Natural Killer Cell Recruitment and Activation Are Regulated by CD47 Expression in the Tumor Microenvironment

CD47 is a transmembrane protein that interacts with several integrins, two signal-regulatory protein (SIRP) family counterreceptors, and the secreted protein thrombospondin-1 (1–3). Malignant cells in some hematopoietic malignancies and solid tumors express elevated levels of CD47 relative to healthy tissues, and elevated CD47 in some cancers is correlated with poor prognosis (4). The prevailing hypothesis is that interaction of tumor cell CD47 with the counterreceptor SIRPα on macrophages inhibits phagocytosis of tumor cells (4). Conversely, antibody-mediated blockade of SIRPα binding or knockdown of CD47 promotes active phagocytosis of implanted human xenograft tumors in mice expressing a variant of SIRPα that binds human CD47 with high affinity (4, 5). In contrast, blocking SIRPα binding to CD47 is generally not sufficient to promote phagocytic clearance of syngeneic tumors in immune-competent mice, which requires additional immune stimuli to induce CD8 T cell–mediated tumor killing (4, 6, 7). CD47 is an inhibitory signaling receptor for thrombospondin-1 on T cells (8, 9). Thrombospondin-1 signaling through CD47 also inhibits antigen presentation by dendritic cells to T cells (10). CD47 is, therefore, an innate and adaptive immune checkpoint.

We found that natural killer (NK) cells express high levels of CD47 mRNA and cell-surface protein, which regulates NK cell homeostasis and their response to viral infection in mice (11). NK cells are also a first line of defense system against malignant cell transformation (12–14).

NK cell development, education, and effector functions are regulated by several families of activating and inhibitory receptors, including the killer cell immunoglobulin-like receptor (KIR) family, the CD94 family, the leukocyte immunoglobulin-like receptor (LIR) family, and NKG2D and the natural cytotoxicity receptors (NCR) NKp30, NKp44, and NKp46 (15). Inhibitory NK receptors recognize MHC class I, as well as non–MHC-I ligands such as PVR (15–19). The regulatory roles of thrombospondin-1 and CD47 in NK cells are less clear. Thrombospondin-1 inhibits early NK cell proliferation and enhances late expansion, but a role for CD47 in these activities was not examined (20). Functioning as a SIRPα counterreceptor, CD47 enables engraftment of NK precursors in mice reconstituted with a human immune system (21). Treatment with a CD47 antibody that inhibits SIRPα and thrombospondin-1 binding increased NK cell killing of head-and-neck squamous carcinoma cells in vitro, but the mechanism remains unclear because NK cells are not known to express SIRPα (22). Depletion of NK cells similarly attenuated the antitumor activity of a SIRPα blocking antibody in a syngeneic murine renal carcinoma model, but the same antibody did not inhibit NK killing of the tumor cells in vitro, suggesting a SIRPα-independent function of CD47 in NK cells (23).

Analysis of The Cancer Genome Atlas (TCGA) data for human melanomas showed positive correlations between CD47 mRNA expression and patient survival and CD47 expression with the expression of NK early activation genes. We identified CD47-dependent inhibitory functions for thrombospondin-1 in human and murine NK cells in vitro. Correspondingly, inhibition of melanoma growth following treatment with a CD47 antibody that inhibits thrombospondin-1 binding was associated with enhanced NK cell activation. In contrast, B16 melanoma tumors grew faster in Cd47^−/− mice, which was associated with reduced NK frequency and enhanced NK exhaustion, similar to that observed during chronic lymphocytic choriomeningitis virus (LCMV) infection of Cd47^−/− mice (11). Gene-expression and functional studies in NK cells identified impaired mitochondrial function and an enhanced proapoptotic signature in Cd47^−/− NK cells when subjected to stress. Therefore, we demonstrated that CD47 has both positive and negative roles in regulating antitumor NK immunity.

Mice were maintained under specific pathogen-free conditions, and the NCI Animal Care and Use Committee approved all treatments under protocol NCI/LP-012. Breeding pairs of wild-type (WT) and B6.129S7-Cd47^tm1Fpl/J (Cd47^−/−) C57BL/6J mice (The Jackson Laboratory) were backcrossed to minimize genetic drift and obtain Cd47^+/− mice. Heterozygous Cd47^+/− were bred to obtain Cd47^+/+, Cd47^+/−, and Cd47^−/− littermate mice. Mouse genotypes were confirmed by PCR using a set of primers targeting the mouse CD47 allele. Littermate and sex-matched mice between 6 and 12 weeks of age were used for experiments unless otherwise indicated.

Banked cryopreserved B16F10 melanoma cells (obtained from ATCC in 2007) were defrosted and cultured in T-75 flask for 48 hours in complete RPMI-1640 medium containing 5% FBS, 1% penicillin and streptomycin antibiotics, and 1 mmol/L L-glutamine. Because banked cells at identical passage were used for all animal studies, their identity was not verified in the last year, but the cells were reverified as negative for known murine viruses and Mycoplasma by the Animal Health Diagnostic Laboratory, Frederick National Laboratory. WT or Cd47^−/− littermate C57BL/6 mice were subcutaneously injected with PBS-washed B16F10 melanoma cells (1 × 10⁶ cells) into the hind limb. Tumor size was measured every other day using calipers for 15 days. In the CD47 blockade studies, doses (intratumoral, 50 μg/mouse) of anti-CD47 miap301 (eBioscience; cat. #16-0471-85) or isotype rat Ig (eBioscience; cat. #02-9602) were administered on days 7 and 15, and tumor volume was measured every other day until day 21.

