Anti-CD47 antibodies were created by immunization of H2L2 transgenic Harbour Mice^TM (Harbour BioMed) with a human CD47-hIgG1Fc fusion protein. Following immunization, a panel of hybridomas expressing human antibodies was obtained that bound specifically to CD47. Based on binding data, high-affinity antibodies were selected for hybridoma expansion, production and characterization. SRF2.3D11 and SRF4.2C11 were selected for further characterization, and SRF2.3D11 was subsequently engineered to have a wild-type (WT) hIgG₄ Fc domain, resulting in SRF231.

Jurkat (E6.1), Raji, HL-60, SU-DHL4 and TK-1 cell lines were obtained from American Type Culture Collection. Ri-1 and OPM-2 were obtained from DSMZ. The Jurkat CD47 knockout (KO) cell lines were generated via CRISPR/CAS9 technology (Applied StemCell). Primary bone marrow-derived acute myeloid leukemia (AML) samples were purchased (Cureline or Conversant). All cell lines were maintained in R10 medium (Roswell Park Memorial Institute [RPMI] 1640 + 10% FBS+penicillin [100 U/mL] and streptomycin [100 µg/mL]) at 37°C, 5% CO₂. RBCs were isolated from fresh human ethylenediaminetetraacetic acid (EDTA) anti-coagulated whole blood (Research Blood Components) via centrifugation and subsequent washing in PBS+2 mM EDTA. RBC suspensions were stored at 4°C at 10% v/v in PBS+2 mM EDTA and used within 3 days.

The following antibodies were generated in-house: antidinitrophenyl hIgG4 isotype control, SRF231 (hIgG₄), SRF231 F(ab’)₂ (full-length antibody incubated on Immobilized FABricator enzyme [Genovis A2-FR2-1000]), Fc byproduct removed by incubation with MabSelect SuRe [GE 17-5438-03] and F(ab’)₂ purified by collection of elution fragments and verified by analytical size exclusion chromatography), SRF231mut (S228P and L235E mutations introduced into SRF231 to reduce Fc-FcR binding and stabilize half-antibody formation)31 and SRF4.2C11 (hIgG₁). B6H12 (mIgG₁) and respective isotype control were purchased from BD Biosciences; polyclonal hIgG was purchased from BioXCell. hIgG₄ or SRF231-Alexa Fluor 647 (AF 647) conjugates were generated using AF 647 Antibody Labeling Kit (ThermoFisher) or custom ordered from BioLegend. For macrophage detection, anti-hCD14-Allophycocyanin (APC), clone M5E5 (BD Biosciences) or anti-mCD11b-AF 647, clone M1/70 (BioLegend) were used. For combination experiments, anti-CD20 antibody (rituximab; anti-hCD20-hIgG₁; Invivogen) was used. For CD32a blocking experiments or flow-based detection, either unlabeled or fluorescein isothiocyanate labeled antibody (clone IV.3; StemCell Technologies), was used.

For whole blood smears, fresh human whole blood was incubated with antibody overnight at 37°C, 5% CO₂. The following day, a drop of blood was smeared onto a glass slide using a glass ‘spreader’ slide. Smears were air-dried and imaged with an EVOS light microscope using the 40X objective. Images were analyzed qualitatively for evidence of RBC agglutination. For flow-based quantitative analyses, purified human RBCs were incubated with antibody for 0.5 hour at 37°C in a 96-well round-bottom plate (Corning), followed by incubation at 4°C overnight and flow cytometry-based assessment of RBC doublets.

To mimic scaffolded antibody conditions, Protein G-coated 96-well plates (Thermo Fisher) were coated with antibody diluted in PBS overnight at 4°C. The following day, unbound antibody was washed away, and cells added at 1×10⁵ cells/well in RPMI +1% FBS. Alternatively, isotype control or soluble antibodies were added to Jurkat cells in the soluble form. Plates were incubated overnight at 37°C. Cells were pipetted off the plate; cell death induction was evaluated using AnnexinV/Propidium Iodide (Life Technologies) via flow cytometry according to the manufacturer’s instructions.

