Targeting Macrophages with CAR T Cells Delays Solid Tumor Progression and Enhances Antitumor Immunity

Immunotherapy is revolutionizing the treatment of many cancers (1). However, most patients with solid tumors do not respond to current immunotherapy agents. Among the factors involved in resistance to therapy are the tumor-associated macrophages (TAM; ref. 2). TAMs are one of the most abundant cell types in the tumor microenvironment (TME) of solid tumors (3) and adopt specific molecular states associated with poor clinical outcome (4). TAMs contribute to the suppression of antitumor immune responses and support tumor growth through a variety of mechanisms, including promoting angiogenesis, providing an immunosuppressive environment, helping to create a barrier that excludes effector T cells from the tumor core, capturing tumor antigens and preventing cross-presentation by cDC1, and promoting tumor invasiveness and metastasis (5).

The contribution of TAMs to the immunosuppressive TME has led to an interest in establishing clinically applicable strategies for depleting or reprogramming TAMs in human tumors (6). One prominent approach has been to target macrophage recruitment and survival by blocking the interaction between CSF1R on macrophages and its ligand CSF1 present in the TME. The CSF1/R axis is critical to the differentiation and survival of macrophages, and inhibitors of this axis have shown reduction of macrophage density in solid tumors in mice and patients with cancer (7). Despite this, CSF1/R inhibition has not been successful in the clinic as a monotherapy (8). Alternative strategies have been developed to modulate or deplete specific macrophage subtypes, such as those expressing TREM2, which are enriched among TAMs (9, 10). Efforts are also being pursued to reprogram TAMs from immunosuppressive to immunostimulatory, including through the use of Toll-like receptor agonists and mRNA nanoparticles (6, 11). These strategies are promising but have not been tested in the clinic yet, and thus there is still an important need to develop clinically applicable strategies to target TAMs and deplete or reprogram them.

One potentially promising, but still little explored, approach for TAM depletion is through the use of chimeric antigen receptor (CAR) T or natural killer (NK) cells (12–14). CAR T cells are genetically modified to express an immunoglobulin-based receptor cognate to a specific surface antigen that triggers T-cell activation in a T-cell receptor (TCR)–independent manner upon contact with an antigen-expressing cell (15). CAR T cells have transformed the treatment of hematologic malignancies, such as acute lymphoid leukemia (ALL; ref. 16) and multiple myeloma (17). In most cases, the CAR T targets a marker specific to the lineage the malignant cells derive from. For example, CAR T cells used in ALL target CD19, a molecule expressed by all normal and leukemic B cells. Continued innovation in the CAR T field, including the implementation of CRISPR/Cas9 technology (18), makes it one of the most active areas of development in cancer immunotherapy (19).

Unfortunately, CAR T therapy has been less successful for the treatment of solid tumors due to a number of issues (20). One of the most significant is the paucity of tumor-specific CAR-targetable antigens. There have been CARs designed against a variety of targets expressed by the cancer cells within solid tumors, such as ERB2, EGFR, and HER2, but none of these are specific to cancer cells, nor uniformly expressed across patients. Though innovative strategies are being developed to help restrict CAR T cell killing to malignant cells (19), the cancer cells’ high capacity for adaptation through gene mutations and epigenetic silencing often leads to loss or reduction of target expression on the cancer cells, enabling them to evade CAR T killing. Additional to the problem of antigen availability is the challenge for the CAR T cells to overcome the highly immunosuppressive and immune excluded environment of solid tumors, which, as noted, is facilitated by TAMs (5).

Here we set out to develop and evaluate CAR T-cell targeting macrophages as a means to achieve antitumor impact and reprogram the immunosuppressive TME of solid organ tumors. The rational for using CAR T cells to target macrophages is to eliminate the immunosuppressive activity of TAMs. To this end, we generated a second-generation CAR targeting the mouse pan-macrophage marker F4/80 (F4.CAR) and show that T cells expressing F4.CAR (F4.CAR-T) specifically killed F4/80⁺ macrophages and eosinophils. Injection of F4.CAR-T into immunocompetent mice bearing lung tumors resulted in slower tumor growth and prolonged survival, despite cancer cells not being directly targeted by treatment. Antitumor benefit was also observed in models of ovarian and pancreatic cancer. Notably, we found that treatment promoted expansion of endogenous CD8 T cells specific for tumor antigen and immune editing of the tumor, and this was dependent on the F4.CAR-T–expressing IFNγ. These studies help to establish macrophage-targeted CAR T cells as a means to potentiate endogenous antitumor immunity and slow tumor growth.

F4/80 hybridoma cells were purchased from ATCC (HB-198). Hybridoma cells were sent to Genscript for Standard Antibody Sequencing for the Variable Domain. The sequence for the identified heavy and light chain portions of the variable domain were synthesized in a single-chain fragment variable (scFv) format, separated by a 3x GGGGS linker (GeneArt Gene Synthesis, Thermo Fisher Scientific). The construct was cloned into an MSGV CD28-CD3z CAR retroviral vector (Addgene #107226). Briefly, an NsiI restriction site was introduced upstream of the backbone's Kozak region by point mutagenesis, and NsiI HF and SalI HF (New England Biolabs R3127S and R3138S) were used to excise the previous CD19 scFv and introduce the new F4/80 scFv. PCR overlap assembly (21) was used to introduce a T2A-GFP sequence downstream of CD3z. Coexpression of GFP and the CAR construct was verified by flow cytometry staining of the CAR using an anti-mouse IgG (H+L) (A21235, Invitrogen). A human CD19-directed CAR was used as a control where indicated. To this end, an SFG-h1928z vector was constructed by stepwise Gibson Assembly using the cDNA of previously described anti-human CD19 scFv (22), Myc-tag sequence (EQKLISEEDL), murine CD28 transmembrane and intracellular domain and murine CD3z intracellular domain into an SFG retroviral vector. CAR expression was verified by flow cytometry identification of Myc-tag positive cells.

