Introduction
CD47 is a ubiquitously expressed transmembrane protein that transmits an antiphagocytic “don’t eat me” signal through binding to its inhibitory receptor signal regulatory protein alpha (SIRPα) on myeloid cells [
1,
2]. A wide variety of cancer cells have been found to exploit this mechanism to escape innate immune surveillance through overexpression of CD47 [
3‐
6]. Previous studies have shown that targeting the CD47/SIRPα axis with anti-CD47 agents (including antibodies and SIRPα-Fc fusion proteins) promotes the phagocytosis of tumor cells by macrophages in vitro, and inhibits tumor growth in many human tumor xenograft models [
7‐
12]. However, antibodies can only bind to tumor cells which express human CD47 in these immunocompromised xenograft models. In contrast, within immunocompetent hosts, the ubiquitous expression of CD47 on healthy cells, especially on red blood cells (RBCs) and platelets, poses a huge challenge for anti-CD47 antibody-based therapies. On the one hand, antibodies binding with CD47 on healthy cells lead to antigen sink effect and attendant poor pharmacokinetic (PK) properties [
13,
14]. On the other hand, antibody binding with CD47 on healthy cells can lead to severe side effects [
15‐
18]. It is therefore possible that i) the antitumor efficacy of anti-CD47 agents might have been overestimated and ii) the likely treatment-related side effects of such therapies might have been overlooked in research using xenograft models.
Several studies in vitro and in xenograft models have demonstrated that additional prophagocytic signals are required to potentiate the antitumor efficacy of anti-CD47 agents following CD47 blockade, including Fc-Fc gamma receptor (FcγR)-mediated effector functions. The antitumor efficacy was very weak when using antibodies having weak or lacking Fc effector function; whereas, CD47 blockade incorporated with Fc that mediates strong effector function substantially promoted the antitumor efficacy [
6,
7,
15,
19,
20]. These results suggest that strong Fc effector function can improve the antitumor efficacy of anti-CD47 therapies. However, anti-CD47 antibodies that mediate stronger Fc effector function have been found to induce more severe side effects in non-human primates, even when treated in a low dosage [
15]. Additionally, several studies in syngeneic mouse models have suggested that stimulation of adaptive immune responses is required for the antitumor effects of anti-CD47 therapies [
21,
22]. Hu5F9 is an anti-CD47 antibody currently in clinical trials, results from its phase I study demonstrated very limited antitumor efficacy in patients with advanced solid tumors [
8,
16]. Further, various side effects such as anemia, hemagglutination, and chills were observed in many patients that received anti-CD47 therapies [
16‐
18].
To overcome these dilemmas in developing anti-CD47 therapies, one attractive strategy is improving the tumor selectivity of CD47-targeting agents. It has been demonstrated that tumor cells preferentially rely on aerobic glycolysis to produce energy (Warburg effect) and excrete a lot of H
+ to acidify the tumor microenvironment [
23‐
25], with pH often in the range of 6.4–6.8 for solid tumors, such as glioblastoma, colon cancer, melanoma, breast cancer, and lymphoma [
26‐
29], which is obviously distinct from physiological-pH (about 7.4) in normal tissues. Thus, we hypothesize that if an antibody binds strongly to CD47 only under acidic conditions, it should selectively bind to CD47 in solid tumors.
Here, we describe the generation of a pH-dependent anti-CD47 antibody (BC31M4) using antibody phage display technology and a pH-dependent selection strategy. BC31M4 binds to CD47 and blocks the CD47-SIRPα interaction with higher efficiency at acidic-pH than at physiological-pH; accordingly, BC31M4 more potently promotes macrophage phagocytosis of tumor cells at acidic-pH than at physiological-pH in vitro, which still requires the Fc-mediated effector functions. Further, BC31M4 selectively accumulates to solid tumors rather than to normal tissues in humanized syngeneic mouse models. Compared to the other tested anti-CD47 antibodies, BC31M4 causes minimal toxicity and exhibits superior PK properties. When converted into an isotype that mediates strong Fc effector function, BC31M4 in combination with adoptive T cell transfer efficiently enhances the antitumor responses of the adaptive immunity in syngeneic mouse models. Thus, our development of a tumor selective, pH-dependent antibody reconciles therapeutic efficacy with safety to support anti-CD47 therapies against solid tumors.
