Engineered SIRPα Variants as Immunotherapeutic Adjuvants to Anticancer Antibodies

See allHide authors and affiliations

Science  05 Jul 2013:
Vol. 341, Issue 6141, pp. 88-91
DOI: 10.1126/science.1238856

Immunotherapy Packs a One-Two Punch

Despite the immune system's best efforts, cancer always seems to be one step ahead. One example of this is that tumor cells express CD47 on their cell surface. CD47 acts as a “don't eat me” signal to phagocytic macrophages. A potential therapeutic strategy could thus be to block this signal. Weiskopf et al. (p. 88, published online 30 May; see the Perspective by Kershaw and Smyth) created variants of the CD47 receptor, SIRPα, that could act as high-affinity antagonists of CD47. Although the antagonists blocked CD47 effectively in tumor-bearing mice, on their own they did not induce macrophages to phagocytose the tumor cells. When paired with a variety of therapeutic antitumor antibodies, however, the CD47 antagonists were very effective in treating several mouse tumor models.


CD47 is an antiphagocytic signal that cancer cells employ to inhibit macrophage-mediated destruction. Here, we modified the binding domain of human SIRPα, the receptor for CD47, for use as a CD47 antagonist. We engineered high-affinity SIRPα variants with about a 50,000-fold increased affinity for human CD47 relative to wild-type SIRPα. As high-affinity SIRPα monomers, they potently antagonized CD47 on cancer cells but did not induce macrophage phagocytosis on their own. Instead, they exhibited remarkable synergy with all tumor-specific monoclonal antibodies tested by increasing phagocytosis in vitro and enhancing antitumor responses in vivo. This “one-two punch” directs immune responses against tumor cells while lowering the threshold for macrophage activation, thereby providing a universal method for augmenting the efficacy of therapeutic anticancer antibodies.

Immune evasion is an emerging hallmark of cancer (1), and therapies that direct immune responses against cancer show promise in experimental and clinical settings (2, 3). Macrophages commonly infiltrate tumors, and their tumoricidal potential can be harnessed to benefit patients (4). CD47 is an antiphagocytic “don’t eat me” signal that distinguishes live cells from dying or aged cells, inhibits Fc receptor–mediated functions by myeloid cells, and is highly expressed by many cancers to avoid detection by macrophages (513). Antibodies that block binding of CD47 to SIRPα, an inhibitory receptor on macrophages, greatly increase phagocytosis of cancer cells (9, 10, 12, 13). However, antibodies have limited tissue distribution and can exert off-target effects (14, 15).

To improve CD47-targeted therapies, we used the single 14-kD binding domain of human SIRPα as a competitive antagonist to human CD47 (fig. S1A). Due to the weak ~1 μM affinity of the native SIRPα-CD47 interaction (1618), however, we found that wild-type SIRPα was a poor antagonist. To improve the affinity of human SIRPα for human CD47, we created mutant libraries of the N-terminal V-set immunoglobulin (Ig) domain of SIRPα (residues 1 to 118) conjugated to Aga2p for yeast surface display (fig. S1B). Using the CD47 Ig superfamily (IgSF) domain as a selection reagent, we conducted two “generations” of in vitro evolution using yeast surface display (figs. S1 and S2 and Fig. 1A). We obtained variants that bound CD47 with dissociation constants (Kd) as low as 34.0 pM and dissociation half-lives (t1/2) as long as 44 min compared with 0.3 to 0.5 μM Kd and 1.8 s t1/2 for wild-type SIRPα (Fig. 1A). The sequences of the high-affinity SIRPα variants converged on a consensus set of mutations. When we grafted these nine conservative substitutions onto the predominant wild-type human SIRPα allele (19) (allele 2), the resulting variant (termed CV1, consensus variant 1) bound human CD47 with an affinity of 11.1 pM. (Fig. 1A).

Fig. 1 Directed evolution of high-affinity SIRPα variants.