All animal studies were conducted according to guidelines established by the internal and external Institutional Animal Care and Use Committee. Female CB17 SCID or non-obese diabetic (NOD)/SCID mice aged 6–8 weeks (Charles River Laboratories) were implanted with either Raji (10×10⁶), OPM-2 (10×10⁶), Ri-1 (10×10⁶), HL-60 (5×10⁶), or human diffuse large B-cell lymphoma cell line (SU-DHL-4 [10×10⁶]) cells in 0.2 mL 50% Matrigel (BD Biosciences) subcutaneously into the right flank. Antibody treatments were initiated when tumors averaged 100–200 mm³. For clodronate studies, mice were treated with 100 µg clodronate liposomes (Thermo Fisher Scientific) intravenously 3 times/week for 2 weeks. One day following the second clodronate treatment, mice were treated intraperitoneally (IP) with either 100 µg isotype control antibody or SRF231 3 times/week for 3 weeks. Mice were monitored, and tumor volume was measured twice weekly by electronic calipers (length × width²)×0.52. Mice received SRF231 IP in 0.2 mL at indicated doses, either once/week for 1 week, once/week for 2 weeks or once/week for 3 weeks depending on the tumor model. Rituximab was administered in 0.2 mL IP once/week for 3 weeks (days 1, 8 and 15).

Mice were terminally bled through cardiac puncture, and blood was collected into EDTA tubes. Samples were immediately centrifuged and aliquoted. Tumor tissues were collected in Lysing Matrix A, orange tube (MP Biomedicals). Lysis buffer was added, and tissues were homogenized using FastPrep (MP Biomedicals). Mouse macrophage inflammatory protein 1α (MIP-1α) and Mouse chemoattractant protein 1 (MCP-1) cytokines were assayed using multiplex V-Plex cytokine kit from (Meso Scale Discovery). Plates were read on a Meso Scale Discovery Imager.

Xenograft tumor tissues were collected, formalin-fixed and paraffin-embedded. Blocks were sectioned (Leica RM2145 microtome) at approximately 4 μm. Immunohistochemistry was performed with rat anti-mouse F4/80 (Bio-RAD). Slides were scanned (Aperio AT2 whole slide scanner). Image analysis was performed on digital slide images (Visiopharm software or QuPath imaging software).32 Tumor tissue regions of interest were identified in whole slide images and image analysis algorithms or tools applied to quantitate either F4/80 expression as a percent positive value within the tumor area or evaluate necrotic tumor area as a percentage of total tumor area (QuPath).

For human macrophages, CD14⁺ monocytes were isolated via negative selection from pooled buffy coats (n=3 donors, Research Blood Components). Primary monocytes were cultured for 5–7 days in the presence of 100 ng/mL human macrophage colony stimulating factor (hM-CSF) (Invitrogen). For mouse macrophages, bone marrow was collected by flushing the femur and tibia from Balb/c mice with cold R10. Bone marrow was passed through a 70 µm filter, spun down and resuspended in R10 with 100 ng/mL mouse macrophage colony stimulating factor (mM-CSF) (Invitrogen) and cultured in a T75 tissue culture treated flask for 6 days.

Macrophages were seeded in 96-well plates (Corning). The following day, target cells were carboxyfluorescein succinimidyl ester (CFSE)-labeled according to manufacturer’s protocol (Life Technologies) and cocultured with macrophages at a 2:1 ratio for 2 hours at 37°C in the presence of antibody. In some cases, macrophages were pretreated with a CD32a (IV.3) blocking antibody prior to addition of target cells. For RBC phagocytosis the target:effector ratio was 10:1, and for primary AML samples it was 1:1. Cells were trypsinized and stained with a macrophage-specific marker (CD14-APC for human and CD11b-APC for mouse macrophages). Flow cytometry was performed, followed by doublet exclusion analysis and phagocytosis reported as a percentage of CD14⁺ macrophages that were CFSE⁺. For analysis of cell death in this assay, a live/dead dye (Life Technologies) was added to the CD14-APC staining cocktail. Cell death was identified within the non-phagocytosed target cell population.

Jurkat cells were cocultured with primary human macrophages in the presence of 10 µg/mL hIgG4 isotype control, SRF231 or SRF231 F(ab’)₂ for 10 min at 37°C. Following antibody treatment, Jurkat cells were separated from macrophages by gentle pipetting, and each cellular fraction was assayed independently according to manufacturer’s instructions from the phospho-immunoreceptor array kit (R&D Systems, Bio-Techne Corporation). Phospho-signals were detected by chemiluminescence and HLImage++ software used to quantitate and normalize phospho-signals.

Jurkat cells were pretreated with either isotype or anti-CD47 antibodies for 20 min at 4°C, washed and incubated with biotinylated human recombinant SIRPα-Fc protein (R&D Systems) for another 20 min at 4°C. After washing, remaining biotinylated SIRPα-Fc was detected with streptavidin-APC (Life Technologies), and geometric mean fluorescence intensity values were captured by flow cytometry excluding dead cells using a Live/Dead dye (Invitrogen). Inhibition of SIRPα-Fc binding to Jurkat cells correlates with decreased fluorescence intensity.

