Development of a Novel Bifunctional Anti-CD47 Fusion Protein with Improved Efficacy and a Favorable Safety Profile Open Access

Cancer cells evade the immune system by overexpressing receptors that maintain immunotolerance, allowing them to escape detection and clearance by innate immune cells (1, 2). A key immune checkpoint involves the interaction between CD47 and SIRPα (3). CD47, a transmembrane glycoprotein expressed on all cell types, including red blood cells (RBC), interacts with SIRPα on macrophages to trigger an inhibitory “don’t eat me” signal preventing phagocytosis (4, 5). In cancer cells, this molecular pathway is frequently exploited by overexpressing CD47, avoiding phagocytosis by macrophages, and is linked to a poor clinical prognosis (6). Consequently, the clinical use of anti-CD47 mAbs aims to block the interaction between CD47 and SIRPα, thereby enhancing immune-mediated cancer cell elimination (3, 5). Despite progress in CD47-targeted therapy (7), most anti-CD47 mAbs cause hematotoxicity, particularly in RBCs, and other poorly understood adverse effects (8, 9). Additionally, anti-CD47 mAbs show limited efficacy in advanced cancer and as monotherapy agents. Thus, blocking the CD47–SIRPα interaction alone is insufficient to achieve effective therapeutic outcomes (10). Therefore, it is desirable to develop safe anti-CD47 proteins that produce clinically relevant responses.

Fortunately, CD47-mediated signaling extends beyond immune evasion and its role as a death receptor adds complexity to its biological significance (11, 12). Certain anti-CD47 mAbs can induce a form of programmed cell death in transformed cells, revealing their underexplored potential (11–13). However, these anti-CD47 mAbs have not been further developed because of their limited functionality or target side effects (10). Notably, the clinically tested anti-CD47 mAbs failed to induce CD47-mediated cell death (7). Combining this direct agonistic effect with reduced side effects represents a promising area of research and therapeutic innovation.

Here, we present the development and functionality of CO-005, a humanized fusion protein derived from the chimeric IgG4 anti-CD47 mAb CO-001. CO-005 exhibits bifunctional activity by promoting potent programmed cancer cell death (PCCD) and enhanced cancer cell phagocytosis without affecting RBCs and other tested normal cells. This advancement suggests potential for more effective and safer CD47-mediated therapies.

The murine precursor of CO-001 was generated against a human T-cell line, as previously described (14), and sequenced by isolating total RNA from the hybridoma cells using the FastPure Cell/Tissue Total RNA Isolation Kit (Vazyme, cat #RC112) according to the manufacturer’s instructions. The variable domains were cloned into human IgG expression vectors to create a chimeric mouse–human CO-001 with a kappa light chain and an IgG4 fragment crystallizable (Fc) by GenScript. The single-chain variable fragment (scFv) of humanized CO-001 (CO-004) and the fusion protein (CO-005) were produced at ATUM. The CO-005 fusion protein featured a bivalent scFv with an IgG4 Fc backbone after removing the constant light and heavy regions. The amino acid sequence of the various CO-001 molecules are described in our patents (15) as follows: Chimeric CO-001 (SEQ ID nos 23 and 24 - Table C), humanized CO-002 (SEQ ID nos 39 and 44 - Table E), and CO-005 (SEQ ID no 53).

The analogues of AO-176, developed by Arch Oncology, and magrolimab/Hu5F9-G4, developed by Gilead Sciences/Forty Seven were produced by GenScript based on the published sequences (13, 16). Human IgG4κ (cat #HG4K, RRID: AB_3094558) and IgG2aκ (cat #HG2K, RRID: AB_3094558) isotype controls were obtained from Sino Biological.

All cell lines were purchased from the ATCC. Jurkat (clone E6-1, RRID: CVCL_0367), MOLT-4 (RRID: CVCL_0013), CCRF-CEM (RRID: CVCL_0207), Reh (RRID: CVCL_1650), and Raji (RRID: CVCL_0511) were cultured in RPMI-1640 medium (Euroclone, cat #ECB2000L-12) and HL-60 (RRID: CVCL_0002) and KG-1a (RRID: CVCL_1824) were cultured in Iscove's modified Dulbecco's medium (Biosera, cat #LM-I1091/500). The murine macrophage cell line RAW264.7 (RRID: CVCL_0493) was cultured in DMEM (Euroclone, cat #ECM0095L-12). Media were supplemented with 10% (v/v) FBS (Thermo Fisher Scientific, Cat# A3160802) and 1% (v/v) penicillin and streptomycin (Thermo Fisher Scientific, cat #15140122), except for HL-60 cells that had 20% (v/v) FBS added. Cell cultures were maintained at 37°C in a humidified atmosphere with 95% air and 5% CO₂. Cell lines were routinely tested for Mycoplasma contamination and authenticated by flow cytometry.