Methods
Cell lines
CHO, Raji, Jurkat, EL4, B16, CT26, MDA-231, and A20 cells were from the Cell Bank of Type Culture Collection (Chinese Academy of Sciences) or ATCC; the FreeStyle 293F were from Life Technologies; the LL/2 cell line was provided by Dr. Li (Beigene); the E.G7 cell line (a derivative of EL4 that expresses OVA) was provided by Dr. Chen (NIBS); the L929 cell line was provided by Dr. Li (NIBS). The CHO-hCD47, E.G7-hCD47, LL/2-hCD47, B16-hCD47, EL4-hCD47, A20-hCD47, and CT26-hCD47 stable cell lines were established by stably expressing full-length human CD47. The 293F-GnTI− cell was generated by knocking out the GnTI gene from the 293F cell using the CRISPR/Cas9 system. All cells were cultured in the recommended conditions (or following the providers’ instructions).
Expression and purification of proteins
The extracellular domain of CD47 (CD47-ECD) was fused to a His(× 6)-Avi-tag, the fusion protein was produced by transient transfection of FreeStyle 293F cells and purified by affinity chromatography. The extracellular domain of SIRPα was fused to the Fc of mouse IgG2a (mIgG2a); the fusion protein was produced by transient transfection of FreeStyle 293F cells and purified by affinity chromatography, after which the purified protein was further biotinylated (bio-SIRPα-Fc) using a biotinylation kit (Thermo Scientific). The full-length IgG antibodies were produced similarly as previously described [
30]. Briefly, the coding sequences of the variable regions of heavy chain (HC) and light chain (LC) were subcloned into corresponding vectors for expressing heavy chains and light chains of human IgG1 (hIgG1), mouse IgG1 (mIgG1), or mouse IgG2a (mIgG2a) isotypes, separately. Antibodies were subsequently expressed by transient transfection of 293F with HC + LC, and purified by protein A or protein G affinity chromatography. The isotype control antibodies (Ctrl. isotype) were specific to a known irrelevant target and were expressed and purified similarly as testing antibodies. The BC31M4-F(ab′)
2 fragment was generated by pepsin (Sigma) digestion of BC31M4-hIgG1 at pH 3.6, and subsequently purified with anion-exchange chromatography and size-exclusion chromatography.
pH-dependent selection and optimization of anti-CD47 antibodies
A human non-immune antibody phage display library was used for panning [
30]. The CD47-ECD protein used for panning was biotinylated by BirA ligase first and then captured on streptavidin-conjugated magnetic M-280 Dynabeads (Life Technologies); the magnetic beads were incubated with phage-displayed single chain antibodies (phage-scFvs) prepared from the library in pH 6.0 buffer for binding, and bound phages were eluted by pH 7.4 buffer; TG1-
E. coli cells were transformed with the eluted phages for ampicillin resistance screening, and subsequently rescued for the next round of panning. After two rounds of panning, single clones were picked and produced as phage-scFv form for enzyme-linked immunosorbent assay (ELISA) analysis, or converted into full-length hIgG1 isotype for SPR or flow cytometry analysis. During the antibody optimization, random mutations were introduced into both the third complementarity-determining region (CDR3) of the heavy chain (HCDR3) and the light chain (LCDR3) to construct phage display sub-libraries. These sub-libraries were subsequently screened using the pH-dependent selection strategy described above.