4,6 diamidino-2-phenylindole (DAPI; cat. #D9542), rat serum (cat. #R9759), and rabbit serum (cat. #R9133) were purchased from Sigma-Aldrich. Aqua live/dead was from Thermo Fisher Scientific (cat. #L3495), UV Zombie was from BioLegend (cat. #423107), and ACK lysis buffer was purchased from Lonza (cat. #10-548E).

Total peripheral NK cell populations from mouse spleens were isolated (>90% purity) using an NK Isolation Kit II (MACS, Miltenyi Biotec; cat. #130-115-818) using MACS LS Columns (Miltenyi Biotec; cat. #130-042-401) following the manufacturer's instructions. Isolated NK cells were cultured in complete RPMI without or with IL15 (5–40 ng/mL).

Purified NK cells from WT and Cd47^−/− mouse spleens were tested for their cytotoxic potential against B16 tumor cells. Freshly isolated NK cells or NK cells cultured with IL15 were incubated with B16 tumor cells with different effector-to-target ratios (E:T; 1:1, 5:1, and 10:1). Cytotoxicity was determined by measuring the fluorescence intensity of the CellTox Green Dye (CellTox Green Cytotoxicity, Promega). Fluorescence intensity correlates with the loss of cell membrane integrity caused by cell death. This assay allows real-time kinetic cytotoxicity measurements because the same well can be measured multiple times, as the fluorescent signal remains constant after 72 hours. B16 cells (5,000 cells/well) were seeded in 96-well black-bottom plates (Greiner). NK cells were added in different E:T ratios to wells containing CellTox Green Dye. Wells without cells were used as a background control. Cell death was determined at 0, 12, 24, and 48 hours of incubation by measuring the fluorescence intensity (485/520 nm excitation/emission) with a TECAN microplate reader. The values obtained from the background control were subtracted from the values obtained from the different experimental conditions.

Cells were stained with annexin V according to the manufacturer's instructions (BD Biosciences; cat. #550474). Cells were washed in PBS and resuspended in 1 × annexin binding buffer containing allophycocyanin-conjugated annexin V. After 15 minutes of incubation at room temperature, cells were diluted in 1 × annexin V binding buffer and analyzed by flow cytometry.

Mitochondrial oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) of CD47-sufficient and CD47-deficient NK cells were compared using a Seahorse Bioscience XF^e24 analyzer in combination with the Seahorse Bioscience XF Cell Mito Stress Test assay kit (Agilent Technologies; cat. #103015-100). Briefly, NK cells from spleens of WT and Cd47^−/− littermate mice were isolated and plated at 0.75 × 10⁶ cells/well in 24-well plates (Seahorse Bioscience) into XF Base Medium (Seahorse Bioscience) with added 10 mmol/L glucose, 1 mmol/L sodium pyruvate, and 2 mmol/L L-glutamine, pH 7.4. Subsequently, analyses of OCR and EACR were performed in a Seahorse analyzer. The OCR and EACR values were obtained during baseline (prior to addition of any Mito Stress Test substances) and after the addition of 0.5 μmol/L oligomycin, 1 μmol/L FCCP, and 0.5 μmol/L rotenone + 0.5 μmol/L antimycin A. Background values only from media were subtracted, and data were corrected by subtracting nonmitochondrial respiration (measured after the injection of rotenone + antimycin A) from all measured OCR values.

Intracellular reactive oxygen species (ROS) of NK cells were quantified using Fluorometric Intracellular ROS Kit (Sigma-Aldrich) according to the manufacturer's instructions. Briefly, isolated NK cells from WT and littermate Cd47^−/− mice were subjected to Seahorse Mito Stress analysis with ATP synthase inhibitor oligomycin (0.5 μmol/L), mitochondrial uncoupler carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone (FCCP; 1 μmol/L), and complex I + II inhibitors rotenone (0.5 μmol/L) + antimycin A (0.5 μmol/L). Cells were then incubated with the ROS detection reagent, and the red fluorescence intensity was measured on a flow cytometer.

RNA was purified using the RNeasy Microkit (Qiagen) or TRIzol (Ambion, a part of Life Technologies) following the manufacturer's instructions. RNA content was quantified by NanoDrop 8000 (Thermo Fisher Scientific), and 500 ng of RNA was reverse transcribed to cDNA using Thermo Scientific Maxima First-Strand cDNA Synthesis Kit (cat. #K1641). Quantitative real-time PCR was performed with SYBR Green using primers for specific genes (Supplementary Table S1) using at least three replicates per group and analyzed on CFX96 Real-Time System (Bio-Rad). Relative transcript abundance was determined by using the ΔΔC_t or ΔC_t method after normalization with β-Actin and Gapdh. All samples were run in triplicate. Error bars represent standard error of mean (SEM).