Phoenix-Eco cells were obtained from ATCC (ATCC CRL-3214) and cultured in Iscove's modified Dulbecco's medium (IMDM; Gibco 12440-053) supplemented with 10% FBS (Gibco 26140-079), 2 mmol/L Glutamine (Gibco 25030-081), and 1% Pen/Strep (Gibco 15140-122). RAW264.7 [ATCC TIB-71, received in 2019, authentication through short tandem repeat (STR) profiling, last profiling February, 2022], A20 (ATCC TIB-208, received in 2019, authentication through STR profiling, last profiling Feb, 2022), HKP1 (Source: Dr.Vivek Mittal, Cornell Center for Immunology, received in 2018, orthotopic tumor growth with tdTomato marker in control tumors as proxy for authentication; ref. 23) and mouse CD8 T cells were cultured in RPMI1640 medium (Gibco 22400-089) containing the previous supplements plus 50 μm β-mercaptoethanol (Gibco 21985-023). ID8_VEGF cells (Source: Dr. Jean J Zhao, Dana-Farber Cancer Institute, received in 2018, Brca1 deficiency and Vegfa overexpression confirmed by RNA sequencing upon reception; orthotopic tumor growth with GFP marker in control tumors served as proxy for authentication; ref. 24) were cultured in RPMI containing the previous supplements plus 500 nmol/L Sodium Pyruvate (Corning 25-000-CI). KPC cells (source: Dr. David Tuveson, Cold Spring Harbor Laboratory, received in 2021, upon receipt PCR confirmed the gender of the cells and that they have the Pdx1-Cre, KrasG12D, and Trp53R172H alleles; ref. 25) were cultured in DMEM supplemented with 5% FBS (Gibco 26140-079), 2 mmol/L Glutamine (Gibco 25030-081), and 1% Pen/Strep (Gibco 15140-122).

After thawing, cell lines intended for tumor inoculation were kept in culture for 7 to 10 days in the indicated cultured media. Cells were trypsinized and reseeded every 2 to 3 days or when 80% to 90% confluence was reached. HKP1 cells were split one last time 24 hours before injection to consistently achieve an approximate 70% confluence and exponential growth at the time of inoculation.

Mycoplasma testing was carried out periodically by PCR on DNA extracts from leftover cells produced during cell passage (Qiagen DNeasy Blood & Tissue Kit: 69504; PCR forward primer: CGCCTGAGTAGTACGTTCGC and reverse primer GCGGTGTGTACAAGACCCGA to amplify DNA myco sequence). No mycoplasma contaminations were detected for the duration of this project.

For killing assays, effector: target: nontarget cocultures (at an X:1:1 ratio) were cultured for 24 hours in 96-well flat-bottom culture plates. To minimize alloreactivity-derived target killing, we used Balb/c CD8 CAR T cells with RAW264.7 macrophages, adding an equal number of A20 B cells as nontarget controls. In separate experiments, C57Bl/6 CAR T cells were cocultured with thioglycate-elicited peritoneal macrophages (see below), using peritoneal neutrophils as nontarget controls. A total of 4 hours before the end of the culture, 3 μg/mL Brefeldin A (Invitrogen 00-4506-51) was added to the cultures for subsequent cytokine staining. Specific lysis was calculated as “([T/NT]_{control_average}-[T/NT]_sample)/([T/NT]_{control_average}*100)”, with T and NT representing the number of target and nontarget cells in each sample.

Generation of thio-elicited macrophages was carried out as described previously (26). Briefly, a 3% solution of Thioglycolate Broth (Thermo Fisher Scientific CM0173B) was prepared in distilled water, autoclaved and stored at 4°C until use. A total of 1.5 mL of thioglycolate solution was injected intraperitoneally into C57Bl/6 mice. A total of 72 hours later, mice were euthanized and 5 mL PBS 2 mmol/L Ethylenediaminetetraacetic acid injected intraperitoneally to collect the peritoneal lavage. The lavage was red blood cell (RBC)-lysed and resuspended in RPMI medium containing 10 ng/mL murine MCSF (Peprotech 315-02) for coculture with CAR or control transduced T cells.

Phoenix-Eco cells were seeded into 10 or 15 cm tissue culture plates (respectively, Falcon #353003 and Thermo Fisher Scientific Nunclon #168381) 24 hours prior to achieve an approximate cell density of 70% at the moment of transfection. Transfection was carried out using the calcium-phosphate method. Briefly, CAR or control plasmid constructs were suspended in 0.1× TE (Maxi Qiagen kit #12362) and 0.25 mol/L CaCl₂ (Sigma-Aldrich #C7902-1KG); one volume of 2× HBS (for 500 mL: 1 mol/L HEPES, i.e., 50 mL Corning #25-060-Cl, 2 mol/L NaCl, i.e., 70.25 mL Fisher Bioreagents #BP358-1; 0.5 mol/L Na₂HPO₄, i.e., 1.5 mL # BP332-500; 378.25 mL Tissue Culture Tested Water Corning #46-000-CV, 5 mol/L NaOH to pH Thermo Fisher Scientific #134070010) was added on a dropwise fashion while continually vortexing, and the resulting solution was immediately added onto Phoenix-Eco cells and allowed to sit overnight. IMDM (Gibco 12440-053) was replaced the next morning and supernatants collected and 0.2 μm filtered 24 to 30 hours after that. Supernatant aliquots were stored at −80°C until use. For control purposes throughout all experiments, we used a third-generation retroviral HSC1 vector coding for GFP under the EF1α promoter, cloned in-house.

CD8 T cells were isolated from mouse spleens using the EasySep Mouse CD8⁺ T Cell Isolation Kit (StemCell 19853). Activation was carried out at a cell density of 1 × 10₆ cells/mL in RPMI medium + 100 U/mL Recombinant Murine IL2 (Peprotech 212-12). Dynabeads Mouse T-Activator CD3/CD28 (Thermo Fisher Scientific 11453D) were used to activate cells at a 1:4 bead-to-cell ratio for 24 hours before transduction. Nontreated culture plates (NUNC, Thermo Fisher Scientific) were coated overnight at 4°C with 20 μg/mL Retronectin (Takara T100B). Viral supernatant was spun for 90 minutes at 2,000 × g and 30°C onto the plated retronectin and half the supernatant volume removed carefully after spinning. T cells were resuspended in fresh RPMI medium with IL2, added onto the supernatant-containing wells (final IL2 concentration 50 U/mL), and allowed to sit for 24 hours. After this, Dynabeads were magnetically removed and T cells were resuspended in fresh RPMI medium at 50 U/mL IL2. New RPMI medium containing 50 U/mL IL2 was added daily to keep T cells at a concentration between 1 and 2.5 × 10⁶ cells/mL until use, typically within 5 days of isolation. Detection of GFP by flow cytometry was employed to quantify the efficiency of transduction.

