Engineered CD47 protects T cells for enhanced antitumour immunity – Nature

Cell lines

The Nalm6 B-ALL cell line was provided by D. Barrett and retrovirally transduced to express GFP and firefly luciferase. 143B osteosarcoma cells were acquired from the American Type Culture Collection (ATCC) and then retrovirally transduced with human CD19. The CHLA-255 neuroblastoma line was obtained and provided by R. Seeger and retrovirally transduced with GFP and firefly luciferase. MG63.3 cells were provided by C. Khanna and retrovirally transduced with GFP and firefly luciferase. D425 cells were provided by S. Chesier and retrovirally transduced to express GFP and firefly luciferase. Nalm6 and MG63.3 cells were originally obtained from ATCC. D425 cells were originally obtained from Sigma-Aldrich. A375 melanoma cells and Jurkat cells (clone E6-1) were obtained from ATCC. The 293GP retroviral packaging line was provided by the Surgery Branch (National Cancer Institute, National Institutes of Health). HEK293T lentiviral packaging cells were obtained from ATCC. Expi293 protein production cells were obtained from ATCC.

D425 cells were maintained in serum-free medium supplemented with B27 (Thermo Fisher Scientific), EGF, FGF (Shenandoah Biotechnology), human recombinant LIF (Millipore) and heparin (StemCell Technologies). Nalm6, 143B, A375, MG63.3, CHLA-255 and Jurkat cells were cultured in RPMI-1640 (Gibco). 293GP and HEK293T cells were cultured in DMEM (Gibco). Expi293 cells were cultured in Expi293 medium (Thermo Fisher Scientific). Cell line culture medium was supplemented with 10% FBS, 10 mM HEPES, 2mM l-glutamine, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin (Gibco), except for the Expi293 medium. Short tandem repeat DNA profiling of all cell lines was conducted once per year (Genetica Cell Line testing). All cell lines were routinely tested for mycoplasma. Cell lines were cultured at 37 °C in a 5% CO₂ environment.

Source of primary human T cells and macrophages

Buffy coats from healthy donors were purchased from the Stanford Blood Center under an IRB-exempt-protocol. Leukopaks from healthy donors were purchased from StemCell Technologies. Primary human T cells were purified by negative selection using the RosetteSep Human T cell Enrichment kit (StemCell Technologies) and SepMate-50 tubes. T cells were cryopreserved at 2 × 10⁷ cells per ml in CryoStor CS10 cryopreservation medium (StemCell Technologies) until use. Primary peripheral monocytes were purified through successive density gradients using Ficoll (Sigma-Aldrich) and Percoll (GE Healthcare). Monocytes were then differentiated into macrophages by 7–9 days of culture in IMDM + 10% AB human serum (Life Technologies).

Viral vector construction

All DNA constructs were visualized using SnapGene software (v.6.0.2; Dotmatics). All retroviral constructs were cloned into the MSGV1 retroviral vector³⁰. The following CAR and TCR constructs used in this study were previously described: B7H3-BBζ³¹, GD2-BBζ¹⁵, CD19-BBζ³², HER2-BBζ³², CD19-28ζ³², HA-28ζ³³ and NY-ESO-1³⁴. B7H3-BBζ was previously generated by fusing, from N to C terminus, a human GM-CSF leader sequence, scFv derived from MGA271 in the VH-VL orientation and (GGGS)₃ linker sequence, CD8α hinge and transmembrane sequence, and human 4-1BB and CD3ζ intracellular signalling domains. GD2-BBζ, HER2-BBζ and CD19-BBζ were generated previously by cloning scFvs derived from 14G2A, 4D5 and FMC63 antibodies, respectively, into the B7H3-BBζ vector in place of the MGA271 scFv. CD19-28ζ was generated previously by replacing the 4-1BB domain in CD19-BBζ with the intracellular signalling domain of human CD28. HA-28ζ was generated previously by replacing the FMC63 scFv with the 14G2a scFv containing a point mutation (E101K) followed by a spacer region derived from the CH2CH3 domains of IgG1. PIP-28ζ and PIP-BBζ were generated by replacing the FMC63 scFv with the 2.5F knottin^16,35 followed by a Flag-tag sequence (DYKDDDDK) in the CD19-28ζ and CD19-BBζ vectors, respectively. The in vivo T cell activation reporter was constructed by cloning a sequence containing firefly luciferase into the pGreenFire1-NF-κB lentiviral vector (System Biosciences) under the NF-κB responsive promoter³². CD47 vectors were generated by inserting codon-optimized CD47 sequences (variant and WT) in place of the CD19-BBζ sequence. For in vivo tracking, CAR-nLuc plasmids were generated by replacing the stop codon in the CD3ζ with a sequence containing a porcine teschovirus-1 2A (P2A) ribosomal skipping sequence, followed by nanoluciferase³². Antares plasmids were generated by inserting the Antares sequence³⁶ in place of the CD19-BBζ sequence. The NY-ESO-1 TCR construct was generated by inserting the NY-ESO-1 α chain, followed by a P2A sequence, followed by the β chain in place of CD19-BBζ.

Virus production

Retroviral supernatant was packaged using 293GP cells and the RD114 envelope plasmid. In brief, 11 μg RD114 and 22 μg of the corresponding MSGV1 transfer plasmid were delivered to 293GP cells grown on 150 mm poly-d-lysine dishes (Corning) to 80% confluency by transient transfection with Lipofectamine 2000 (Thermo Fisher Scientific). The medium was replenished every 24 h. Virus production was performed side by side for comparable CAR, TCR and CD47 constructs. The retroviral supernatant was collected 48 and 72 h after transfection. The supernatants from replicate dishes were pooled, centrifuged to deplete cell debris and stored at −80 °C until use. Third-generation, self-inactivating lentiviral supernatant was similarly produced with HEK293T cells using 7 μg pMD2.G (VSVg) envelope, 18 μg pMDLg/pRRE (Gag/Pol), 18 μg pRSV-Rev and 20 μg of the corresponding transfer plasmids. All of the constructs were retroviral, except for the T cell NF-κB activation construct, which was lentiviral.

CAR T and TCR T cell manufacturing

At day 0, primary human T cells were thawed and activated with anti-CD3/CD28 human T-Expander Dynabeads (Thermo Fisher Scientific) at a 3:1 or 1:1 bead to cell ratio. On day 2, virus-coated culture plates were prepared on non-tissue-culture-treated 12-well plates that had been precoated with RetroNectin (Takara Bio) according to the manufacturer’s instructions, by incubating with 1 ml of retroviral supernatant (2 × 10⁷–5 × 10⁷ TU ml⁻¹) and centrifugation at 3,200 rpm and 32 °C for 2 h. The supernatant was subsequently aspirated off of the wells and 0.5 × 10⁶–1 × 10⁶ T cells were added in 1 ml of T cell medium comprising AIM V (Thermo Fisher Scientific), 5% FBS, 100 U ml⁻¹ penicillin (Gibco), 100 mg ml⁻¹ streptomycin (Gibco), 2 mM l-glutamine (Gibco), 10 mM HEPES (Gibco) and 40 U ml⁻¹ rhIL-2 (Peprotech). After addition of the T cells, the plates were gently centrifuged at 1,200 rpm for 2 min then incubated for 24 h at 37 °C under 5% CO₂. This transduction process was repeated on day 3 and day 4 (if necessary). Dynabeads were removed on day 4 or day 5 by magnetic separation. Cells were maintained between 0.4 × 10⁶ and 2 × 10⁶ cells per ml and expanded until day 10–12. Typically, T cells were transduced with CAR or TCR and Antares (if used) on day 2, and then CD47 variants on days 3 and 4.