Tumors were cut into small pieces and enzymatically dissociated with Collagenase/Dispase (Roche; cat. #269638; final concentration 1 mg/mL) and DNase 1 (Sigma; cat. #D4527; final concentration 100 μg/mL). The suspension was centrifuged at 50 × g for 5 minutes to separate tissue debris, and supernatant cell suspension was filtered through a 70-μm strainer, followed by centrifugation with 330 × g for 5 minutes. Red blood cells (RBC) were lysed using ACK buffer, and cells were suspended in FACS buffer followed by staining or sorting.

Single-cell suspensions from organs and tissues were stained with optimized antibody dilutions for 15 minutes at 4°C (Supplementary Table S2). Antibodies were directly conjugated to Brilliant Ultraviolet (BUV)395, Brilliant Violet (BV)786, BV711, BV650, BV605, Pacific Blue (PB), fluorescein isothiocyanate (FITC), phycoerythrin (PE), PE-Cy5.5, PE-Texas Red, peridinin–chlorophyll–protein complex (PerCP)-Cy5.5, PE-Cy7, allophycocyanin (APC), APC-Alexa 700, or biotin. Biotinylated antibodies were revealed with streptavidin APC-eFluor780. Cells were resuspended in FACS buffer (1% BSA + 0.01% NaN₃ in PBS1x, filtered) and incubated with rat plus rabbit serum, followed by incubation with antibody cocktail against surface molecules. For intracellular staining, cells were fixed (IC Fixation Buffer, eBioscience) and permeabilized (Permeabilization Buffer 10×, eBioscience). Dead cells were excluded through DAPI uptake or Aqua live/dead staining. Doublets were excluded through forward scatter–height by forward scatter–width and side scatter–height by side scatter–width parameters. Cells were acquired on an LSRFortessa SORP (BD Biosciences), and data were analyzed using FlowJo (Tree Star). Cell sorting was performed on a FACSAria II (BD Biosciences), and cell purity after sorting was >99%.

RNA sequencing (RNA-seq) library preparation and sequencing were performed as described previously (11). Briefly, Lin⁻NK1.1⁺NKp46⁺ cells were sorted from spleens of naïve and B16F10 tumor–bearing mice. Total RNA was isolated from sorted cells using TRIzol, and DNase treatment was performed using the RNase-free DNase set (Qiagen). RNA quality was checked using an Agilent 2100 Expert bioanalyzer (Agilent Technologies). RNA samples falling below the threshold of 8.0 (scored from 1–10) were omitted from the study. Ribosomal RNA was removed using biotinylated, target-specific oligos combined with Ribo-Zero rRNA removal beads. The RNA was fragmented and copied into first-strand cDNA, followed by second-strand cDNA synthesis using DNA polymerase I and RNase H. The resulting double-strand cDNA was used as the input to a standard Illumina library prep with end repair, adapter ligation, and PCR amplification being performed. The final purified product was quantitated by qPCR before cluster generation and sequencing. Paired-end sequencing (2 × 150 bp) of stranded total RNA libraries was performed on one HiSeq run with Illumina HiSeq3000/4000 chemistry pair end sequencing. The samples have 99 to 148 million pass filter reads, with a base call quality of above 92% of bases with Q30 and above.

HiSeq Real-Time Analysis software (RTA 1.18) was used for processing image files. Illumina bcl2fastq1.8.4 was used to demultiplex and convert binary base calls and qualities to fastq format. The sequencing reads were trimmed of adapters and low-quality bases using Trimmomatic (version 0.30), and the trimmed reads were mapped to mouse reference genome (GRCm38/mm10) and Gencode annotation M9 using STAR (version 2.5) with two-pass alignment option. RSEM (version 1.2.22) was used for transcript quantification.

The reference genome and annotation from GENECODE Mouse (mm10) was used for further downstream analyses. An in-house pipeline (Pipeliner: https://github.com/CCBR/Pipeliner) was used for data analysis. Briefly, to remove the adapter sequences, cutadapt/1.14 was used followed by fastQ Screen v0.9.3 to screen the library for its composition. For quality control, the tool fastqc/0.11.5 was used, and the spliced transcripts alignment to a reference genome and counting the number of reads per gene were performed by STAR/2.5.2b. The mapped reads in the genomic features were counted by subread/1.5.1, and the distribution of these reads was calculated by rseqc/2.6.4 along with the estimation of strandness of the reads. The library complexity was estimated by preseq/2.0.3. To mark duplicate reads and estimate the gene coverage, the program picard/1.119 was used, and for the overall statistics of input data sets, samtools/1.5 was used. For differential expression analysis, the low abundant gene threshold was set to include genes with ≥0.5 counts per million in at least ≥ 4 samples. In order to generate the gene lists based on different contrast, the R Bioconductor package DESeq2 was used. Principal component analysis (PCA) was performed using Partek Genomics Suite (v 6.5), and gene set enrichment analysis (GSEA) was performed using software provided by the Broad Institute. RNA-seq data are deposited on the NCBI Sequence Read Archive (GSE113876).