C57Bl/6, CD45.1⁺ C57Bl/6, Ifng^−/− C57Bl/6, and Balb/c female mice were obtained from Jackson laboratories (Jackson ID 000664, 002014, 002287, and 000651, respectively) and kept in the Mount Sinai vivarium during use. All mouse experiments were carried out under institutional Institutional Animal Care and Use Committee approval. All mice were randomized before treatment administration.

For tumor inoculation, HKP1 cells were trypsinized, passed through a 70 μm cell strainer, and resuspended in ice-cold sterile PBS at a concentration of 5 × 10⁵ cells/mL. Mice were heated for 3 to 5 minutes using a heating lamp and 1.5 × 10⁵ cells were injected into the tail vein in a volume of 300 μL using a 27G needle. Anti–PD-1 (RMP1-14, BioxCell) or anti-Ly6C (clone Monts1, kindly gifted by Dr. Jordi Ochando, PhD; ref. 27) were diluted in PBS to a concentration of 1 mg/mL for intraperitoneal administration of the amounts shown in the respective figures. Blood samples were drawn where indicated using maxillary vein puncture and stained for flow cytometry. Mice were euthanized at the indicated timepoints using CO₂ inhalation. Lungs and spleens were extracted, weighed, and processed for immunofluorescence or flow cytometry.

T cells were injected intravenously via tail vein injection in a volume of 300 μL PBS using 27G needles. The number of T cells to inject in each experimental group was adjusted on the basis of GFP transduction efficiency to have 5 × 10⁶ GFP⁺ control or F4.CAR-T cells per injection.

For ID8 tumor inoculation, cells were trypsinized and resuspended in PBS at a concentration of 5 × 10⁶ cells/mL. A total of 10⁶ cells were injected intraperitoneally into female mice. A total of 14 days later, 5 × 10⁶ CAR T or control cells were injected intravenously into mice. Mice weight was recorded twice weekly from tumor inoculation. We monitored mice visually for the development of peritoneal ascites and photographed mice at day 24 posttumor inoculation.

For the experiments involving pancreatic tumors, we used the KPC cell line FC1245 that were established from a tumor in the KPC mouse model as described previously (25). Briefly, female C57Bl/6 mice were anesthetized with ketamine/xylazine. A total of 5 × 10⁴ tumor cells were suspended in 25 μL matrigel (Corning Matrigel Growth Factor Reduced Basement Membrane Matrix, #356230) and surgically inoculated into the pancreas. CAR T cells were administered in 200 μL PBS by retroorbital (r.o.) intravenous injection 7 days later. At day 14 posttumor inoculation, mice were euthanized and pancreatic tumors obtained, separated from healthy tissue, weighed, and digested in collagenase IV/DNAse I (Sigma-Aldrich, # C5138-1G) for flow cytometry.

Where indicated, preconditioning mouse lymphodepletion was carried out in an X-ray irradiator (RS-2000 Biological System, Radsource) at the Mount Sinai facility. Mice were exposed to a total body irradiation (TBI) dose of 2 Gy (1.25 Gy/minute for 100 seconds). Lymphodepletion took place no longer than 6 hours before administration of CAR T or control cells.

Mice were subjected to anesthesia using inhaled isofluorane and received 100 μL 15 mg/mL luciferin (PerkinElmer #122799) through retroorbital injection 5 minutes before image acquisition. Luciferase-based in vivo bioluminescence was acquired in an IVIS Imaging System (Perkin Elmer) at the BioMedical Engineering and Imaging Institute at Mount Sinai (New York, NY). Images were processed using Living Image software (Perkin Elmer).

Lungs were minced and digested for 30 minutes in Collagenase IV (Sigma-Aldrich, # C5138-1G) before being mashed through a 70 μm cell strainer. RBCs were lysed using RBC Lysis Buffer (eBioscience 00-4300–54). Extracellular stainings were carried out for 10 minutes at 4°C in PBS containing 5% BSA and 2 mmol/L EDTA (Flow Buffer). H2Db-LMYRFEEEL tetramers were generated by the NIH Tetramer Core Facility and allowed to stain at a 1/150 dilution for 30 minutes at 4°C. For intracellular cytokine staining, 3 μg/mL Brefeldin A (eBioscience #00-4506-51) was added 3 to 5 hours before staining. Cells were sequentially stained for extracellular antigens, fixed and permeabilized using BD Cytofix/Cytoperm Fixation/Permeabilization Solution Kit (BD Biosciences # 51-2090KZ) before being resuspended in Flow Buffer for acquisition. The cell per organ calculation, where applicable, was carried out using AccuCheck counting beads (Life Technologies # PCB100). Samples were acquired using FACS Diva in a BD Fortessa flow cytometer and analyzed using FlowJo (BD). The full list of flow antibodies used can be found in Supplementary Table S1.

Lungs were fixed overnight in periodate-lysine-paraformaldehyde buffer (in-house) containing 1% paraformaldehyde (PFA; Thermoscientific, ref:J19943-K2), 75 mmol/L L-Lysin (Sigma, ref L5501) and 10 mmol/L NaIO₄ (Sigma, ref S1878-100g) in pH 7.4 phosphate buffer. After fixation, samples were dehydrated in an increasing sucrose gradient, included in optimal cutting temperature compound (Thermo Fisher Scientific 23-730-571) and stored at −80°C. A total of 10 μm sections were prepared on a cryostat (Thermo Fisher Scientific HM525) and placed on microscopy slides (Thermo Fisher Scientific 12-550-15) for staining. Nonspecific binding was blocked using PBS containing 2% FBS and 0.5% BSA. Staining of AF647-conjugated F4/80 (1/200, Invitrogen MF48021) was carried out overnight at 4°C in PBS containing 1% FBS and 0.25% BSA. DAPI counterstaining (1/5,000, Thermo Fisher Scientific 62248) was added as a 5-minute incubation after antibody binding, and cover slides added for acquisition. Images were obtained at 10× magnification using a Zeiss LSM780 confocal microscope from the Microscopy Core at Mount Sinai (New York, NY). Images were generated using Zeiss Zen software and processed and analyzed using ImageJ and QuPath software. The Cell Detection function within QuPath was used for quantification of macrophages per area of tissue, based on F4/80 staining intensity.

The left lobes of mouse lungs were fixed overnight at 4°C in 4% PFA (Thermo Fisher Scientific, #J19943-K2) and then immersed in 70% EtOH (Thermo Fisher scientific # Molecular Grade Ethanol: BP2818-4) before processing. Paraffin embedding, generation of 10 μm sections and hematoxylin and eosin (H&E) staining were carried out by the Biorepository and Pathology Core at Mount Sinai (New York, NY). H&E sections were scanned in a Leica Aperio AT2 whole-slide scanner. Quantification of tumor burden on H&E slides was calculated on lung whole slide scans using QuPath software, defining tumor nodules by morphologic features. Histologic tumor burden was calculated as the sum of the areas of all tumor nodules divided by the total lung area and expressed as a percentage.