CRISPR–Cas9 KO of CD47 and AAVS1

Ribonucleoprotein (RNP) was prepared using synthetic sgRNA with 2′-O-methyl phosphorothioate modification (Synthego) diluted in TE buffer at 120 μM. A total of 5 μl sgRNA was incubated with 2.5 μl duplex buffer (IDT) and 2.5 μg Alt-R Streptococcus pyogenes Cas9 Nuclease V3 (IDT) for 30 min at room temperature. Reactions (100 μl) were assembled with 5 million T cells or Jurkat cells, 90 μl P3 buffer (Lonza) and 10 μl RNP. Cells were pulsed with protocol EO115 using the P3 Primary Cell 4D-Nucleofector Kit and 4D-Nucleofector System (Lonza). Cells were recovered immediately with warm medium for 6 h before transduction with CAR or TCR. Cells were electroporated with RNP on day 2 after thaw and transduced later the same day. Guide sequences were as follows: CD47, 5′-AUGCUUUGUUACUAAUAUGG-3′; AAVS1, 5′-GGGGCCACUAGGGACAGGAU-3′.

Flow cytometry analysis of mammalian cells

Cells were washed with FACS buffer (2% FBS in PBS) before staining. Staining was performed in FACS buffer for 30 min at 4 °C. In certain experiments, cells were first stained with Fixable Viability Dye eFluor 780 (eBioscience, 1:2,000) in PBS for 10 min at room temperature before washing with FACS buffer and staining with other antibodies. After staining, cells were then washed once with FACS buffer and analysed on the BD Fortessa system. FACSDiva (v.8.0.1; BD) software was used for data collection and FlowJo software (v.10.8.1; BD) was used for data analysis (gating strategies are shown in Supplementary Fig. 2).

Recombinant B7H3-Fc and HER2-Fc (both R&D systems, 1:400 dilution) were used to detect B7H3 and HER2 surface CAR, respectively. Likewise, anti-FMC63 idiotype antibody (Genscript, 1:400) was used to detect CD19 CARs, while anti-14G2a idiotype antibody (National Cancer Institute, 1:400) was used to detect GD2 and HA CARs. CAR detection reagents were fluorescently labelled using the DyLight 650 Microscale Antibody Labelling Kit (Thermo Fisher Scientific). Anti-DYKDDDDK tag (Flag tag, APC, L5, BioLegend, 1:400) was used to detect the PIP CAR. NY-ESO-1 TCR was detected with antibodies specific for Vβ13.1 (APC, H131, BioLegend, 1:100), the beta chain of the NY-ESO-1 TCR. CD47 was detected with B6H12^37,38 (BV711 and PE, B6H12, BD, 1:100; APC, B6H12, Invitrogen, 1:100; or unlabelled, Bio X Cell, concentrations are listed in the figures), TJC4²⁰ (unlabelled, produced in-house, concentrations are listed in the figures), Hu5F9¹² (unlabelled, produced in-house, concentrations are listed in the figures), CV-1-Fc (unlabelled, ALX Oncology, concentrations are listed in the figures), mSIRPα-Fc (unlabelled, Sino Biological, concentrations are listed in the figures) or hSIRPα-Fc (unlabelled, Sino Biological, concentrations are listed in the figures), followed by detection with polyclonal anti-mouse or anti-human IgG antibodies (AF488 and AF647, polyclonal, Invitrogen, 1:500). mIgG1 isotype control antibodies (unlabelled, MPOC-21, Bio X Cell, 1:100 and PE, B11/6, Abcam, 1:100) were used as controls for B6H12 staining. The following antibodies were used for detection of cell surface proteins: calreticulin (PE, FMC 75, Abcam, 1:100); human CD4 (BUV 395, SK3, BD, 1:200); human CD8 (BUV 805, SK1, BD, 1:400); human CD45 (Per-CP-Cy5.5, HI30, Invitrogen, 1:50); human CD69 (BV421, FN50, BioLegend, 1:100); human CD39 (BV605, A1, BioLegend, 1:100); human TIM3 (BV510, F38-2E2, BioLegend, 1:100); human LAG3 (PE, 3DS223H, Invitrogen, 1:100); human PD1 (PE-Cy7, J105, Invitrogen, 1:100); human CD45RA (BV785, HI100, BioLegend, 1:100); human CD62L (BV605, DREG-56, BD, 1:100); human CD3 (BUV 737, SK7, BD, 1:100); mouse CD45 (BUV 805, I3/2.3, BD, 1:100); F4/80 (BV605, BM8, BioLegend, 1:100); CD11b (APC, M1/70, BioLegend, 1:50 and BUV 395, M1/70, BD, 1:100); human CD19 (BUV 496, SJ25C1, BD, 1:100). Annexin V was detected using the eBioscience Annexin V Apoptosis Detection Kit (Invitrogen) according to the manufacturer’s instructions.

BLI analysis

Mice were administered 200 μl of 15 mg ml⁻¹ d-luciferin for firefly luciferase imaging or a 1:40 dilution of Nano-Glo substrate (Promega, diluted in DPBS) for Antares and nanoluciferase imaging by intraperitoneal injection. Images were acquired on the IVIS (Perkin Elmer) or Lago (Spectral Instruments Imaging) imaging system 4 min after injection for fLuc and 8 min after injection for nLuc/Antares using 30 s exposures and medium binning. If saturated pixels were detected in the image, an additional image was acquired using the auto-expose setting. Total flux was measured using Living Image (v.4.7.3; Perkin Elmer) or Aura (v.4.0.7; Spectral Instruments Imaging) software with a region of interest around the body of each mouse. Only non-saturated images were used for quantification of BLI. Mice were randomized before T cell administration to ensure uniform distribution of tumour burden between groups. At the end of the experiment, all of the images were collected into a single sequence on Aura and set to the same luminescence scale.

Recombinant protein cloning and production

The gWIZ vector with a BM40 signal peptide was used for protein expression. DNA encoding the Hu5F9 (magrolimab¹²) heavy chain with an hIgG1 Fc domain, Hu5F9 light chain, TJC4 (lemzoparlimab²⁰) heavy chain with an hIgG1 Fc domain and TJC4 light chain were ordered from IDT. Heavy and light chains were individually cloned into an AscI/BamHI-digested gWIZ vector using Gibson assembly. Plasmids were transfected into Expi293F cells (Thermo Fisher Scientific) at a 1:1 ratio of heavy chain:light chain using ExpiFectamine according to the manufacturer’s instructions. Then, 5 days after transfection, the supernatant was collected, adjusted to pH 8.0 and sterile-filtered. Hu5F9 and TJC4 were then purified using recombinant Protein A-Sepharose 4B (Thermo Fisher Scientific) buffer-exchanged into PBS and concentrated using Amicon Centrifugal Filters (Millipore Sigma). To assess CD47 binding, cells were stained with Hu5F9 or TJC4 and then stained with labelled anti-human secondary antibodies (AF488 or AF647, Invitrogen, 1:500). B6H12^37,38 and mIgG1 isotype control (MOPC-21) were acquired from Bio X Cell. CV-1 variants (ALX-222, CV-1-hIgG1 Fc; and ALX-90, CV-1-hIgG1 dead Fc) were acquired from ALX Oncology. Human SIRPα-mFc and mouse SIRPα-hFc were acquired from Sino Biologic.

