The First Wave of CD47 Immunotherapy: Biological Promise, Clinical Failure, and the Path Forward

Biological Rationale: A Compelling but Complex Target

The CD47–signal regulatory protein alpha (SIRPα) axis occupies a central position in innate tumor immune evasion. CD47, a transmembrane glycoprotein overexpressed across the spectrum of hematologic and solid malignancies, binds SIRPα on macrophages and dendritic cells to deliver a canonical "don't eat me" signal, actively suppressing phagocytosis-mediated tumor clearance. In healthy tissues, this interaction protects normal cells from inappropriate immune destruction; in cancer, it is co-opted to create a permissive microenvironment for malignant proliferation and metastasis 1.

The theoretical elegance of blocking this axis was substantial. Anti-CD47 antibodies or SIRPα-Fc (fragment crystallizable) fusion proteins were predicted to restore macrophage-mediated phagocytosis while simultaneously enhancing antigen cross-presentation to cytotoxic T lymphocytes—thereby bridging innate and adaptive antitumor immunity. Preclinical models between 2015 and 2020 validated this dual mechanism convincingly, generating robust investor enthusiasm and spawning a rich pipeline of clinical candidates. By 2020, multiple phase 1b and early phase 2 trials were enrolling patients across myelodysplastic syndromes (MDS), acute myeloid leukemia (AML), non-Hodgkin lymphoma (NHL), and solid tumors 17. Within four years, this promise had substantially collapsed.

The First-Wave Clinical Record: Failures and Lessons

Magrolimab: The Defining Setback

No agent better encapsulates the arc of first-wave CD47 therapy than magrolimab (Gilead Sciences), a humanized anti-CD47 monoclonal antibody (mAb). Early phase 1 results were genuinely encouraging: in higher-risk MDS combined with azacitidine, overall response rates (ORRs) reached 75% and complete remission (CR) rates 33%, substantially exceeding historical azacitidine monotherapy benchmarks. In TP53-mutated MDS and AML, CR rates of 38% and 59%, respectively, attracted widespread attention 1.

These early signals did not translate into pivotal trial success. In February 2024, an interim analysis of the phase 3 ENHANCE-3 trial—evaluating magrolimab combined with venetoclax and azacitidine in newly diagnosed, treatment-ineligible AML (n=378)—demonstrated futility and a numerically increased risk of death in the experimental arm (hazard ratio 1.173; 95% confidence interval [CI] 0.819–1.679). Treatment-related mortality was 18.5% in the magrolimab arm versus 10.9% in controls, driven predominantly by infections and respiratory failure 7. The U.S. Food and Drug Administration (FDA) subsequently placed a full clinical hold on all magrolimab studies in AML and MDS, citing consistent patterns of increased mortality across three pivotal phase 3 trials—ENHANCE (higher-risk MDS, HR 1.254; 95% CI 0.980–1.605), ENHANCE-2 (TP53-mutated AML, HR 1.191; 95% CI 0.744–1.906), and ENHANCE-3 7. Gilead simultaneously paused enrollment in solid tumor programs and ultimately announced discontinuation of magrolimab development in hematologic malignancies 78.

Evorpacept and Maplirpacept

Evorpacept (ALX148, ALX Oncology), an engineered SIRPα–Fc fusion protein designed with an inactive Fc domain to minimize erythrocyte binding, showed acceptable tolerability in early AML and MDS studies. The ASPEN-05 phase 1/2 trial reported full CD47 target occupancy and signals of activity in small cohorts of newly diagnosed AML (100% response rate, n=3 evaluable) and relapsed/refractory (R/R) AML (4/10 responses) 5. The ASPEN-02 phase 1/2 trial in MDS reported an ORR of 36.5% in newly diagnosed higher-risk patients and 15.4% in R/R patients 5. These results were modest relative to early expectations, and ALX Oncology terminated evorpacept development in MDS and AML in August 2023 citing poor efficacy—a paradox that revealed a critical mechanistic trade-off: the Fc-silent design that reduced hematologic toxicity also attenuated antibody-dependent cellular phagocytosis (ADCP) and antibody-dependent cell-mediated cytotoxicity (ADCC), weakening antitumor potency 1.

Maplirpacept (TTI-622/PF-07901801, Pfizer/Trillium Therapeutics), a SIRPα-Fc fusion protein, similarly encountered development challenges. Pfizer terminated a phase 2 trial in diffuse large B-cell lymphoma (DLBCL) after enrolling only six patients since August 2023, citing recruitment difficulties 910.

Dissecting the Failure Modes

The convergence of failures across mechanistically distinct agents points to structural, rather than agent-specific, liabilities. Four interacting failure modes account for most of the clinical disappointment.

1. On-Target, Off-Tumor Hematologic Toxicity

Erythrocytes constitutively express high CD47 levels, protecting senescent red blood cells (RBCs) from premature phagocytic clearance. Anti-CD47 antibodies disrupt this protection, triggering accelerated macrophage-mediated RBC destruction and hemolytic anemia. In the magrolimab plus rituximab plus gemcitabine plus oxaliplatin (R-GemOx) trial in R/R DLBCL, 60.6% of patients developed grade ≥3 anemia 15. Developers attempted to mitigate this through priming-dose strategies—administering a lower initial antibody dose to selectively deplete senescent RBCs with high prophagocytic calreticulin surface expression, allowing compensatory reticulocytosis before therapeutic doses. While this approach reduced the severity of anemia, it added dosing complexity and did not eliminate the underlying liability 1.

2. Systemic Immune Dysregulation and Infection

Critically, the magrolimab phase 3 failures were not primarily driven by anemia but by infection and respiratory failure. Broad CD47 blockade in already immunocompromised AML/MDS patients may dysregulate myeloid homeostasis, impairing protective macrophage functions and predisposing patients to bacterial and fungal infections. This finding reframes the toxicity concern from a manageable hematologic problem to a potentially fatal systemic immune perturbation 7.