Data were presented as mean±SD, and statistical analysis was performed using Prism V.8 software. A Student’s t-test was used to determine the p value at the indicated day, and p values were considered significant below 0.05 (*p≤0.05; **p≤0.01; ***p≤0.001; ****p≤0.0001).

SRF2.3D11 was identified from a panel of IgG transgenic hybridomas expressing antibodies with human variable regions and was subsequently engineered as a hIgG₄ and named SRF231. SRF231 specifically binds to CD47 as indicated by surface staining of CD47-expressing WT Jurkat cells but not CD47 KO Jurkat cells (figure 1A). Multiple human tumor cell lines of hematopoietic lineage are sensitive to SRF231-mediated phagocytosis by primary human macrophages (figure 1B). Additionally, significant enhancement of phagocytosis of primary AML bone marrow cells is observed in the presence of SRF231 (figure 1C). As expected, SRF231 cooperates with the anti-CD20 antibody (rituximab) resulting in enhanced antibody-dependent cellular phagocytosis (ADCP) of malignant B-cell lymphoma cell lines, Raji and SU-DHL-4 (figure 1D).

To evaluate the distinct mechanism of action of SRF231-mediated phagocytosis, SRF231 (whole IgG) and SRF231 F(ab’)₂ fragments were evaluated in phagocytosis assays using Jurkat target cells with either human or mouse macrophage effectors. Despite the ability of SRF231 to block the human CD47/human SIRPα interaction (figure 2A), pure blockade of the CD47/SIRPα interaction using a SRF231 F(ab’)₂, capable of saturating CD47 binding sites (online supplementary figure S1), was insufficient for phagocytosis induction in a fully human system (figure 2B). In contrast, pure blockade of the human CD47/mouse SIRPα interaction using the SRF231 F(ab’)₂ was sufficient to induce mouse macrophage-mediated phagocytosis (figure 2C). Additionally, a ~10-fold shift in the potency of SRF231 was observed when mouse versus human macrophage effector cells were used (with mouse effectors being more potent), highlighting species considerations in probing CD47/SIRPα biology. The affinity of the interaction between WT mouse SIRPα and human CD47 is substantially weaker than the affinity between the NOD SIRPα variant and human CD47, while the human CD47/human SIRPα affinity is intermediate to the two.33 When macrophages from different mouse backgrounds were used as effectors in the phagocytosis assay, differences were observed in the potency of the ability of SRF231 to induce phagocytosis of human targets (Jurkat cells). Bone marrow-derived macrophages from Balb/c mice were more sensitive to the induction of phagocytosis by SRF231, in full length or F(ab’)₂ form, than macrophages from NOD mice (online supplementary figure S2A). However, this difference was not observed when mouse targets (TK-1 cells) and an anti-mouse CD47 antibody (clone miap410) were used (online supplementary figure S2B), implicating the affinity difference between WT and NOD SIRPα for human CD47 in the differences observed.

Since CD47/SIRPα blockade is sufficient to induce phagocytosis in a mouse/human system yet insufficient in a fully human system, other prophagocytic signal(s) are likely necessary for the phagocytosis-inducing effects of SRF231 in the human setting. To search for additional signals that may mediate the effects of SRF231, phospho-immunoreceptor array screening was conducted within the context of the phagocytosis assay. Jurkat cells were separated from human macrophages after SRF231 treatment, and each fraction assayed independently. This screening approach led to the identification of FcγRIIa (CD32a) phosphorylation on treatment with SRF231 in both Jurkat and macrophage fractions (figure 2D). This signaling event was not observed with the SRF231 F(ab’)₂ fragment (figure 2D, bottom panel), which emphasizes the requirement for the antibody Fc region for CD32a phosphorylation and suggests that CD32a phosphorylation may be linked to Fc engagement of the FcR and subsequent phagocytosis induction. The 2 cell types used within this assay, Jurkat cells and human macrophages, were profiled for surface expression of CD32a. As expected, macrophages showed robust CD32a expression while Jurkat cells were negative (online supplementary figure S3A). Since the Jurkat fraction contained <10% macrophage contamination (data not shown) in the phospho-array screening assays, it is possible that the Jurkat subset is enriched for a fraction of activated macrophages that have already undergone conjugate formation with Jurkat cells in the presence of SRF231. The requirement for CD47 on the target cell for both CD32a phosphorylation and phagocytosis induction by SRF231 was evaluated using Jurkat WT and Jurkat CD47 KO cells. These data reveal that CD47 is required for both CD32a phosphorylation and phagocytosis induction, as CD47 Jurkat KO cells are insensitive to SRF231 in both assays (online supplementary figure S3B,C). The functional significance of CD32a in the phagocytosis assay was evaluated by blocking this receptor on macrophages prior to their coculture with Jurkat target cells. Pre-incubation of macrophages with a CD32a-specific function-blocking antibody led to abrogation of SRF231 activity within the phagocytosis assay (figure 2E) reinforcing the functional significance of CD32a in SRF231-mediated phagocytosis. In summary, while pure CD47/SIRPα blockade is sufficient for phagocytosis using mouse macrophage and human targets, pure blockade of the CD47/SIRPα interaction is insufficient in a fully human system where Fc/FcR engagement is required for functional activity.