Animal models

NSG mice (NOD.Cg-Prkdc^scidIl2rgt^m1Wjl/SzJ) were purchased from the Jackson Laboratory and bred in-house under Stanford University APLAC-approved protocols. Healthy male and female mice that were used for in vivo experiments were aged between 6 and 10 weeks at tumour or T cell engraftment and were drug naive, and not involved in previous procedures. The mice were housed in sterile cages in a barrier facility at Stanford University at 22 °C and 50% humidity under a 12 h–12 h light–dark cycle. Veterinary Services Center (VSC) staff at Stanford University monitored the mice daily. Mice were euthanized when they manifested persistent hunched posture, persistent scruffy coat, paralysis, impaired mobility, greater than 20% weight loss, if tumours significantly interfered with normal bodily functions or if they exceeded limits designated in APLAC-approved protocols of 1.70 cm in any direction. According to the recommendations of VSC staff, mice with morbidities were supported with 500 μl subcutaneous saline, diet gel (DietGel 76A, ClearH2O) and wet chow. For all experiments, no sample size calculations were performed, but group sizes were determined by experience with well-established, previously published models^31,32,34,38. Cages of mice that were previously engrafted with tumour were randomly assigned CAR T cell and anti-CD47 conditions for infusion, ensuring approximately equal distributions of tumour size between groups before treatment. Tumour engraftments and T cell infusions were performed by a technician who was blinded to treatments and expected outcomes.

143B osteosarcoma tumour model

0.5 × 10⁶ or 1 × 10⁶ 143B or 143B-CD19 cells (143B cells engineered to over-express CD19; 143B cells do not naturally express CD19) in 100 μl DPBS were injected into the tibial periosteum of six- to ten-week-old NSG male or female mice (engraftment dose indicated below for each specific study)³². Generally, 5 days after tumour implantation and after visual confirmation of tumour formation, mice were treated with HER2-BBζ CAR T cells, followed by two doses of B6H12. Tumour progression was monitored by measurement using callipers. Mice were euthanized according to the criteria described in the ‘Animal models’ section. Specifics for different iterations of the model presented are as follows:

CAR T cell + B6H12 studies (Fig. 1a, Supplementary Fig. 1a and Extended Data Fig. 1a,h): mice engrafted with 0.5 × 10⁶ 143B-CD19 cells were treated with 10 × 10⁶ Her2-BBζ CAR T cells by tail-vein injection on day 5. Mice were then treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on day 6 and day 10. T cells were quantified in the blood by flow cytometry on day 12.

PIP CAR T cell survival study (Fig. 3a): mice engrafted with 0.5 × 10⁶ 143B were treated with 10 × 10⁶ CD19-BBζ (non-tumour-targeting control), HER2-BBζ (tumour targeting control), PIP-28ζ or PIP-BBζ CAR T cells on day 5 by tail-vein injection.

PIP CAR T cell serum cytokine study (Fig. 3b): non-tumour-bearing mice or mice engrafted with 0.5 × 10⁶ 143B-CD19 were treated with 5 × 10⁶ CD19-BBζ (tumour-specific control), PIP-28ζ or PIP-BBζ CAR T cells, or mock T cells by tail-vein injection on day 4. Blood was collected for serum cytokine analysis on day 8 (4 days after CAR T cell administration).

47_E CAR T cell studies with high-dose B6H12 (Fig. 5e–g, Supplementary Fig. 1p and Extended Data Fig. 10a–h): mice engrafted with 1 × 10⁶ 143B-CD19 cells were treated with 4 × 10⁶ HER2-BBζ Antares CAR T cells with endogenous CD47 KO and overexpressing either CD47 WT (47_WT) or CD47(Q31P) (47_E), or an equivalent number of mock-Antares T cells intravenously by tail-vein injection on day 5. Mice were then treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on day 7 and day 11. T cells were quantified by nanoluciferase BLI before (day 7) and after (day 13) anti-CD47 treatment and in the blood by flow cytometry on day 14.

47_E CAR T cell studies with low-dose B6H12 (Supplementary Fig. 1q and Extended Data Fig. 10i,j): mice engrafted with 0.5 × 10⁶ 143B-CD19 cells were treated with 4 × 10⁶ HER2-BBζ CAR T cells with endogenous CD47 KO and overexpressing either 47_WT or 47_E, or an equivalent number of mock T cells intravenously by tail-vein injection on day 5. The mice were then treated twice with B6H12 (75 μg or 25 μg) or PBS by intraperitoneal injection on day 6 and day 10. T cells were quantified in the blood by flow cytometry on day 12. Only those mice that were treated with 47_E CAR T cells were evaluated for antitumour efficacy in combination with B6H12.

A375 melanoma tumour model

A total of 3 × 10⁶ A375 cells in 100 μl DPBS was injected into the flanks of NSG male or female mice aged 6–10 weeks³⁴. Generally, 7 to 14 days after tumour implantation and after visual confirmation of tumour formation, mice were treated with NY-ESO-1 TCR T cells, followed by two or three doses of B6H12. Tumour progression was monitored by measurement using callipers. Mice were euthanized according to the criteria described in the Animal Models section. Specifics for different iterations of the model presented are as follows:

Low-dose NY-ESO-1 TCR T cell + B6H12 studies (Fig. 1d,e, Supplementary Fig. 1d and Extended Data Fig. 1i): mice were treated with 2 × 10⁶ NY-ESO-1 TCR T cells or an equivalent number of mock T cells intravenously by tail-vein injection on day 9 after tumour implantation. Mice were then treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on day 10 and 15. T cells were quantified in the blood by flow cytometry on day 17.

High-dose NY-ESO-1 TCR T cell + B6H12 studies (Supplementary Fig. 1d and Extended Data Fig. 1j,k): mice were treated with 5 × 10⁶ NY-ESO-1 TCR T cells or an equivalent number of mock T cells intravenously by tail-vein injection on day 7 after tumour implantation. Mice were then treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on day 9 and 13. T cells were quantified in the blood by flow cytometry on day 16.

47_E NY-ESO-1 TCR T cell quantification studies (Fig. 5h, Supplementary Fig. 1t and Extended Data Fig. 11g,h): 7 days after tumour implantation, mice were treated with 2.75 × 10⁶ NY-ESO-1-Antares TCR T cells with endogenous CD47 KO and overexpressing 47_E, or an equivalent number of mock-Antares T cells intravenously by tail-vein injection. Mice were then treated three times with B6H12 (250 μg) or PBS by intraperitoneal injection on days 9, 11 and 14. T cells were quantified by nanoluciferase BLI before (day 9) and after (day 14) anti-CD47 treatment and in the blood by flow cytometry on day 15.

