CD47–signal regulatory protein-α (SIRPα) interactions form a barrier for antibody-mediated tumor cell destruction

We postulated that interactions between CD47, expressed broadly on normal and tumor cells, and the myeloid inhibitory immunoreceptor SIRPα would negatively regulate phagocyte-mediated ADCC induced by cancer therapeutic antibodies, and that targeting of CD47–SIRPα interactions would comprise a generic strategy to improve antibody therapy against cancer. In line with this, Chao et al. (27) have recently shown that antibodies against human CD47 synergize with rituximab in the elimination of non-Hodgkin lymphoma cells in immunodeficient mice and in in vitro phagocytosis experiments. Instead, we used mutant mice lacking the SIRPα cytoplasmic tail (21) to investigate whether inhibitory signaling via SIRPα could regulate the antibody-mediated elimination of syngeneic tumor cells in immunocompetent mice. In particular, we used the well-established mouse metastatic B16 melanoma model, in which the therapeutic antibody TA99, directed against the melanoma gp75 tumor antigen, has shown prominent beneficial effects in tumor cell clearance (28). First, B16F10 cells that expressed surface CD47 (Fig. 1A) were injected i.v., in the absence of therapeutic TA99 antibody, into wild-type and SIRPα-mutant mice, and this resulted in a similar tumor formation in both strains of mice (Fig. 1B), indicating that SIRPα signaling did not affect tumor cell metastasis and outgrowth per se. Next, these experiments were performed in mice that were treated with suboptimal concentrations of TA99 antibody. TA99 antibody treatment resulted only in a minimal reduction in tumor cell outgrowth in wild-type mice, but tumor formation was essentially abrogated in SIRPα-mutant animals under these conditions (Fig. 1C). This demonstrated directly that SIRPα-derived signals can form a limitation for antibody-dependent tumor cell elimination in vivo.

In line with the above, we hypothesized that CD47–SIRPα interactions were restricting the clinical efficacy of trastuzumab in the treatment of patients with Her2/Neu-positive breast cancer. To test this hypothesis, we explored a possible relationship between CD47 expression and breast cancer pathological features and clinical trastuzumab responsiveness. To do so, we analyzed breast cancer tissue CD47 mRNA expression in our cohort of 353 breast cancer patients as well as in a public data set (29). CD47 mRNA was overexpressed in many tumors, and expression correlated with poor-prognosis molecular subtypes (i.e., basal, Her2/Neu⁺) (Fig. 2A) and with adverse pathological features [i.e., high-grade, estrogen receptor (ER)⁻, progesterone receptor (PR)⁻]. Furthermore, analysis of a relatively small public data set (29) of Her2/Neu-positive breast cancer patients treated with trastuzumab plus vinorelbine revealed an inverse correlation between CD47 expression level and pathological response to the therapy (Fig. 2B), with significantly lower CD47 expression in complete responders. Although the latter finding clearly requires confirmation in a larger and independent patient cohort, it is consistent with an adverse role of CD47 in the trastuzumab-mediated elimination of breast cancer cells.

To directly investigate whether CD47–SIRPα interactions play a role in the trastuzumab-dependent destruction of breast cancer cells by phagocytes, we established an in vitro ADCC assay using trastuzumab-opsonized human SKBR-3 breast cancer cells expressing surface Her2/Neu and CD47 (Fig. 3A) as targets and human neutrophils as effector cells. Trastuzumab-mediated ADCC by neutrophils was potently and synergistically enhanced by F(ab′)₂ fragments of the B6H12 mAb that blocks CD47 binding to SIRPα (30) (Fig. 3 B–E). The enhancing effect of blocking anti-CD47 F(ab′)₂ was observed at different effector:target (E:T) ratios (Fig. 3C) and appeared to act by both decreasing the threshold as well as by increasing the magnitude of killing (Fig. 3D). Importantly, in the absence of trastuzumab, no detectable tumor killing effect of anti-CD47 F(ab′)₂ was observed, suggesting that CD47–SIRPα interactions do not control antibody-independent mechanisms of killing. This observation is in apparent contrast with the results of Chao et al. (27, 31), who also reported significant effects on lymphoma phagocytosis with the anti-CD47 mAb B6H12 alone. The latter may possibly relate, at least in part, to their use of intact B6H12 mAb that according to our own results can indeed cause direct ADCC in SKBR-3 cells (Fig. S1).

In the numerous independent experiments (n > 50) that were performed with neutrophils as effector cells for killing of trastuzumab-opsonized SKBR-3 cells a consistent enhancing effect of anti-CD47 F(ab′)₂ was observed, although the degree of killing (with trastuzumab alone) varied considerably for different effector cell donors (Fig. 3B). The latter appeared to be related to a factor(s) intrinsic to the effector cells, including individual differences in the expression of FcγRI and FcγRIIIb receptors that is pivotal for the induction of ADCC (Fig. S2).

To further study a regulatory role of CD47–SIRPα interactions in ADCC, siRNA-mediated knockdown of CD47 expression was performed in SKBR-3 target cells. This yielded cells with 80–90% reduced surface CD47 expression (Fig. 4A). These cells were significantly more sensitive toward neutrophil-mediated ADCC, consistent with a role for CD47–SIRPα interactions in restricting tumor cell killing (Fig. 4B). The increase was comparable to levels seen with wild-type SKBR-3 cells in the presence of blocking anti-CD47 F(ab′)₂.

