Gilead CD47 Trials: Why Magrolimab Failed in Phase 3

Gilead Sciences made one of the most expensive bets in immuno-oncology when it acquired Forty Seven, Inc. in 2020, paying roughly $4.9 billion to gain control of magrolimab, a first-in-class antibody designed to block a protein called CD47 on cancer cells. The idea was elegant: strip away the molecular disguise that tumors use to hide from the immune system. Early-phase trials generated genuine excitement, particularly in hard-to-treat blood cancers. But every pivotal Phase 3 trial of magrolimab failed to show a survival benefit, and some showed increased risk of death. Gilead ultimately abandoned magrolimab for hematologic cancers entirely, leaving the broader field of CD47-targeting therapy at a crossroads.

What CD47 Does and Why Oncology Cared

CD47 is a protein found on the surface of virtually every cell in the body. Its primary job is to send a “don’t eat me” signal to immune cells called macrophages. When macrophages encounter a cell, they check for surface markers. CD47 binds to a receptor on macrophages called SIRPα, and that interaction tells the macrophage to leave the cell alone and move on.¹ This is how healthy red blood cells, platelets, and other normal tissues avoid being destroyed by the body’s own cleanup crew.

Cancer cells exploit this system. Many tumors dramatically increase CD47 on their surface, essentially waving a bigger and louder “don’t eat me” flag than normal cells do.² The overexpression has been documented across a wide range of cancers, from blood malignancies to solid tumors including mesothelioma and salivary gland cancers.³ The more CD47 a tumor displays, the more effectively it hides in plain sight.

The effects of CD47-SIRPα signaling go beyond simply blocking macrophages from engulfing tumor cells. Research has shown that this pathway also interferes with the maturation of dendritic cells, impairs antigen presentation, sustains the function of immune-suppressive myeloid cells, and prevents neutrophils from migrating into tumors.⁴ In other words, CD47 doesn’t just protect individual cancer cells from being eaten. It helps create a broader environment in which the immune system struggles to mount an effective response at all.

Magrolimab and the Logic of Combination Therapy

Magrolimab, originally known as 5F9, was designed as a humanized monoclonal antibody that binds to CD47 on cancer cells and blocks the signal to SIRPα.⁵ With that signal blocked, macrophages can recognize and engulf the cancer cells. The theory was straightforward: remove the disguise, and the immune system does the rest.

Researchers quickly recognized that magrolimab would likely work best in combination with other drugs rather than alone. The most prominent pairing was with azacitidine, a chemotherapy drug commonly used in blood cancers like myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML). Azacitidine increases the expression of “eat me” signals on cancer cells, which are the opposite of CD47’s “don’t eat me” message. Blocking CD47 while simultaneously boosting “eat me” signals was expected to create a synergistic one-two punch.⁶

Early-phase results supported the optimism. In a Phase 1b study of patients with TP53-mutated AML, a particularly aggressive and treatment-resistant form of the disease, magrolimab plus azacitidine produced an objective response rate of about 49%, with a third of patients achieving complete remission. Median overall survival reached 10.8 months in that group.⁷ For a population with notoriously few options and typical survival measured in months, those numbers looked promising. Gilead invested heavily in moving to Phase 3.

The Phase 3 Collapse

Gilead launched three large, randomized Phase 3 trials to confirm what the earlier studies suggested. All three failed. The disappointment was comprehensive.

The ENHANCE trial tested magrolimab plus azacitidine against azacitidine alone in patients with untreated higher-risk MDS. The complete remission rate was actually slightly lower in the magrolimab group (about 21% versus 24% with azacitidine alone), and median overall survival was shorter: 15.9 months versus 18.6 months. Neither primary endpoint was met. Worse, patients who received magrolimab experienced more frequent severe side effects.⁸

The ENHANCE-2 trial focused on TP53-mutated AML, the same population where earlier data had been most encouraging. In the main comparison arm (patients ineligible for intensive chemotherapy), median overall survival was 4.4 months with magrolimab plus azacitidine versus 6.6 months with venetoclax plus azacitidine. The trial met its pre-specified definition for futility at interim analysis and was terminated.⁹ In the smaller intensive-therapy arm, the results trended in the same wrong direction: 7.3 months survival with magrolimab versus 11.1 months with standard chemotherapy.