Statistical analyses were carried out using GraphPad Prism 9 software. Log-rank analysis was used to compare survival curves. Student t test comparison was used for column analyses, except where indicated: Mann–Whitney rank-based analyses were used to include placeholder datapoints representing mice that succumbed to tumor development before sacrifice (indicated in figures with an “x” datapoint symbol).

BioRender was used for the elaboration of illustrative guides.

The data generated in this study are available upon request from the corresponding authors.

We set out to generate a CAR with specific tropism for macrophages. We aimed to target macrophages broadly as a proof of principle for this approach. We chose to target the Adgre1 gene product, F4/80, which is highly and specifically expressed by mouse macrophages, as well as eosinophils and certain populations of monocytes (28). We sequenced the immunoglobulin variable domains of the F4/80 hybridoma (29) and cloned the heavy and light chain fractions of the variable region as a scFv into a CD28/CD3z-based CAR retroviral vector (30). A GFP moiety, separated from the CAR by a T2A self-cleaving peptide, was cloned downstream of the CAR coding region to mark transduced cells (Fig. 1A). We isolated CD8 T cells from wild-type mice and transduced them with the F4.CAR vector or a control retroviral vector encoding GFP. Expression of the CAR and GFP by transduced T cells was confirmed by flow cytometry (Fig. 1B).

To test the activity of F4/80 CAR T cells (F4.CAR-T), we cocultured them with primary macrophages ex vivo. To this end, we generated peritoneal thioglycolate-elicited macrophages and cocultured them with F4.CAR-T CD8 T cells to interrogate target-specific lysis. Peritoneal neutrophils acted as nontarget internal controls (Fig. 1C). GFP⁺ F4.CAR-T but not control T cells produced IFNγ in this setting (Fig. 1D). This was accompanied by the specific lysis of macrophages (Fig. 1E). We confirmed the specificity of F4.CAR-T activity using RAW264.7 cells, a mouse macrophage cell line which constitutively expresses F4/80. In this setting, we used A20 cells, an F4/80-negative B cell line, as an internal control for the quantification of specific lysis (Supplementary Fig. S1A). Similar to peritoneal macrophages, RAW264.7 induced IFNγ production by F4.CAR-T cells (Supplementary Fig. S1B) and were specifically killed by these cells (Supplementary Fig. S1C). These results indicate that the F4.CAR construct is functional and facilitates lysis of macrophages by F4.CAR-expressing CD8 T cells.

We next evaluated F4.CAR-T activity in vivo. Although CAR T injection is normally preceded by lymphodepleting conditioning to facilitate CAR T expansion, we reasoned that F4.CAR-T may not require conditioning because of the ample availability of target cells (macrophages), which themselves are antigen-presenting cells. To test this, we injected 5 × 10⁶ F4.CAR-T intravenously into syngeneic healthy C57Bl/6 mice and measured expansion in peripheral blood by flow cytometry detection of GFP⁺ cells. As a control, we injected mice with T cells transduced with a retroviral vector encoding GFP. By 6 days, F4.CAR-T reached 5% to 15% of circulating CD45⁺ cells and contracted shortly afterward, whereas control T cells did not expand (Fig. 1F). Mice receiving F4.CAR-T cells did not show signs of stress and no lethality was observed from treatment. To evaluate F4.CAR-T functionality, we collected the spleens from the mice, stained for a panel of immune cell markers and analyzed by flow cytometry 12 days after CAR T administration. In mice injected with control T cells, the frequency of macrophages in the spleen averaged 4 × 10³ cells per mg of tissue (Fig. 1G). In contrast, in mice that received the F4.CAR-T, macrophage frequency was reduced by 80%. Eosinophils, which express F4/80, were similarly eliminated in the F4.CAR-T–treated mice (Fig. 1H). These results indicate that F4.CAR-T cells were effective in target cell killing in vivo.

Macrophages are abundant in lung cancer lesions, and considerable evidence indicates that they play a key role in promoting immunosuppression in the TME. To model lung cancer in mice, we utilized HKP1 cells (23), which carry an activating Kras allele (Kras^G12D) and a deletion of p53 (Trp53^−/−), both common genetic mutations in human lung cancer. HKP1 cells also express luciferase and tdTomato, which enables tumors to be tracked in vivo. HKP1 were injected intravenously to seed lung tumors. At 19 days postinjection, there was abundant macrophage accumulation in the tumor lesions (Fig. 2A). We performed flow cytometry analysis of healthy and tumor-bearing lungs and observed that, in tumor-bearing lungs, monocyte-derived macrophages (Mo-Mac) were recruited to lungs and outnumbered resident alveolar macrophages (AM; Fig. 2B; Supplementary Fig. S2), fitting with previous findings in human cancer lesions and mouse models (31, 32). We quantified tumor burden histologically by calculating the percentage of tumor tissue area in lung H&E-stained sections. The total number of Mo-Macs in the lungs of tumor-bearing mice was proportional to histologic tumor burden, calculated as the sum of the areas of all tumor nodules divided by the total lung area and expressed as a percentage, underlining the association between tumor growth and macrophage accumulation (Fig. 2C).

We next set out to determine whether F4.CAR-T could deplete the macrophages in the lung tumor lesions. Seven days after HKP1 injection, we treated mice with control T cells or F4.CAR-T cells (Supplementary Fig. S3A). Similar to healthy mice, we detected F4.CAR-T cell expansion in the blood, followed by rapid contraction within 9 days of injection (Supplementary Fig. S3B). At 12 days post-T cell transfer (19 days after tumor inoculation), we collected lungs and examined tissue by immunofluorescence confocal microscopy and flow cytometry. Despite contraction of the transferred cells in the blood, microscopy analysis indicated F4.CAR-T had highly and preferentially infiltrated lung tumor lesions and colocalized with F4/80⁺ cells at tumor sites (Fig. 2D and E). There was a substantial reduction in the number of macrophages and eosinophils in the tumor-bearing lungs of F4.CAR-T–treated animals compared with controls, whereas the number of dendritic cells (DC) and patrolling monocytes remained the same (Fig. 2E). F4/80 expression among TAMs was homogeneous and comparable between control and F4.CAR-T–treated mice (Supplementary Fig. S4), indicating that the remaining macrophages were not a subpopulation of F4/80^neg/low cells that escaped killing due to CAR target reduction. Furthermore, we observed an increase in the number of inflammatory monocytes in the lungs, which could reflect an ongoing homeostatic response to replenish the depleted macrophage population.