(A) Summary of sequences and surface plasmon resonance (SPR) affinity measurements of engineered SIRPα variants. The position of the mutated residues and their corresponding sequence in wild-type allele 1 is denoted at the top of the table. Red text color indicates the consensus mutations, and blue shading indicates contact residues in the consensus. Representative SPR sensorgrams of wild-type SIRPα and high-affinity variant FD6 binding immobilized CD47 are shown to the right. RU, response units. (B) The crystal structure of the FD6:CD47 complex depicted as transparent surfaces containing ribbon representations of FD6 (orange) and CD47 (blue). (C) Superimposition of the wild-type (magenta) and high-affinity (green) SIRPα:CD47 complexes. (Insets) Mutated contact residues in the SIRPα C′D loop (sticks) and the binding interface of CD47 (top, surface; bottom, sticks).

We determined the crystal structure of the high-affinity variant FD6 bound to the human CD47 IgSF domain (Fig. 1B, fig. S3, and table S1). The FD6:CD47 complex superimposed with the wild-type SIRPα:CD47 complex (17) with a root mean square deviation of only 0.61 Å, indicating a high degree of structural similarity and validating our efforts to preserve the geometry of the wild-type interaction (Fig. 1C). The overlapping binding modes of FD6 and wild-type SIRPα indicate that they would compete for the same CD47 epitope, providing maximal antagonism. As notable differences, the FD6 binding interface appears stabilized by three mutations in the C′D loop (Fig. 1C, lower inset) and the Ile31Phe mutation (Fig. 1C, upper inset). These structural studies imply that the high-affinity SIRPα variants could serve as efficacious CD47 antagonists.

When we assessed their ability to antagonize CD47 on the surface of human cancer cells, we found that SIRPα variants with increased CD47 affinity exhibited greater potency in binding (fig. S4, A and C) and blocking cell-surface CD47 (Fig. 2A and fig. S4B). As single-domain monomers (fig. S5A), both FD6 and CV1 exhibited potent antagonism relative to wild-type SIRPα. Subsequently, we evaluated the ability of high-affinity SIRPα variants to increase phagocytosis in vitro by coculturing macrophages and tumor cells in the presence of CD47-blocking agents. As fusion proteins to the Fc fragment of human IgG4 (hIgG4) (fig. S5, A and B), the high-affinity SIRPα variants led to increased phagocytosis of cancer cells as measured by microscopy and flow cytometry (Fig. 2B, fig. S6, and movies S1 and S2). No substantial phagocytosis was observed upon treatment with high-affinity SIRPα monomers, with dimers lacking Fc chains, or with B6H12 Fab fragments at concentrations that maximally antagonize CD47 (Fig. 2, B to D, fig. S5, C and D, and fig. S7). Thus, phagocytosis was only observed when CD47 was blocked in the presence of antibody Fc chains, which contribute a necessary prophagocytic stimulus.

Fig. 2 High-affinity SIRPα variants lower the threshold for macrophage phagocytosis.

(A) Dose-response curves of CD47 antagonism on Raji lymphoma cells with wild-type SIRPα allele 1 (WTa1), Fab fragments produced from the blocking antibody to CD47 clone B6H12, or two high-affinity SIRPα variants as monomers (FD6 and CV1). Cells stained with varying concentrations of CD47 blocking agents in competition with 100 nM Alexa Fluor 647-conjugated wild-type SIRPα tetramer. (B) Phagocytosis of GFP+ tumor cells by donor-derived human macrophages as assessed by flow cytometry. All protein treatments used at 100 nM. (C) Phagocytosis of GFP+ DLD-1 cells with vehicle [phosphate-buffered saline (PBS)], nonspecific isotype control (mIgG1), nonblocking antibodies to CD47 (2D3), or to EpCam. All antibodies were used at 20 μg/mL. WTa1 SIRPα or high-affinity SIRPα variant FD6 monomers were combined as indicated. (D) Phagocytosis of GFP+ SK-BR-3 breast cancer cells with vehicle, WTa1 SIRPα, or high-affinity SIRPα monomers alone or in combination with 1 μg/mL trastuzumab. (A to D) Phagocytosis assays were performed with macrophages derived from a minimum of three independent blood donors. The percentage of GFP+ macrophages was normalized to the maximal response by each donor for each cell line. Data are mean ± SD. (C and D) All SIRPα variants were used at 1 μM. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001 versus wild-type SIRPα variants, or as indicated otherwise, by two-way [(B) and (C)] or one-way (D) analysis of variance with Bonferroni correction.