The ENHANCE-3 trial, which evaluated a three-drug combination (magrolimab plus azacitidine plus venetoclax) in AML, was also stopped after an independent data monitoring committee found both futility and an increased risk of death, primarily from infections and respiratory failure. Following these results, the FDA placed all studies of magrolimab in AML and MDS on full clinical hold. Gilead announced it would discontinue the drug’s development in hematologic cancers.

Why Did the Promising Early Data Not Hold Up?

The gap between Phase 1b excitement and Phase 3 failure is one of the most discussed issues in the CD47 field, and there are several possible explanations. Single-arm Phase 1b trials are notoriously unreliable for estimating survival in blood cancers, because patient selection, lead-time bias, and the lack of a comparison group can all inflate apparent benefits. The response rates that looked encouraging in small studies may have partially reflected favorable patient characteristics rather than a true drug effect.

There’s also the question of whether blocking CD47 on cancer cells can be separated cleanly from blocking it everywhere else. CD47 is present on red blood cells, platelets, and many healthy tissues. An antibody like magrolimab doesn’t just block the signal on tumors. It blocks it system-wide, which contributes to significant toxicity, particularly anemia. While the trial designs included a priming dose strategy (giving a low initial dose to clear older red blood cells before ramping up to full dose), this didn’t eliminate the problem. Grade 3 or higher anemia occurred in roughly a quarter of magrolimab-treated patients across the Phase 3 trials.¹⁰

The mechanism behind the anemia turns out to be more complex than simple red blood cell destruction. Research has shown that anti-CD47 antibodies, despite being an IgG4 subtype (a class of antibody not usually associated with cell destruction), can trigger macrophages to eat red blood cells through specific Fc receptor interactions. This finding was somewhat surprising and adds a wrinkle to future antibody design.¹¹

The Transfusion Problem

An underappreciated headache with anti-CD47 drugs involves blood transfusions, which patients with blood cancers frequently need. Because CD47 sits on the surface of all red blood cells, anti-CD47 antibodies in a patient’s blood will bind to donor red blood cells during standard pretransfusion compatibility testing, causing false-positive results across the board.¹² This interference makes it difficult for blood banks to determine whether a patient has other clinically significant antibodies that would make a particular blood unit unsafe to transfuse.

Laboratory testing has confirmed that all anti-CD47 drugs tested interfere significantly with standard compatibility assays at clinically relevant concentrations, affecting multiple testing platforms.¹³ Researchers are developing workarounds, including antigen-masking techniques that can prevent the drug from binding during testing, but these require additional time and expertise at the blood bank. For patients in crisis who need blood urgently, this delay is not trivial. Any future CD47-targeting drug that reaches the clinic will need to contend with this issue.

Magrolimab in Solid Tumors

Gilead’s abandonment of magrolimab was specific to hematologic cancers. There was also some exploration in solid tumors, though the results were modest. A Phase 1b/2 study combined magrolimab with cetuximab (an established antibody used in colorectal cancer) in patients with KRAS wild-type colorectal cancer and other solid tumors. Only two out of 32 evaluable colorectal cancer patients achieved a partial response, putting the overall response rate at about 6%. The disease control rate (meaning the cancer at least stopped growing) was 50%, and the two responses that did occur lasted a median of about 10 months.¹⁴ These are not practice-changing numbers, but the durable responses in a couple of heavily pretreated patients hint that CD47 blockade might benefit a narrow subset of solid tumor patients. Identifying who those patients are remains an open question.

Next-Generation CD47-Targeting Approaches

The failure of magrolimab has not killed interest in the CD47 pathway. If anything, it has redirected the field toward strategies designed to avoid the problems that sank Gilead’s program. The core challenge is selectivity: how do you block CD47 on tumor cells without causing widespread effects on healthy tissues?

