A mask that did not hold: zero responses in 95 patients on TAK-186

What was tried

TAK-186, also called MVC-101, is an EGFR by CD3 Conditional Bispecific Redirected Activation protein, the platform Maverick Therapeutics branded COBRA. Takeda agreed to acquire Maverick on 9 March 2021 for up to 525 million dollars in upfront and milestone payments, with TAK-186 as the lead asset. The design goal was spatial selectivity. The molecule circulates masked and inactive and is meant to assemble into an active T-cell engager only after cleavage by tumour-enriched proteases, so that killing happens at the tumour rather than on EGFR-positive skin, gut and liver.

The first-in-human study (NCT04844073, protocol CP-MVC-101-01) was an open-label Phase 1/2 dose escalation run by Takeda. Eligible participants had unresectable advanced or metastatic solid tumours considered to express EGFR, performance status 0 or 1, and measurable disease by RECIST v1.1. The expansion cohort enrolled non-small cell lung cancer, with head and neck and colorectal cohorts to follow.

Primary outcomes were adverse events, cytokine release syndrome by ASTCT grading, and dose-limiting toxicities, with response and survival as secondary. The study ran from 8 March 2021 to 17 June 2025, enrolled 95 participants across ten treatment groups, and posted full results. The record was updated on 24 September 2026 with status TERMINATED, the reason given as a sponsor decision based on limited anti-cancer activity.

The biological hypothesis

EGFR (ENSG00000146648) is one of the best-validated targets in oncology. This was a basket across three tumour types, so the pair is scored at the neoplasm level, where Open Targets gives EGFR 0.6512. Against non-small cell lung carcinoma alone, the cohort that expanded, the score is 0.8503. Open Targets records 82 drug or clinical candidates. The problem has never been whether EGFR is a real target. It is that EGFR sits on normal epithelium, which caps how hard any EGFR-directed agent can be pushed. Open Targets lists diarrhoea, skin rash, ALT increase, AST increase and hepatotoxicity among the target's safety liabilities.

T-cell engagers sharpen that problem, because redirected cytotoxicity needs no high antigen density and spares no normal tissue carrying the antigen. Panchal and colleagues described COBRA in 2020 as a conditionally active engager with a protease-cleavable mask, so that potency is recovered in tumour and suppressed in circulation. If the mask held, the therapeutic index would come from the protease environment rather than from antigen selectivity.

What actually happened

No participant responded. Across all ten treatment groups the objective response rate was 0 percent, and in the 31-participant non-small cell lung cancer expansion the 90 percent confidence interval ran from 0 to 9.2 percent. Duration of response was not estimable. Median progression-free survival in that cohort was 7.9 weeks and median overall survival 23.6 weeks. Of the 95 who started, 6 completed and 49 died.

The safety profile was not quiet. Every participant had a treatment-emergent adverse event and 75 of 95 had a serious one. Cytokine release syndrome was recorded at grade 1 in 27 participants, grade 2 in 37, grade 3 in 9 and grade 4 in 1, and the serious form affected 49. Twelve dose-limiting toxicities occurred among 51 dose-evaluable participants, and escalation used prophylactic dexamethasone, then double dexamethasone, then multi-step ramping. The recommended Phase 2 dose was 10 micrograms per kilogram.

The most common non-serious events were diarrhoea in 58 participants, nausea in 54, vomiting in 48, alanine aminotransferase increase in 44 and aspartate aminotransferase increase in 41, the gut and liver signature Open Targets lists for EGFR. Among 89 participants evaluable for anti-drug antibodies, 64 were negative at baseline and became positive. No openFDA FAERS record names TAK-186 or MVC-101.

Failure mechanism, best guess

The archetype is efficacy failure, and the sponsor said so plainly. The useful question is why a conditionally active engager against a validated target produced zero responses in 95 patients. The toxicity pattern argues that the mask did not deliver its intended selectivity. Diarrhoea, vomiting and transaminase elevations in roughly half the cohort are on-target, off-tumour EGFR effects in tissues the design was meant to protect, and cytokine release syndrome in most participants shows systemic T-cell activation. If activation had been confined to tumour, the dose-limiting problem should have looked different.

The dose ceiling follows. Escalation ended at 10 micrograms per kilogram with 12 dose-limiting toxicities and a dexamethasone regimen that had to be doubled. Prophylactic steroids suppress lymphocytes, so the measure that made the dose tolerable plausibly cut the effector function the drug depends on.