3. Fc Optimization Paradox

Preclinical work established that anti-CD47 antibodies require Fc–FcγR (Fc gamma receptor) interactions for optimal antitumor activity through ADCP and ADCC. Yet first-wave agents were predominantly designed on immunoglobulin G4 (IgG4) scaffolds—chosen specifically to minimize Fc-dependent effector functions and reduce off-tumor toxicity. This design choice inadvertently weakened antitumor potency, creating an optimization trap: enhancing Fc function improved efficacy but increased toxicity; silencing Fc reduced toxicity but compromised efficacy 13.

4. Combination Dependency Without Biomarker Guidance

Anti-CD47 monotherapy consistently produced insufficient single-agent activity; meaningful responses required combinations. Azacitidine synergized with magrolimab in MDS and AML by inducing calreticulin surface expression on tumor cells, creating an "eat me" signal that complemented CD47 blockade. Triple regimens incorporating venetoclax achieved higher response rates than doublets in select subgroups 1. Yet not all combinations worked: magrolimab plus rituximab plus acalabrutinib in R/R DLBCL yielded only 28% ORR and was terminated early 1. Without validated predictive biomarkers, the field pursued broad and underpowered combinatorial trials, generating inconsistent results and premature discontinuations.

Indication-Level Lessons

The comparison across indications reveals a consistent pattern: NHL—particularly B-cell NHL where CD47 blockade can be anchored by co-administered anti-CD20 antibodies—appears more amenable to clinical benefit than AML/MDS, where immunosuppression, disease biology, and macrophage dysfunction converge to limit therapeutic response 511. Solid tumors remain largely resistant, partly due to complex immunosuppressive tumor microenvironments (TMEs), M2-polarized macrophages, and the absence of the calreticulin-inducing synergy provided by hypomethylating agents 4.

Emerging Approaches: What Comes Next

The failure of first-wave agents has catalyzed a transition toward more sophisticated design strategies and indication-specific development.

Lemzoparlimab and the RBC-Sparing Paradigm

Lemzoparlimab (I-Mab), engineered via human-derived bacteriophage technology to achieve reduced erythrocyte binding through glycosylation near its binding epitope, represents the most clinically advanced attempt to decouple antitumor activity from RBC toxicity. In a phase 2a dataset of newly diagnosed higher-risk MDS patients treated with lemzoparlimab plus azacitidine (n=29 treated ≥4 months), ORR reached 86.2% with a CR rate of 31%; 56% of patients achieving CR also attained measurable residual disease (MRD) negativity 15. While these data are promising, they derive from a single-arm Chinese dataset with limited follow-up maturity, and cross-trial comparison with magrolimab's early phase 1 signals warrants caution.

Bispecific Formats and the IBI322 Experience

IBI322 (Innovent Biologics), a CD47/PD-L1 recombinant bispecific antibody, achieved an ORR of 47.8% and a disease control rate (DCR) of 91.3% in 23 efficacy-evaluable patients with anti-PD-1/PD-L1-resistant classic Hodgkin lymphoma (cHL), including a 57.1% ORR in patients with primary PD-1/PD-L1 resistance 12. By restricting phagocytic activation to PD-L1-expressing tumor cells, this format addresses both the antigen-sink problem inherent to nonselective CD47 blockade and the need for adaptive immune co-engagement. A rapidly expanding pipeline of PD-L1/CD47 and tumor-antigen/CD47 bispecifics—predominantly in China but increasingly in the United States—reflects field-wide recognition that contextual tumor targeting is essential for therapeutic advance 6.

Evorpacept in NHL: A Remaining Positive Signal

Notwithstanding its failure in AML/MDS, evorpacept retained a meaningful efficacy signal in B-cell NHL. In a phase 1/2 investigator-sponsored trial combining evorpacept with rituximab and lenalidomide (R2) in R/R B-cell NHL (n=20), the ORR was 90% and CR rate 80% (16/20 patients), with a two-year progression-free survival (PFS) of 69% and overall survival (OS) of 84% at a median follow-up of 28 months 511. No dose-limiting toxicities were observed. This result, interpreted against a historical CR rate of approximately 34% for R2 alone, underscores that CD47 blockade anchored to a tumor-engaging antibody partner in a relatively immune-competent disease setting can generate clinically meaningful outcomes 11.

Conclusion: A Valid Target, A Failed Strategy

The first wave of CD47 immunotherapy failed not because the target is invalid but because the initial clinical translation strategy was flawed. Broad CD47 blockade with nonselective antibodies in highly immunocompromised hematologic malignancy populations, combined with inadequate Fc optimization, absence of predictive biomarkers, and insufficient understanding of combination synergy requirements, created conditions for consistent failure. Magrolimab's phase 3 discontinuation and FDA clinical hold represent the most consequential regulatory signal in the field's history—but they should be read as indictments of first-generation design choices, not of the CD47–SIRPα axis itself.

The path forward is visible in the current pipeline: tumor-selective bispecific formats that restrict phagocytic activation to antigen-defined malignant cells; RBC-sparing antibody engineering to decouple efficacy from hematologic liability; SIRPα-directed alternatives that approach the axis from the receptor rather than the ligand; and rational, biomarker-guided combination strategies that exploit the calreticulin-inducing properties of hypomethylating agents or the phagocytic anchoring provided by co-administered therapeutic antibodies. Whether this second wave will achieve durable clinical benefit across hematologic and solid tumor indications remains to be demonstrated in randomized trials. What is clear is that the CD47 checkpoint continues to represent a biologically compelling innate immune target—one that demands more sophisticated clinical translation than the first wave provided 1567.