TAK-280: a conditional B7-H3 x CD3 T-cell engager that reached cytokine release but not response

What was tried

Takeda ran a Phase 1/2 first-in-human study (NCT05220098) of TAK-280, an intravenous B7-H3 by CD3 epsilon T-cell engager, in adults with unresectable locally advanced or metastatic solid tumors. TAK-280 came out of Takeda's 2021 acquisition of Maverick Therapeutics and used the COBRA design, a conditional bispecific engineered to stay masked in circulation and to become active only after protease cleavage inside the tumor microenvironment. Dosing ran on days 1, 8, 15, and 22 of a 28-day cycle. The trial opened in April 2022, escalated through 10 dose levels labeled A through J, and treated 69 patients before Takeda closed it. Its listed primary endpoints were dose-limiting toxicities and treatment-emergent adverse events, the standard first-in-human safety readouts. The posted reason for stopping was direct: "Terminated due to limited anti-cancer activity."

The biological hypothesis

B7-H3, encoded by CD276, is expressed broadly across solid tumors with comparatively restricted expression on healthy tissue, which is why it has drawn antibody-drug conjugates, CAR-T cells, and T-cell engagers. A CD3 engager forces a synapse between a cytotoxic T cell and a tumor cell regardless of the T cell's native specificity. The COBRA twist was meant to solve the central problem of CD3 engagers in solid tumors, which is that systemic T-cell activation causes cytokine release and dose-limiting toxicity before tumor control is reached. By masking the CD3 arm and requiring tumor-localized protease activation, the design aimed to widen the therapeutic window. The rationale was published before the trial in Maverick's COBRA platform paper and reinforced by reviews of conditionally active engagers. Open Targets gives CD276 a direct association score of only 0.128 for cancer, and that score is driven almost entirely by literature and expression rather than by human genetics, so the target rests on biology of convenience rather than causal genetic anchoring.

What actually happened

The study behaved like a drug that engaged its target but did not treat the disease. Cytokine release syndrome was the most common serious adverse event, recorded in 14 of 69 treated patients, with serious alanine aminotransferase and aspartate aminotransferase increases in 6 and 5 patients. Serious adverse events of any kind affected 45 of 69 patients. That toxicity profile is exactly what on-target CD3 engagement produces, so the molecule was pharmacologically live. Escalation continued across all 10 dose levels, and the largest cohort of 26 patients sat at dose level I, which is where a program parks patients when it is chasing a response signal near the top of the range. The signal did not come. Primary completion was recorded on 28 July 2025, and Takeda deprioritized the Maverick-derived bispecifics, listing the reason as limited anti-cancer activity rather than safety.

Failure mechanism, best guess

This is a pharmacology-without-efficacy failure, the signature outcome for solid-tumor CD3 engagers. The presence of cytokine release confirms that T cells were activated somewhere, but activation did not convert into tumor regression. Several non-exclusive mechanisms fit. B7-H3 surface density is heterogeneous and the protein is shed, so redirected killing may have lacked a dense, uniform antigen to work against. The immunosuppressive tumor microenvironment and poor T-cell infiltration blunt engager activity even when antigen is present. The conditional masking that was meant to improve the window may also have lowered the concentration of active engager reaching the tumor, trading efficacy for tolerability. The cleanest external contrast is that B7-H3 antibody-drug conjugates deliver a cytotoxic payload rather than relying on host T cells, and one such agent, ifinatamab deruxtecan, reported a 54.8 percent objective response rate in extensive-stage small-cell lung cancer. The target is reachable. Redirected T-cell killing was the modality that did not deliver.

How to prevent this next time

Two quantitative levers would have changed the odds before dose level I. First, a Bayesian interim futility rule tied to the predictive probability of success, computed as

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

with a pre-specified stopping threshold, would have forced an explicit go or no-go once the early expansion cohorts produced cytokine release but no responses, rather than filling a 26-patient cohort on hope. Second, honest historical base-rate adjustment sets the prior: oncology programs advance from Phase 1 to approval about 3.4 percent of the time, and patient selection by a validated biomarker raises the probability of success by roughly 13 percentage points on average (Wong, Siah and Lo, 2019). A B7-H3 immunohistochemistry density cutoff, paired with a translational-validation gate requiring ex vivo evidence that cleaved TAK-280 kills patient-derived tumor at achievable exposures, would have concentrated the trial on the tumors where the mechanism could plausibly work. The single highest leverage change would have been enrolling only B7-H3-high tumors under a pre-registered Bayesian futility boundary, so the program spent its expansion cohort confirming activity in an enriched population instead of documenting toxicity across an unselected one.

What this means for similar programs

B7-H3 is a credible target for payload-delivering modalities and a hard one for T-cell redirection in solid tumors. Conditional and masked engager designs still owe a direct demonstration that peripheral cytokine release corresponds to intratumoral killing, because a wide therapeutic window is worthless if the active drug never concentrates on the tumor. Programs in this class should gate dose escalation on pharmacodynamic proof of tumor-cell killing, not on tolerability alone.

Open questions

Did cleaved, active TAK-280 reach tumor tissue at cytotoxic concentrations, and was that ever measured? Were enrolled tumors screened for B7-H3 density, and would an enriched population have responded? Would costimulatory support, such as a B7-H3 by CD28 bispecific, have rescued redirected killing? And did the masking strategy suppress efficacy more than it suppressed toxicity?

Sources

1. • ClinicalTrials.gov, NCT05220098, study record and posted results (accessed 2026-07-07): https://clinicaltrials.gov/study/NCT05220098 . Source for 69 treated patients, 10 dose levels, cytokine release syndrome in 14, serious ALT and AST increases in 6 and 5, serious adverse events in 45 of 69, primary completion 28 July 2025, and the stated reason for termination. - Root A. R. et al. COBRA: a highly potent conditionally active T cell engager engineered for the treatment of solid tumors. mAbs, 2020.- Cattaneo G. et al. Conditionally active T cell engagers for the treatment of solid tumors: rationale and clinical development. Expert Opinion on Biological Therapy, 2022.- Zhou W. T. and Jin W. L. New frontiers in immune checkpoint B7-H3 (CD276) research and drug development. Molecular Cancer, 2023.- Kontos F. et al. B7-H3/CD276: An Emerging Cancer Immunotherapy. Frontiers in Immunology, 2021.- Ifinatamab deruxtecan primary analysis, IDeate-Lung01, extensive-stage small-cell lung cancer (objective response rate 54.8 percent at 12 mg/kg). Journal of Clinical Oncology, 2025.- Wong C. H., Siah K. W., Lo A. W. Estimation of clinical trial success rates and related parameters. Biostatistics, 2019.- Open Targets Platform, target CD276 (ENSG00000103855), direct disease association scores (accessed 2026-07-07): https://platform.opentargets.org/target/ENSG00000103855.