Immunogenicity compounds both. A masked construct with engineered non-germline cleavage sequences presents novel epitopes, and 64 of 89 evaluable participants seroconverted. Antibodies that clear or neutralise the molecule would erode exposure over a weekly schedule, though no supporting data were posted.

A fourth possibility cannot be separated out on public data. Enrolment required EGFR expression by literature report rather than measured density, so the cohort may never have carried the antigen levels the mechanism needs.

How to prevent this next time

Posted data include response rates, confidence intervals and safety counts, but no exposure, protease activity or antigen density by participant, so no exposure-response model can be built. The levers below are qualitative.

Measure the conditional activation itself. A masking platform makes a testable claim about the ratio of active to masked molecule in tumour versus blood. Without a validated assay in early cohorts, the central premise was never put at risk, and the trial tested the molecule instead of the idea.

Treat on-target toxicity as a platform readout, not a tolerability nuisance. Gastrointestinal and hepatic events at low doses were early evidence of unmasking outside the tumour, long before the expansion cohort.

Enrich on measured antigen. Enrolling on a literature-based expectation of expression leaves a negative result uninterpretable.

Pre-specify a futility rule on response rather than tolerability. Stopping at the first 15 participants in the expansion without a response would have reached the same conclusion with half the exposure.

The single highest leverage change would have been a validated tumour-versus-plasma assay of the unmasked, active form, with a pre-specified activation ratio the platform had to clear before any expansion cohort opened.

What this means for similar programs

EGFR now carries three Claidex entries. The recomputed mechanism risk score for this pair is 56, in the orange band, from phase burden 10.63, archetype severity 15.80, recency 9.39, genetic deficit 5.23 and saturation 15.00. Saturation sits at its ceiling because 82 Open Targets programmes plus 3 Claidex entries give 85 distinct programmes against this target. The genetic term is larger than it would be for an EGFR-mutant lung cancer pair, because enrolling on expression across three tumour types buys a weaker genetic case. The score is about crowding and therapeutic index, not whether EGFR matters.

The read-across runs to every masked or conditionally active biologic, not to EGFR. Protease-activated designs convert a selectivity problem into a protease biology problem, and that only helps if the protease environment is measured in humans. The other two EGFR entries in the graph, an imaging analogue stopped for insufficient tumour uptake and a chimeric degrader retired for strategy, point at the same issue of getting enough active agent into tumour and not elsewhere. The target was right, the construct reached patients, and the window never opened.

Open questions

Was unmasked TAK-186 measurable in tumour biopsies, and did the tumour-to-plasma ratio match the design target?

Did anti-drug antibody status track with exposure or time on treatment? No relationship was posted.

Sources

1. • ClinicalTrials.gov, NCT04844073, protocol record and posted results retrieved 25 September 2026. https://clinicaltrials.gov/study/NCT04844073 - Takeda. Takeda to Acquire Maverick Therapeutics to Advance T-Cell Engager Therapies for Solid Tumors and Expand Novel Immuno-Oncology Portfolio. 9 March 2021. https://www.takeda.com/newsroom/newsreleases/2021/takeda-to-acquire-maverick-therapeutics-to-advance-t-cell-engager-therapies-for-solid-tumors-and-expand-novel-immuno-oncology-portfolio/ - Panchal A, et al. COBRA: a highly potent conditionally active T cell engager engineered for the treatment of solid tumors. MAbs.

2. - Open Targets Platform, EGFR (ENSG00000146648), associations with neoplasm (MONDO_0005070, score 0.6512) and non-small cell lung carcinoma (MONDO_0005233, score 0.8503), 82 drug and clinical candidates, safety liabilities, retrieved 25 September 2026. https://platform.opentargets.org/target/ENSG00000146648 - ChEMBL, CHEMBL5315000 (TAK-186), maximum phase 1, retrieved 25 September 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL5315000/ - openFDA drug adverse event endpoint, queries for TAK-186 and MVC-101 returned no matching records, 25 September 2026. https://open.fda.gov/apis/drug/event/ - Segal NH, et al. CEA-CD3 bispecific antibody cibisatamab with or without atezolizumab in patients with CEA-positive solid tumours. Nat Commun.

3. - Paz-Ares L, et al. Tarlatamab, a First-in-Class DLL3-Targeted Bispecific T-Cell Engager, in Recurrent Small-Cell Lung Cancer. J Clin Oncol.

4. - Claidex failure graph entries fpi-2107-egfr-cmet-nsclc-dosimetry-uptake-failure and bg-60366-egfr-cdac-nsclc-strategic-shutdown. https://app.claidex.com/claim/fpi-2107-egfr-cmet-nsclc-dosimetry-uptake-failure.