47_E NY-ESO-1 TCR T cell antitumour efficacy studies (Fig. 5i, Supplementary Fig. 1u and Extended Data Fig. 11i,j): 7 days (T cell donor experiment 1; Extended Data Fig. 11j) or 14 days (T cell donor experiment 2; Fig. 5i and Extended Data Fig. 11i) after tumour implantation, mice were treated with 1 × 10⁶ NY-ESO-1-Antares TCR T cells with endogenous CD47 KO and overexpressing 47_E, or an equivalent number of mock-Antares T cells intravenously by tail-vein injection. Mice were then treated either: three times with B6H12 (250 μg) or PBS by intraperitoneal injection on days 9, 11 and 14 (experiment 1); or twice with B6H12 (250 μg) or PBS by intraperitoneal injection on days 15 and 19 (experiment 2).

MG63.3 osteosarcoma tumour model

A total of 1 × 10⁶ MG63.3 cells in 100 μl DPBS was injected into the tibia periostea of NSG male or female mice aged 6–10 weeks³¹. Starting 15 days after tumour implantation and after visual confirmation of tumour formation, the mice were treated with 400 μg of B6H12 or PBS three times per week by intraperitoneal injection. On day 21, the mice were treated intravenously with 10 × 10⁶ GD2-BBζ or B7H3-BBζ CAR T cells or no T cells. Tumour progression was measured using digital callipers twice per week. Mice were euthanized according to the criteria described in the ‘Animal models’ section (Supplementary Fig. 1b). For T cell quantification experiments, mice engrafted orthotopically with 1 × 10⁶ MG63.3 cells were treated intravenously with 10 × 10⁶ B7H3-BBζ CAR T cells on day 15 with or without 3 doses of B6H12 treatment (400 µg per dose; intraperitoneal). Blood and tumours were collected on day 30 after tumour engraftment.

D425 medulloblastoma tumour model

Mice (aged 6–10 weeks) were anaesthetized with 3% isoflurane (Minrad International) in an induction chamber³¹. Anaesthesia on a stereotactic frame (David Kopf Instruments) was maintained at 2% isoflurane delivered through a nose adaptor. D425 medulloblastoma cells were injected at coordinates 2 mm posterior to lambda on midline and 2 mm deep using a blunt-ended needle (75 N, 26 s gauge/2 inch/point style 2, 5 μl; Hamilton). Using a microinjection pump (UMP-3; World Precision Instruments), 0.2 × 10⁶ D425-GFP-fLuc cells were injected in a volume of 3 μl at 30 nl s⁻¹. After leaving the needle in place for 1 min, it was retracted at 3 mm min⁻¹. Then, 4 days after tumour implantation and after confirmation of tumour formation by bioluminescence, mice were randomized and treated with no T cells (B6H12 only group), or 10 × 10⁶ B7H3-BBζ CAR⁺ T cells or an equivalent number of non-tumour targeting CD19-BBζ CAR⁺ T cells intravenously by tail-vein injection. Starting on day 4, the mice were also treated with 400 μg of B6H12 or PBS three times per week by intraperitoneal injection. Tumour progression was monitored by firefly luciferase BLI (Supplementary Fig. 1c). In Extended Data Fig. 1c,d, CD19-BBζ and B7H3-BBζ treatments are reproductions of previously published data³¹, included for comparison with B7H3-BBζ + B6H12, as these arms were all run simultaneously in the same experiment.

Nalm6 leukaemia tumour models

A total of 1 × 10⁶ Nalm6-GFP-fLuc cells in 200 μl DPBS was implanted by tail-vein injection into NSG male or female mice aged 6–10 weeks³². Generally, four days after tumour implantation and after confirmation of tumour formation by BLI, mice were treated with CD19-BBζ or CD19-28ζ CAR T cells, followed by doses of anti-CD47. Tumour progression was monitored by fLuc BLI measurement. Mice were euthanized according to the criteria described in the ‘Animal models’ section. Specifics for different iterations of the model presented are as follows:

High-dose CAR T cell + B6H12 studies (Supplementary Fig. 1e and Extended Data Fig. 1l,n,p,s): mice engrafted with 1 × 10⁶ Nalm6-GFP-fLuc cells were treated with B6H12 (400 μg) or PBS by intraperitoneal injection three times per week, starting on day 3. Mice were then treated with 1 × 10⁶ CD19-28ζ-nLuc CAR T cells by tail-vein injection on day 4. T cells and tumours were quantified by BLI weekly.

Low-dose CAR T cell + B6H12 studies (Supplementary Fig. 1f and Extended Data Fig. 1m–o,q,r): mice engrafted with 1 × 10⁶ Nalm6-GFP-fLuc cells were treated with 0.15 × 10⁶ CD19-28ζ-nLuc CAR T cells by tail-vein injection on day 4. Mice were treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on day 5 and 7. T cells and tumours were quantified by BLI weekly.

High-dose CAR T cell + CV-1 studies (Fig. 1f,g, Supplementary Fig. 1g and Extended Data Fig. 2a,b,e): mice engrafted with 1 × 10⁶ Nalm6-GFP-fLuc cells were treated with 1 × 10⁶ CD19-BBζ-nLuc CAR T cells by tail-vein injection on day 4. Mice were treated with CV-1-Fc (ALX-90; dead Fc; 400 μg) or PBS by intraperitoneal injection three times per week starting on day 5. T cells and tumours were quantified by BLI weekly.

Low-dose CAR T cell + CV-1 studies (Supplementary Fig. 1h and Extended Data Fig. 2c–f): mice engrafted with 1 × 10⁶ Nalm6-GFP-fLuc cells were treated with 0.1 × 10⁶ CD19-28ζ-nLuc CAR T cells by tail-vein injection on day 4. Mice were treated with CV-1-Fc (ALX-90; dead Fc; 400 μg) or PBS by intraperitoneal injection three times on days 5, 7 and 10. T cells and tumours were quantified by BLI twice weekly.

Low-dose 47_KO CAR T cell studies (Fig. 1h,i and Extended Data Fig. 2h,i): mice engrafted with 1 × 10⁶ Nalm6-GFP-fLuc cells were treated with 0.15 × 10⁶ CD19-28ζ-nLuc CAR T cells with endogenous CD47 KO (47_KO), CD47 KO with overexpression of CD47 WT (47_WT), or an equivalent number of mock T cells by tail-vein injection on day 4. Mice were treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on days 5 and 7. Tumours were quantified by BLI weekly. T cells were quantified by BLI on day 11.

Low-dose 47_E CAR T cell + B6H12 studies (Supplementary Fig. 1s and Extended Data Fig. 11e,f): mice engrafted with 1 × 10⁶ Nalm6-GFP-fLuc cells were treated with 0.15 × 10⁶ CD19-28ζ-nLuc CAR T cells with endogenous CD47 KO and overexpressing either 47_WT or 47_E, or an equivalent number of mock T cells by tail-vein injection on day 4. Mice were treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on days 5 and 7. Tumours were quantified by BLI weekly.