RBC isolation was performed by collecting human whole blood into heparin-coated tubes (approximately 5 mL) which was then diluted in wash buffer [0.05% (w/v) BSA and 1 mmol/L EDTA in PBS] to a total volume of 50 mL, followed by centrifugation at 1800 × g for 10 minutes and resuspension in wash buffer to a total volume of 50 mL. Following that, the RBC pellet was resuspended in PBS to create a 2% (v/v) RBC solution.

Peripheral blood mononuclear cells (PBMC) were isolated using density gradient centrifugation with Lymphoprep (STEMCELL Technologies, cat #07801) according to the manufacturer’s instructions. Briefly, the peripheral blood from healthy human donors was diluted 1:1 in RPMI supplemented with 2% (v/v) FBS and 1% (v/v) penicillin/streptomycin, followed by gently pipetting the solution over a layer of Lymphoprep. After centrifugation (800 × g for 20 minutes, deceleration 0), the PBMC layer was retained for downstream experiments.

Blood collection for primary cell isolation was performed under written informed consent by healthy human donors, and the study was approved by the Regional Committee for Medical and Health Research Ethics in the South-Eastern Norway Regional Health Authority under REK ID number 635222.

PBS-washed Jurkat cells were seeded into a round-bottom 96-well plate (5 × 10⁵ cells/mL). The cells were incubated with 100 μg/mL of anti-CD47 mAbs CO-001, B6H12 (Thermo Fisher Scientific, cat #16-0479-85, RRID: AB_1234977), 2D3 (Thermo Fisher Scientific, cat #14-0478-82, RRID: AB_837150), or the isotype control for 1 hour on ice with slight agitation. Next, the cells were washed twice with PBS and resuspended in 10 μg/mL of FITC-conjugated SIRPα (Sino Biological, cat #30014-H08H) for an additional 30 minutes on ice. Each sample was prepared in duplicates. The cells were then washed twice in cold PBS and kept on ice until analysis by flow cytometry using a NovoCyte (Agilent Technologies) equipped with three lasers (405, 488, and 605 nm) and 13 detection channels. Excitation wavelength for FITC was 488 nm and detection wavelength was 530/30 nm.

Epitope mapping was performed by Deeptope using deep mutational scanning as described previously (17).

RAW264.7 mouse macrophage cells were seeded in six-well plates at 4 × 10⁵ cells/well and allowed to adhere overnight. The RAW264.7 cells were rinsed once with DMEM supplemented with 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin and stained with 40 nmol/L Vybrant DiO Cell-Labeling Solution (Thermo Fisher Scientific, cat #V22886) diluted in DMEM for 20 minutes at 37°C. After staining, the cells were rinsed three times with DMEM. Target cancer cells were collected by centrifugation and stained with 1 μL/mL CellTrace Violet Cell Proliferation Dye according to the manufacturer’s protocol (Thermo Fisher Scientific, cat #C34557). Target cells were added to the RAW264.7 cell culture plates and treated with the indicated antibodies for 2 hours in a standard cell incubator, followed by two washes in PBS containing 1 mmol/L EDTA. The cells were analyzed by flow cytometry on a MACSQuant X (Myltenyi Biotec) equipped with three lasers (405, 488, and 640 nm) and eight detection channels. The percentage of cells staining positive for DiO and CellTrace represented phagocytosed target cells. Excitation wavelength for DiO was 488 nm and detection wavelength was 525/50 nm, whereas excitation wavelength for CellTrace Violet was 405 nm and detection wavelength was 450/50 nm.

For the phagocytosis assay utilizing human-derived effector cells, freshly isolated PBMCs were plated in RPMI medium supplemented with 10% (v/v) FBS, 1% (v/v) penicillin/streptomycin, and 100 ng/mL human macrophage colony-stimulating factor (Sino Biological, cat #11792-HNAH) and distributed into six-well plates. The plated cells were incubated for 3 to 4 days in a standard cell incubator allowing monocytes to adhere to the plate. Floating cells were removed, fresh medium containing macrophage colony-stimulating factor was added, and the plates were further incubated for another 3 to 4 days to facilitate monocyte differentiation into macrophages. The same staining procedure and experimental setup as described for the murine macrophages were employed.

Cells were diluted to 5 × 10⁵ cells/mL in supplemented medium and seeded into 24-well plates at 1 mL per well. They were then incubated with various concentrations of CO-candidates or IgG4 isotype control for different durations (30 minutes, 1 hour, and 3 hours) under standard cell culture conditions. In some experiments, anti-CD47 mAbs, AO-176, B6H12, and magrolimab (Hu5F9-G4) were used at the indicated concentrations. Following incubation, the cells were harvested and stained with annexin V eFlour 450 and 7-aminoactinomycin D (7-AAD) according to the manufacturer’s protocol (Thermo Fisher Scientific, cat #88-8006-74). The cells were analyzed immediately by flow cytometry on the MACSQuant X. Excitation wavelength for annexin V was 405 nm and detection wavelength for annexin V was 450/50 nm, whereas 7-AAD was excited at 488 nm and detected at 655 to 730 nm.

