Command Palette

Search for a command to run...

Evorpacept plus cetuximab and pembrolizumab stopped in MSS colorectal cancer after two fatal immune activation events

OncologySafetyJuly 27th, 2026·6 min read·10.5281/zenodo.20479005

A phase 2 safety run-in of the CD47 blocker evorpacept with cetuximab and pembrolizumab in refractory microsatellite stable metastatic colorectal cancer halted accrual at 16 treated patients. One treatment-related grade 5 hemophagocytic lymphohistiocytosis and one treatment-related grade 5 cytokine release syndrome ended a trial designed to expand to 42.

Mechanism Risk Score

ComponentPoints
Phase-weighted failure burden4.7 / 40
Archetype severity8.5 / 25
Temporal recency4.3 / 15
Genetic evidence deficit13.4 / 15
Programmatic saturation2.5 / 5

For CD47 in Refractory microsatellite stable metastatic colorectal cancer, the Mechanism Risk Score is 33/100 (yellow band). The score is a failure-burden index derived from Claidex post-mortems on this target–disease pair, not a probability of approval.

MRS 33/100 (YELLOW). 1 programs across CD47 have been documented for CD47 in Refractory microsatellite stable metastatic colorectal cancer: 0 Phase 3, 0 Phase 2, 0 Phase 1 — of which 0 were efficacy failures, 1 safety, 0 biomarker, and 0 operational (enrollment, sponsor, or funding). The most informative failure on file is Evorpacept plus cetuximab and pembrolizumab stopped in MSS colorectal cancer after two fatal immune activation events. This score quantifies the documented failure burden; the Open Targets association score of 0.11 reflects weak genetic anchoring, compounding the documented failure record. The MRS is not a prediction of future trial outcomes — it is a structured summary of the empirical record, recomputed live from the Claidex claims table, and intended to flag mechanisms where any new program must explicitly resolve each prior failure mode before pursuit is justified.

This score does not predict whether the next trial will succeed. It flags how heavy the documented mechanistic failure record is before a new program is justified.

Primary figure supporting this claim (Evorpacept (ALX148) / CD47 / Refractory microsatellite stable metastatic colorectal cancer): Evorpacept plus cetuximab and pembrolizumab stopped in MSS colorectal cancer after two fatal immune activation events

What was tried

Evorpacept, also called ALX148 and registered in ChEMBL as CHEMBL4297880, is a CD47 blocking fusion protein built on an inactivated Fc scaffold. The University of Colorado sponsored a phase 2 study of evorpacept with cetuximab and pembrolizumab in refractory microsatellite stable metastatic colorectal cancer, with ALX Oncology, Merck Sharp and Dohme, Eli Lilly and the Academic GI Cancer Consortium as collaborators.

The design was open label, non-randomized and sequential. Stage 1 was a safety run-in to set a recommended dose, and stage 2 was to expand to a planned 42 patients and test objective response rate against a historical control. Eligible patients had an ECOG performance status of 0 or 1, unresectable metastatic colorectal adenocarcinoma, and progression on at least two prior lines. EGFR expression by immunohistochemistry was not required. Evorpacept was dosed weekly at 15 mg/kg or 10 mg/kg alongside standard cetuximab and pembrolizumab. The study ran from 28 July 2022 to 30 October 2024, and results were posted on 22 July 2026.

The biological hypothesis

CD47 signals through SIRPα on macrophages to suppress phagocytosis, and blocking that interaction should release myeloid cells onto antibody-coated tumour cells. The trial stacked three steps of one chain. Cetuximab opsonizes EGFR-bearing colorectal cells and supplies the pro-phagocytic signal, evorpacept removes the CD47 brake so macrophages act on it, and pembrolizumab sustains the T cell response that follows. In humanized immune system patient-derived xenograft mice, the triplet slowed tumour growth and raised tumour-infiltrating CD8 positive T cells relative to its components.

Genetic support is weak. Open Targets scores CD47 against colorectal carcinoma at 0.109, with literature evidence at 0.898 the only contributing datatype and genetic association contributing nothing. The design choice that matters most is the inactive Fc, intended to avoid the red cell clearance seen with Fc-competent CD47 antibodies. Work in Cancer Cell in 2023 reported that the antitumour activity of anti-CD47 antibodies requires Fc to Fc gamma receptor interactions, a known tension at the centre of this molecule.