A limitation of flow cytometry in the study of orthotopic lung tumors is the inability to parse cells originating from healthy and tumorous tissue. To verify whether depletion of target cells was occurring in the tumor lesions, we generated lung tissue sections from control-treated and F4.CAR-T–treated mice bearing HKP1 tumors and analyzed defined tumor regions to quantify the number of F4/80⁺ cells per unit of area within tumors. This analysis showed that the density of macrophages in the TME of F4.CAR-T–treated mice was markedly reduced across specimens, to approximately half of the control-treated counterparts (Fig. 2F).

Flow cytometry of single-cell suspensions from the same lungs indicated that the total number of CAR T cells detected per lung was directly proportional to the histologic tumor burden in those mice (Fig. 2G). This is most likely related to the accumulation of macrophages during tumor growth (Fig. 2C), and it corroborates the observed preference for F4.CAR-T cells to localize to tumor sites.

Together, these results demonstrate that F4.CAR-T were able to infiltrate lung tumor lesions, eliminate macrophages in the TME, and reduce TAM density in lung tumors.

We next set out to assess whether F4.CAR-T administration could impact lung tumor growth. We injected mice intravenously with HKP1 lung cancer cells, followed 7 days (day 7) later by injection of control T cells or F4.CAR-T cells (Fig. 3A). We monitored tumor growth noninvasively by luciferase-based luminescence using an IVIS system, and detected comparable tumors at the moment of injection of F4.CAR-T. However, 10 days after treatment (day 17), we found a 51% reduction in tumor burden in mice that received F4.CAR-T, as measured by photon detection (Fig. 3A).

In successive experiments, we sacrificed mice 19 days after tumor cell inoculation, extracted the lungs and performed H&E staining to quantify tumor burden by histologic analysis (Fig. 3B). We consistently observed significantly reduced tumor burden in F4.CAR-T–treated mice compared with mice injected with control T cells (Fig. 3B). Importantly, this delay in tumor growth translated into a survival benefit for treated mice, increasing the median survival by 28% from 23 to 30 days (Fig. 3C). These results show that in vivo targeting of macrophages by F4.CAR-T cells has an antitumor effect in the orthotopic HKP1 model of NSCLC.

We wanted to determine whether an increase in the number F4.CAR-T cells might provide further benefit. We tested two different approaches to boost CAR T numbers. In one group, we preconditioned lung tumor-bearing mice with mild lymphodepletion (2 Gy TBI), and injected with control or F4.CAR-T cells, leading to greater expansion of the adoptively transferred T cells compared with unconditioned mice. In another group, we injected mice with two sequential doses of F4.CAR-T cells, once at day 7 and again at day 14, posttumor inoculation. In both cases, using either conditioning or repeat administration, the F4.CAR-T cells mediated a significant reduction in tumor burden compared with control-treated animals (Fig. 3D; Supplementary Fig. S5A). However, the degree of macrophage reduction was similar to single CAR T treatment without conditioning or with single F4.CAR-T treatment (Supplementary Fig. S5B), suggesting the F4.CAR-T numbers are not a limiting factor in the antitumor effect.

As tumors grow, the protumorigenic macrophages are derived from monocytes migrating from the blood. Drugs are being developed to reduce TAMs by blocking recruitment of Mo-Macs into tumors (6). Anti-Ly6C can be used to deplete monocytes and limit monocyte-derived cell recruitment into tumors (33). To understand how this approach compares with macrophage-targeted CAR T and whether the two approaches could synergize, we injected lung tumor-bearing mice with anti-Ly6C, F4.CAR-T, or a combination of the two (Supplementary Fig. S5C). Although F4.CAR-T treatment was able to extend animal survival, there was not a significant benefit from anti-Ly6C treatment alone compared with control, and combined treatment did not provide additional benefit over F4.CAR-T alone (Supplementary Fig. S5D).

Though tumor burden was significantly reduced, we were still able to detect tumors in F4.CAR-T–treated mice. When we examined these tumors by florescence microscopy, we observed that there was a major reduction in tdTomato detection in HKP1 tumors from F4.CAR-T–treated mice (Fig. 4A). Using flow cytometry, we quantified the mean florescence intensity (MFI) for tdTomato in HKP1 cells and confirmed a strong reduction in F4.CAR-T–treated mice (Fig. 4B). We also found a direct correlation between the expression of tdTomato and tumor burden in untreated mice: mice with higher tumor burden displayed a higher per-cell expression of tdTomato (Fig. 4C), suggesting that naturally occurring pressure against tdTomato in this tumor model can delay its development.

In previous studies, using another visualizable protein, GFP, as a cancer-associated antigen, we had observed an inverse correlation between MHC and GFP expression on cancer cells, which was indicative of tumor evolution against an antigen-directed T-cell response (34). To assess whether similar dynamics may be occurring upon treatment with F4.CAR-T, we stained single-cell suspensions of mouse lungs for MHC class I (MHC-I) and MHC class II (MHC-II) and analyzed by flow cytometry. We observed that HKP1 cells from high tumor burden control mice presented high tdTomato expression and little to no detectable MHC-I or MHC-II. In contrast, in F4.CAR-T–treated mice and a fraction of low tumor burden untreated mice, HKP1 cells showed starkly elevated expression of MHC-I and MHC-II, coinciding with reduced per-cell expression of tdTomato and reduced tumor burden (Fig. 4D; Supplementary Fig. S5E). These observations were consistent across >5 independent experiments.

The reduction of tdTomato and increased MHC expression in tumor cells, associated with a reduced tumor burden phenotype, suggested F4.CAR-T treatment may be promoting an endogenous T-cell response against tdTomato as a tumor-associated antigen. Though a tetramer is not available for tdTomato-specific T cells, one is available for an immunodominant epitope of luciferase [H2Db-LMYRFEEEL (35)], which is also expressed by HKP1 cells. We injected mice with HKP1 cells, followed by control or F4.CAR-T, as above, and after 19 days, collected lungs, made single-cell suspensions, and stained with this tetramer to detect luciferase-specific T cells by flow cytometry. Consistent with our hypothesis, there was a significant increase in the frequency of luciferase-specific CD8 T cells in the F4.CAR-T–treated mice compared with control animals (Fig. 4E).