Statistical analysis of groups with limited variance was performed using GraphPad Prism 7 (Version 7.01). Comparison between two groups was done via two-sided Student t test. Differences with a P value less than 0.05 were considered statistically significant. Error bars indicate SEM unless otherwise indicated. cBioPortal tools were used to calculate P values for TCGA data.

Consistent with the hypothesis that CD47 protects tumor cells from innate immune surveillance, elevated expression of CD47 in some solid tumors and hematologic malignancies is correlated with a poorer prognosis (24–26). However, analysis of CD47 mRNA expression data in TCGA revealed the opposite for a set of 469 melanomas with RNA-seq data, including 461 with survival data (Fig. 1A). Segregating the melanoma RNA-seq data with a mean cutoff showed that patients with melanoma with tumor CD47 mRNA expression above the mean had significantly longer overall survival than those with expression below the mean (111 months and 62.7 months median survival, respectively, log-rank test P = 0.0017). The 65 patients with CD47 expression >1 SD above the mean showed no further improvement (median survival of 103.2 months, log-rank test P = 0.14). However, the 85 cases with CD47 mRNA expression more than 1 SD below the mean had a shorter median overall survival of 44.5 months (P = 6.8 × 10⁻⁶ compared with the remaining 374 tumors). Disease-free survival was also significantly shorter in the cases with the lowest CD47 expression (36 months vs. 56 months, log-rank test P = 0.002; Supplementary Fig. S1A). Thus, increased CD47 gene expression in human melanomas is dose dependently associated with improved survival.

Because these data were inconsistent with the prevailing “don't eat me” hypothesis, a prediction that tumors with elevated CD47 have poorer survival because they are resistant to phagocytosis by macrophages, we examined coexpressed genes in the TCGA data to identify potential immune mechanisms through which increased intratumoral CD47 mRNA expression could improve melanoma survival. Among the relevant gene sets examined, we identified the highest significance for NK cell–related genes (Fig. 1B). Because significant correlations between expression of these genes and CD47 were generally lacking in other cancers (Supplementary Fig. S1B), this association appears to be specific for melanoma. Using reported gene signatures for naïve versus activated NK cells (27), the highest correlations were observed for an early NK effector signature, and greater than mean expression of one or more of these signature genes was strongly correlated with increased overall survival (148 months vs. 48 months; Fig. 1C, P = 6.9 × 10⁻⁹). Lack of correlation with overall survival in other cancers indicated that the relationship between NK early activation gene expression and outcome is also specific for melanomas (Supplementary Fig. S1C).

A strong positive correlation between CD47 and IL15 mRNA expression in human melanomas suggested that increased expression of this cytokine could contribute to activation of NK cells in melanomas that express elevated CD47 (Fig. 1D). Consistent with this hypothesis, IL15 mRNA expression greater than the mean was associated with improved overall survival (P = 1.5 × 10⁻⁸) and disease/progression-free survival for the patients with melanoma (P = 2.3 × 10⁻³; Supplementary Fig. S1D and S1E).

CD47 is an inhibitory receptor for thrombospondin-1 on vascular cells and T cells (28–31). Interaction between CD47 and thrombospondin-1 inhibits T-cell receptor (TCR) signaling in T cells and VEGF-stimulated endothelial cell proliferation (9, 30, 32). Similarly, a 2 nmol/L concentration of thrombospondin-1, which is sufficient to engage CD47, but not other known thrombospondin-1 receptors, significantly reduced proliferation of isolated murine NK cells as assayed by Ki-67 staining (Fig. 2A). Further, thrombospondin-1 treatment of WT NK cells, but not Cd47^−/− NK cells, inhibited basal and IL15-stimulated proliferation as assessed using an MTS assay (Fig. 2B; Supplementary Fig. S2A). Both mRNA and protein expression of CD69 and protein expression of CD11b and CD11c, which are induced upon NK activation (27), were similarly downregulated in LPS-activated NK cells treated with thrombospondin-1 (Fig. 2C and D; Supplementary Fig. S2B and S2C).

We confirmed the inhibitory activity of thrombospondin-1 using human NK cells. NK-92 cells cultured with IL2 showed significantly inhibited proliferation in the presence of thrombospondin-1. Such inhibition was reversed when CD47 was blocked using the CD47 antibody B6H12, which similarly stops inhibitory thrombospondin-1 signaling in T cells (ref. 33; Fig. 2E). Therefore, the inhibitory activity of thrombospondin-1 requires CD47 in murine and human NK cells.

The CD47 antibody miap301 similarly blocks inhibitory thrombospondin-1 signaling in murine T cells and enhances syngeneic tumor cell killing by murine CD8 T cells (29, 34). Correspondingly, treating murine NK cells with miap301 for 48 hours in culture enhanced cell-surface CD69 expression and intracellular IFNγ. However, intracellular granzyme B (GzmB) was unchanged (Fig. 3A). The response of anti-CD47 treatment was comparable to anti-NK1.1, but not to IL15, treatment in terms of CD69 expression, GzmB, and IFNγ (Fig. 3A). Lack of these responses in Cd47^−/− NK cells confirmed specificity (Fig. 3B).

