Magrolimab in classic Hodgkin lymphoma: a responding trial closed when the CD47 program collapsed

What was tried

This Phase 2 study (NCT04788043) paired magrolimab with pembrolizumab in relapsed or refractory classic Hodgkin lymphoma. It was an investigator-led trial from Ranjana Advani at Stanford, with Merck as collaborator. The logic was to combine two checkpoints from different arms of immunity: pembrolizumab releasing the adaptive PD-1 brake, magrolimab releasing the innate macrophage brake. Eight patients were treated before the study was terminated. The posted reason has nothing to do with the eight patients. It reads: discontinuation of magrolimab development.

Magrolimab (Hu5F9-G4, CHEMBL4297985) is a first-in-class anti-CD47 IgG4 antibody. Pembrolizumab is already standard in relapsed Hodgkin lymphoma, which made it a rational backbone. The trial was a bet that adding innate immune engagement to an established adaptive checkpoint would deepen responses in a disease built around its immune microenvironment.

The biological hypothesis

CD47 is the dominant "don't eat me" signal on the cell surface. It binds SIRPalpha on macrophages and delivers an inhibitory message that blocks phagocytosis. Tumors overexpress CD47 to hide from innate clearance. Blocking the CD47-SIRPalpha axis lets macrophages recognize and engulf malignant cells, and it can prime downstream T cell responses (van Duijn et al, J Immunother Cancer 2022, Veillette et al, Nat Rev Cancer 2026).

Classic Hodgkin lymphoma is an unusually good setting for this idea on paper. The malignant Reed-Sternberg cells are rare, sitting inside a dense reactive infiltrate packed with tumor-associated macrophages. A therapy that redirects those macrophages against the tumor, layered on a PD-1 inhibitor that already works in this disease, is a coherent hypothesis. The genetics tell a weaker story. Open Targets scores the CD47 to classic Hodgkin lymphoma association at 0.08, with contributions from literature and clinical evidence and no genetic anchor. CD47 is a pan-tumor phagocytosis switch, not a Hodgkin-specific driver, and the association reflects that.

What actually happened

The small cohort actually responded. Among the eight treated patients, overall response was 6 of 8 and complete response was 3 of 8, per the posted results. Those are 75 percent and 38 percent, though with eight patients the Wilson 95 percent confidence intervals are wide, spanning 41 to 93 percent and 14 to 69 percent. There were two serious adverse events and one on-study death. The trial recorded 70 magrolimab-related adverse events across the cohort. Read narrowly, this was a combination with a response signal.

The trial died from the outside. On February 7, 2024, Gilead discontinued the Phase 3 ENHANCE-3 study of magrolimab with venetoclax and azacitidine in acute myeloid leukemia after an independent committee found futility and an increased risk of death driven by infections and respiratory failure. The FDA placed all magrolimab AML and MDS studies on full clinical hold. The higher-risk MDS study ENHANCE had already been stopped for futility in 2023, and the TP53-mutant AML program showed the same futility and mortality pattern (Daver et al, Blood 2025, Zeidner et al, Blood 2025). Gilead exited magrolimab in hematologic malignancies. Every trial riding on that asset, including this Hodgkin study, closed with it.

Failure mechanism, best guess

The Hodgkin trial did not fail on efficacy. The magrolimab program failed on the balance of benefit and harm in the diseases that mattered most to the sponsor. The likely driver is the class liability of CD47 blockade. Red blood cells express CD47, so blocking it triggers on-target anemia and drives a priming-and-maintenance dosing scheme to manage hemolysis. Stack that hematologic toxicity onto azacitidine and venetoclax in older AML patients and the infection and mortality signal that stopped ENHANCE-3 becomes predictable rather than surprising. The FAERS record for magrolimab is consistent, with 100 serious reports led by febrile neutropenia, neutropenia, and catheter-related and bloodstream infections. The molecule engaged its target. The therapeutic window in frontline myeloid disease was too narrow to survive a randomized comparison, and the portfolio decision that followed swept up the smaller combinations.

How to prevent this next time

Two quantitative disciplines would have protected the eight Hodgkin patients and the resources behind them. The first is base-rate humility about the combination environment. Development-wide, the probability of success from Phase 1 to approval is roughly 13.8 percent, and oncology sits below that average (Wong et al, Biostatistics 2019). A first-in-class innate checkpoint with an on-target anemia liability and no genetic anchor should carry a conservative prior, and satellite investigator combinations should sequence behind the pivotal randomized readout, not run in parallel with it.

The second is a Bayesian predictive-probability gate tied to the parent program:

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

Updating that integral with the ENHANCE and ENHANCE-2 futility interims would have shown the predictive probability of a positive ENHANCE-3 collapsing well before February 2024, and a portfolio governed by that number would have paused new combination starts rather than opening them.

The single highest leverage change would have been to gate all magrolimab combination trials on the predictive probability of the lead randomized ENHANCE readout, so that satellite studies never opened while the core efficacy and safety question was still unresolved.

What this means for similar programs

The surviving CD47 and SIRPalpha field should read this as a window problem, not a wrong-target verdict. Evorpacept, lemzoparlimab, letaplimab, and the SIRPalpha agents inherit the same on-target anemia and the same need for a genuine therapeutic margin. The isolated Hodgkin responses suggest the biology can do something in a macrophage-rich, immune-inflamed tumor. The lesson is to prove the window in a disease that tolerates it before betting a portfolio on frontline myeloid combinations.

Open questions

Was the Hodgkin response signal real or a small-sample artifact? Would a next-generation CD47 agent with less anemia have cleared the ENHANCE-3 safety bar? And is there a macrophage-rich, immune-inflamed indication where the CD47 window is wide enough to matter?

Sources

1. ClinicalTrials.gov, Study of Magrolimab and Pembrolizumab in Relapsed or Refractory Classic Hodgkin Lymphoma, NCT04788043, https://clinicaltrials.gov/study/NCT04788043.

2. Gilead Sciences, Statement on Discontinuation of Phase 3 ENHANCE-3 Study in AML, February 2024, https://www.gilead.com/company/company-statements/2024/gilead-statement-on-discontinuation-of-phase-3-enhance-3-study-in-aml.

3. Daver N, et al, The ENHANCE-3 study, venetoclax and azacitidine plus magrolimab or placebo for untreated AML, Blood, 2025,.

4. Zeidner JF, et al, Magrolimab plus azacitidine versus physician's choice for untreated TP53-mutated acute myeloid leukemia, Blood, 2025,.

5. van Duijn A, Van der Burg SH, Scheeren FA, CD47/SIRPa axis, bridging innate and adaptive immunity, Journal for ImmunoTherapy of Cancer, 2022,.

6. Veillette A, et al, Targeting phagocytosis checkpoints for cancer immunotherapy, Nature Reviews Cancer, 2026,.

7. Jiang C, et al, Targeting the CD47/SIRPa pathway in malignancies, recent progress, difficulties and future perspectives, Frontiers in Oncology, 2024,.

8. Wong CH, Siah KW, Lo AW, Estimation of clinical trial success rates and related parameters, Biostatistics, 2019,.

9. Open Targets Platform, CD47 to classic Hodgkin lymphoma association, https://platform.opentargets.org/.

10. openFDA Drug Adverse Event API (magrolimab), https://api.fda.gov/drug/event.json.

11. ChEMBL, Magrolimab (CHEMBL4297985), https://www.ebi.ac.uk/chembl/.