Taken together, these findings indicate treatment with F4.CAR-T not only eliminated tumor macrophages, but also promoted expansion of tumor-antigen associated T cells and immune editing of tumor cells. This helps provide insight into the mechanisms by which F4.CAR-T mediate reduction of tumor burden.

HKP1 lung tumors have been previously shown to be partially responsive to PD-1 blockade (36). We sought to assess how F4.CAR-T treatment compares to anti–PD-1, and to evaluate whether they might synergize. We injected HKP1 lung tumor-bearing mice with F4.CAR-T, anti–PD-1, or both anti–PD-1 and F4.CAR-T (Fig. 4F). Anti–PD-1 trended to reduce tumor burden, and F4.CAR-T mediated a similar reduction, indicating equivalent efficacy of the two approaches (Fig. 4G). Combination treatment of anti–PD-1 and F4.CAR-T also led to a decrease in tumor burden, but it did to the same extent as each of the monotherapies (Fig. 4G). When we analyzed lungs by flow cytometry, we once again found that treatment with F4.CAR-T led to a significant decrease in tdTomato^high HKP1 cells, and we also discovered this same process was facilitated by PD-1 blockade (Fig. 4H). Combined treatment did not further cause reduction in tdTomato expression by HKP1 cells.

These observations suggests that CAR T–mediated macrophage depletion and PD-1 blockade both delayed HKP1 tumor growth by facilitating endogenous T cell–mediated elimination of the most highly antigenic tumor cells. This mechanistic redundancy could provide an explanation for the lack of synergy between both strategies in this tumor model.

Our observations that treatment with F4.CAR-T led to increased MHC-I and MHC-II expression on cells in the tumor led us to consider how this might be mediated. Among the strongest inducers of MHC is IFNγ, which is one of the dominant cytokines produced by activated T cells. We thus hypothesized that F4.CAR-T–mediated release of IFNγ may be playing a role in sensitizing the tumors to cellular immunity.

To determine whether F4.CAR-T–produced IFNγ was playing a significant role in tumor growth delay, we isolated CD8 T cells from Ifng^−/− mice, transduced them with the F4.CAR vector, and injected them into mice bearing HKP1 lung tumors (Fig. 5A). As a comparison control, mice were injected with wild-type F4.CAR-T cells. In mice injected with Ifng^−/− F4.CAR-T cells, the delay in tumor growth mediated by wild-type F4.CAR-T was completely abrogated (Fig. 5B). Furthermore, HKP1 tumor cells from mice treated with IFNγ-deficient CAR T cells failed to upregulate expression of MHC-I, MHC-II, or PD-L1 (Fig. 5C). Coinciding with the absence of upregulated MHC-I and MHC-II, use of Ifng^−/− CAR T cells did not lead to a decrease in tdTomato^high cells, as occurred in F4.CAR-T–treated mice (Fig. 5D). This was not because IFNγ directly dampens tdTomato expression, as in vitro treatment of HKP1 cells with IFNγ did not affect tdTomato expression (Supplementary Fig. S6).

Other immune cells in the TME were also found to show increased expression of MHC-II and PD-L1 in F4.CAR-T–treated mice (Supplementary Fig. S7A). These changes were again dependent on IFNγ from F4.CAR-T cells because upregulated MHC-II and PD-L1 was not observed in tumor-bearing mice injected with Ifng^−/− F4.CAR-T (Supplementary Fig. S7B). Ifng^−/− F4.CAR-T were not defective in their ability to infiltrate tumor-bearing lungs, as they reached comparable numbers to their wild-type counterparts (Supplementary Fig. S7C). Moreover, Ifng^−/− F4.CAR-T cells retained their ability to lyse target populations (Mo-Macs and eosinophils) comparable with that of wild-type F4.CAR-T cells (Supplementary Fig. S7D).

These studies indicate that F4.CAR-T–mediated control of tumor burden and immune editing is dependent on release of IFNγ. Tumor macrophages may thus serve as a fixed target for CAR T accumulation into the TME and release of IFNγ to facilitate an endogenous antitumor response.

One study employing a CAR targeting folate receptor beta (FRB)-expressing macrophages demonstrated delay of ID8 ovarian tumor growth in mice (14). To determine whether F4.CAR-T could also be effective in this tumor setting, we evaluated their efficacy in an aggressive ID8 ovarian tumor model in which the ID8 overexpress VEGF-A and Defensin Beta 29 (24). When injected into the peritoneal cavity, ID8 cells form disseminated tumors in the omentum and on the diaphragm, which are abundantly surrounded by macrophages (37). In this model, tumor growth is associated with the development of peritoneal ascites. We injected mice with ID8 cells intraperitoneally and treated them 2 weeks later with control T cells or F4.CAR-T (Fig. 6A). Though tumors grew in both groups, treatment with F4.CAR-T cells delayed weight gain and the development of tumor-induced ascites in treated mice (Fig. 6B and C).

We next evaluated F4.CAR-T cells activity in the context of pancreatic cancer, a typically macrophage-rich, T-cell excluded malignancy with few therapeutic options (38, 39). We utilized the KPC pancreatic cancer model in which the cancer cells carry Kras^G12D and Trp53R172H (25), two of the most common mutations in human pancreatic cancer (40). KPC cells were surgically inoculated into the pancreas of mice, and control T cells or F4.CAR-T cells were injected intravenously 7 days afterward. At day 14 after tumor inoculation, mice were euthanized and pancreas lesions weighed as a readout for antitumor effect and processed for flow cytometry and histology (Fig. 6D). By flow cytometry, we observed F4.CAR-T cells successfully infiltrated tumor masses (Fig. 6E), reduced the density of macrophages and eosinophils in the tumor lesions (Fig. 6F) and led to upregulation of MHC-I and MHC-II on tumor cells, as well as PD-L1 (Fig. 6G). Importantly, F4.CAR-T treatment reduced the mass of the primary pancreatic lesions at the time of sacrifice (Fig. 6H). These studies demonstrate that F4.CAR-T can mediate antitumor effects in different tumor types, including aggressive, non–T cell–inflamed tumor models.

We show here that targeting macrophages with CAR T cells can delay tumor growth in mouse models of lung, ovarian, and pancreatic cancer. Antitumor activity against lung tumors was associated with the expansion of tumor antigen-specific T cells and tumor immune editing and was dependent on IFNγ production by the CAR T cells, which stimulated upregulation of antigen presentation by tumor and myeloid cells in the TME. Some of these mechanistic traits could be also observed in the ovarian and pancreatic tumor models.