In addition to promoting macrophage-mediated phagocytosis, SRF231 is capable of inducing tumor cell death. This cell death phenotype was initially observed within the context of the phagocytosis assay, in which both SRF231 and another anti-human CD47 mAb, B6H12, induced phagocytosis of Jurkat target cells, commensurate with the loss of tumor cell viability in the non-phagocytosed target cell population in a dose-dependent manner (figure 3A). This phenomenon was also observed in phagocytosis assays using mouse macrophages (online supplementary figure S4A). As macrophage-derived CD32a had been shown to be essential for SRF231-mediated phagocytosis by human macrophages, the significance of CD32a was also evaluated with respect to SRF231-induced cell death observed within macrophage/Jurkat coculture. CD32a blockade abrogated SRF231-mediated tumor cell death in the non-phagocytosed Jurkat population (figure 3B) as did the lack of CD47 expression on the target population (online supplementary figure S4B). Transwell analyses also revealed that this cell death phenomenon is both contact dependent and independent of the phagocytosis event (data not shown). These data implicate a dual role for CD32a in both SRF231-mediated phagocytosis and direct tumor cell death induction. To further assess the importance of CD32a-expressing macrophages in SRF231-mediated cell death, Jurkat cells were treated with soluble SRF231 and cell death was evaluated. Unlike anti-CD47 mAb clone CC2C6, which induced Jurkat cell death in the soluble form,18 soluble SRF231 did not induce Jurkat cell death (figure 3C, top panel). While CD32a engagement leads to downstream signaling within macrophages that may be involved in SRF231-mediated phagocytosis, it may also serve as a scaffold for SRF231, permitting a particular antibody orientation that allows for CD47 cross-linking on the tumor cells leading to tumor-intrinsic apoptosis. To mimic the scaffolding function of macrophages, SRF231 or hIgG4 isotype control antibody were immobilized on protein G-coated plates prior to addition of Jurkat cells. Under these conditions, scaffolded SRF231 was capable of inducing tumor cell death (figure 3C, bottom panel). Scaffolded SRF231-induced cell death was not limited to Jurkat cells and was also observed in AML and ovarian cell lines (online supplementary figure S4C). A panel of primary, non-malignant cells were evaluated for sensitivity to scaffolded SRF231-mediated cell death. CD47-high Jurkat cells were more susceptible to cell death compared with non-malignant cells, which uniformly expressed lower levels of surface CD47 and showed little to no susceptibility to cell death induction by SRF231 (online supplementary figure S4D).

CD47 perturbation has been implicated in a unique caspase-independent cell death pathway implicating phospholipase C gamma 1.6 15–17 To probe the molecular mechanism of SRF231-induced cell death, Jurkat cells were pretreated with a pan-caspase inhibitor (Z-VAD-FMK) prior to incubation with scaffolded SRF231. The addition of 100 µM Z-VAD-FMK (a pan-caspase inhibitor) did not appreciably abrogate SRF231-induced Annexin V staining (figure 3D), in contrast to doxorubicin-induced cell death, which is partially dependent on caspase induction (data not shown).

SRF231 binds to CD47-expressing RBCs in a dose-dependent manner (figure 4A), but does not cause hemagglutination (figure 4B,C) in contrast to B6H12 and SRF4.2C11, a tool antibody generated in a similar fashion to SRF231. Similarly, SRF231 does not induce RBC phagocytosis as is observed with B6H12 (figure 4D). Despite binding to RBCs, SRF231 displays a distinct, RBC-sparing functional profile in vitro.