Consequently, we hypothesized that treatment with high-affinity SIRPα monomers would enhance phagocytosis in the presence of tumor-specific monoclonal antibodies. Thus, we performed phagocytosis assays using antibodies targeting DLD-1 colon cancer cells. Indeed, high-affinity SIRPα monomers significantly augmented phagocytosis when combined with a nonblocking antibody to CD47 (20) or to EpCam, but not an isotype control (Fig. 2C).

To demonstrate the clinical implications of this principle, we examined the ability of high-affinity SIRPα monomers to enhance the efficacy of established antibodies currently used as cancer therapies. Phagocytosis assays were performed using Her2/neu+ SK-BR-3 breast cancer cells (21). Baseline phagocytosis was observed with SIRPα monomers alone, whereas treatment with the therapeutic antibody to Her2/neu, trastuzumab, alone or with wild-type SIRPα monomer led to moderate increases in phagocytosis (Fig. 2D). However, the combination of trastuzumab with high-affinity SIRPα monomers FD6 or CV1 resulted in maximal levels of phagocytosis that were higher than the additive effect of either agent administered alone. Similar effects were observed with the antibody to EGFR, cetuximab, using DLD-1 cells and with Raji lymphoma cells treated with varying concentrations of rituximab, an antibody to CD20 (fig. S7, A and B). Using a panel of recombinant antibodies to CD20 that contain the rituximab variable region and differing heavy-chain isotypes, we found that CV1 monomer was able to augment phagocytosis in response to all human IgG subclasses (fig. S7C). Therefore, the high-affinity SIRPα monomers could act as universal adjuvants to tumor-specific antibodies.

We next evaluated these principles in vivo by investigating the activity of the high-affinity SIRPα variants in mouse tumor models. We engrafted human tumors into NSG (NOD scid gamma) mice, which lack functional T, B, and natural killer (NK) cells but retain functional macrophages (10, 12, 22). NSG mice express a SIRPα allele that binds human CD47 and inhibits macrophage activity, thereby enabling in vivo evaluation of human CD47 blockade (10, 12, 19, 2325). We tested CV1-hIgG4 in a model of human bladder cancer to determine whether a single molecule combining a high-affinity SIRPα domain with a prophagocytic signal could act as a single agent for cancer. Green fluorescent protein (GFP)–luciferase+ 639-V bladder cancer cells were subcutaneously engrafted into NSG mice. Treatment with CV1-hIgG4 inhibited tumor growth (Fig. 3, A and B) and produced a significant benefit in survival even after discontinuing treatment (Fig. 3C). In CV1-hIgG4–treated mice, palpable stromal tissue developed around the sites of tumor engraftment that contained small tumor nodules embedded in an inflammatory infiltrate containing macrophages with evidence of phagocytosis (fig. S8).

Fig. 3 High-affinity SIRPα-Fc fusion proteins are effective as single agents for cancer.

(A) Growth of GFP-luciferase+ 639-V bladder cancer cells in the dorsal subcutaneous tissue of NSG mice upon daily treatment with vehicle control (PBS) or 200 μg high-affinity SIRPα-Fc (CV1-hIgG4) as evaluated by bioluminescence imaging. Ten mice per group were treated and analyzed, and each data point represents a measurement from an independent mouse. Bars indicate median values for each treatment group. Black arrows depict start and stop of treatment. ns, not significant; **P < 0.01; ***P < 0.001 by Mann-Whitney test. (B) Representative bioluminescence images of 639-V–engrafted mice from each treatment group on day 37 after engraftment. (C) Survival of mice engrafted with GFP-luciferase+ 639-V cells. Black arrows indicate the start and stop of treatment. Statistical analysis performed by Mantel-Cox test.