Extending the in vitro enhancement of NK cell function by miap301 to study NK cell immunity in a mouse tumor model is complicated by the dual function of this antibody to block both thrombospondin-1 and SIRPα binding. The anti-human CD47 B6H12 is efficacious in multiple xenograft tumor models, which is attributed to enhancing SIRPα-dependent phagocytic clearance by macrophages, but miap301 generally does not inhibit growth of syngeneic tumors in immune-competent mice (25). Therefore, we tested the effect of miap301 on NK cell function in the syngeneic B16 model. Intratumoral anti-CD47 treatment with miap301 slowed the growth of B16 tumors compared with tumors in mice treated with isotype-matched rat Ig (Fig. 3C). Treatment of tumor-infiltrating NK cells isolated from B16 tumors treated with miap301 but not the control antibody significantly increased intracellular IFNγ after 48 hours of treatment (Fig. 3D–F). Such in vitro treatment increased intracellular IFNγ and GzmB levels in B16-infiltrating NK cells, but not in infiltrating CD8 T cells (Fig. 3G–I). Therefore, our previously reported increase in intratumoral GzmB expression following CD47 knockdown in several syngenic tumor models may be produced primarily by NK cells (6).

IL15 dose dependently increased the frequency of NK1.1⁺ cells in cultured WT and Cd47^−/− splenocytes, but the response to IL15 was greater in Cd47^−/− (Fig. 4A). Similarly, cell activation and proliferation, assessed by CD69 and Ki-67 staining, respectively, was enhanced in Cd47^−/− NK1.1⁺ cells (Fig. 4B). Intracellular GzmB staining of splenic NK cells from naïve Cd47^−/− mice tended to be higher in frequency and count but did not achieve significance (Supplementary Fig. S3A). The mean fluorescence intensity (MFI) of IFNγ and GzmB were comparable between WT and Cd47^−/− total splenocytes and NK1.1⁺ populations within the total splenocytes cultured with increasing dosages of IL15 (Supplementary Fig. S3B and S3C). NK cells isolated from the spleens of WT and Cd47^−/− mice cultured with IL15 for 24 hours exhibited downregulated CD44 and upregulated CD62L surface expression (Fig. 4C; Supplementary Fig. S3D). However, Cd47^−/− NK cells were more CD44 positive and less CD62L positive compared with the WT NK cells (Fig. 4C). An increasing trend of GzmB, T-bet, KLRG1, and TNFα was evident in IL15 treated Cd47^−/− NK cells relative to similarly treated WT NK cells (Fig. 4D; Supplementary Fig. S3E). IL15 treatment significantly upregulated Eomes and T-bet expression in Cd47^−/− NK cells relative to that in WT NK cells (Fig. 4E). However, in vitro cytotoxic activities of WT and Cd47^−/− NK cells, both with and without IL15 treatment, were comparable when cocultured with target B16 melanoma cells (Supplementary Fig. S3F).

After continuous culture with IL15 for 72 hours, a decreasing trend of IFNγ and GzmB protein labels was evident in Cd47^−/− total splenocytes relative to WT counterparts (Supplementary Fig. S4A). Analysis of similarly cultured Cd47^+/− NK cells revealed that the decrease of Ifng and Gzmb transcripts was CD47 dose dependent (Supplementary Fig. S4B).

Global transcriptome analysis of WT and Cd47^−/− NK cells from spleens of naïve mice identified 2,585 genes to be significantly upregulated (P < 0.05) in naïve Cd47^−/− versus WT NK cells, including the effector genes Ifna2 (IFNα2), Ifna4 (IFNα4), Ifna13 (IFNα13), Gzmc (granzyme C), Il2ra (CD25), Mapk13 (MAPK13), and Jun (c-Jun; Supplementary Fig. S5A; Supplementary Table S3). As expected, Cd47 was significantly decreased in Cd47^−/− NK cells (Supplementary Fig. S5A; Supplementary Table S3). An unsupervised GSEA showed upregulation of early effector, sustained effector, interferon, and proliferation gene signatures, whereas naïve and memory gene signatures were downregulated in naïve Cd47^−/− NK cells compared with WT counterparts (Supplementary Fig. S5B; Supplementary Table S4).

Pathway analysis showed altered responses of NK cells in the absence of Cd47. An upregulation of immune-related pathways, including immune response (G:0002376, GO:0002682, GO:0050776, R-HAS-168249, and GO:0006955), cytokine production, and signal transduction (go:0007165; Supplementary Fig. S5C), as well as significant downregulation of NK-mediated cytotoxicity and MAPK, EGFR, and DAP12 signaling pathways were evident in Cd47^−/− NK cells (Supplementary Fig. S5C). Cytoscape functional and molecular interaction networks in NK cells revealed that the immune- and inflammation-related genes were significantly altered upon Cd47 depletion (Supplementary Fig. S5D).