ELISA
The ELISA binding assays followed a previously described method [
30]. Briefly, CD47-ECD was coated on 96-well plates (MaxiSorp, Nunc). For analyzing phage-scFvs, phage-scFvs were added to the CD47-ECD-coated plates, and the binding of phage-scFvs to CD47-ECD was subsequently detected using a mouse anti-M13-HRP antibody (GE Healthcare). For analyzing full-length IgGs, serially diluted IgGs were added, and the binding of IgGs to CD47-ECD was subsequently detected using a mouse anti-human IgG Fc-HRP antibody (Thermo Scientific). These assays were performed in buffers of different pH. Specifically, for testing phage-scFvs during pH-dependent selection, the assays were performed in pH 6.0 and 7.4; for testing phage-scFvs during site saturation mutagenesis of histidines, the assays were performed in pH 6.5 and 7.4; for testing IgGs during binding confirmation, the assays were performed in pH 6.8 and 7.4.
Binding kinetic analysis by surface plasmon resonance (SPR)
Kinetic analyses of antibody binding to CD47-ECD were measured with a Biacore T200 instrument (GE Healthcare) at 25 °C. Anti-human IgG was immobilized on a CM5 sensor chip using a Human Antibody Capture kit following the manufacturer’s instructions (GE Healthcare). All antibodies analyzed were in hIgG1 form, and captured at similar levels on the chip. Twofold serially diluted CD47-ECD was injected over the surface of the chip. This experiment was performed at pH 7.4 and pH 6.8 buffers. The binding kinetic parameters were determined by fitting the sensograms to a 1:1 binding model using BIAcore T200 evaluation software.
Flow cytometry-based binding and blocking assays
For the binding assays, serially diluted antibodies (in hIgG1 form) were incubated with CHO-CD47 or tumor cells, and the binding of antibodies to the cells was detected using a goat anti-human IgG-FITC antibody (Thermo Scientific). The binding activity of antibodies is shown as the percentage of binding by normalizing the binding at the highest concentration as 100% binding. For the blocking assays, serially diluted antibodies were incubated with CHO-CD47 or tumor cells in the presence of bio-SIRPα-Fc, and the binding of bio-SIRPα-Fc to the cells was detected using streptavidin-FITC (Sigma). The blocking activity of antibodies is shown as the percentage of inhibition by normalizing the value of ‘bio-SIRPα-Fc only’ as 0% inhibition. Specifically, for blocking analysis of BC27, the assay was performed at pH 6.0, 6.5, 6.9, and 7.4; for binding and blocking analysis of BC31M4 and BC31M5, the assay was performed at pH 6.8 and 7.4. Specimens were analyzed by a flow cytometry instrument (BD, LSR II).
For the measurement of antibody binding to primary human T cells, human peripheral blood mononuclear cells (PBMCs) were incubated with serially diluted antibodies (in hIgG1 form), and the binding of antibodies to the cells was detected using the goat anti-human IgG-PE antibody (Thermo Scientific). Anti-CD8 (clone SK1, Biolegend), anti-CD4 (clone OKT4, Biolegend), and anti-CD3 (clone SK7, Biolegend) were used to identify T cells in PBMCs. The assay was performed at pH 6.8 and 7.4 separately. T cells were defined as follows: CD4+ T cells: CD3+CD4+, CD8+ T cells: CD3+CD8+. Specimens were analyzed by flow cytometry.
Crystallization and solving of the BC31M5-CD47 complex structure
The Fab of BC31M5 (with a heavy chain C-terminal His(× 6)-tag) was expressed by transiently transfection of HEK293 cells. The human CD47-ECD (residues 1–118) with a C15G mutation and a C-terminal His(× 6)-tag was expressed by transiently transfection of Expi293F-GnTI
− cells. The secreted BC31M5 and CD47 proteins were separately purified by Ni–NTA chromatography (Qiagen). BC31M5 and CD47 were mixed at ratio of 1:1.2 in pH 6.0 HBS (10 mM HEPES pH 6.0, 150 mM NaCl). The BC31M5-CD47 complex was purified using a Superdex S200 column (GE Healthcare), and was concentrated to 10 mg/mL for crystallization. Crystals were obtained by addition of proteins to an equal volume of 0.2 M Zinc acetate dihydrate, 0.1 M Sodium cacodylate trihydrate pH 6.5, 18% w/v PEG 8000. The diffraction data were collected at the Shanghai Synchrotron Radiation Facility (BL17B) and, integrated and scaled using XDS [
31]. The crystals were of the P2
12
12
1 space group, and were solved by molecular replacement with Phaser using the crystal structures of CD47 (PDB ID 5TZT) and Fab (PDB ID 4JPK) as search models. Two closely related complexes were found in the asymmetric unit, and the model was iteratively built in Coot [
32] and refined in PHENIX [
33].
