T cell–mediated antibody-dependent cellular cytotoxicity provides a therapeutic approach for solid cancers

Extensive research spanning over the past several decades has highlighted the important role of IgG receptors in facilitating protective immunity and in pathologies. Experimental mouse models as well as clinical data strongly suggest that alterations in Fcγ receptor (FcγR) signaling, germline mutations, and polymorphism drastically affect the progression of a wide variety of pathologies including autoimmunity and cancer (1, 2). In human, there are 8 IgG receptors: the type I receptors FcγRIIa, FcγRIIc, FcγRIIIa, FcγRIIIb, the high-affinity FcγRI, and the inhibitory receptor FcγRIIB; and the type II receptors CD209 and CD23. These receptors differ markedly in their expression patterns on leukocytes, their affinity for IgG antibodies, and the biological process they promote.

Since the signaling of all type I FcγRs is conserved, the divergent immunological processes that IgG promotes stem largely from the pattern of receptors it occupies on leukocytes. Therefore, dissecting the distribution of FcγR expression across immune cell subsets and the subsequent process that they induce has been the focus of extensive studies. FcγRI is composed of an α chain, which binds the IgG Fc and a γ chain co-receptor, which contains immunoreceptor tyrosine activation motifs (ITAMs) that trigger signaling upon phosphorylation (1, 3). Clustering and crosslinking of FcγRs by IgG immune complexes induces phosphorylation of tyrosine residues within the ITAM on the FcγRI γ chain (4). The phosphorylated ITAM then serves as a docking site for Src-family kinases (SFK) and for cytoplasmatic Syk tyrosine kinases (5, 6). In turn, they activate the mitogen-activated protein kinase/extracellular signal–regulated kinase (MAPK/ERK) and phosphoinositide 3-kinase/protein kinase C (PI3K/PKC) pathways, and induce transcription of inflammatory mediators, antibody-dependent cell cytotoxicity (ADCC), and antibody-dependent phagocytosis (7, 8).

While expression of FcγRs on T cells has been reported since the 1970s (9–12), the lack of sufficient tools to definitively discriminate between various subsets of lymphocytes made these observations questionable. Since subsequent studies attributed the receptors to NK cells and γ/δ T cell populations (13), the central dogma remained that T cells do not express FcγRs (2, 14). Yet, multiple publications have continued to challenge this paradigm, demonstrating that distinct subsets of activated T cells do express FcγRs (13, 15–19). Clémenceau and colleagues were the first to suggest T cell–mediated ADCC as a therapeutic strategy against tumor cells, by demonstrating that T cells armed with chimeric FcγRIIIa elicit ADCC against B lymphoblastoid cells coated with anti-CD20 (16). This concept was further supported by the observation that a subset of effector memory α/β T cells naturally express FcγRIIIa and are capable of executing ADCC (20). Subsequent studies further developed that concept by fusing the FcγRIIIa (native, or high affinity) to the CD3ζ transmembrane domains, including, or in combination with, 41BB/CD28 co-stimulatory molecules (21–23). While this strategy is still being tested, initial attempts to exploit T cell ADCC in the clinic were hampered by a significant toxicity profile.

Recently, our group described a mixed population of exhausted CD4⁺ T cells that express the high-affinity FcγRI augmented in mouse and human tumors (18). We demonstrated that the natural scaffold of the high-affinity IgG receptor, FcγRI, can be successfully used to elicit strong ADCC by both CD4⁺ and CD8⁺ α/β T cells. Yet, several key mechanistic questions regarding the activation of FcγR in T cells have remained unclear. In this work, we identify the physiological process by which CD8⁺ T cells acquire FcγRI and show these cells can utilize it to mediate a transient TCR-independent ADCC. Mechanistically, we show that FcγRI signaling in T cells is governed by a strict antigen density threshold and requires immune complex clustering, thereby preventing activation by soluble IgG. By preserving the native FcγRI signaling architecture, engineered T cells achieve selective killing of high-antigen-density tumor cells while sparing cells with normal antigen expression, overcoming a key limitation of chimeric antigen receptor–based (CAR-based) approaches. While T cell–mediated ADCC has been described by us and others (16, 24, 25), the present study provides a deeper insight into the mechanisms underlying density dependency, free-IgG unresponsiveness, and the imitations of using Fcγ-based receptors to elicit cytotoxicity.

Overall, this work defines the unique constraints and considerations involved in engineering T cell–mediated ADCC. In addition, it highlights how antigen density, rather than antigen identity, can serve as a key determinant governing selective cancer immunotherapy. Altogether, these findings position engineered T cell–mediated ADCC as a promising therapeutic strategy against solid tumors.