CD47 controls mitochondrial homeostasis and metabolism and, in T cells, apoptosis pathways involving the mitochondrial regulators Drp1 and BNIP3 (32, 35–37). Mitochondrial electron transport can generate ROS that control intrinsic apoptosis pathways (38). Global gene-expression analysis demonstrated that the top upregulated pathways in Cd47^−/− NK cells include metabolic regulation and ion transport, whereas the top downregulated pathways include mitochondrial protein regulation and mitochondrial organization (Supplementary Fig. S5C). To investigate whether CD47 regulates mitochondrial and glycolytic metabolism in NK cells, we measured the mitochondrial OCR and ECAR in NK cells subjected to mitochondrial stress. Sequentially blocking key enzymes of the respiratory chain of mitochondria using oligomycin, FCCP, and a mix of rotenone and antimycin A was performed to measure ATP production, maximum respiration, and nonmitochondrial respiration, respectively. Basal mitochondrial flux assessed by OCR values was similar for WT and Cd47^−/− NK cells. In contrast, basal glycolytic flux assessed by ECAR values was higher in Cd47^−/− NK cells (Fig. 5A and B). The ATP turnover values in WT and Cd47^−/− NK cells were comparable (56.6 ± 19.4 vs. 67.2 ± 0.3). However, the proton leak values in Cd47^−/− NK cells (38.7 ± 13.5) were higher than in WT NK cells (14.5 ± 0.5). In addition, the spare respiratory capacity in Cd47^−/− NK cells (191.9 ± 20.7) was higher than that in WT NK cells (151.1 ± 9.2). Therefore, WT and Cd47^−/− NK cells differ in glucose metabolism and respond differently to inhibition or uncoupling of mitochondrial respiratory pathways.

We measured ROS in WT and Cd47^−/− NK cells after 1 and 24 hours of cell culture with oligomycin, FCCP, rotenone, and antimycin A. The frequency of intracellular ROS-positive Cd47^−/− NK cells was higher than in WT NK cells subjected to 1 hour of mitochondrial stress (Fig. 5C). Incubation for 24 hours with the mitochondrial stressors upregulated intracellular ROS accumulation both in WT and Cd47^−/− NK cells, with a significantly higher mean intracellular ROS intensity in Cd47^−/− NK cells than in WT NK cells (Fig. 5C).

The higher metabolic activity and mitochondrial stress responses of CD47-deficient NK cells results in accumulation of more intracellular ROS, suggesting that CD47 deficiency may increase NK cell apoptosis. Annexin V staining of untreated NK cells from WT and Cd47^−/− mice was comparable. However, mitochondrial stress induced by treating NK cells with oligomycin, FCCP, rotenone plus antimycin A for 24 hours resulted in higher annexin V–positive staining of Cd47^−/− NK cells (Fig. 5D–F).

We previously reported that combining antisense CD47 morpholino knockdown and local tumor irradiation or local irradiation of B16 tumors in Cd47^−/− mice significantly delays the growth of B16 melanoma relative to irradiation alone in immune-competent C57BL/6 mice (6, 39). However, CD47 knockdown in the absence of irradiation only minimally affected B16 tumor growth (6, 39). Implanted B16 tumors grew faster in Cd47^−/− than in WT mice (Fig. 6A–C).

B16 cells elicit a weak cytotoxic T-cell response due to downregulated MHC class I expression (40), suggesting that the reported CD8 T cell–dependent tumor ablation could also involve NK cells. Frequencies of NK1.1⁺ and NKp46⁺ NK cells significantly dropped within the spleens of Cd47^−/− mice on day 15 after B16 melanoma implantation (Fig. 6D–G). Flow-cytometric analysis of the NK1.1⁺ compartment in spleens showed a significant increase of CD44 and decrease of CD62L levels in Cd47^−/− NK cells compared with WT counterparts (Fig. 6H and I). Reduction of the CD62L level in Cd47^−/− NK cells following in vitro stimulation (Fig. 4C) as well as in tumor-bearing mice indicates reduced functionality of Cd47^−/− NK cells. Concomitantly, increased expression of the exhaustion marker TIGIT and the ATPase CD39 within the CD62L⁺ population of CD47-deficient NK cells further indicated an exhausted and apoptotic phenotype (Fig. 6J and K). Eomesodermin (Eomes) is the master transcriptional regulator of NK cell development, maturation, and function (41) and is downregulated in exhausted NK cells in leukemic mice (42). Similarly, splenic NK cells from Cd47^−/− tumor–bearing mice exhibited a decreasing trend of Eomes (Fig. 6L). CD127 expression on NK cells correlates with lower GzmB level and a defect in activation (43), and the CD127 level was significantly increased in CD47-deficient NK cells of tumor-bearing mice (Fig. 6M). KLGR1 expression differentiates effector and memory NK cells (44). However, KLRG1 expression did not differ between B16-bearing WT and Cd47^−/− mice at day 15 (Fig. 6N). Tumor-experienced NK cells express significantly higher levels of PD-1 and PD-L1, suggesting NK exhaustion as well as inhibitory immune regulation (45, 46). Upregulation of PD-1 was also associated with decreased degranulation and impaired cytokine production (45, 47). Consistent with these studies, Cd47^−/− NK cells expressed significantly higher levels of PD-1 and PD-L1 and significantly less intracellular GzmB compared with WT NK cells (Fig. 6O–Q).