SRF231 does not cross-react with mouse CD47 (data not shown). Therefore, the therapeutic activity of SRF231 was evaluated in multiple hematological xenograft models, including Burkitt’s lymphoma (Raji), multiple myeloma (OPM-2), B-cell lymphoma (Ri-1) and AML (HL-60) (figure 5A–D). In all models evaluated, SRF231 showed significant monotherapy antitumor activity (p<0.0001), and sustained tumor regressions were observed. To address whether Fc receptor interactions were important for SRF231 antitumor activity in the mouse, a variant of SRF231 with a mutation that results in reduced Fc-receptor binding (termed SRF231mut) was engineered. When tested in the Raji xenograft model, SRF231mut showed reduced antitumor activity compared with SRF231, suggesting that Fc-receptor engagement, at least in part, contributes to the antitumor activity of SRF231 (online supplementary figure S2C). To confirm that the reduced activity with SRF231mut was not due to reduced antibody exposure, a pharmacokinetic analysis was performed. Serum exposures were comparable for both SRF231 and SRF231mut across the time points evaluated (data not shown). Consistent with in vitro phagocytosis assays (online supplementary figure S2), efficacy was more readily achieved in CB17 SCID (Balb/c background) recipients of Raji tumors as opposed to NOD SCID animals, possibly due to the increased affinity of NOD SIRPα for human CD47 (online supplementary figure S2D). Collectively, these data suggest that in a mouse-human interaction, despite the absence of CD32a Fc-receptor engagement is important for optimal macrophage activation and phagocytic activity.

The efficacy of SRF231 in combination with the anti-CD20 antibody rituximab (Invivogen) was also evaluated in 2 lymphoma xenograft models. SU-DHL-4 is a diffuse large B-cell lymphoma model of the germinal center subtype containing a 14:18 chromosomal translocation resulting in Bcl-2 over-expression. As figure 5E (left panel) indicates, both SRF231 and rituximab showed single-agent antitumor activity in this model (p=0.0004 and 0.0016, respectively, versus isotype control on day 20). The combination of SRF231 with rituximab had a trend toward greater antitumor activity than SRF231 alone (p=0.056 on day 28 with 7/10 tumor-free animals in the SRF231 +rituximab combination group vs 1/10 or 0/10 tumor-free animals in the rituximab or SRF231 single agent arms, respectively, on day 24 onward). Increased antitumor activity was also observed in the Raji Burkitt’s lymphoma model when SRF231 and rituximab were co-administered (figure 5E; p=0.0272 for SRF231 vs SRF231+rituximab on day 65). In this case, complete regression of tumors in mice treated with the combination of SRF231 and rituximab was observed.

Since macrophages were involved in both in vitro antitumor mechanisms induced by SRF231, namely tumor cell phagocytosis and induction of cell death, the in vivo role of macrophages as effectors of SRF231 activity was investigated. Macrophages were depleted using clodronate liposomes in a Raji xenograft efficacy study. Treatment of Raji tumor-bearing animals with clodronate alone resulted in minimal tumor growth inhibition relative to the control group (p=0.0299; day 22). However, clodronate treatment prior to SRF231 administration reduced the antitumor effects of SRF231 (p=0.0451; day 22) (figure 6A). To determine whether macrophages were acting directly in the tumor, immunohistochemistry (IHC) was performed on Raji tumors from mice treated with a single dose of isotype control antibody or SRF231 and collected at different time points after treatment was performed. Tumors were stained with the murine macrophage marker F4/80. Treatment of Raji xenografts with SRF231 resulted in recruitment of F4/80⁺ macrophages into the tumor (figure 6B, left and middle panels). This macrophage infiltration was most prominent at the tumor margin and in the tumor core. In addition to infiltration of macrophages, SRF231 antitumor activity was also associated with significantly increased (p=0.0035) areas of tumor necrosis compared with isotype control treated animals (figure 6B, right panel). MCP-1 is a well-described monocyte-attracting chemokine that contributes to the recruitment of blood monocytes into tumors and sites of inflammation.34 To determine if SRF231 could induce MCP-1 in vivo, Raji tumor-bearing mice were treated with a single dose of isotype control antibody or SRF231 and plasma was collected and analyzed at various timepoints following treatment. As shown in figure 6C, mouse MCP-1 levels spiked in plasma from SRF231 treated mice approximately 6 hours after dosing and decreased to baseline thereafter. Macrophage cytokines were also evaluated in an HL-60 xenograft model in which tumor lysates were generated at 27 and 48 hours after a single administration of isotype control antibody or SRF231. In this study, both MCP-1 and a related macrophage chemoattractant, MIP-1α, were increased in tumor lysates from SRF231 treated mice by 27 hours and still elevated at 48 hours (figure 6D). Collectively, these data suggest that SRF231 may promote macrophage recruitment and activation that contributes to antitumor activity.