Previous reports have explored CAR T and NK targeting of TAMs or myeloid-derived suppressor cells (MDSC) in mouse tumor models. Ruella and colleagues identified CD123 as a shared antigen between Hodgkin lymphoma cells and M2-like TAMs and showed that tumor-directed CD123 CAR T cells could also be active against TAMs, avoiding immune suppression (12). Parihar and colleagues developed CAR NK-cell targeting NKG2D ligands expressed on tumor-associated MDSCs, achieving significant antitumor efficacy in a xenograft neuroblastoma model in mice (13). Rodriguez-Garcia and colleagues described the use of macrophage-targeted CAR T cells to treat ID8 ovarian tumors in mice (14). In this work, they used a CAR directed against FRB, which is upregulated on a subset of inflammatory macrophages, including many TAMs. They showed that FRB-targeted CAR T cells depleted macrophages in the peritoneal cavity and extended the survival of ID8 tumor-bearing mice. Promisingly, both Parihar and Rodriguez-Garcia showed that combining myeloid-targeted CAR T or NK with cancer cell–targeting CAR T cells synergistically enhanced mouse survival in the models tested (13, 14). Here, we expand on these findings by showing that CAR T targeting of macrophages can additionally delay tumor growth in models of lung (HKP1 tumor model) and pancreatic cancer (KPC tumor model), and by providing mechanistic insight into the processes leading to tumor growth delay in these models.

When analyzing HKP1 tumor lesions, we observed a striking decrease in tdTomato abundance within tumors that had persisted in F4.CAR-T–treated animals. KP tumors, an established lung tumor model (41), are not very immunogenic, but the incorporation of tdTomato acts a “neoantigen” when implanted (42). Because the HKP1 tumors grow aggressively in the mice and in most cases do not lose tdTomato, this indicates the tumor normally evades an immune response against it. This is likely to happen, at least partially, through downregulated expression of MHC-I, as indicated in our analyses. This is not unlike what happens in patient tumors which can carry different neoantigens or tumor associated antigens, such as NY-ESO-1 or MAGE, but downregulate antigen presentation (43). Conversely, we found increased expression of MHC-I and MHC-II on HKP1 cancer cells in F4.CAR-T–treated mice and observed that these remaining cells now presented reduced expression of tdTomato. We and others reported previously this can occur as a consequence of antigen editing, as cancer cells that lose or downregulate antigens are no longer under selective pressure to downregulate the antigen presentation machinery (34, 44, 45). Our studies therefore suggest that treatment with macrophage-targeted CAR T cells promoted T-cell immunity against antigen-expressing HKP1 cell clones and selected for the outgrowth of tumor cells expressing lower amounts of antigen. Though we could not directly measure tdTomato-specific T cells for lack of available MHC tetramers, we did find an expansion of luciferase-specific T cells (35) in the F4.CAR-T–treated mice, which further supports the notion that treatment promoted antitumor T-cell responses.

We show that release of IFNγ by F4.CAR-T cells was necessary for therapeutic benefit and immune editing of tdTomato^high cancer cells. Although Ifng^−/− F4.CAR-T cells did not delay tumor growth, they still mediated killing of macrophages and eosinophils. This does not rule out a potential contribution of macrophage reduction to the antitumor effect of F4.CAR-T. However, it does mean that this reduction is not sufficient, and that further TME conditioning by IFNγ is required. IFNγ has been established to play key roles in the cancer-immunity cycle and in the treatment efficacy of CAR T-cell therapy in solid tumors (46). Our comparison between wild-type and Ifng^−/− F4.CAR-T–treated mice indicated that the IFNγ released from the CAR T cells was reprogramming the immunophenotype of the tumor, as we detected upregulation of MHC-I and MHC-II on tumor and tumor-associated myeloid cells. There were likely additional effects of IFNγ, which induces expression of >150 hallmark genes, including the immunoproteasome components PSMB8 and PSMB9 (47), and the T-cell chemoattractants CXCL9 and CXCL10. It is probable that IFNγ-mediated upregulation of antigen presentation would have increased the sensitivity of antigen-expressing cancer cells to cellular immunity. This likely happens with cancer cell–directed CAR T cells as well, which would also release IFNγ (48); however, as noted above, a challenge for applying CAR T to solid tumors is a paucity of target molecules specific to the cancer cells (49). Cancer cells are also prone to mutations and epigenetic remodeling that leads to escape of target-low or -negative cells. Though there may not be an absolutely TAM-specific target, single-cell RNA sequencing (scRNA-seq) has identified surface molecules highly expressed by TAMs which are quite restricted to myeloid cells (see below). Thus, TAM can serve as a common and stable Trojan horse target with CAR T to condition the TME to facilitate endogenous tumor immunity, or to improve activity of cancer-targeted CAR T (14).

In this work, we have utilized CD28-based CAR T cells, which have been described to exert stronger cytolytic activity and exhibit shorter persistence than 4-1BB–based CAR T cells (50). This CAR configuration might be favorable in this context, as long-term persistence of anti-macrophage CAR T may not be essential or desired, so that macrophage depletion is transient. However, studies will be needed to evaluate whether a different CAR configuration, such as 4-1BB, may improve the therapeutic effect of the treatment.

HKP1 lung tumors had been previously shown to be partially responsive to PD-1 blockade (36). When tested side by side, F4.CAR-T were able to mediate a similar or greater tumor reduction as PD-1 blockade, but there was no synergistic enhancement of antitumor effects by combining F4.CAR-T cells and PD-1 blockade. Notably, we found that PD-1 blockade, similar to F4.CAR-T cells, resulted in enrichment of tdTomato^low/neg cancer cells. This outgrowth of poorly immunogenic tumor clones could provide an explanation for the ultimate tumor progression in mice treated with either agent, and for the lack of synergy between the two in this model. That is, both immunotherapies induced an anti-tdTomato endogenous immune response which cleared tdTomato^high cells, but then tdTomato^low/neg cells evaded this response and tumor growth continued. This suggests that a “ceiling” of response may have been reached in the specific models evaluated here which is dependent on the immunogenicity of the cancer cells. Whether this will occur in patients will reflect factors such as the antigenicity of the patient's tumors.