PBS-washed cancer cells or normal cells were resuspended in 10% (v/v) FBS and 1% (w/v) sodium azide in PBS and incubated with 2.5 μg/mL Human BD Fc Block (BD Biosciences, cat #564220) for 10 minutes before being plated into a round-bottom 96-well plate (2.5 × 10⁵ cells per well). The cells were then incubated with increasing concentrations of anti-CD47 mAbs or appropriate isotype controls, as indicated, for 1 hour on ice with slight agitation. All samples were prepared in duplicate. Secondary staining was performed using 5 μg/mL of PE-conjugated goat anti–human IgG Fc secondary antibody (Thermo Fisher Scientific, cat #12-4998-82, RRID: AB_465926) on ice for 30 minutes. The scFv molecule (CO-004) was conjugated with FITC using the Lightning-Link FITC conjugation kit from Abcam (cat #Ab188285). The cells were then washed three times with cold washing buffer and analyzed by flow cytometry on the NovoCyte. Excitation wavelength for PE was 488 nm and detection wavelength was 572/28 nm.

Anti-CD47 mAbs or isotype controls were added to round-bottom 96-well plates at increasing concentrations. Next, a solution of 2% (v/v) freshly isolated RBCs was added to each well and incubated for 30 to 60 minutes until the cells settled at the bottom of the well. A diffuse hazy pattern indicates hemagglutination, whereas a small punctate circle indicates no hemagglutination (13).

The affinity of antibodies to the extracellular domain of human recombinant CD47 was determined by surface plasmon resonance on a Biacore S200 system at 25°C and PBS buffer supplemented with 0.05% v/v Tween 20. For CO-001 and CO-005, recombinant CD47 was immobilized in 10 mmol/L sodium acetate buffer pH 5.0 to a level of 40 to 60 resonance units onto a Series S CM5 chip (Cytiva, cat #29149603) using the amine coupling reaction kit (Cytiva, cat #BR100050). For CO-004, recombinant CD47 with a biotin tag was immobilized in 10 mmol/L sodium acetate buffer pH 5.0 to a level of 110 to 120 resonance units onto a Series S Streptavidine SA chip (Cytiva, cat #BR100531). CO-001, CO-004, and CO-005 were run in single-cycle mode for a nine-point 1:1 dilution series with increasing concentrations from 39 pmol/L to 10 nmol/L, with a flow rate of 30 μL/minute. The process of molecular binding (time, 120 seconds) and dissociation (1,800 seconds) was recorded, and the kinetics and affinity data were analyzed with a 1:1 binding model. For 1-minute regeneration between individual runs, 10 mmol/L glycine buffer with pH 2 was used.

Immune cell phenotyping was performed on PBMCs isolated from healthy human donors. All antibodies and reagents used for the phenotyping experiments are listed in Supplementary Table S1. Freshly isolated cells were washed, resuspended in 10% (v/v) FBS and 1% (w/v) sodium azide in PBS, and plated in a 96-well plate. The cells were stained with LIVE/DEAD fixable dye and subsequently surface-stained with cell type–specific antibodies (Supplementary Table S1). Additionally, a fluorescence minus one control was prepared for each cell marker.

For compensation, one test of the conjugated primary antibody was incubated with one drop of UltraComp eBeads for 30 minutes on ice with slight agitation. Beads were washed, resuspended in flow cytometry staining buffer, and analyzed using the NovoCyte.

For antibody binding assays, the cells were treated with 10 μg/mL of CO-001, CO-005, magrolimab, or IgG4 isotype control and incubated for 1 hour on ice with slight agitation. The cells were washed and stained with a cell marker cocktail and PE-conjugated goat anti–human Fc secondary antibody for 30 minutes on ice with gentle agitation. The cells were then washed three times with a cold flow cytometry staining buffer and analyzed by flow cytometry on the NovoCyte. Excitation wavelength for PE was 488 nm and detection wavelength was 572/28 nm.

The blood loop assay was conducted using blood from healthy donors as described previously (18) at Immuneed. Anti-CD47 mAbs at different concentrations were added to each loop. The loops were incubated at 37°C for 4 hours and maintained in continuous motion by attaching them to a rotating wheel. Blood samples were collected at the start (baseline) and after 4 hours for hematologic parameter analysis. At each sampling point, EDTA was added to the blood at a final concentration of 10 mmol/L to prevent coagulation. Counts of RBCs, white blood cells (WBC), lymphocytes, and neutrophils, as well as the hematocrit ratio before and after treatment with anti-CD47 antibodies were analyzed using a hematology analyzer XP-300 (Sysmex).