T cell depletion model

NSG male or female mice (aged 6–10 weeks) were implanted with 2 × 10⁶ or 5 × 10⁶ CD19-28ζ-nLuc CAR T cells with endogenous CD47 KO and overexpressing either 47_WT, 47_E or no additional protein (47_KO) by tail-vein injection (day 0). Mice were then treated twice with B6H12 (250 μg) or PBS by intraperitoneal injection on days 3 and 5. T cells were quantified by nanoluciferase BLI before (2 × 10⁶ dose, day 2; 5 × 10⁶ dose, day 3) and after (2 × 10⁶ dose, day 9; 5 × 10⁶ dose, day 7) anti-CD47 treatment and in the blood by flow cytometry (2 × 10⁶ dose, day 7; 5 × 10⁶ dose, day 6). For isotype control studies (Extended Data Fig. 1f,g), mice were implanted with 5 × 10⁶ CD19-28ζ CAR T cells by tail-vein injection (day 0) and were then treated with B6H12 (250 μg), mIgG1 isotype control (250 μg) or PBS by intraperitoneal injection on day 1. T cells were quantified in the blood by flow cytometry on day 5. Mice were euthanized according to the criteria described in the ‘Animal models’ section at the conclusion of the experiment (Supplementary Fig. 1i,m).

PIP CAR toxicity model

PIP CAR vectors were made as described in the viral vector construction section. NSG male or female mice (aged 6–10 weeks) were treated with the PIP CAR T or control CD19-BBζ, HER2-BBζ or mock T cells at the dose indicated in the figure (10 × 10⁶, 5 × 10⁶ or 2 × 10⁶ CAR T cells) by tail-vein injection. Mice in the PIP-28ζ and PIP-BBζ groups experienced rapid onset of toxicity (within 1–5 days, depending on dose) observed clinically as a hunched posture, scruffy coat, slow movement, dehydration and weight loss. Treatment-related toxicity was monitored by weight change, which was measured before T cell administration and 1–2× per week thereafter. The percentage weight change was calculated as follows: percentage weight change = ((weight at time x/initial weight) − 1) × 100. Mice died from toxicity or were euthanized if they reached 20% weight loss or showed clinical signs of severe toxicity, as described in the ‘Animal models’ section. For assessment of T cell localization and activation, CD19-28ζ-nLuc or PIP-28ζ-nLuc CAR T cells were transduced with a firefly luciferase reporter under control of an NF-κB-inducible promoter. Mice were implanted with 5 × 10⁶ CAR T cells and imaged daily with Nano-GLO substrate (nLuc; total CAR T) and luciferin (fLuc; active CAR T), with each substrate dose separated by 12 h. For organ BLI analysis, 4 days after treatment with 5 × 10⁶ CD19-28ζ-nLuc or PIP-28ζ-nLuc, the mice were injected with either Nano-GLO substrate (nLuc; total CAR T) or luciferin (fLuc; active CAR T) and euthanized 10 min after injection. Organs were collected according to standard procedures and imaged using BLI on the IVIS machine (Perkin Elmer) in six-well plates. For safety-switch models, mice were dosed with 2 × 10⁶ PIP-28ζ or CD19-28ζ CAR T cells intravenously. Mice treated with PIP-28ζ CAR T cells were administered 250 µg B6H12 or PBS over three consecutive days (days 2, 3 and 4), 2 days after CAR administration (day 0). Blood was collected for serum cytokine analysis on day 4 after CAR administration (Supplementary Fig. 1k).

CHLA-255 neuroblastoma metastatic tumour model

Six- to ten-week-old NSG male or female mice were implanted with 1 × 10⁶ CHLA-255-GFP-fLuc cells by tail-vein injection³⁸. The, 7 days after tumour implantation and after confirmation of tumour formation by BLI, mice were randomized and treated with 2 × 10⁶ B7H3-BBζ-nLuc CAR T cells with endogenous CD47 KO and overexpressing 47_WT or 47_E or an equivalent number of mock (non-transduced) T cells intravenously by tail-vein injection. Mice were then treated three times with B6H12 (250 μg) or PBS by intraperitoneal injection on days 7, 9 and 13. Tumour progression was monitored by firefly luciferase BLI. T cells were quantified by nanoluciferase BLI after anti-CD47 treatment on day 14 and in the blood by flow cytometry on day 15. Mice were euthanized according to the criteria described in the ‘Animal models’ section (Supplementary Fig. 1r).

CAR T cell GvHD model

NSG male or female mice (aged 6–10 weeks) were implanted with 1 × 10⁶ Nalm6-GFP-fLuc cells by tail-vein injection. The mice were then treated with 10 × 10⁶ CD19-BBζ CAR T cells on day 4. Half of the cohort of mice received three doses of B6H12 (250ug) over 3 days after CAR T cell administration. Mice were monitored for tumour growth by BLI and signs of GvHD, such as alopecia, dyskeratosis and weight loss^39,40. Spleens and skin were extracted surgically. Skin sections were prepared as slides and stained with H&E using the standard method (Supplementary Fig. 1l).

Isolation of T cells from spleens and tumours

Spleens and tumours were collected and mechanically dissociated using the gentleMACS dissociator (Miltenyi). Single-cell suspensions were made by passing spleens and tumours through a 70 μm cell strainer (Thermo Fisher Scientific), depleting red blood cells by ACK lysis (Quality Biological), and further filtration through flow cytometry filter tubes with 35 μm cell strainer caps (Falcon). Single-cell suspensions were then frozen in CryoStor buffer (StemCell Technologies) in liquid nitrogen, or stained and run directly on the flow cytometry system.

Quantification of T cells and cytokines from the blood

Mouse blood was collected from the retro-orbital sinus into Microvette blood collection tubes with EDTA (Thermo Fisher Scientific). Red blood cells were depleted by ACK lysis (Quality Biological), followed by two washes with FACS buffer (PBS + 2% FBS). The samples were stained and mixed with CountBright Absolute Counting beads (Thermo Fisher Scientific) before flow cytometry analysis. IL-2 and IFNγ cytokine levels in blood were quantified using LEGENDPlex immunoassays (BioLegend) according to the manufacturer’s instructions from serum collected after centrifuging blood samples at 3,000 rpm for 10 min. Negative cytokine values were set to 0.

IncuCyte tumour killing assays, cytokine analysis and T cell activation marker detection

A total of 5 × 10⁴ GFP-labelled tumour cells was cocultured with 5 × 10⁴ CAR T cells in 200 μl RPMI supplemented with 10% FBS, 10 mM HEPES, 2mM l-glutamine, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin. For conditions with B6H12, a concentration of 10 µg ml⁻¹ was used. Triplicate wells were plated in 96-well flat-bottom plates for each condition. Tumour fluorescence was monitored every 2–3 h with a ×10 objective using the IncuCyte S3 Live-Cell Analysis System (Sartorius), housed in a cell culture incubator at 37 °C and 5% CO₂, set to take 4 images per well at each timepoint. The total integrated GFP intensity was quantified using the IncuCyte basic analyzer software feature (IncuCyte S3 v.2019B Rev2; Sartorius). Data were normalized to the first timepoint and plotted as the fold change in tumour fluorescence over time. For cytokine secretion and T cell marker analysis, cocultures were set up as described above except in 96-well round-bottom plates. After approximately 24 h, the plates were centrifuged to pellet cells and 150 µl of supernatant was collected and stored at −20 °C until analysis, while the cell pellets were immediately processed for flow cytometry. IFNγ and IL-2 levels in the coculture supernatants were quantified by ELISA (Human ELISA MAX Deluxe, BioLegend) according to the manufacturer’s instructions. Negative cytokine values were set to 0. Absorbance values were measured using the Synergy H1 Hybrid Multi-Mode Reader with Gen5 software (v.2.00.18; BioTek). For analysis of T cell markers after activation by tumour cells, pellets from centrifuged plates were pooled together for triplicate wells, stained and analysed using flow cytometry. Coculture experiments were set-up using day 10 T cells.