We performed global transcriptome analysis of intratumoral and splenic NK cells at day 15 after implantation of B16 melanoma in WT and Cd47^−/− mice. Intratumoral NK cells from WT and Cd47^−/− mice showed a low RNA integrity value (Supplementary Fig. S6A) and high divergence by PCA, which precluded further analysis. Therefore, we focused on systemic changes in splenic NK cells. Delineation of population relationships with the three most informative principal components showed segregation of the population into four discrete clusters (Fig. 7A). Comprehensive pairwise comparison of NK cells showed upregulation of around 3,000 genes in WT tumor–bearing compared with WT-naïve mice. However, many fewer genes were upregulated in Cd47^−/− NK cells (total number = 705) than in WT NK cells (total number = 3,012) from tumor-bearing mice (Fig. 7B; Supplementary Table S3). Differential upregulation of inhibitory genes, including Cish (CIS), Mmp9 (MMP-9), Lag3 (CD223), Havcr2 (Tim-3), Tigit (TIGIT), and Pdcd1 (PD-1), and downregulation of activation genes, including Sell (CD62L), Mcam (CD146), Klrk1 (NKG2D), and Klrc2 (NKG2C), were evident in NK cells from tumor-bearing Cd47^−/− mice relative to WT counterparts (Fig. 7C; Supplementary Fig. S6B–S6D).

As observed in NK cells from naïve mice (11), significant negative enrichments of naïve and memory NK cell signatures were evident in Cd47^−/− NK cells from tumor-bearing mice, but sustained effector and proliferation gene signatures were positively enriched in Cd47^−/− NK cells from tumor-bearing mice (Fig. 7D; Supplementary Table S4). Consistent with our earlier observation (Fig. 5D–F), a proapoptotic gene signature (Qiagen GeneGlobe: Apoptosis and Cell Death, species = mouse) was also upregulated in splenic NK cells from tumor-bearing Cd47^−/− mice (Fig. 7D). Similarly, pathway analysis revealed upregulation of gene ontologies involved in negative regulation of cell-cycle phase transition (GO:0045786 and GO:0010948), apoptosis (R-HSA-109581, GO:0042981), programmed cell death (R-HSA-5357801), intrinsic apoptotic signaling (GO:0008630) and downregulation of signal transduction (GO:0009966), TCR signaling (cd8tcrpathway), response to endogenous stimulus (GO:0009719), mTOR signaling (hsa04150), and response to hypoxia (GO:0001666) pathways in Cd47^−/− NK cells from tumor-bearing mice (Supplementary Fig. S7). These data suggest a globally impaired NK cell immune response in B16 melanoma–bearing Cd47^−/− mice.

Cytotoxic NK cells are key effectors in cancer immune surveillance, transplantation rejection, and viral immunity (48). TCGA data identified positive correlations for CD47 expression with patient survival and enrichment of NK cell activation markers. Among the cancers analyzed, these correlations are unique to melanomas. Using mouse models and antibody blockade, we identified positive and negative roles for CD47 in regulating NK cell recruitment, activation, proliferation, and survival in the tumor microenvironment. CD47 in NK cells is a cell-intrinsic inhibitory signaling receptor for thrombospondin-1 that limits NK proliferation and activation, consistent with the known CD47-dependent inhibitory functions of thrombospondin-1 in T cells and vascular cells (8, 30). Thrombospondin-1 inhibits TCR-induced proliferation and CD69 induction in a CD47-dependent manner in T cells (8, 30) and inhibited LPS-induced Ki-67 and CD69 expression in NK cells. Treatment with a CD47 blocking antibody reversed thrombospondin-1 inhibition in a human NK cell line, inhibited tumor growth in mice bearing B16 melanoma, and increased tumoral NK cell recruitment and expression of GzmB and IFNγ.

Conversely, we observed a CD47-dependent increase of exhaustion markers in splenic NK cells of melanoma tumor-bearing mice. Immune cell exhaustion has been associated with chronic infection and cancer (49, 50), and sustained infection and tumor growth may lead to NK cell exhaustion (48, 51–55). Downregulation of NK-activating receptors, including NKG2D, CD16, CD146, and NCRs and/or upregulation of certain NK inhibitory receptors such as PD-1, LAG3, TIM-3, or TIGIT are hallmarks of NK cell exhaustion (56). The activating receptor NKG2D is frequently downregulated on NK cells in patients with cancer as well as chronic virus-infected patients (27, 57). Similarly, the inhibitory receptor Tim-3 was upregulated in patients with cancer and chronic hepatitis B and significantly decreased cytotoxicity of peripheral NK cells (58, 59). The cytokine-inducible SH2-containing protein (CIS) has been identified as a critical negative regulator in NK-mediated tumor immunity (60). CIS is a key suppressor of IL15 signaling, promoting NK cell death and limiting proliferation, IFNγ production, and cytotoxicity toward B16 melanoma (60). Similarly, Mmp9 mediates NK cell dysfunction and augments tumor resistance to NK cytotoxicity within the tumor microenvironment (61, 62). Phenotypic changes and altered expression of these genes consistent with NK exhaustion and dysfunction were found in CD47-deficient NK cells isolated from the spleen of B16-bearing mice, indicating that CD47 systemically protects NK cells from exhaustion. However, our data do not exclude additional contributions of previously reported altered angiogenic and adaptive immune functions in Cd47-deficient mice to the faster growth of B16 tumors.