Lentiviral vectors containing genes coding for firefly luciferase and enhanced GFP were produced by transfecting HEK293T with 8.3 μg of each of the following plasmids: pMD2.G envelope plasmid, pCMVΔ8.91 packaging plasmid, and pSLIEW transfer plasmid (19). The cells were cultured to 70% to 90% confluence on the day of transfection. The cell culture media were changed approximately 1 hour before transfection. Transfection mixtures were prepared using the Calcium Phosphate Transfection Kit (Thermo Fisher Scientific, cat #K278001) according to the manufacturer’s instructions. Cell culture medium was changed 4 hours after transfection. Viral supernatants were collected after 2 days and concentrated using Lenti-X Concentrator (Takara Bio, cat #631231-CLO) at 4°C overnight before the viruses were collected by centrifugation at 1,500 × g for 45 minutes at 4°C. Pellets were suspended in a small volume (<1 mL) of cell culture medium and stored at −80°C. Frozen lentiviral stocks were titrated using Reh cells by using the standard transduction protocol (see the following section).

Reh cells (5 × 10⁵ cells per well) were seeded into 48-well plates with 4 mg/mL polybrene (Merck, cat #TR-1003-G) present in the cell culture media. Lentiviral concentrates were added to the cells, and spinfection was performed by centrifugation of the plates at 900 × g for 50 minutes at 34°C. After spinfection, the plates were transferred to a standard cell incubator (at 37°C and 95% air + 5% CO₂ in a humidified atmosphere) for 2 days before removing the viral particles by two repeated washes at 300 ×g for 10 minutes at 4°C. A small aliquot was taken from these cells to analyze the number of enhanced GFP–positive cells by flow cytometry, and the remaining cells were kept for establishing the xenograft model.

The xenograft model of B-cell precursor acute lymphoblastic leukemia (BCP-ALL) was established with transduced Reh cells (2 × 10⁵ cells) that were injected intra-tibially into 6 to 8-week-old female NOD/SCID IL2Rγ^null (NSG) mice (The Jackson Laboratory, strain #005557, RRID: IMSR_JAX:005557) as described previously (20, 21). Cancer progression was monitored by noninvasive in vivo imaging using an IVIS Spectrum CT from PerkinElmer. D-Luciferin (150 mg/kg, PerkinElmer, cat #122799) substrate was administered by intra-peritoneal injection, and after 10 minutes, the image was recorded using the autoexposure setting. All mice injected with the transduced Reh cells were included in the study and were randomized into different treatment groups around day 10 after IT injection when the mice had reached an average luminescence signal of 10⁷ to 10⁸ photons per second.

In the lymphoma xenograft model, 8 to 10-week-old female mice were inoculated subcutaneously in the right flank with 1.5 × 10⁶ Raji cells in 0.1 mL of a 1:2 suspension of PBS:VitroGel. Treatments were initiated when the tumor size averaged 100 mm³, and tumor volume was measured in two dimensions (length and width) using a standard caliper by an investigator not familiar with the prior treatment that the animal had received. Animals that reached a tumor volume of 2,500 mm³ or above, or developed ulcerations on their tumors, were sacrificed.

In all experiments, mice were euthanized when showing heavy engraftment or symptoms above a predetermined humane end point such as weight loss or limping of hind legs. All animals included in the study finished the experiment, i.e., there was no attrition. Sample size was determined based on previous knowledge about variation in tumor engraftment (20), and power analysis was therefore not performed. Mice were housed under specific pathogenic-free conditions with food and water ad libitum. Health status was monitored daily, and all animal procedures were conducted according to the approval by the Norwegian Food Safety Authority under identification number 29016.

To deplete neutrophils, NSG (The Jackson Laboratory, strain #005557, RRID: IMSR_JAX:005557) mice were exposed to 2 Gy of ionizing radiation (IR). Bone marrow was collected 1, 2, 3, 4, 10, 15, 18, and 21 days after IR, and each sample was stained with 0.2 μg of PE-conjugated Ly-6G/Ly-6C mAb (RB6-8C5; Thermo Fisher Scientific, cat #12-5931-82, RRID: AB_466045) for 10 minutes.

Subsequently, a new set of mice was irradiated with 2 Gy (day 1). On day 2, the mice were injected intra-tibially with transduced BCP-ALL cells (0.5 × 10⁶ cells per mouse). IVIS imaging was performed on day 3 after intra-tibial injection. The mice were then randomized and treated with a single intraperitoneal injection of 1 mg/kg of the indicated mAbs on the same day.

Graphs are presented as the mean values from at least three independent experiments as specified in the figure legends. Error bars represent the SD, and statistical analyses were performed using ANOVA. To determine the statistical differences between different animal treatment groups, the Mann–Whitney U test was employed. All statistical tests were two-sided and the significance level was set at 5% (P < 0.05). Curve fitting for binding assays was performed using four-parameter nonlinear regression in GraphPad Prism. All graphs and statistical analyses were performed using the GraphPad Prism 9 software (GraphPad Software, RRID: SCR_002798).

Data and materials from this study can be made available upon reasonable request to the corresponding author.