Macrophage depletion and peritoneal lavage

NSG male or female mice (aged 6–10 weeks) were pretreated with intravenous injection with 200 µl of clodronate liposomes (Liposoma), followed by 400 µg of anti-mouse-CSF1R (AFS98; Bio X Cell) by intraperitoneal injection³⁸. Mice were treated with 400 µg of anti-CSF1R three times per week for the duration of the experiment. Then, 6 days after clodronate treatment, the mice were administered with 2 × 10⁶ CD19-28ζ-nLuc CAR T cells intravenously, followed by 250 µg B6H12 intraperitoneally on day 7. T cells were quantified by nanoluciferase BLI before (day 7) and after (day 9) anti-CD47 treatment. Peritoneal lavage was performed on day 13 with 10 ml of FACS buffer and a 25 gauge needle. Peritoneal cells were collected, washed with FACS buffer and stained, before being run on the flow cytometry system (Supplementary Fig. 1j).

Phagocytosis assay

For all flow-based in vitro phagocytosis assays, T cells and human macrophages were co-cultured at a ratio of 2:1 (for example, 100,000 T cells:50,000 macrophages) in ultra-low-attachment 96-well U-bottom plates (Corning) in serum-free RPMI (Thermo Fisher Scientific). T cells were labelled with CFSE (Invitrogen) by suspending cells in PBS (5 µM working solution) according to the manufacturer’s instructions for 20 min at 37 °C protected from light and washed twice with 20 ml of FBS-containing medium before co-culture. Cells were then either incubated alone or in the presence of anti-CD47 (B6H12; Bio X Cell) or mIgG1 isotype control (MOPC-21; Bio X Cell) at a concentration of 10 μg ml⁻¹. T cells and antibodies were incubated for 30 min in a humidified 5% CO₂ incubator at 37 °C. Plates were washed twice; human macrophages were added to the plate; and plates were incubated for 1–2 h at 37 °C. Phagocytosis was stopped by washing with 4 °C PBS and centrifugation at 1,450 rpm before the cells were stained with Live/Dead stain and anti-CD11b (APC, M1/70, BioLegend, 1:50). Assays were analysed by flow cytometry, and phagocytosis was measured as the number of CD11b⁺ and CFSE⁺ macrophages, quantified as a percentage of the total CD11b⁺ macrophages and normalized to the control condition.

For IncuCyte-based in vitro phagocytosis assays, T cells and human macrophages were co-cultured at a ratio of 2:1 (for example, 100,000 T cells:50,000 macrophages) in 96-well flat-bottom plates (Corning) in RPMI supplemented with 10% FBS, 10 mM HEPES, 2mM l-glutamine, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin. T cells were labelled with pHrodo Red dye (Invitrogen) by incubating T cells at 1 × 10⁶ cells per ml with a working concentration of pHrodo Red of 30 ng ml⁻¹ in PBS for 1 h at 37 °C in the dark in a humidified 5% CO₂ incubator. The labelling reaction was quenched and excess dye was washed away by washing twice with complete medium. Cells were then either incubated alone or in the presence of anti-CD47 (B6H12; Bio X Cell) at a concentration of 10 μg ml⁻¹ in serum-free RPMI. T cells and antibodies were incubated for 30 min in a humidified 5% CO₂ incubator at 37 °C, before being washed twice with complete medium. Macrophages were added to each well and allowed to adhere for 2 h in a humidified 5% CO₂ incubator at 37 °C. After 2 h, labelled T cells were added to the plate at a 2:1 T cell:macrophage ratio. pHrodo Red fluorescence due to phagocytosis was monitored after 3 h with a ×10 objective using the IncuCyte S3 Live-Cell Analysis System (Sartorius), housed in a cell culture incubator at 37 °C and 5% CO₂, set to take four images per well at each timepoint. Total integrated red fluorescence intensity was quantified using the IncuCyte basic analyzer software feature (IncuCyte S3 v.2019B Rev2; Sartorius).

Confocal microscopy of T cell–macrophage interactions

CD19-28ζ CAR T cells were labelled with pHrodo Red dye as described above. pHrodo-Red-labelled T cells were then labelled with DiO Vybrant lipophilic dye (Invitrogen) according to the manufacturer’s instructions in PBS for 2 min at 37 °C in the dark. Macrophages were labelled with DiD Vybrant dye (Invitrogen) according to the manufacturer’s instructions in PBS for 15 min at 37 °C in the dark. After dye labelling, cells were washed three times with complete medium to remove excess dye. T cells were then incubated with B6H12 at 10 μg ml⁻¹ for 20 min at 37 °C in PBS, followed by two washes with complete medium. Labelled T cells and macrophages embedded in a collagen matrix (Cellmatrix type I-A, FUJIFILM Wako chemicals) at a ratio of 2:1 (for example, 1,000,000 T cells:500,000 macrophages) in a 24-well glass-bottom culture plate (Mattek). Four-dimensional (x,y,z,t) live confocal imaging was performed on a confocal laser-scanning microscope (Zeiss LSM900). Images were analysed using Imaris software (v.10.0; Oxford Instruments).

Quantification of CD47 expression on tumour and T cells using QuantiBrite

CD47 expression was quantified using an anti-CD47-PE antibody (B6H12, BD, 1:20) and a QuantiBrite PE Quantitation Kit (BD) according to the manufacturer’s instructions⁴¹. CD19-28ζ CAR T cells were produced as described above, except that cells were kept in culture 1 day after thawing before activation with anti-CD3/CD28 beads. T cells were analysed by flow cytometry on day 0 (before activation; 1 day after thaw), day 4 (immediately after removal from bead activation), day 7 and day 11 (average time of transfer in vivo). T cells were stained with anti-hCD4 (BUV 395, SK3, BD, 1:200), anti-hCD8 (BUV 805, SK1, BD, 1:400), anti-hCD47 or mIgG1 isotype control (PE, B11/6, Abcam, 1:20), anti-hCD45RA (BV785, HI100, BioLegend, 1:100) and anti-hCD62L (BV605, DREG-56, BD, 1:100) antibodies. T cell differentiation subtypes were defined as follows: T naive (CD45RA⁺CD62L⁺), T central memory (CD45RA⁻CD62L⁺), T effector memory (CD45RA⁻CD62L⁻) and T effector memory re-expressing CD45RA (CD45RA⁺CD62L⁻). Tumour cells were stained with only anti-hCD47 or mIgG1 isotype control. Molecules of CD47 were calculated according to the QuantiBrite kit instructions using extrapolation from MFI signals of BD QuantiBrite-PE beads with known quantities of PE. The degree of labelling for anti-CD47-PE (BD, 2040745) was determined experimentally as 0.842 molecules of dye per antibody, using the maximum absorbance at 566 nm, the extinction coefficient for PE (1,863,000 M⁻¹ cm⁻¹) and the listed antibody concentration.

Imaging of patient CSF samples

A CSF cytospin preparation was collected from a patient treated with axicabtagene ciloleucel (axi-cel) CD19-28ζ CAR T cell therapy, stained with Wright-Giemsa and imaged by microscopy at ×1,000 magnification, capturing histiocytes with engulfed lymphocytes⁴² (raw images are shown in Supplementary Fig. 3).

