A PD-L1 and CTLA-4 bispecific was stopped for lack of efficacy in MSS colorectal cancer on an endpoint the mechanism had never moved

What was tried

Peking University Cancer Hospital and Institute ran NCT05985109, a Phase 2 multi-cohort study of KN046 with regorafenib in microsatellite stable metastatic colorectal cancer. The registry describes KN046 as a recombinant humanized PD-L1 and CTLA-4 bispecific single-domain antibody, listed in ChEMBL as erfonrilimab, CHEMBL4650483, at maximum phase 3. Regorafenib is a multikinase inhibitor and an established third-line option here.

The design was non-randomized and open label. Three cohorts ran together: Arm A without active liver metastasis, Arm B with BRAF V600E mutation, and Arm C for patients unable or unwilling to take standard first-line or second-line therapy. Eligibility required colorectal adenocarcinoma, a measurable lesion by RECIST 1.1, ECOG status 0 or 1 and available tumor tissue. The primary endpoint was objective response rate by investigator assessment, with disease control rate, progression-free survival and overall survival secondary.

The study opened on 26 October 2023 and recorded completion on 30 September 2026, with actual enrollment of 20 across all three cohorts. The registry was updated on 3 November 2026 with the stopped reason recorded as the study being stopped early due to lack of efficacy. No results and no per-cohort allocation appear in the record.

The biological hypothesis

The combination rested on a clinical observation rather than human genetics. Fukuoka and colleagues reported REGONIVO, a Phase 1b study of regorafenib plus nivolumab in which 9 of 25 colorectal cancer patients responded, a 36% objective response rate with median progression-free survival of 7.9 months, almost all microsatellite stable. Regorafenib was thought to reduce macrophage-mediated suppression and make a cold tumor responsive to checkpoint blockade, and a PD-L1 and CTLA-4 bispecific was meant to deepen that effect.

Two later datasets refined it. Xiao and colleagues treated 22 patients with refractory non-liver metastatic disease using regorafenib, ipilimumab and nivolumab, reporting a 36.4% response rate with 8 responders and median overall survival of 27.5 months. Regorafenib plus sintilimab in 103 patients gave a 21.4% response rate, 95% confidence interval 13.3% to 29.4%, and median overall survival of 14.1 months. Restricting to patients without liver metastasis, as Arm A did, was evidence-based.

Target-level evidence is thinner. Open Targets, queried on 4 November 2026, scores CD274 against colorectal carcinoma at 0.145916, resting on text-mined literature at 0.9932 and clinical precedence at 0.1655, with no genetic association evidence at all. The platform lists 13 clinical-stage or approved agents against CD274.

What actually happened

The sponsor stopped the study for lack of efficacy and posted no numbers. What can be established is the precision the design could deliver. Using the registry enrollment of 20 across three cohorts, a balanced allocation gives roughly 6 to 7 patients per cohort, and Claidex computed exact Clopper-Pearson intervals from those counts. A cohort of 7 with zero responders yields a 95% confidence interval of 0% to 41.0%, which contains the 36.4% benchmark from the non-liver metastatic regorafenib, ipilimumab and nivolumab cohort. A cohort of 7 with one responder yields 0.4% to 57.9%. Only the pooled set of 20 with zero responders gives an interval, 0% to 16.8%, that excludes the benchmark, and pooling three distinct cohorts discards the stratification the design existed to test.

The same inputs give the probability of seeing no responses when the true rate equals the benchmark: 0.042 for a cohort of 7, and 0.0001 for 20 pooled. A futility call inside one cohort was compatible with an active regimen roughly one time in 24.

An openFDA FAERS query on 4 November 2026 returned 2 reports naming KN046, too few for any disproportionality calculation, against 2,106 for regorafenib. Spontaneous reports cannot give incidence, and neither figure bears on why this study stopped.

Failure mechanism, best guess

The archetype is efficacy failure as recorded, and the mechanism most consistent with the evidence is an endpoint and sample size mismatch rather than a refuted biological hypothesis.

The decisive comparison is the Canadian Cancer Trials Group CO.26 study, which tested the same dual axis, durvalumab plus tremelimumab, in 180 patients with refractory colorectal cancer. Median overall survival was 6.6 months against 4.1 for best supportive care, a hazard ratio of 0.72 with a 90% confidence interval of 0.54 to 0.97 and P equal to 0.07. Progression-free survival was 1.8 months against 1.9, a hazard ratio of 1.01. In the microsatellite stable subgroup overall survival improved with a hazard ratio of 0.66 and P equal to 0.02, and among those with plasma tumor mutational burden of at least 28 variants per megabase, 21% of that subgroup, the hazard ratio was 0.34.

Dual PD-L1 and CTLA-4 blockade in this disease has shown a survival effect with no progression-free survival effect, concentrated in a minority identified by a blood biomarker. A study powered on objective response rate read the one dimension on which the randomized evidence predicts no separation, in cohorts too small to measure even that, and without the biomarker that selected the responsive minority.