Mice
NOD-SCID and BALB/c mice were purchased from Vital River. C57-hCD47/hSIRPα mice were generated as previously described [
34], and BALB/c-hCD47/hSIRPα mice were generated using the same strategy. Briefly, using CRISPR/Cas9 gene editing method, genes (exon 2 of
CD47 and exon 2 of
SIRPα) coding the IgV domains of both CD47 and SIRPα—which are responsible for the CD47-SIRPα interaction—were replaced with the corresponding orthologous human sequences. OT-I transgenic mice were provided by Dr. Chen (NIBS). All mice were maintained and bred under SPF conditions. All animal experiments were conducted following the National Guidelines for the Housing and Care of Laboratory Animals in China and performed under approved IACUC protocols at NIBS, Beijing.
Antibody-dependent cellular phagocytosis (ADCP)
Bone marrow-derived macrophages (BMDMs) from C57-hCD47/hSIRPα mice were used as effector cells in this assay. To prepare BMDMs, mouse bone marrow cells were collected from the tibia and femurs of C57-hCD47/hSIRPα mice, the cells were subsequently stimulated by adding L929-cell-culture-supernatants (containing granulocyte macrophage-colony-stimulating factor (GM-CSF) that secreted by L929 cells) to the medium, and cultured on a 24-well tissue culture plate for 7 days. Tumor cells were labeled with carboxyfluorescein succinimidyl ester (CFSE) following the manufacturer’s instructions (Thermo Scientific), and used as target cells. The BMDMs were labeled with anti-mouse F4/80-Alex Fluor647 (Thermo Scientific) prior to incubation with tumor cells. The CFSE-labeled tumor cells were incubated with different antibodies at room temperature for 15 min and then added to the labeled BMDMs using an effector-to-target ratio of about 1:1. Cells were incubated at 37 °C for 2 h in RPMI1640 medium supplemented with 10% heat-inactivated FBS. During the phagocytosis assay, the pH of the medium was adjusted to pH 7.4 and 6.8 using HEPES and PIPES, respectively. The phagocytosis of tumor cells by macrophages was measured via confocal microscopy.
Tumor models
For human tumor xenograft models, 6–8 weeks old female NOD-SCID mice were inoculated subcutaneously (s.c.) with 1 × 106 Raji cells on the right lower flank. When tumors reached about 50 mm3, mice were intraperitoneally (i.p.) injected with anti-CD47 antibodies (10 mg/kg, hIgG1 isotype) or PBS as control started on day 8 after inoculation, 2 doses per week for 3 weeks.
For syngeneic mouse models, 6–8-week-old male and/or female C57-hCD47/hSIRPα or BALB/c-hCD47/hSIRPα mice were inoculated subcutaneously (s.c.) with 5 × 105 E.G7-hCD47, 2 × 106 LL/2-hCD47, 2 × 105 B16-hCD47, 5 × 105 A20-hCD47, or 3 × 105 CT26-hCD47 cells on the right lower flank. In C57-hCD47/hSIRPα syngeneic mouse models testing anti-CD47 antibody treatment alone, mice were i.p. injected with antibodies (20 mg/kg, mIgG2a isotype) or PBS as control started on day 6 after tumor inoculation, dosing every 3 days for a total of 4 or 6 doses. In BALB/c-hCD47/hSIRPα syngeneic mouse models adopting anti-CD47 antibody treatment alone, mice were i.p. injected with a priming dose of antibodies (1 mg/kg, mIgG2a isotype) or PBS as control on day 1 after inoculation, followed by maintenance doses (10 mg/kg) started on day 3, dosing every 4 days for a total of 5 doses.