CO-001 demonstrated pan-reactivity across all the tested human cell lines, normal human PBMCs, and erythrocytes, suggesting its specificity for CD47. Given that CD47 interacts with SIRPα to promote the “don’t eat me” signal (5), we aimed to evaluate whether CO-001 could inhibit this interaction and potentially disrupt the inhibitory signaling. To assess the ability of CO-001 to directly inhibit the interaction between CD47 and SIRPα, Jurkat cells were pre-incubated with saturating concentrations of CO-001 or IgG4 isotype control. The mAb B6H12 is known to inhibit the CD47–SIRPα interaction whereas 2D3 does not (22); thus, they were chosen as the positive and negative controls, respectively. After pre-incubation, the cells were incubated with FITC-conjugated recombinant SIRPα and analyzed using flow cytometry. The data showed that CO-001 and B6H12 similarly blocked the binding of SIRPα to CD47 on Jurkat cells (Fig. 1A). In contrast, 2D3 did not block this interaction. The specific epitope recognition of CO-001 on CD47 was further verified through sequential epitope mapping using deep mutational scanning. The epitope mapping indicated that CO-001 binds to CD47 with amino acids Q19, N45, R121, T120, E122, and G123 being the most critical residues (Supplementary Table S2).

Given the established effect of CD47–SIRPα blockade by CO-001, the efficacy of CO-001 in inducing cancer cell phagocytosis was assessed using a flow cytometry–based in vitro assay with RAW264.7 macrophages (22). DiO-labeled macrophages were co-cultured with Jurkat cells pre-stained with CellTrace in the presence of CO-001 or anti-CD47 mAb B6H12 or sequence analogues of AO-176 and magrolimab. The results demonstrated that CO-001 effectively enhanced the phagocytosis of Jurkat cells by macrophages, comparable with magrolimab at 1 μg/mL. Notably, at lower dosages of 0.03 and 0.1 μg/mL, CO-001 exhibited superior phagocytosis-inducing capabilities compared with the other tested anti-CD47 antibodies (Fig. 1B). Conversely, the AO-176 analogue exhibited limited overall phagocytosis activity and was the least effective among the tested anti-CD47 mAbs, as previously shown (13).

Some antibodies directed against human CD47 have been shown to directly induce a unique form of PCCD in transformed cells (11, 13, 23). To further investigate this unique feature of the prototypic chimeric CO-001 mAb, Jurkat T cells were treated with CO-001 at various time points. Cell death was determined by the exposure of phosphatidylserine (PS) using the annexin V assay combined with 7-AAD staining specific for late apoptotic/dead cells and analyzed either by IncuCyte Live Cell Imaging or flow cytometry. As shown in Fig. 1C, CO-001 treatment induced rapid cell aggregation and PS exposure as early as 1 hour after the addition of the treatment, which was supported by flow cytometry showing cells stained positive for annexin V, thus undergoing PCCD (Fig. 1D). Consequently, CO-001 effectively triggered cancer cell death in Jurkat cells in a dose-dependent manner (Fig. 1E). CO-001–induced cell death was comparable in cells treated for 30 minutes or 3 hours (Fig. 1E); therefore, a 3-hour incubation time was used as the standard experimental setup for subsequent experiments with CO-001 or other anti-CD47 mAbs. CO-001 also shows superiority in inducing PCCD over AO-176 in Reh cells (Fig. 1F), which is the only CD47-targeting antibody that has reached clinical development claiming direct induction of PCCD (13). The induction of PCCD and phagocytosis by CO-001 in various hematologic cell lines is summarized in Supplementary Table S3.

To assess CO-001 binding to cancer and normal RBCs, staining was performed with increasing concentrations of CO-001, followed by incubation with a PE-conjugated secondary antibody and flow cytometry. CO-001 exhibited high specificity for all examined cell lines, with EC₅₀ values as indicated in Fig. 1G. Although the binding of CO-001 to RBCs was significantly lower compared with cancer cells (Fig. 1G), increased binding to RBCs was observed at higher concentration (10 μg/mL), raising concerns about potential off-target effects.

Given that some anti-CD47 antibodies induce hemagglutination of RBCs (12, 13, 24), we also evaluated the ability of CO-001 to cause RBC hemagglutination by incubating RBCs derived from healthy human donors with increasing doses of CO-001 and anti-CD47 sequence analogues AO-176 and magrolimab. Hemagglutination is indicated by a hazy pattern, whereas a small punctate circle indicates no hemagglutination (13, 24). As shown in Fig. 1H, CO-001 induced more pronounced hemagglutination than magrolimab and AO-176.

Although CO-001 effectively exhibits bifunctional activity by promoting both phagocytosis and PCCD of cancer cells in vitro, there is a need to humanize and engineer it to reduce potential clinical issues such as anemia, which is a known complication of anti-CD47–mediated cancer treatment (7). Thus, we developed a panel of humanized versions of CO-001 to identify a version with a high ability to trigger PCCD and cancer cell phagocytosis while avoiding RBC aggregation. The complementarity-determining regions regions of CO-001 were inserted into the human V gene with an IgG4 Fc. Additionally, 12 other humanized CO-001 variants were generated by different back mutations of amino acids and assessed for their possible impact on mAb stability and function.

