Acasunlimab in melanoma: a PD-L1x4-1BB bispecific shelved before the question could be tested

What was tried

Genmab opened ABBIL1TY MELANOMA-07 (NCT06984328), a randomized, open-label Phase 2 study of acasunlimab as monotherapy and in combination with pembrolizumab. The population was relapsed or refractory, unresectable locally advanced or metastatic cutaneous melanoma that had progressed on or after an approved checkpoint inhibitor. The primary endpoint was objective response rate measured up to roughly 12 months.

Acasunlimab, previously known as GEN1046 and BNT311, is an Fc-inert DuoBody bispecific antibody that binds PD-L1 on one arm and 4-1BB (CD137, gene TNFRSF9) on the other. The design intent is to block the PD-L1 checkpoint and deliver 4-1BB costimulation in the same molecule. Per the ClinicalTrials.gov record, the study was terminated with one participant enrolled, the actual primary completion date was May 15, 2026, and the sponsor stated the decision followed strategic portfolio prioritization and was not related to safety.

The biological hypothesis

4-1BB is a costimulatory receptor in the TNF-receptor superfamily expressed on activated T cells and NK cells. Agonism promotes T-cell proliferation, survival, and memory formation, which makes it an attractive partner for checkpoint blockade. The class has a long record of disappointment. The monospecific agonist urelumab produced dose-limiting hepatotoxicity that capped its dose below the active range, and utomilumab was well tolerated but only weakly agonistic (Blood 2018, 10.1182/blood-2017-06-741041; Eur J Immunol 2016, 10.1002/eji.201445388).

The bispecific premise is conditional agonism. By requiring PD-L1 for 4-1BB clustering, costimulation should concentrate in the PD-L1-rich tumor microenvironment rather than the liver, widening the therapeutic window while the same molecule relieves PD-L1 suppression. Preclinical and translational work reported that acasunlimab combined with PD-1 blockade gave complementary immune modulation and antitumor activity (J Immunother Cancer 2025, 10.1136/jitc-2024-011377). The genetic anchor for this target in melanoma is thin. The Open Targets association score for TNFRSF9 and cutaneous melanoma is 0.0083, supported by literature mining rather than human genetic evidence (Open Targets).

What actually happened

No efficacy signal came out of MELANOMA-07. The trial enrolled a single patient before Genmab discontinued the entire acasunlimab program following a portfolio review in early 2026, redirecting resources toward late-stage assets epcoritamab and petosemtamab (Genmab corporate disclosures; AACR Cancer Discovery News, 2026). BioNTech had returned its rights to the asset in 2024, after which Genmab held full development control. The companion Phase 3 non-small cell lung cancer program was shelved in the same review. The termination here is administrative rather than data-driven, and the central scientific question the trial was built to answer was never tested in melanoma.

The pharmacovigilance footprint is correspondingly small. FAERS reports associated with acasunlimab are sparse and dominated by hepatitis and febrile neutropenia terms, which are consistent with known 4-1BB hepatic and myeloid liabilities but rest on very few cases and cannot support a disproportionality claim (openFDA FAERS).

Failure mechanism, best guess

This was a capital-allocation outcome, not a mechanistic verdict. With one enrolled patient and no readout, nothing about conditional 4-1BB agonism in checkpoint-refractory melanoma was confirmed or refuted. The honest interpretation is that the asset lost an internal competition for resources against programs closer to revenue.

The program did carry real scientific risk into that competition. The 4-1BB class has not yet produced a controlled efficacy win in any indication, the bispecific architecture reduces but does not provably eliminate the efficacy-tolerability bind, and post-checkpoint melanoma is a crowded setting with a low base rate for novel immuno-oncology combinations. Those facts make the asset easy to deprioritize when a portfolio tightens, because the expected value of a hard melanoma arm is low relative to assets with established late-stage signal.

How to prevent this next time

Two changes would have protected the science from the portfolio math. First, a Bayesian go/no-go framework with an explicit predictive probability of success would have made the melanoma arm conditional on a cleaner upstream proof of mechanism. The predictive probability is the standard quantity:

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

Second, a power and sample-size calculation anchored to a realistic base rate would have flagged the arm as fragile before activation. Post-checkpoint melanoma responds to novel combinations at roughly 10 to 15 percent in historical experience, so a single-arm Phase 2 powered to distinguish a 25 percent response rate from a 10 percent null at 80 percent power and a one-sided 0.05 alpha needs on the order of 40 to 50 evaluable patients. With a number-needed-to-screen inflated by mandatory PD-L1 expression and prior checkpoint progression, the operational cost of that arm was high and its decision value was contingent on the lung program. Pre-committing the melanoma activation to a lung readout, rather than opening both in parallel, would have kept the asset from being orphaned mid-stream.

The single highest leverage change would have been to gate the melanoma Phase 2 on a pre-specified Bayesian proof-of-mechanism threshold from the lead indication before enrolling any patients.

What this means for similar programs

For PD-L1×4-1BB and other conditional-agonist bispecifics, the lesson is sequencing discipline. The hardest indication should sit behind a clean mechanistic readout, not run alongside it on shared funding. The Claidex Mechanism Risk Score for TNFRSF9 is 36 out of 100, in the yellow band, and the dominant component is the genetic-evidence deficit of 14.9 out of 15 rather than accumulated failures. A weakly anchored costimulatory target in a competitive setting is exactly the profile that gets cut first when a pipeline is rationalized, which means program teams should treat genetic-anchor strength as a survival variable, not only a probability-of-efficacy variable.

Open questions

Does PD-L1-conditional 4-1BB agonism actually localize costimulation to tumor and spare the liver in patients, as the design predicts? Would acasunlimab plus pembrolizumab have produced measurable intratumoral T-cell expansion in checkpoint-refractory melanoma, where reinvigoration is the central problem? Is there a biomarker-defined melanoma subset, for example high PD-L1 with residual T-cell infiltrate, where conditional agonism could clear the historical base rate? And does the broader 4-1BB bispecific field still justify late-line solid-tumor trials given the absence of a controlled class win to date?

Sources

1. • ClinicalTrials.gov, NCT06984328 (ABBIL1TY MELANOMA-07): study design, single-patient enrollment, termination reason, primary completion date. https://clinicaltrials.gov/study/NCT06984328 - J Immunother Cancer 2025, acasunlimab Fc-inert PD-L1×4-1BB mechanism with PD-1 blockade. https://- Blood 2018, immunotherapy targeting 4-1BB, mechanistic rationale and clinical results. https://- Eur J Immunol 2016, CD137 (4-1BB) signaling in T-cell costimulation. https://- ESMO Open 2020, CD137/4-1BB as an emerging immunotherapy target. https://- Open Targets Platform, TNFRSF9 (ENSG00000049249) to cutaneous melanoma (EFO_0000389) association score 0.0083. - openFDA FAERS, adverse-event term counts for acasunlimab. - AACR Cancer Discovery News and Genmab corporate disclosures, 2026, acasunlimab program discontinuation following portfolio prioritization.