In combination therapy that comprised antibodies and OT-I T cells, C57-hCD47/hSIRPα mice were inoculated s.c. with E.G7-hCD47 cells on the right lower flank as described above, two antibody administration schedules were adopted: i) mice were i.p. injected with a priming dose of antibodies (1 mg/kg, mIgG2a or mIgG1 isotype) or PBS as control on day 3 after inoculation, followed by two maintenance doses (10 mg/kg) on days 5 and 11; ii) mice were i.p. injected with a priming dose of BC31M4 (1 mg/kg, mIgG2a isotype) or PBS as control on day 4 after inoculation, followed by four maintenance doses (1 or 10 mg/kg) on days 6, 10, 14, and 16. OT-I T cells (5 × 106) were transfused intravenously (i.v.) on day 8. To prepare OT-I T cells, spleen cells from OT-I mice were stimulated by adding IL-2 (3SBio) and OVA257-264-peptide (Sigma) containing RPMI1640 medium supplemented with 10% FBS and 0.05 mM 2-mercaptoethanol, followed by culturing and passaging for 4 days before injection.
For the tumor rechallenge experiments, C57-hCD47/hSIRPα mice were inoculated s.c. with E.G7-hCD47 cells on the right lower flank as described above, mice were i.p. injected with a priming dose of 1 mg/kg BC31M4 or BC31M5 (both have mIgG2a isotype) or PBS on day 3 after inoculation, followed by two maintenance doses (10 mg/kg) on days 5 and 11, OT-I T cells were transfused i.v. on day 8. Mice that survived from the combination therapy were inoculated s.c. with 5 × 105 E.G7-hCD47 or 2 × 105 EL4-hCD47 cells on the left lower flank, at about 4 months after the initial tumor inoculation. As controls, age-matched naïve C57-hCD47/hSIRPα mice were inoculated with the same tumor cells.
In all tumor models, tumor volumes were calculated using the modified ellipsoid formula (length × width2 × π / 6) based on caliper measurements.
Hematotoxicity analysis
Healthy C57-hCD47/hSIRPα mice (male and female, about 12-week-old) were injected i.p. with a single dose of anti-CD47 antibodies (20 mg/kg, mIgG2a isotype) or PBS. Blood was drawn from the retro-orbital plexus and collected in dipotassium-EDTA anticoagulation tubes at 3 h after injection. The hematological analyses were performed using the ADVIA 2120 Hematology System (Siemens) to assess the complete blood count. This analysis was carried out at the Vital River Labs (Beijing).
In vitro hemagglutination analysis
Antibodies (mIgG2a isotype) were threefold serially diluted in pH7.4 PBS from 600 nM in a U-bottom shaped 96-well tissue culture plate; RBCs from C57-hCD47/hSIRPα mice were resuspended in pH7.4 PBS and added at 1:1 volume ratio to the diluted antibodies (the final RBC density was 6 × 106 cells/well). The plate was incubated at room temperature for 2 h. The RBCs were further diluted and analyzed by a flow cytometry instrument (BD, LSR II); cell aggregation was assessed by the increase of FSC-A and SSC-A values in dot plots, compared to the PBS control.
Measurement of antibody distribution
C57-hCD47/hSIRPα mice (female) were inoculated with 5 × 105 E.G7-hCD47 cells s.c. on the right lower flank as described above. Antibodies (mIgG2a isotype) were labeled with Cy7 NHS Ester (Amersham, GE) following the manufacturer’s instructions. When tumors reached volumes of about 500 mm3, mice were i.p. injected with a priming dose of antibodies (1 mg/kg) or PBS as control, followed by giving a single maintenance dose (5 mg/kg) two days later. For the in vivo antibody distribution and persistence analysis, mice were monitored by in vivo fluorescence imaging using the IVIS Lumina III Imaging System (PerkinElmer) with excitation at 745 nm and emission measured at 800 nm; measurement was conducted at 3, 24, and 72 h after the maintenance dose. The total radiant efficiency was quantified for in vivo fluorescence imaging. For the ex vivo antibody distribution analysis, tumors and organs (spleens, livers, kidneys, and lungs) were isolated from mice for fluorescence imaging at 3 h and 72 h after the maintenance dose. The average radiant efficiency was quantified for ex vivo fluorescence imaging. The radiant efficiency was quantified using Living Image Analysis Software (PerkinElmer).