In addition to lung cancer, we showed that F4.CAR-T treatment can lead to benefit in models of ovarian and pancreatic cancer, and it did so while showing analogous mechanistic traits like the induction of MHC-I expression on tumor cells. Though we did achieve benefit in each model, the F4.CAR-T did not cure animals. As noted above, this appeared to be due to escape by neoantigen negative/low cells, at least in the HKP1 lung cancer model. The lack of cure was not unexpected, as our approach did not target cancer cells themselves, and the main rationale for developing this approach was to enhance existing treatment strategies. Though in the model, we evaluated there was no synergy with PD-1 blockade, using a macrophage-targeted CAR T in conjunction with a cancer cell–targeted CAR T, as Powell and colleagues reported previously (14), may be the most effective approach. The IFNγ-dependent induction of MHC-I and MHC-II on tumor cells would also suggest that this approach could sensitize tumors to antitumor TCR-based adoptive cell transfer (51), a therapeutic strategy that has already produced some positive results in models analogous to the ones presented here (45, 52).

Preconditioning lymphodepletion is typically employed prior to adoptive T-cell therapies, including CAR T cells (53). This is used to facilitate T-cell expansion in the recipient. In our studies, F4.CAR-T cells greatly expanded in mice without the need for prior lymphodepletion. This may have been possible because the cells being targeted, macrophages, are professional antigen presenting cells, which express many immunostimulatory genes, including costimulatory molecules and proinflammatory cytokines, that can enhance T-cell activation. Indeed, two previous studies used mRNA or amph-ligand vaccines to promote expression of a CAR target in DCs and macrophages as a means to expand CAR T cells in vivo (54, 55).

We designed a CAR construct directed against the mouse pan-macrophage marker F4/80 to be able to achieve broad targeting of macrophages. It is important to note that F4/80 is not intended as a clinical target, and that clinical translational will be best served by using a more TAM-specific target, such as FRB. We chose a pan-macrophage approach for proof-of-principle studies to probe the effectiveness of macrophage-targeting CAR T and gain insight into its mechanisms of action. It is worth noting that broad but transient macrophage depletion is tolerated, as demonstrated in studies using pan-macrophage targeting drugs such as CSF1/R inhibitors and clodronate liposomes (8, 56). In our studies, we did not observe acute toxicity or lethality following treatment with the F4.CAR-T, though longer-term studies are necessary to make safety conclusions. It will be particularly important to assess the brain, where CAR T cells can traffic if their target is expressed (57), and ensure microglia are not a major target of the CAR. Use of TAM-specific CARs will undoubtedly provide a better safety profile, and this will require identification of targetable surface proteins that are as restricted to TAMs as possible. FRB is an excellent target, as it is not highly expressed by most tissue-resident macrophages. Though whether it is sufficiently expressed by enough human TAMs will need to be established. Work from several labs, including our own, have sought to use scRNA-seq to identify genes uniquely expressed by TAMs (9, 10, 31, 58, 59). These efforts have resulted in the identification of tumor-enriched macrophage profiles across different cancer types and have identified a number of genes that are relatively restricted to TAMs, such as TREM2 and GPNMB. However, both molecules are constitutively expressed on microglia, so their use as a target needs to be considered. Because immunosuppressive TAMs share phenotypic similarities with anti-inflammatory macrophages (60), identifying a completely TAM-specific target may not be feasible. However, the inclusion of suicide genes into the CAR T could be used as a means to ensure targeting and depletion is transient if needed (61). If it is determined that IFNγ delivery is sufficient for antitumor effect, it may also be feasible to engineer anti-TAM CAR T cells that do not kill target cells (62), but still serve as a Trojan horse for delivery of cytokines or other factors to reprogram the TME and facilitate tumor elimination. Spatial genomics using new technologies may help in identifying optimal pathways to target to remodel the TME in a manner that best favors tumor immunity (63).

In summary, we have here shown that targeting of macrophages using CAR T cells can achieve antitumor efficacy as a monotherapy in different models of solid organ tumors, bypassing the need for expression of CAR targets on tumor cells. This approach has promising features to become a new tool in the cancer immunotherapy repertoire, making it possible to deliver CAR T cells and IFNγ into solid tumors in a tumor antigen-agnostic manner.

M. Merad serves on the scientific advisory board Innate Pharma Inc., DBV, Inc. Genenta Inc. M. Merad also serves on the scientific advisory board and holds stock from Compugen Inc., Myeloid Therapeutics Inc., Morphic Therapeutic Inc., Asher Bio Inc., Dren Bio Inc., Nirogy Inc., Oncoresponse Inc., Owkin Inc., Pionyr Inc., OSE Inc., and Larkspur Inc. M. Merad receives funding for contracted research from Regeneron Inc. and Boerhinger Ingelheim Inc. M. Merad is a named co-inventor on an issued patent for multiplex IHC to characterize tumors and treatment responses. The technology is filed through Icahn School of Medicine at Mount Sinai (ISMMS) and is currently unlicensed. This technology was used to evaluate tissue in this study and the results could impact the value of this technology. No disclosures were reported by the other authors.

A.R. Sánchez-Paulete: Conceptualization, investigation, methodology, writing–original draft, writing–review and editing. J. Mateus-Tique: Conceptualization, investigation, methodology, writing–original draft, writing–review and editing. G. Mollaoglu: Methodology. S.R. Nielsen: Methodology. A. Marks: Methodology. A. Lakshmi: Methodology. J.A. Khan: Resources, methodology. C.M. Wilk: Methodology. L. Pia: Methodology. A. Baccarini: Methodology. M. Merad: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing. B.D. Brown: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing.

We thank Ivan Reyes-Torres, Maxime Dhainaut, Abishek Vaidya, Chittampalli Yashaswini, and Achuth Nair at Mount Sinai for helpful discussions. We also thank Alan Soto and Frances Avila, from the Biorepository and Pathology Core at Mount Sinai; and Yu Zhou, from the Biomedical Engineering and Imaging Institute at Mount Sinai, for technical assistance. We thank the Center for Comparative Medicine and Surgery at Mount Sinai for animal care. We also thank Vivek Mittal (Weill Cornell) for the HKP1 cells, Jean J Zhao (Dana Farber) for ID8 cells, Dave Tuveson (CSHL) for KPC cells, and Jordi Ochando (Icahn School of Medicine at Mount Sinai Hospital) for Ly6C-depleting antibodies. B.D. Brown was supported by NIH (R01CA257195) and a grant from the Alliance for Cancer Gene Therapy. M. Merad. was supported by NIH (R01CA254104). The project was also supported by a grant from the Applebaum Foundation. A.R. Sánchez-Paulete was supported by a grant from Fundación Alfonso Martín Escudero (Spain).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.