Single-cell analysis of patient samples

Two datasets were reanalysed: ref. ¹⁴ (GSE168940), including scRNA-seq data collected from nine patients with LBCL treated with axi-cel CD19-28ζ CAR T cell therapy, where 50,000–70,000 CAR T cells (single live CD4⁺ or CD8α⁺CD235a⁻CAR⁺ events) were FACS sorted to ≥95% purity; and ref. ¹⁵ (GSE186802), including scRNA-seq data collected from four patients with DMG treated with GD2.BBζ CAR T cell therapy, with cells derived from CSF. Both datasets were analysed on the 10x Genomics platform^14,43. Where indicated, previously annotated CAR-mRNA-expressing cells were used.

Histology of tissue samples

The tissues assessed include skin and lung. Tissues were collected and immersion-fixed in 10% neutral-buffered formalin. After fixation, tissues were routinely processed, embedded in paraffin, sectioned at 5.0 μm and routinely stained with haematoxylin and eosin (H&E). Tissues were visualized using the Olympus BX43 upright bright-field microscope, and images were captured using the Olympus DP27 camera and cellSens software (v.3.2; Olympus Life Science).

Yeast surface display vectors

A DNA sequence encoding the CD47 Ig-like domain (Gln19–Ser135) was cloned into the pCTCON2 yeast-surface display vector (Addgene) using the NheI and BamHI sites. The pFreeNTerm (pFNT) vector was based on the pCL backbone⁴⁴, designing an intrinsic NheI cutsite into the Aga2p signal sequence as the 5′ cloning site and using a MluI cutsite prior to a 3×Gly₄Ser linker as the 3′ cloning site. Variants of the CD47 Ig-like domain (Gln19–Ser135) were cloned into the pFNT yeast-surface display vector using these NheI and MluI sites.

Flow cytometry analysis of yeast surface display constructs

Saccharomyces cerevisiae (strain, EBY100; ATCC) yeast were transformed with pCTCON2 or pFNT plasmids and selected on SD-CAA-Agar plates. Yeast (~100,000 per sample) were grown and induced in SG-CAA, and binding was set up over a range of soluble ligand concentrations in PBS containing 1 mg ml⁻¹ BSA (BPBS), taking into account ligand depletion and equilibrium time⁴⁵. After incubation with binding partner, yeast cells were washed once with BPBS, then incubated with a 1:5,000 dilution of chicken anti-MYC antibody (polyclonal, Invitrogen) for pCTCON2 displayed proteins, and incubated for 30 min at 4 °C in the dark. After primary addition, the samples were washed once with BPBS, and secondary antibodies were added. Expression was detected with a 1:500 dilution of goat anti-chicken Alexa Fluor 488 (polyclonal, Invitrogen) or Alexa Fluor 647 (polyclonal, Abcam). For pFNT displayed proteins, co-displayed GFP was used to monitor expression. Binding of proteins with mouse Fc domains (hSIRPα, B6H12) was detected with a 1:500 dilution of goat anti-mouse Alexa Fluor 488 or Alexa Fluor 647 (polyclonal, Invitrogen). Binding of proteins with a human Fc domain (CV-1 (ALX-222), mSIRPα, Hu5F9, TJC4) was detected using a 1:500 dilution of goat anti-human Alexa Fluor 488 or Alexa Fluor 647 (polyclonal, Invitrogen). Secondary antibodies were incubated for 15 min at 4 °C in the dark. After secondary incubation, the samples were washed once with BPBS, pelleted and left pelleted on ice until analysis. The samples were analysed by resuspending them in 50 μl of BPBS and running flow cytometry on the BD Accuri C6 (BD Biosciences) system. Accuri C6 software (v.1.0.264.21; BD) was used for data collection and FlowJo software (v.10.8.1; BD) was used for data analysis (gating strategies are shown in Supplementary Fig. 2). The samples were gated for bulk yeast cells (forward scatter (FSC) versus side scatter (SSC)) and then for single cells (FSC-height versus FSC-area). Expressing yeast were determined and gated through the C-terminal MYC tag or GFP detection. The geometric mean of the binding fluorescence signal was quantified from the expressing population and used as a raw binding value. When comparing binding signals, the average fluorescence expression signal was quantified for different protein variants and used to normalize binding signal. To determine the ‘fraction bound’, binding signals were divided by the signal derived from the highest concentration of binding partner used, or that derived from binding to WT CD47. To calculate K_d values, data were analysed in Prism (v.9.5.1, GraphPad) using a nonlinear regression curve fit.

Yeast surface display library generation, sorting and sequencing

CD47 was expressed in S. cerevisiae (strain, EBY100; ATCC) as a genetic fusion to the agglutinin mating protein Aga2p in the pCTCON2 vector. An error-prone PCR library was created using the CD47 Ig-like domain (Gln19 to Ser135) as a template and mutations were introduced using the Gene Morph II random mutagenesis kit (Agilent) according to the manufacturer’s instructions. Separate PCRs were performed using various concentrations of Mutazyme II enzyme. Products from these reactions were purified by gel electrophoresis, pooled and amplified with standard PCR using Phusion polymerase (New England BioLabs). Purified mutant DNA and linearized pCTCON2 plasmid were electroporated into EBY100 yeast, where they were assembled in vivo through homologous recombination. We estimated 5 × 10⁷ variants for the library, determined by dilution plating and colony counting. Yeast were grown in SD-CAA medium and induced for CD47 protein expression by growth in medium containing 90% SG-CAA and 10% SD-CAA overnight⁴⁵. Yeast displaying CD47 variants were isolated by FACS using the SONY SH800S cell sorter (SONY; SH800S cell sorter software v.2.1.5) and analysed using the BD Accuri C6 flow cytometer (BD Biosciences; BD Accuri software v.1.0.264.21). Gating strategies are shown in Supplementary Fig. 2. Data were analysed using FlowJo software (v.10.8.1, BD). Screens were performed using equilibrium binding conditions where yeast were incubated at room temperature in BPBS with the following concentrations of B6H12 or CV-1 (ALX-222) for 2 h. For negative sorts to B6H12, the CD47-expressing, but non-binding populations of yeast were collected. For positive sorts to CV-1, the CD47-expressing and binding populations of yeast were collected (Extended Data Fig. 6g). Sort 1, negative sort, 500 pM B6H12; sort 2, negative sort, 5 nM B6H12; sort 3, positive sort, 20 nM CV-1; sort 4, negative sort, 20 nM B6H12; sort 5, negative sort, 50 nM B6H12; sort 6, positive sort, 10 nM CV-1. After incubation with B6H12 or CV-1, yeast was pelleted, washed and labelled with fluorescent antibodies as described above before sorting. Sorted yeast clones were propagated, induced for CD47 expression and subjected to iterative rounds of FACS as described above. After each round of screening, plasmid DNA was recovered using the Zymoprep yeast plasmid miniprep I kit (Zymo Research), transformed into DH10B electrocompetent cells (Thermo Fisher Scientific) and isolated using the GeneJET plasmid miniprep kit (Thermo Fisher Scientific). Sequencing was performed by ELIM Biopharmaceuticals.

CD47 structure modelling

CD47 structures were downloaded from the Protein Data Bank (PDB) and analysed using PyMol (v.2.5.8; Schrödinger). The CD47–hSIRPa structure used was PDB 2JJS (ref. ¹⁸). The CD47–B6H12 structure used was PDB 5TZU (ref. ¹⁹).