PK analysis
Healthy C57-hCD47/hSIRPα mice were i.p. injected with a priming dose of antibodies (1 mg/kg, mIgG2a isotype), followed two days later by a single dose of 20 mg/kg. Blood was collected at different time points (from 15 min to 29 day) after the 20 mg/kg dose. Antibody concentrations in serum were measured by ELISA (in pH6.5 buffer). The PK data were evaluated with WinNonlin software.
Measurement of body temperature and treatment-related-death
C57-hCD47/hSIRPα mice used in this study were treated with different antibodies adopting different treatment strategies. Side effects in these mice were recorded after antibody treatment, which assessed by monitoring body temperature and treatment-related-death. Mice body temperatures were measured at about 3 h after antibody injections using an infrared thermometer. The treatment-related-death of mice includes the following events: deaths of mice due to the side effects within 24 h after antibody treatment; mice endured continuous temperature drop and lethargy, or disability, for more than 24 h after antibody treatment that were euthanized; mice endured weight loss (more than 20%) and lethargy after antibody treatment that were euthanized. Mice in the repeated experiments (not shown) were included. In the summary of the body temperature of C57-hCD47/hSIRPα mice, 30 mice shown in the PBS group were randomly selected from the total mice using RAND function in Excel. The body temperature was summarized as temperature drop (compared to the average temperature in the corresponding control group). The mice examined for the antibody distribution and PK analyses were excluded from side effect assessment; additionally, mice used in the hematotoxicity analysis were excluded from treatment-related-death assessment.
Statistical analysis
All statistical analyses were performed using GraphPad Prism. For the data from in vitro experiments, Ordinary one-way ANOVA was used for comparisons of three or more groups, and unpaired Student’s t tests was used for comparisons of two groups. For the data from in vivo experiments, comparisons between the antibody treated groups and the corresponding PBS control were assessed using two-way ANOVA for tumor growth and, log-rank (Mantel Cox) tests for survival; Tumor volumes are shown as mean ± SEM. In all statistical analyses, the P values (* P < 0.05, ** P < 0.01, *** P < 0.01, **** P < 0.0001) were considered significant.
Discussion
The overexpression of antiphagocytic molecule CD47 on various tumor cells has made it a promising therapeutic target. However, the ubiquitous expression of CD47 on healthy cells poses a substantial hurdle for the development of safe and effective anti-CD47 therapies. In the present study, we aimed to overcome this dilemma by improving the tumor selectivity of anti-CD47 antibodies, and our approach was to exploit the known acidic microenvironment of solid tumors. We developed a pH-dependent anti-CD47 antibody that selectively binds to cells in solid tumors but sparing cells in normal tissues in immunocompetent syngeneic mouse models, which exhibits a favorable safety profile. When combined with adoptive T cell transfer, BC31M4 efficiently promotes adaptive antitumor immune responses as well as the development of immune memory.
Anti-CD47 antibodies have exhibited potent antitumor efficacy in many human tumor xenograft models, specifically by promoting the tumoricidal activity of macrophages [
3], and similar results were observed in our study. However, several limitations of using xenograft models to study anti-CD47 therapy have been highlighted in previous studies [
21,
22]. The mice used in these xenograft models are immunocompromised, lacking adaptive immune function but retaining functional macrophages that are responsible for the antitumor effects under CD47 blockade. It is therefore highly notable that the antibodies examined in these studies only target tumor cells expressing human CD47 in these models. Moreover, almost all antibodies used in these models are known to mediate strong Fc effector function that potentiates the antitumor activity. As a consequence, the antitumor efficacy of these antibodies may have been overestimated; and the treatment-related side effects caused by binding with CD47 on healthy cells has almost certainly been overlooked and/or underestimated. Consistent with this notion, we found that BC31M4, BC31M5 and Hu5F9 examined in this study confer potent antitumor efficacy in xenograft models without any side effects. However, these antibodies did not exhibit antitumor effect in syngeneic mouse models.