How to prevent this next time

Endpoint-level data were never released, so no posterior probability or conditional power can be computed, and none is claimed. The levers below use the sourced comparator data and the registry enrollment count.

Match the endpoint to the mechanism's demonstrated axis of benefit. Where the only randomized evidence shows a survival effect with a progression-free survival hazard ratio of 1.01, a single-arm response rate study cannot confirm or refute activity. A randomized design with survival as the primary measure answers the question.

Set the sample size against a prespecified benchmark rather than convenience. The comparators were published before this study opened in October 2023 and ranged from 21.4% to 36.4%. A cohort able to exclude a 36.4% benchmark at conventional confidence needs far more than 7 patients, and the intervals above show how fast precision collapses below that.

Enrich on the biomarker the randomized data identified. Plasma tumor mutational burden of at least 28 variants per megabase selected the CO.26 subgroup with a hazard ratio of 0.34, and the eligibility criteria here required tumor tissue but set no such threshold.

The single highest leverage change would have been to power the non-liver metastatic cohort as a randomized survival comparison with prespecified plasma tumor mutational burden enrichment, rather than as a 7-patient response rate cohort.

What this means for similar programs

The Claidex Mechanism Risk Score for CD274 now stands at 73 of 100, in the orange band, recomputed across all four CD274 claims in the graph after this insert. Saturation contributes 14.09 of a possible 15 points from 14 distinct programs against the target, genetic deficit 12.81 of 15 because the association score is 0.146, phase burden 18.25 of 30 and archetype severity 17.22 of 25. Any new CD274 program here enters a crowded field with no genetic support.

The graph already holds terminated colorectal programs against CD47 and LAG3, each a checkpoint or innate immune axis tested in unselected patients. A medRxiv preprint posted on 23 October 2026 reports site-specific immune remodeling underlying resistance to dual checkpoint blockade in metastatic colorectal cancer, consistent with the non-liver metastatic selection in Arm A and suggesting anatomic site deserves prospective stratification.

Open questions

How were the 20 participants distributed across the three cohorts, and did any produce a response? What futility rule triggered the stop, and was it prespecified? Was tumor mutational burden measured on the archival samples the protocol required? Did the BRAF V600E cohort behave differently from Arm A?

Sources

1. • ClinicalTrials.gov. A Phase II Multi-cohort Clinical Study Evaluating The Efficacy and Safety of KN046 in Combination With Regorafenib for Metastatic Microsatellite-Stable Colorectal Cancer. NCT05985109. Record updated 3 November 2026. https://clinicaltrials.gov/study/NCT05985109 - Open Targets Platform. CD274 (ENSG00000120217) and colorectal carcinoma (MONDO_0024331) association evidence. Queried 4 November 2026. https://platform.opentargets.org/evidence/ENSG00000120217/MONDO_0024331 - ChEMBL. Erfonrilimab, CHEMBL4650483. Queried 4 November 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL4650483/ - Fukuoka S, Hara H, Takahashi N, et al. Regorafenib plus nivolumab in patients with advanced gastric or colorectal cancer: an open-label, dose-escalation, and dose-expansion phase Ib trial (REGONIVO, EPOC1603). J Clin Oncol. 2020;38(18):2053-2061. https://- Chen EX, Jonker DJ, Loree JM, et al. Effect of combined immune checkpoint inhibition vs best supportive care alone in patients with advanced colorectal cancer: the Canadian Cancer Trials Group CO.26 study. JAMA Oncol. 2020;6(6):831-838. https://- Xiao A, Li X, Wang C, Ye J, Fakih M. Updated survival outcome of regorafenib, ipilimumab, and nivolumab in refractory microsatellite stable non-liver metastatic colorectal cancer: a phase I nonrandomized clinical trial. Eur J Cancer. 2024;213:115111. https://- Regorafenib plus sintilimab as a salvage treatment for microsatellite stable metastatic colorectal cancer: a single-arm, open-label, phase II clinical trial. Nat Commun.

2. - Tauriello DVF, Palomo-Ponce S, Stork D, et al. TGFbeta drives immune evasion in genetically reconstituted colon cancer metastasis. Nature. 2018;554(7693):538-543. - Kennel KB, Greten FR. The immune microenvironment of colorectal cancer. Nat Rev Cancer. 2025;25(12):945-964. - Chen E, Ling X. Immunotherapy in microsatellite-stable colorectal cancer: strategies to overcome resistance. Crit Rev Oncol Hematol. 2025;212:104775. - Site-specific immune remodeling underlies resistance to dual immune checkpoint blockade in metastatic colorectal cancer. medRxiv preprint posted 23 October 2026. - openFDA drug event (FAERS) API. Queries for KN046 and regorafenib, 4 November 2026. https://api.fda.gov/drug/event.json.