In vitro assessments demonstrated that humanized CO-001 versions retained the ability to induce phagocytosis in cancer cells (Supplementary Fig. S1A). However, they were surprisingly less potent than chimeric CO-001 in inducing PCCD (Supplementary Fig. S1B). Additionally, no significant improvement in the reduction of RBC aggregation was observed (Supplementary Fig. S2). Following this, a humanized scFv containing CO-001–derived complementarity-determining regions, termed CO-004, was developed to explore the functional responses to monovalent CD47 ligation. Similarly, a bivalent scFv–Fc fusion protein CO-005 was designed. The structural representations of the distinct CO candidates are shown in Fig. 2A. First, we assessed the antigen-binding affinities of CO-004 and CO-005 using surface plasmon resonance and compared them with those of the full IgG4 chimeric CO-001. CO-001 exhibited an equilibrium constant of 38 pmol/L with the human CD47 antigen. Similarly, our findings revealed that CO-004 and CO-005 displayed high affinity toward CD47 with equilibrium constant values of 59 and 80 pmol/L, respectively (Supplementary Table S4). This reveals that CO-001, CO-004, and CO-005 bind specifically to the human CD47 antigen with comparable affinities within picomolar range.

In contrast to parental chimeric CO-001, CO-004 and CO-005 did not induce hemagglutination even at concentrations as high as 100 μg/mL (Fig. 2B; Supplementary Fig. S3) or bind to RBCs (Fig. 2C). We further investigated the functionality of humanized CO-004 and CO-005 and compared it with CO-001. Both CO-004 and CO-005 maintained the capability to trigger PCCD in several cancer cell lines, mirroring the initial observation with CO-001 (Fig. 2D; Supplementary Fig. S4A). Moreover, CO-005 enhanced the phagocytosis of Jurkat cells by 40% (Fig. 2E) and other hematologic cell lines (Supplementary Fig. S4B). As expected, CO-004 did not demonstrate phagocytosis-inducing activity characteristic of CO-001 and CO-005 because of lack of the Fc region (Fig. 2E). Phagocytosis was also evaluated using macrophages differentiated from human PBMCs (Supplementary Fig. S4C). Collectively, CO-005 demonstrated no binding or adverse impact on RBCs but retained the potent dual functionality of CO-001.

Having established the efficacy of CO-005 in killing cancer cells in vitro, we examined the anticancer activity of CO-005 in a BCP-ALL xenograft model using immunodeficient NSG mice (19, 20) and compared it with that of parental CO-001. As shown in Fig. 3A, a single dose of CO-001 or CO-005 (1 mg/kg) significantly eradicated cancer cells from the mice, as shown by a marked reduction in the luminescence signal compared with the control group treated with IgG4 isotype control.

In a lymphoma xenograft model, the efficacy of CO-005 was assessed in a head-to-head comparison with magrolimab in mice that had been engrafted with Raji B-cell non–Hodgkin lymphoma cells. The mice received a total of six injections of 5 mg/kg of CO-005, magrolimab, or human IgG4 isotype control. Tumor growth was measured using a digital caliper, and animals that reached a tumor volume of 2,500 mm³ or developed ulcerations at the tumor site were sacrificed. Interestingly, both CO-005 and magrolimab demonstrated potent efficacy in eradicating lymphoma cells (Fig. 3B). Notably, CO-005 significantly reduced tumor burden following a singular injection, whereas magrolimab, required multiple injections to achieve similar effects.

Previous research has demonstrated that anti-CD47 agents that solely enhance phagocytosis have limited efficacy as monotherapy (25). In this study, we observed that magrolimab required multiple injections to achieve effects comparable with those of CO-005 (Fig. 3B). This led us to hypothesize that the rapid efficacy of CO-005 is due to its unique ability to induce PCCD, which distinguishes it from other anti-CD47 agents. To test this hypothesis, BCP-ALL xenograft mice were injected with a single dose of 1.33 nmol/L of various CO variants to compare them: CO-005 (a phagocytosis and PCCD inducer), CO-004 (a PCCD-only inducer), and a human IgG4 isotype control. As anticipated, CO-005 significantly inhibited tumor progression. Notably, CO-004, despite exclusively inducing PCCD, also significantly inhibited tumor progression compared with the human isotype control (Fig. 4A), suggesting that PCCD plays a significant role in the potent and rapid onset of CO-005 efficacy in vivo. To validate the role of PCCD in vivo, we investigated the involvement of neutrophils in the efficacy of anti-CD47 mAbs in NSG mice (16). This is particularly relevant, as dendritic cells and macrophages exhibit reduced functionality in NSG mice (26), suggesting that the observed efficacy may depend on alternative cellular immune defenses. Notably, neutrophils constitute the majority of the remaining immune cells in the peripheral blood of NSG mice (27). Therefore, to assess the efficacy of CO-005 in the absence of neutrophils, we depleted neutrophils in NSG mice by IR and monitored the kinetics of post-radiation depletion and repopulation. We observed a reduction in neutrophil numbers between days 3 and 15 after IR, with complete reconstitution by day 20 (Fig. 4B). One day after neutrophil depletion, the NSG mice were injected with 0.5 million transduced BCP-ALL cells. On day 3 after engraftment (day 4 after IR), the mice were randomized and treated with a single dose of 1 mg/kg of either the human IgG4 isotype control or CO-005. Imaging revealed that CO-005 significantly inhibited tumor progression between days 5 and 10 compared with the control (Fig. 4C). This period coincided with neutrophil depletion, indicating that the antitumor effect of CO-005 in this setup was independent of neutrophils in NSG mice (Fig. 4B) and may be attributed to the agonistic effect of CO-005.