Although the above strongly supported the idea that CD47–SIRPα interactions regulate ADCC in vitro and tumor elimination in vivo, it was important to confirm these findings with blocking antibodies against SIRPα. In fact, because of its much more restricted expression (12, 16), we anticipate that SIRPα, rather than the ubiquitous CD47, constitutes the preferred target for potential future therapeutic intervention. Because the previously reported antibodies against human SIRPα available to us either lacked the proper specificity or the ability to block interactions with CD47, we generated unique blocking mAbs against SIRPα₁. One antibody, designated 1.23A, was generated by electrofusion technology following negative selection on CHO cells expressing the myeloid-specific SIRP family member SIRPβ₁, whereas the other, designated 12C4, was generated by conventional hybridoma technology. Both of the two SIRPα polymorphic variants predominating in the Caucasian population, SIRPα₁ and SIRPα_BIT, as well as the highly homologous myeloid SIRPβ₁ and nonmyeloid SIRPγ family members were recognized by 12C4, but the 1.23A mAb exclusively recognized the SIRPα₁ variant (Fig. S3 A and B). Staining of leukocytes from SIRPα-genotyped individuals was consistent with this specificity (Fig. S3C), with the mAb 1.23A selectively recognizing monocytes and neutrophils from both α_1/α₁-homozygous and α_1/α_BIT-heterozygous individuals. Both mAbs effectively inhibited the binding of CD47-coated beads to CHO cells expressing SIRPα₁ and/or SIRPα_BIT (Fig. 5A) and promoted trastuzumab-mediated ADCC toward SKBR-3 cells by neutrophils from individuals with different genotypes (Fig. 5 B and C). For the 1.23A mAb, enhanced killing was only observed when neutrophils from α_1/α₁-homozygous individuals were used. When α_BIT/α_BIT-homozygous or α_1/α_BIT-heterozygous donor cells were used, 1.23A did not enhance SKBR-3 killing by trastuzumab, suggesting that the presence of a single functional allele of SIRPα is sufficient to restrict ADCC and that both alleles have to be inhibited simultaneously to achieve a beneficial effect accordingly.

C57BL/6 mice with a targeted deletion of the SIRPα cytoplasmic region have been described previously (21). These mice, originally generated onto the 129/Sv background and backcrossed onto C57BL/6 mice for 10 generations, were bred and maintained under specific pathogen-free conditions, together with wild-type C57BL/6 mice from the same genetic background, and used between 8 and 12 wk of age. Age-matched wild-type and SIRPα-mutant mice were injected i.v. with 1.5 × 10⁵ B16F10 tumor cells in 100 μL of HBSS on day 0. Mice were injected i.p. with a suboptimal dose of 10 μg of TA99 antibody (or PBS as control) on days 0, 2, and 4. At day 21 the mice were killed. Their lungs were excised and scored for the number of metastases and tumor load as described (28).

Antibodies, cell lines, culture conditions, procedures for the production of monoclonal antibodies, the CD47 bead binding assay, and flow cytometry are described in SI Methods.

We analyzed CD47 mRNA expression in 353 invasive breast carcinomas and 11 normal breast samples profiled (36) using whole-genome Affymetrix oligonucleotide microarrays (Gene Expression Omnibus accession no. GSE21653). Only two of the probe sets representing CD47, 211075_s_at and 213857_s_at, mapped exclusively to constitutively transcribed CD47 exons according to NetAffx, RefSeq, and the University of California Santa Cruz Genome Browser (27). Their expression strongly correlated (Spearman correlation, 0.87). We retained that with the highest variance (211075_s_at). Before analysis, the CD47 expression level for each tumor was centered by the average expression level of the normal breast samples. We analyzed the correlation between CD47 expression and patients’ age (≤/>50 y), pathological tumor size (≤/>2 cm), axillary lymph node status (negative/positive) and grading (I/II/III), immunohistochemisty estrogen and progesterone receptor status (negative/positive; positivity threshold 10% of tumor cells), and molecular subtypes (luminal A/luminal B/basal/Her2/Neu⁺/normal-like), defined as described (37). We also analyzed a public (http://caarraydb.nci.nih.gov/ccarray) expression data set of Her2/Neu-positive breast cancers treated with primary trastuzumab plus vinorelbine weekly for 12 wk followed by surgery (29). Pathological complete response was defined as the absence of invasive cancer in the breast and axillary lymph nodes at the time of surgery.

Neutrophils were isolated by density centrifugation from heparinized blood obtained from healthy volunteers using isotonic Percoll (Pharmacia) followed by red cell lysis with hypotonic ammonium chloride solution. Cells were cultured in complete RPMI medium in the presence of 10 ng/mL clinical grade G-CSF (Neupogen; Amgen) and 50 ng/mL recombinant human IFN-γ (PeproTech) at a concentration of 5 × 10⁶ cells/mL for 16–20 h. Monocytes were isolated from the peripheral blood mononuclear cells fraction by magnetic cell sorting using anti-CD14-coated beads according to the manufacturer's instructions (Miltenyi Biotec) or by counterflow elutriation. Washed tumor cells (5–8 × 10⁶ cells) were collected and labeled with 100 μCі ⁵¹Cr (PerkinElmer) in 1 mL for 90 min at 37 °C. The cells were preincubated with anti-CD47 and/or the therapeutic antibodies, as indicated, and washed again. The target cells (5 × 10³ per well) and effector cells were cocultured in 96-well U-bottom tissue culture plates in complete medium in an E:T ratio of 50:1, unless indicated otherwise, for 4 h at 37 °C in 5% CO₂ in RPMI with 10% FCS medium. Aliquots of supernatant were harvested and analyzed for radioactivity in a gamma counter. The percent relative cytotoxicity was determined as [(experimental cpm − spontaneous cpm)/(total cpm − spontaneous cpm)] × 100%. All conditions were measured in triplicate.

Statistical differences were determined using ANOVA or Student's t test as indicated.