CD47 regulates NK cell transcriptional responses to activating stimuli, and its absence results in a defect in mitochondrial metabolism. CD47 is a known regulator of mitochondrial homeostasis and metabolism (32, 37). Consistent with those studies, Cd47^−/− NK cells exhibit an increased spare respiratory capacity. The increased mitochondrial proton leak in NK cells and increased intracellular ROS production in Cd47^−/− NK cells subjected to mitochondrial stress is consistent with the observed apoptotic phenotype. Induction of apoptosis in cytotoxic NK cells after activation is an essential mechanism to prevent autoimmunity (63–65). Mitochondria release proapoptotic factors including ROS and cytochrome-c that activate apoptosis programs in cytolytic lymphocytes, CD8 T cells, and NK cells during the contraction phase (66–68). A resulting systemic depletion of NK cells in Cd47^−/− mice may contribute to the augmented chronic LCMV infection (11) and B16 melanoma growth. Thus, CD47 has a cell-intrinsic protective function in NK cells that sustains their survival under chronic stress conditions.

Integrins regulated by CD47 play positive roles in T-cell migration and in immune synapse formation (69). Further studies are required to delineate whether defective integrin activation contributes to the impaired NK responses to B16 melanoma in Cd47^−/− mice. Integrin signaling also limits apoptosis, which could contribute to the apoptotic signature of Cd47^−/− NK cells in challenged mice (70). Although the absence of CD47 enhances mature NK cell numbers in naïve mice (11), the increased apoptotic signature in challenged mice could result in surface markers that, in the absence of a CD47 don't-eat-me signal, could lead to increased phagocytic clearance of Cd47^−/− NK cells.

Gene set enrichment suggests that CD47 mediates distinct effects on gene expression in naïve, effector, and memory NK cells. Furthermore, because all immune cell types express CD47, we do not expect that NK cells exclusively mediate the observed functions of CD47 in the tumor microenvironment that regulate melanoma growth.

Our findings have implications for the ongoing clinical development of humanized CD47 antibodies and other biologics that target CD47. Most of the preclinical tumor models supporting these trials were performed using xenogeneic NOD/SCID mice that lack T and B cells but have NK cells and macrophages. Unlike macrophages, NK cells do not express SIRPα but express more CD47 than other lymphocytes (11). If these experimental therapeutics also disrupt the thrombospondin-1-CD47 interaction, they might promote the antitumor activity of NK cells. CD47 antibody treatment could thereby promote activation of systemic and tumor-associated NK cells. The resulting enhanced production of GzmB and IFNγ by tumor-infiltrating NK cells could contribute to the antitumor efficacy of experimental CD47 therapeutics by increasing direct NK cytotoxicity in addition to the known effects on tumor-associated macrophages.

In summary, these results identify important roles of CD47 in NK cell immune responses to human and murine melanomas. Some of the observed regulation of NK cell activities involves CD47 functioning as an inhibitory signaling receptor for thrombospondin-1, but additional CD47 signaling in NK cells may be mediated by its lateral interactions with integrins and other signaling receptors. Given the essential role of NK cells in tumor immunity, additional studies are needed to define how experimental therapeutics affect CD47–SIRPα versus thrombospondin-1–CD47 signaling in NK cells and to assess their direct effects on antitumor activities of NK cells. Understanding how CD47 signaling in NK cells affects tumor cell killing when combined with cytotoxic or targeted agents may lead to more effective application of these therapeutics.

A.L. Schwartz is Chief Executive Officer at and has ownership interest (including patents) in Morphiex Biotherapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P.R. Nath, A.L. Schwartz, D.D. Roberts

Development of methodology: P.R. Nath, D. Pal-Nath, A.L. Schwartz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.R. Nath, A.L. Schwartz

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.R. Nath, D. Pal-Nath, A. Mandal, M.C. Cam, A.L. Schwartz, D.D. Roberts

Writing, review, and/or revision of the manuscript: P.R. Nath, D. Pal-Nath, M.C. Cam, A.L. Schwartz, D.D. Roberts

Study supervision: D.D. Roberts

Other (bioinformatic data analysis): A. Mandal

The authors thank Dr. Ferenc Livak in the NCI flow cytometry core for assisting with cell analysis and sorting. The authors thank the Office of Science and Technology Resources sequencing facility for performing RNA sequencing. The authors also thank Mr. John M. Sipes for technical support and the rest of the members of Roberts lab for valuable discussions. This work was supported by the Intramural Research Program of the NCI (D.D. Roberts).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.