One approach is bispecific antibodies, molecules engineered to bind two targets at once. For instance, researchers have developed antibodies that target both CD47 and CD20 (a marker on certain lymphoma cells). These bispecifics have reduced affinity for CD47 on its own, meaning they don’t grip red blood cells or platelets very tightly. But when they encounter a cell displaying both CD47 and CD20, they lock on and trigger phagocytosis.¹⁵ The selectivity comes from requiring two simultaneous binding events, which tumor cells provide and healthy red blood cells do not.

Another strategy is to target SIRPα, the receptor on macrophages, instead of CD47 on the tumor. This flips the equation: rather than blocking the signal where it originates, you block it where it’s received. A pan-allelic anti-SIRPα antibody called hAB21 has been developed that binds to multiple variants of human SIRPα with high affinity and prevents CD47 from engaging it.¹⁶ Because SIRPα is mostly restricted to myeloid immune cells rather than distributed everywhere like CD47, this approach could in theory limit off-target effects on red blood cells. SIRPα-Fc fusion proteins, which are engineered molecules that act as decoys to soak up CD47 signals, are also being explored.¹⁷

Then there are small molecules. Antibodies are large proteins that must be given intravenously and cannot easily penetrate solid tumors. Small-molecule inhibitors of the CD47-SIRPα interaction could potentially be taken orally and might reach parts of tumors that antibodies struggle to access.¹⁸ One particularly intriguing small molecule, called SMC18, was designed to block both the CD47-SIRPα interaction and the PD-1/PD-L1 checkpoint at the same time, hitting two immune evasion pathways with a single drug. In laboratory experiments, it restored macrophage phagocytosis and activated T cells simultaneously.¹⁹ These are still in early development, but they represent a fundamentally different design philosophy from the brute-force approach of a systemically distributed antibody like magrolimab.

CD47 Beyond Cancer

One of the more surprising chapters in CD47 research involves atherosclerosis, the buildup of fatty plaques in arteries that drives heart attacks and strokes. It turns out that diseased vascular tissue also upregulates CD47, and this shields the damaged cells from being cleared by macrophages through a process called efferocytosis. In mouse models, blocking CD47 with antibodies reversed this defective cleanup, restored normal clearance of diseased tissue, and reduced atherosclerosis across multiple experimental setups.²⁰

Follow-up work has added more detail to this picture. In mice lacking apoE (a standard model for atherosclerosis), anti-CD47 treatment reduced lesion size by about 25%, decreased the dead necrotic core within plaques by roughly 35%, and nearly halved the ratio of free-floating dead cells to macrophage-engulfed dead cells, indicating that macrophages were doing their cleanup job much more efficiently.²¹ These are animal studies, and no one is testing anti-CD47 antibodies for heart disease in humans yet. But the findings demonstrate that CD47’s role as an immune-evasion signal extends well beyond cancer, and future therapeutic applications of the pathway might eventually reach into cardiovascular disease.

CD47 Expression as a Diagnostic Marker

Separate from treatment, researchers are investigating whether measuring CD47 levels on tumor cells and the immune cells surrounding them might be useful for diagnosis or treatment selection. In salivary gland cancers, for instance, CD47 expression varies substantially across tumor subtypes. Mucoepidermoid carcinomas showed the lowest infiltration of CD47-positive tumor cells (averaging about 11%), while other subtypes displayed much higher levels. Interestingly, in low-grade tumors, the immune cells around the tumor’s edges expressed more CD47 than the tumor cells themselves, a pattern that was less pronounced in the tumor’s center.²²

These expression patterns matter because they suggest that not all tumors would respond to CD47-targeting therapy in the same way. A tumor where CD47 is low might not be hiding behind that shield much at all, while one saturated with CD47 might be a strong candidate. This kind of biomarker profiling could eventually help oncologists select patients most likely to benefit from next-generation CD47-targeting drugs, avoiding the one-size-fits-all approach that arguably contributed to magrolimab’s disappointing performance in unselected populations.