Hematotoxicity is a major concern with anti-CD47 therapies. We found that several of the side effects reported from the phase I study of Hu5F9, were also observed in C57-hCD47/hSIRPα mice, such as anemia, thrombocytopenia, hemagglutination, neutropenia, and chills (severe body temperature drop) [
16,
17]. Several studies have reported the development of anti-CD47 antibodies (or SIRPα-Fc fusion protein) that bind minimally to CD47 on the surface of RBCs or some other normal cells, which have exhibited good safety profiles when tested in preclinical and clinical studies [
42‐
44], However, it is noteworthy that CD47 is also ubiquitously expressed on normal tissues, so the impact(s) of anti-CD47 agents on other healthy cells should be thoroughly investigated. In this study, we showed that BC31M4 exhibits favorable safety profile owing to its selective binding to cells in solid tumors in immunocompetent syngeneic mouse models.
It has been demonstrated that blockade of the CD47/SIRPα signaling alone is insufficient to inhibit tumor growth in the absence of additional pro-phagocytic signals (
e.g., Fc-FcγR-mediated effector function). Similarly, our in vitro phagocytosis analysis results demonstrated that the Fc-mediated effector functions of BC31M4 is required to promote macrophages phagocytosis of tumor cells. Accordingly, most of the antibodies examined in the present study were in hIgG1 or mIgG2a isotypes that mediate strong Fc effector function. We did observe in vivo that BC31M4 in mIgG2a isotype efficiently promoted antitumor immunity when combined with adoptive T cell transfer; however, conversion of BC31M4 into the mIgG1 isotype (thereby reducing the strength of its Fc effector function) abrogated its antitumor effects. Our results suggest that strong Fc effector function is required to maximize the antitumor efficacy of anti-CD47 therapy in immunocompetent hosts. However, paradoxically, strong Fc effector function could lead to severe side effects. In our study, the side effects of Hu5F9 were more severe in mice than those observed in patients, which may be a result of the different isotypes used. Specifically, the Hu5F9 used in patients is the hIgG4 isotype, which mediates weak Fc effector function [
37], while the Hu5F9 used in mice herein was the mIgG2a isotype. We observed no side effects when Hu5F9 was converted into mIgG1. This may explain reports from clinical trials indicating that most of the anti-CD47 agents with weak Fc effector function can be tolerated in patients [
16,
46,
47].
Although a few studies have shown that an anti-CD47 antibody monotherapy was able to inhibit tumor growth in some syngeneic mouse models [
21], most studies in syngeneic mouse models and in clinical trials reported that anti-CD47 antibody alone did not exert significant antitumor effects [
16,
22,
48,
49]. Consistently, our data also showed that anti-CD47 antibodies (including BC31M4, BC31M5 and Hu5F9) monotherapy did not confer antitumor activity in the syngeneic mouse tumor models although they conferred potent antitumor activity in xenograft models. Moreover, most of the recent clinical trials involving anti-CD47 agents are examining combination modalities that include other antitumor agents. Several studies in syngeneic mouse models have reported that stimulation of adaptive immune responses is required to obtain an antitumor benefit from certain anti-CD47 therapies [
21,
22]. Consistently, we found that BC31M4 efficiently promoted antitumor immunity of both transferred T cells and native T cells when combined with adoptive T cell transfer. We speculate that BC31M4 promoted the phagocytosis of tumor cells through blocking the CD47-SIRPα interaction and engaging the activating FcγRs on phagocytes, after which such phagocytes may be activated to prime T cell immunity and to induce immune memory. However, BC31M4 monotherapy did not exhibit antitumor effects, perhaps owing to the inhibited and/or exhausted phenotypes of T cells in the tumor microenvironment [
50]. Accordingly, combining BC31M4 with additional immune-modulating agents to activate antitumor T cells in tumors may be an attractive approach for future development.
