positive result should read out first, because a positive result there protects the mechanism while a negative result in a low-prior indication can end it.

Separate the mechanism decision from the formulation decision. KEYFORM-007 tested a fixed-dose combination, which couples the fate of the antibody to one product presentation. A programme-level stopping rule should distinguish evidence that the target is wrong from evidence that one configuration in one tumour type is not commercially viable.

Adjust the base rate for cold-tumour co-blockade. Across the LAG3 class the only approval sits in melanoma, an immunologically hot tumour. Entering a microsatellite stable colorectal population with a co-blockade hypothesis carries a prior that should have been written down and used to size the downside of a miss.

Post results from terminated trials. MK-4280-003 enrolled 137 participants across three lymphoma histologies and posted nothing. The randomised favezelimab versus pembrolizumab monotherapy sub-part in Hodgkin lymphoma is the single most informative comparison the programme generated, and it remains unreported.

The single highest leverage change would have been running the Hodgkin lymphoma Phase 3 to a randomised readout before, rather than alongside, the microsatellite stable colorectal Phase 3, so that the mechanism was judged where the prior was strongest.

What this means for similar programs

LAG3 is not refuted. Relatlimab is approved, and four other programmes remain at Phase 2 or beyond, which is why the saturation component of this target's Mechanism Risk Score is the largest single contributor at 10.48, with a total of 35 in the yellow band. A high saturation score with a low archetype score is the signature of a crowded target losing programmes to portfolio decisions rather than to biology.

For teams working on LAG-3 or any second checkpoint, the transferable lesson is that co-blockade assets are unusually exposed to the indication in which they are first tested at scale. Response-rate signals in a favourable histology do not protect an asset if the registrational bet was placed elsewhere.

Open questions

What did the randomised favezelimab versus pembrolizumab monotherapy sub-part of MK-4280-003 show? That comparison is the only randomised evidence the programme produced on favezelimab contribution in Hodgkin lymphoma, and it has not been reported.

Did the diffuse large B-cell and indolent non-Hodgkin cohorts show activity, or was the Hodgkin signal isolated?

Was there a biomarker, LAG-3 expression or MHC class II status, that separated the 13 of 27 participants with major tumour reduction from the rest?

Sources

1. • ClinicalTrials.gov, NCT03598608, MK-4280-003, full record. https://clinicaltrials.gov/study/NCT03598608 (accessed 2026-07-29) - Open Targets Platform, LAG3 (ENSG00000089692), target record, clinical candidates, and association with Hodgkin lymphoma (MONDO_0004952), score 0.11076. https://platform.opentargets.org/target/ENSG00000089692 (accessed 2026-07-29) - ChEMBL, favezelimab, CHEMBL4297784, antibody, maximum phase 3. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL4297784/ (accessed 2026-07-29) - openFDA drug event API, medicinal product FAVEZELIMAB, 38 reports, 38 serious, 3 with death outcome. https://api.fda.gov/drug/event.json (accessed 2026-07-29) - Merck. Merck Provides Update on Phase 3 KEYFORM-007 Trial. 25 September 2024. https://www.merck.com/news/merck-provides-update-on-phase-3-keyform-007-trial-evaluating-investigational-fixed-dose-combination-of-favezelimab-and-pembrolizumab-for-patients-with-previously-treated-pd-l1-positive-microsatellite/ - Merck. Merck Provides Update on KeyVibe and KEYFORM Clinical Development Programs. 16 December 2024. https://www.merck.com/news/merck-provides-update-on-keyvibe-and-keyform-clinical-development-programs-evaluating-investigational-vibostolimab-and-favezelimab-fixed-dose-combinations-with-pembrolizumab/ - Armand P, Zinzani PL, Timmerman J, Johnson NA, et al. Estimating efficacy of favezelimab plus pembrolizumab relative to pembrolizumab in anti-PD-1-refractory Hodgkin lymphoma. Blood Adv. 2025;9(19):4987-4995.- Garralda E, et al. A first-in-human study of the anti-LAG-3 antibody favezelimab plus pembrolizumab in previously treated, advanced microsatellite stable colorectal cancer. ESMO Open.

2. - Mishra AK, et al. CryoEM structure of a therapeutic antibody (favezelimab) bound to human LAG3. Structure.

3. - Wakamatsu E, et al. Indirect suppression of CD4 T cell activation through LAG-3-mediated trans-endocytosis of MHC class II. Cell Rep.

4. - Tian X, et al. The upregulation of LAG-3 on T cells defines a subpopulation with functional exhaustion and correlates with disease progression. J Immunol.

5. - Jimenez O, et al. PD-1 and LAG-3 expression in EBV-associated pediatric Hodgkin lymphoma has influence on survival. Front Oncol.

6. .