Although wild-type human SIRPα does not bind mouse CD47, the high-affinity SIRPα variants acquired cross-reactivity with mouse CD47 (fig. S10). Thus, our in vivo models allow for evaluation of efficacy in the presence of a large “antigen sink” and toxicity due to CD47 expression on normal mouse cells. Examination of the blood of treated mice revealed that CV1-hIgG4 coated all mouse blood cells (fig. S9A) and resulted in the development of chronic anemia as a side effect of therapy (fig. S9, B and C). No toxicity to other blood lineages was observed (fig. S9C). Red blood cell loss has also been observed with antibodies to mouse CD47 (12), consistent with these findings.

We further examined the safety of the high-affinity SIRPα variants in a toxicity study with cynomolgus macaques, which express a CD47 ortholog that is nearly identical to human CD47 (fig. S11, A and B). A single low-dose [1.5 mg per kg of weight (mg/kg)] injection of high-affinity SIRPα-Fc into two monkeys produced a substantial drop in red blood cells (fig. S11, C and D), similar to our findings in mice. By contrast, no hematologic toxicity was observed in a monkey treated with a dose escalation of high-affinity SIRPα monomer from 0.3 mg/kg to 10 mg/kg. No toxicity to other blood lineages or organ systems was detected in any of the monkeys, nor did we detect evidence of anaphylaxis (fig. S11, D and E). These findings further demonstrate that a single molecule that blocks CD47 and contains a prophagocytic stimulus, such as antibody Fc chains, produces toxicity. We therefore surmised that the high-affinity SIRPα monomers, along with use of an independent antibody, may offer an alternative and improved strategy for targeting CD47.

To explore the in vivo potential of the high-affinity SIRPα monomers, combination with rituximab was evaluated in a localized model of Raji cell lymphoma. Beginning 8 days after engraftment (fig. S12A), mice received a 3-week course of daily treatment with either vehicle, CV1 monomer alone, rituximab alone, or a combination of rituximab plus CV1 monomer. Treatment with CV1 monomer or rituximab alone only slowed tumor growth, whereas the combination therapy completely eliminated tumors in the majority of mice, and the effects persisted after treatment was stopped (Fig. 4, A to D, and fig. S12B). During the course of treatment, no toxicity to red blood cells or other hematologic lineages was observed (fig. S13). The effects of each therapy translated to respective trends in survival (Fig. 4D). The combination therapy remained effective against large Raji tumors (fig. S14, A and B). Additionally, we administered the high-affinity SIRPα monomers in combination with alemtuzumab (antibody to CD52), a second therapeutic antibody targeting Raji lymphoma cells (26). Combination therapy significantly inhibited tumor growth and increased survival (fig. S15). Together, these findings validate our strategy as a safe and effective approach to antagonizing CD47 and provide a proof-of-concept demonstration that the high-affinity SIRPα monomers can augment the efficacy of therapeutic antibodies in vivo.

Fig. 4 High-affinity SIRPα monomers enhance the efficacy of therapeutic antibodies in vivo.

(A) Growth of GFP-luciferase+ Raji lymphoma tumors upon daily treatment with PBS, 200 μg CV1 monomer, 200 μg rituximab, or rituximab plus CV1 monomer (200 μg each), as evaluated by bioluminescence imaging. Ten to 15 mice per group were treated and analyzed, and each data point represents a measurement from an independent mouse. Bars indicate median values. (B) Representative bioluminescence images of mice on last day of treatment (day 29 after engraftment). Red circles indicate sites of primary tumors; red arrow indicates site of metastases to axillary lymph nodes. (C) Tumor volume measurements of treated mice. Data are mean ± SD. (D) Survival of treated mice over time. (E) Representative pretreatment image of Her2/neu+ BT474M1 human breast tumors engrafted into the mammary tissue of NSG mice. Tumors were allowed to grow ~2 months before initiating treatment. Scale bar, ~1 cm. (F) Logarithmic fold-change in breast tumor volume upon treatment with the indicated therapeutic regimens. Eight to 10 mice per group were treated and analyzed, and each data point represents a measurement from an independent mouse. Tumor volumes for all cohorts were measured at the same time points (data are staggered for clarity). Bars indicate median values. (A to F) Black arrows indicate the start and stop of the treatment period. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 for antibody alone versus combination therapy, or as indicated otherwise, by Kruskal-Wallis with Dunn’s correction [(A) and (C)] or Mann-Whitney test with Bonferroni correction for the indicated comparisons (F).