143B in vivo phagocytosis model

A total of 1 × 10⁶ 143B cells in 100 μl DPBS was injected into the tibia periosteum of NSG mice (aged 6–10 weeks). Then, 20 days after tumour implantation, mice were treated with 3 × 10⁶ CFSE-labelled HER2-BBζ CAR T cells intratumourally. Before administration, T cells were labelled with CFSE (5 µM working solution), according to the manufacturer instructions, with cells labelled at a concentration of 10⁷ cells per ml in PBS for 5 min. Cells were washed in complete medium, and then incubated with or without 10 µg ml⁻¹ B6H12 in PBS for 20 min at 37 °C. Cells were resuspended in 80 µl of PBS/tumour for intratumoural injection. Immediately after T cell administration, the mice were then treated with B6H12 (250 µg) or PBS by intraperitoneal injection. Tumours were collected 16 h later on day 21 after tumour implantation. Tumours were mechanically dissociated using the gentleMACS dissociator (Miltenyi). Single-cell suspensions were made by passing tumours through a 70 μm cell strainer (Thermo Fisher Scientific), depleting red blood cells by ACK lysis (Quality Biological) and further filtration through flow cytometry filter tubes with 35 μm cell strainer caps (Falcon). Single-cell suspensions were subsequently stained for flow cytometry to detect CFSE⁺ T cells and macrophages from dissociated tumours (see below) (Supplementary Fig. 1n).

143B correlative study and tumour dissociation

A total of 1 × 10⁶ 143B cells in 100 μl DPBS was injected into the tibia periosteum of NSG mice (aged 6–10 weeks). Then, 13 days after tumour implantation and after visual confirmation of tumour formation, the mice were treated with 4 × 10⁶ HER2-BBζ CAR T cells with endogenous CD47 KO and overexpressing either 47_WT or 47_E, an equivalent number of mock T cells intravenously by tail-vein injection, or no T cells. The mice were then treated twice with B6H12 (250 µg) or PBS by intraperitoneal injection on day 15 and day 19. Tumours were collected at day 21 after tumour implantation (day 8 after CAR T cell treatment). Tumours were split with a razor, with one section being fixed in 10% paraformaldehyde, and the other mechanically dissociated as described above, before being stained for flow cytometry and FACS. Formaldehyde-fixed tumour had paraformaldehyde removed after 24 h and replaced with 70% ethanol for long-term storage. Tumour sections were then formalin-fixed and paraffin-embedded according to the standard protocol (Supplementary Fig. 1o).

Flow cytometry and IHC analysis of dissociated tumours

For flow cytometry, tumours were collected as described above. Single-cell suspensions of dissociated tumours were stained for CAR (APC, HER2-Fc, R&D, 1:400), hCD19 (BUV 496, SJ25C1, BD, 1:100), CD11b (BUV 395, M1/70, BD, 1:100), F4/80 (BV605, BM8, BioLegend, 1:100), hCD45 (PerCP-Cy5.5, HI30, Invitrogen, 1:50), hCD3 (BUV 737, SK7, BD, 1:100), mCD45 (BUV 805, I3/2.3, BD, 1:100) and hCD47 (BV711, B6H12, BD, 1:100), and with Fixable Viability Dye (eFluor 780, Invitrogen, 1:2000) for 30 min in PBS + 2% FBS (FACS Buffer) before being analysed by flow cytometry.

For IHC analysis, tumours were collected as described above. Formalin-fixed, paraffin-embedded xenograft tumour sections were used. F4/80 (D2S9R, Cell Signaling Technology, 1:200) staining was performed manually, and hCD3 (SP7, Abcam, 1:100) and ARG1 (D4E3M, Cell Signaling Technology, 1:250) staining was performed using the Ventana Discovery platform. In brief, tissue sections were incubated in either 6 mM citrate buffer (F4/80, 1:200 dilution) or Tris EDTA buffer (CD3/ARG1, 1:100 and 1:250 dilution respectively; cell conditioning 1 standard) at 100 °C for 25 min (F4/80) or 95 °C for 1 h (CD3/ARG1) to retrieve antigenicity, followed by incubation with the respective primary antibody for 1 h. Bound primary antibodies were incubated with the respective secondary antibodies (goat anti-rabbit, polyclonal, F4/80: Vector Laboratories, undiluted or CD3/ARG1: Jackson ImmunoResearch, 1:500), followed by UltraMap HRP (Roche, F4/80) or ChromoMap DAB (Roche, CD3/ARG1) detection. For IHC analysis, tumour regions were identified on the basis of histology. F4/80, CD3 and ARG1 positivity were analysed for each tumour region. F4/80, CD3 and ARG1 IHC positivity scores were automatically quantified in the regions of interest using Aperio ImageScope software (v.12.3.2.8013). Regions of interest were randomly selected within the tumour to exclude macrophages present in the normal tissue around the tumour.

Single-cell analysis of dissociated 143B tumours

Dissociated tumours from the 143B osteosarcoma model described above were sorted for live cells using a Live/Dead stain (Invitrogen) at the Stanford Shared FACS facility (the gating strategy is shown in Supplementary Fig. 2). scRNA-seq libraries were prepared using the Chromium Next GEM Single Cell 5’ v2 platform (10x GENOMICS). Libraries were sent to Novogene for sequencing on a NovoSeq S4 lane (PE150) with approximately 30,000 mean reads per cell. Reads were aligned and quantified using CellRanger (v.6.0; 10x GENOMICS) using the standard workflow, with the reference transcriptomes GRCh38 for human and mm10 for mouse. The CellRanger output was imported into R (v.4.2.2) using Seurat (v.4.2.0). The following filters were applied using the subset function to select for live cells: nFeature_RNA > 200 and nFeature_RNA < 5000; percent mitochondrial reads < 5%. After filtering, the eight biological samples ranged from 7,658–9,327 mean unique molecular identifiers per cell. The data matrix was normalized with NormalizeData and scaled with Seurat. Cell types were assigned using SingleR automated cell type recognition. Differential expression analysis, clustering and UMAP dimensionality reduction analysis were performed on the resulting data matrix using Seurat⁴⁶. Pathway analysis was performed using Enrichr⁴⁷, with the NCI-Nature 2016 gene set collection queried for human T cells and the KEGG Human Pathway collection queried with converted mouse gene IDs for mouse macrophages.

Statistical analyses

The specific statistical tests used are indicated in the figure legends. Statistical analyses were performed using R (v.4.2.2), Excel (v.16.64; Microsoft) or Prism (v.9.3.1, GraphPad). For comparisons between two groups, statistical significance was assayed using two-tailed unpaired Student’s t-tests or Mann–Whitney U-tests. For comparison within in vivo studies and between grouped studies, two-way ANOVA combined with Tukey’s multiple-comparison test for post hoc analysis was performed. Significance for survival data was calculated using the log-rank Mantel–Cox test. Sample sizes were determined on the basis of the variability of tumour models used, determined by experience with well-established, previously published models^31,32,34,38. Tumour-bearing animals were assigned to the treatment groups randomly to ensure an equal distribution of tumour sizes between groups. Data are represented as mean ± s.d. (in vitro studies) or mean ± s.e.m. (some in vivo studies). For all statistical analyses, P values are indicated in each figure panel.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.