What actually happened

Nineteen patients enrolled and 16 received study drug. Three never started, two because the trial was halted and one who withdrew consent. Stage 1 treated nine at 15 mg/kg and three at 10 mg/kg, and stage 2 treated four at 15 mg/kg. One patient at 15 mg/kg in the run-in had a grade 5 dose-limiting toxicity of hemophagocytic lymphohistiocytosis, after which the protocol was amended to evaluate both dose levels.

Activity was minimal. One of 16 patients responded, a rate of 6.3 percent with a posted 95 percent interval of 0.2 to 30.2 percent, and that response lasted 13.2 months. Disease control was recorded in two of 16. Median progression-free survival was 2.3, 2.2 and 2.8 months across the three cohorts, and the publication reports 2.3 months overall with a median overall survival of 10.9 months.

Toxicity stopped the study. Serious adverse events occurred in 11 of 16 patients and eight of 16 died of any cause. The posted serious event table lists single occurrences of hepatic failure, pneumonitis, pneumonia, hypoxia, respiratory failure, stroke, thromboembolic event, serious anemia, hemophagocytic lymphohistiocytosis and cytokine release syndrome. The publication attributes one grade 5 hemophagocytic lymphohistiocytosis and one grade 5 cytokine release syndrome to treatment. The registered reason reads "Accrual was terminated due to safety concerns."

Failure mechanism, best guess

This is a safety signal, and specifically not the one the CD47 class is known for. Anemia, the on-target consequence of removing the red cell survival signal, was mild: three of nine patients at 15 mg/kg in stage 1 and one of four in stage 2 had non-serious anemia, with one serious case. The inactive Fc appears to have controlled red cell destruction and not the macrophage-driven cytokine amplification that killed two patients.

The likely chain runs forward from the intended mechanism rather than away from it. Cetuximab-coated tumour cells plus CD47 blockade produce a burst of phagocytosis and antigen presentation, pembrolizumab removes the brake on the T cells that respond, and the feedback loop between activated macrophages and activated T cells is the substrate for both fatal syndromes. The trial's correlative analyses identified innate and adaptive immune activation, the first half of that chain observed directly.

The doublet comparator sharpens the point. Magrolimab with cetuximab in NCT02953782 reached a 6.3 percent response rate and a 50.0 percent disease control rate in the KRAS wild-type refractory cohort, with a median overall survival of 9.5 months and no deaths attributed to treatment. The third agent bought no additional response and coincided with two fatal immune events.

How to prevent this next time

Sixteen treated patients and one response cannot support a Bayesian posterior or a power calculation. The interval around 6.3 percent runs from 0.2 to 30.2 percent, wide enough to contain both a dead regimen and a useful one, so the levers that apply are qualitative.

Base rate first. The magrolimab doublet had already reported 6.3 percent response in this setting before the trial read out. Treating that as the prior sets a low ceiling on what a third agent can add.

Biomarker enrichment second. Neither EGFR nor CD47 expression was required for entry. ALX Oncology has since presented CD47 overexpression as a predictive biomarker for evorpacept response in HER2 positive gastric cancer, announced in October 2025. The same logic was available here.

Red team the immune arithmetic third. Stacking an opsonizing antibody, a myeloid checkpoint blocker and a T cell checkpoint blocker is a bet on controlled amplification, and the two fatal syndromes here are the named failure modes of amplification that is not controlled. Pre-specified ferritin, soluble CD25, triglyceride and fibrinogen monitoring from cycle 1, with mandatory holding rules, would have made the first event a detected signal rather than a death.

The single highest leverage change would have been to run the triplet only after a randomized doublet comparison had shown that adding a T cell checkpoint blocker to CD47 blockade plus cetuximab improved response, rather than committing patients to three-agent immune amplification on the strength of a xenograft model.

What this means for similar programs

CD47 in refractory microsatellite stable metastatic colorectal cancer now carries a modelled risk score of 33 out of 100 in the yellow band, built from this single phase 2 safety termination, an Open Targets score of 0.109, and two known clinical programs against the target. The score is computed per target and disease pair, so the separate CD47 record in classic Hodgkin lymphoma stays at 27 and is not raised by this entry. A new CD47 program should state in advance which recorded failure mode it resolves.

The lesson for anyone building on CD47 and SIRPα is that solving the Fc problem does not solve the amplification problem. Immune activation syndromes come from a different part of the mechanism.

Open questions

Did the two fatal events cluster by dose in a way a larger run-in would have separated? Both occurred at 15 mg/kg, but with three patients at 10 mg/kg there is no basis for a dose-response claim.

Would CD47 expression have selected the single responder? The posted results do not report expression by patient, and the publication reports no CD47-high subgroup.

Sources

Related failure claims

Linked claims sharing target, indication, or failure mechanism.