We performed immunohistochemical staining of the large Raji tumors after treatment to examine macrophage infiltration. Compared with rituximab alone, tumors receiving combination treatment showed extensive macrophage infiltration (fig. S14, C and D). The extent of macrophage infiltration and therapeutic efficacy in vivo paralleled the degree of phagocytosis in vitro by NSG mouse macrophages (fig. S14E). Together, these findings indicate that macrophages are effector cells for the CV1 monomer/rituximab therapy, as has previously been confirmed for rituximab and antibodies to CD47 (9, 10).

In a third model, Her2/neu+ BT474M1 breast cancer cells were engrafted into the mammary tissue of female NSG mice, and tumors were allowed to grow to ~1 cm in diameter (~200 mm3) (Fig. 4E). Treatment with CV1 monomer alone had no effect on tumor growth, whereas trastuzumab alone was able to reduce tumor volume over time (Fig. 4F). The addition of CV1 monomer to the antibody regimen enhanced tumor regression (Fig. 4F), consistent with the results observed with rituximab and alemtuzumab.

Our findings support a new model for the action of CD47-targeting agents (fig. S16). As observed with high-affinity SIRPα monomers, blockade of CD47 alone does not induce macrophage activation but instead lowers the threshold for phagocytosis. The combination of high-affinity SIRPα monomers with an independent, tumor-specific antibody directs macrophage attack against cancer cells without toxicity to normal cells expressing CD47. These findings are consistent with previous studies showing that disruption of the CD47/SIRPα axis potentiates responses to therapeutic antibodies (6, 10, 11). Due to their immense potential as targeted therapies, over 100 antibodies are under clinical investigation (27). Although antibodies have demonstrated considerable clinical success thus far, they often elicit limited responses, and relapse is common after treatment (2830). We have developed reagents that broadly enhance the efficacy of tumor-specific antibodies and, thus, could act as universal adjuvants to antibody therapies. Overall, this study deepens our knowledge of macrophage responses to malignant cells and supports further evaluation of the high-affinity SIRPα reagents as immunotherapies for cancer.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S16

Tables S1

Movies S1 and S2

References (3138)

References and Notes

  1. Acknowledgments: The authors wish to thank members of the Weissman and Garcia laboratories for helpful advice and discussions. The authors thank R. Majeti, J. Liu, and the CD47 Disease Team for discussions and for providing antibodies to CD47. The authors thank T. Storm, L. Jerabek, H. Contreras-Trujillo, A. McCarty, S. Jaiswal, B. di Robilant, S. Varma, T. Naik, S. Willingham, H. Kohrt, N. Goriatcheva, D. Waghray, and S. Fischer for technical assistance, discussions, and reagents. The data presented in this paper are tabulated in the main paper and the supplementary materials. Atomic coordinates and structure factors of the FD6:CD47 complex (PDB code 4KJY) have been deposited in the Protein Data Bank. Research reported in this publication was supported by the National Cancer Institute (F30CA168059 to K.W. and CA139490 to I.L.W.), the National Institute of Diabetes and Digestive and Kidney Diseases (F30DK094541 to A.M.R.), the Stanford Medical Scientist Training Program (NIH-GM07365 to K.W. and A.M.R.), the Stanford University SPARK Program (to K.W. and A.M.R.), the Deutsche Forschungsgemeinschaft (VO 1976/1 to A.K.V.), the Joseph and Laurie Lacob Gynecologic/Ovarian Cancer Fund (to I.L.W), the Virginia and D. K. Ludwig Fund for Cancer Research (to I.L.W.), and the Howard Hughes Medical Institute (to K.C.G.). The content of this manuscript is solely the responsibility of the authors. K.W., A.M.R., A.M.L., I.L.W., and K.C.G. have filed a patent describing the high-affinity SIRPα reagents.
View Abstract

Navigate This Article