Comparing BC31M4 (high pH-dependence), BC31M5 (weak pH-dependence) and Hu5F9 (no pH-dependence), the absolute accumulation of BC31M4 in tumors is much higher than BC31M5 and Hu5F9, which might have contributed to the better antitumor effect of BC31M4 when combined with adoptive T cells transfer in syngeneic mouse models. Temporarily disregarding BC31M5’s severe side effects, it is noteworthy that although BC31M5 and Hu5F9 have similarly poor PK properties, BC31M5 exhibited higher antitumor efficacy than Hu5F9 when combined with adoptive T cell transfer. Given the weak pH-dependence of BC31M5, BC31M5 still exhibits higher relative intratumoral accumulation than Hu5F9, highlighting the apparently advantage of tumor selectivity for anti-CD47 therapy. Several studies have attempted to improve the tumor selectivity of anti-CD47 agents by generating bispecific antibodies (i.e., an antibody which recognizes two different antigens on tumor cells simultaneously), which have been demonstrated to minimize side effects [
34,
51]. Moreover, a recent study sought to improve therapeutic efficacy safety for ovarian cancer by engineered an oncolytic herpesvirus to express anti-CD47 antibodies in tumors [
52]. However, the selectivity of these antibodies is restricted to tumors that expressed two specific targets. As to BC31M4 we developed, its pH-dependent binding relies on the existence of an acidic condition in tumors. Although solid tumors often have a pH in the range of 6.4–6.8, pH can vary from 5.8–7.6 depending on tumor type, size, location, and metabolic state [
28,
53,
54]. Thus, the clinical application of BC31M4 would be for solid tumors with an acidic microenvironment.
Several studies have reported the generation of “reversed” pH-dependent antibodies, which bind to antigen with high affinity at physiological-pH but with low affinity at acidic-pH; these antibodies are generated by introducing histidines into the variable regions of the corresponding parental antibodies [
55‐
58]. A recent study reported the generation of a pH-dependent anti-human Her2 antibody that has selective binding in acidic conditions (again based on introducing histidines into the parental antibody), and demonstrated tumor inhibition ability only in vitro [
59]. In our study, we generated BC31M4 with high pH-dependence by employing antibody phage display technology and a pH-dependent selection strategy. By solving the co-crystal structure of CD47 and its close variant antibody BC31M5, and site saturation mutagenesis of histidines in the CDRs of BC31M4, we determined the structural basis of the pH-dependent binding property of BC31M4. Histidines H38 and H107 in CDRs of the light chain contribute to the pH-dependent binding of BC31M4, which rely largely on their protonation state switch that occurs around pH 6.8. BC31M4 and BC31M5 have only one amino acid (A108 in BC31M4, and T108 in BC31M5) difference, but the pH-dependence of BC31M4 is higher (about 22-fold as examined by SPR) than BC31M5; the weak pH-dependence of BC31M5 is likely due to the strong polar contacts formed by T108, which can apparently compensate for the loss of electrostatic contacts formed by H38 and H107 at physiological-pH, as evidenced by the higher affinity of BC31M5 at pH 7.4.
Many anti-CD47 antibodies are currently under preclinical investigations and clinical trials. The major challenges for these antibodies are the side effects associated with CD47 blockade and the weak therapeutic efficacy in solid tumors. There are other challenges, including tumor heterogeneity and an immunosuppressive tumor microenvironment. Our results suggest that, compared to an anti-CD47 monotherapy, combination therapies designed both to selectively target tumor cell killing and to promote adaptive immune responses should be more efficacious for treating solid tumors in patients. Our development of a tumor-selective, pH-dependent anti-CD47 antibody confirms that the acidic tumor microenvironment is an exploitable characteristic for effective deployment of antibodies to treat solid tumors. More generally, our study illustrates a strategy for generating antibodies against solid tumor antigens that are also expressed by healthy tissues.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.