The pan-expression of CD47 on all cell types complicates the use of therapeutic anti-CD47 mAbs because of the risk of nonspecific binding to normal cells (7, 10). To assess CO-005 binding to normal immune cells, PBMCs from three healthy human donors were isolated and immunophenotyped. PBMCs prestained with surface marker antibody cocktails were incubated with 10 μg/mL of CO-005, CO-001, magrolimab, or IgG4 isotype control, and their binding to each cell type was assessed using goat anti–human Fc PE. CO-001 and magrolimab showed high binding to all cell types (Fig. 5A), whereas CO-005 exhibited significantly reduced binding to normal cells. Notably, CO-005 did not bind to RBCs at the tested concentrations, in contrast to magrolimab and CO-001, which bound to RBCs with high affinity. Furthermore, we observed no binding to normal B lymphocytes with CO-005 and two- to threefold less binding to other leukocytes at 10 μg/mL compared with magrolimab. These observations indicate that the design of the fusion protein CO-005 has reduced binding to CD47 expressed on normal RBCs and WBCs compared with CO-001 and magrolimab.

To assess the hematologic safety profile of CO-005 under physiologic conditions, whole blood samples from three healthy donors were evaluated using an ex vivo test system. Blood was circulated at 37°C, and hematologic parameters were measured using a hematology analyzer. The parameters assessed included RBC counts, hematocrit ratio, WBC counts, neutrophil counts, and lymphocyte counts. Blood was incubated for 4 hours with CO-005 at concentrations of 10, 100, and 500 μg/mL and compared with magrolimab (100 μg/mL), and IgG4 isotype control (500 μg/mL). CO-005 exhibited a negligible reduction in the RBC counts and hematocrit ratio, compared with a 99% drop with magrolimab (Fig. 5B and C). Similarly, as shown in Fig. 5D, no significant impact on the WBC count was observed with CO-005 at any concentration. In contrast, magrolimab caused an 18% decrease in the WBC count, primarily because of a 20% reduction in neutrophils and a 17% reduction in lymphocytes (Fig. 5E and F), whereas CO-005 displayed minimal effect on neutrophils and lymphocytes.

The clinical challenges associated with CD47-targeting therapies are exemplified by magrolimab, one of the most extensively studied anti-CD47 antibodies (16). Although it showed promise in combination therapies for several malignancies, its use has been associated with reversible anemia, particularly during priming doses, and hemagglutination (28–30). Consequently, these adverse effects ultimately led to the discontinuation of magrolimab’s development as a potential anti-CD47 therapeutic drug (https://clinicaltrials.gov/study/NCT04313881). This outcome not only underscores the critical need for safer alternatives in this therapeutic space but also highlights the inherent limitations of current approaches which rely predominantly on enhancing phagocytosis through macrophage activation (31). Moreover, the broader limitations observed with immune checkpoint therapies suggest that activating the innate immune system alone is insufficient to achieve durable antitumor immunity (32). These findings emphasize the necessity of incorporating additional cytotoxic mechanisms to address these gaps and improve therapeutic outcomes (33).

To address these challenges, we developed CO-005, a fully humanized bivalent scFv fusion protein with an IgG4 Fc backbone. CO-005 overcomes the sink effect by avoiding binding to RBCs and other normal cells while simultaneously inducing potent PCCD and maintaining phagocytic functionality. This bifunctional mechanism offers a potentially more effective strategy for CD47-targeted therapy. Coupled with its improved safety profile, CO-005 addresses both the efficacy and safety challenges associated with existing anti-CD47 agents.

CO-005 induces direct and potent PCCD that is characterized by a rapid exposure of PS, which acts as a robust “eat me” signal, empowering macrophage-mediated clearance and potentially activating the adaptive immune response (34, 35). Interestingly, although certain anti-CD47 mAbs have demonstrated the potential to induce PCCD, none are currently undergoing clinical development (11, 36, 37). These antibodies seem to mediate cell death through a caspase-independent pathway (11), similar to what was observed for CO-005. The exact cancer killing mechanism of CO-005 is currently under investigation. Furthermore, the induction of PCCD by CO-005 does not require ligation of CD47 by two antigen-binding sites; instead, monovalent binding, as observed with CO-004, is sufficient to initiate unique intracellular signaling cascades that drive PCCD. Notably, the efficacy of CO-004 highlights the critical role of PCCD as an antitumor mechanism in vivo even in the absence of macrophage engagement. This is particularly evident as CO-004, which lacks an Fc domain, cannot facilitate phagocytosis, underscoring the standalone contribution of PCCD to its therapeutic effects.

CO-005 exhibits strong antitumor activity at low doses (1 or 5 mg/kg) as a monotherapy, which can be attributed to its dual mechanism of action. These lower doses have been used to demonstrate superior antitumor activity compared with other anti-CD47 antibodies, which primarily rely on enhancing tumor phagocytosis activity and typically require higher doses (e.g., 10 mg/kg) to achieve similar effects (13, 16, 24). However, it is important to note that these antibodies do not bind to the murine homolog of CD47 (Supplementary Fig. S5). Therefore, the efficacy of anti-CD47 mAbs in these wild-type mouse models may be overestimated, as the absence of an antibody sink effect could artificially enhance their apparent effectiveness. To address this limitation, syngeneic mouse models will be used to further evaluate the efficacy of CO-005.

The specificity of CO-005 in targeting cancer cells while sparing normal cells is a significant advantage. Currently, the primary side effects of CD47-blocking antibodies include anemia and antibody sink effects, which hinder their clinical application (7). These limitations arise from the widespread expression of CD47 on both tumor and normal cells, particularly on RBCs (3, 7). To date, several strategies are being explored to address these problems. Some antibodies are designed to bind negligibly to CD47 on RBCs, whereas others are engineered as bispecific antibodies, reducing CD47 binding through the high affinity of another target. Another approach involves modifying the binding mode, potentially through orientation or conformational epitope changes (13, 24, 38, 39). The design of CO-005 incorporates modifications to its CD47-binding interface, reducing its binding affinity to RBCs and other normal cells without compromising its antitumor activity. The underlying mechanism responsible for selective binding of CO-005 to tumor cells over normal cells, including RBCs, is currently under investigation. This selective binding is expected to confer a more favorable pharmacokinetic and safety profile in humans compared with other CD47-targeting antibodies that exhibit high affinity for normal cells, including RBCs.

In conclusion, CO-005 represents a significant advancement in CD47-targeted cancer therapies. Its dual mechanism of action, selective binding profile, and reduced toxicity address the limitations of current approaches while offering the potential for enhanced antitumor efficacy. These properties may position CO-005 as a promising candidate for clinical development, capable of overcoming the challenges associated with CD47-targeting agents and paving the way for safer and more effective immunotherapies.

S. Matar reports a patent for WO2024074730A2 pending and a patent for WO2024074724A1 pending and employment with and eligibility for share options at Caedo Oncology. S. Skah reports a patent for WO2024074724A1 pending and a patent for WO2024074730A2 pending and employment with and eligibility for share options at Caedo Oncology. K. Hestdal reports personal fees from Caedo Oncology outside the submitted work, as well as a patent for WO2024074724A1 pending and a patent for WO2024074730A2 pending. R.D. Pettersen reports a patent for WO2024074724A1 pending and a patent for WO2024074730A2 pending and being a shareholder and board member in Caedo Oncology. N. Richartz reports a patent for WO202407472A1 pending and a patent for WO2024074730A2 pending, as well as employment with and share options at Caedo Oncology. L.E. Diomande reports current employment with Caedo Oncology but holds no share options or patents. No disclosures were reported by the other authors.

S. Matar: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Skah: Data curation, formal analysis, investigation, writing–review and editing. L.E. Diomande: Data curation, methodology, writing–review and editing. T. Buss: Conceptualization, data curation, supervision, validation, writing–review and editing. H.R. Hagland: Conceptualization, supervision, writing–review and editing. A. Yadav: Data curation, formal analysis, investigation, methodology, writing–review and editing. R.J. Forstrøm: Data curation, formal analysis, methodology, writing–review and editing. B. Dalhus: Data curation, formal analysis, investigation, writing–review and editing. K. Hestdal: Conceptualization, resources, supervision, funding acquisition, validation, methodology, project administration, writing–review and editing. R.D. Pettersen: Conceptualization, supervision, investigation, methodology, writing–original draft, project administration, writing–review and editing. N. Richartz: Conceptualization, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We would like to acknowledge the Norwegian Core Facility for Human Pluripotent Stem Cells at the Norwegian Center for Stem Cell Research for providing access to their cell culture facilities and equipment for the transfection and transduction of lentiviral constructs and the OUS Core Facility for Structural Biology for surface plasmon resonance analysis (Project 2015-095).