A rare population of FcγRI⁺CD8⁺ T cells is present in mouse tumors. In past projects, our group identified a mixed population of exhausted infiltrating CD4⁺ T cells that express FcγRI in mouse and human tumors (18). To determine whether a similar population exists among CD8⁺ T cells, we analyzed CD8⁺ T cells from draining lymph nodes (DLNs) and tumors of mice bearing day 10 B16F10 melanomas by cytometry by time-of-flight (CyTOF). After excluding non-T cell lineages, unbiased uniform manifold approximation and projection (UMAP) analysis identified a small population of TCRβ⁺CD8⁺ T cells expressing FcγRI. These cells co-expressed CD44, T-bet, perforin, Sca-1, CD40, CCR7, and CX3CR1, while lacking CD27 and CD62L. Unlike naive, central memory, and early activated CD8⁺ T cells, FcγRI⁺CD8⁺ T cells were detected exclusively in tumors, together with perforin^hi effector cells (Figure 1, A–C, and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.205984DS1).

Figure 1

A rare population of FcγRI⁺CD8⁺ T cells is present in mouse tumors. (A) UMAP plot of CD8⁺ T cells from tumors and DLNs of mice bearing day 10 B16F10 tumors (n = 3). (B) Heatmap of median marker intensities in CD8⁺ T cell clusters. (C) Differential abundance test results and normalized proportions for CD8⁺ T cell clusters obtained using the diffcyt package. (D) Mean percentages of FcγRI expression on T cells in mice bearing day 10 B16F10 (cold) tumors (n = 5) and YUMMER1.7 (hot) tumors (n = 8). (E) Immunofluorescence of FcγRI in histological sections from day 10 B16F10 (cold) tumor–bearing mouse and YUMMER1.7 (hot) tumor–bearing mouse. Original magnification, ×400. Arrowheads indicate positively stained cells. (F) Percentages of apoptotic B16F10 melanoma 24 hours after incubation with T cells from tumor-bearing mice (n = 4). CM, central memory; EM, effector memory; TRM, tissue-resident memory; GZB, granzyme B. Shown is the summary of 1 experiment out of at least 3 independent experiments. ****P < 0.00001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

FcγRI⁺CD8⁺ T cells were more frequent in immunogenic YUMMER1.7 tumors than in B16F10 tumors (~2.5% vs. ~0.5%) and were not detected in the DLNs or blood (Figure 1D and Supplemental Figure 1B). Confocal microscopy confirmed FcγRI expression on the CD8⁺ T cell membrane (Figure 1E). To determine whether the FcγRI was functional, infiltrating FcγRI⁺CD8⁺ T cells were isolated from OT-I mice bearing B16F10 tumors and selected for an irrelevant OVA-specific T cell receptor (TCR) to exclude TCR-mediated tumor recognition. These cells efficiently killed B16F10 cells only in the presence of the tumor-binding anti-gp75 antibody TA99, demonstrating functional FcγRI-mediated ADCC (Figure 1F).

FcγRI⁺CD8⁺ T cells are present at sites of inflammation in humans. We next examined whether FcγRI⁺CD8⁺ T cells exist in human tumors. Reanalysis of CyTOF data from patients with head and neck squamous cell carcinoma identified FcγRI expression almost exclusively among activated CD8⁺ T cells characterized by CD69 and CD38 expression and reduced TCF-1 and CD45RA. FcγRI⁺ cells also expressed FcγRIII (Figure 2, A and B, and Supplemental Figure 1, C and D). Flow cytometric analysis of freshly resected tumors detected FcγRI⁺CD8⁺ T cells in all 5 melanoma patients examined, but not in 9 patients with pancreatic ductal adenocarcinoma (PDAC). Confocal microscopy confirmed FcγRI⁺CD8⁺ T cells within melanoma tumors, frequently adjacent to FcγRI⁺ non-T cells (Figure 2, C and D, and Supplemental Figure 1E). In addition, we checked for the existence of FcγRI⁺CD8⁺ T cells in inflammatory settings. FcγRI⁺CD8⁺ T cells were also detected in tonsils from 4 of 10 patients with recurrent tonsillitis, including 2 patients with circulating FcγRI⁺CD8⁺ T cells, and in the blood of all 6 patients who had recovered from acute COVID-19 (Figure 2E and Supplemental Figure 1, F and G).

Figure 2

FcγRI⁺CD8⁺ T cells are present at sites of inflammation in humans. (A) UMAP plot of CD8⁺ T cells from tumors, metastatic DLNs, and normal LNs of patients with head and neck squamous cell carcinoma (n = 10). (B) Heatmap of median marker intensities in CD8⁺ T cell clusters. CM, central memory; EM, effector memory; TRM, tissue-resident memory; GZB, granzyme B. (C) Mean percentages of T cells expressing FcγRI, out of total T cells (n = 5 for healthy patients’ blood, n = 10 for PDAC patients, n = 5 for melanoma patients). (D) Immunofluorescence of FcγRI in histological sections from patient with breast cancer. Original magnification, ×400. Arrowheads indicate positively stained cells. (E) Mean percentages of FcγRI-expressing T cells from freshly resected tonsils of patients suffering from chronic tonsillitis, blood of COVID-19–recovered patients, and blood of healthy patients (n = 5 for healthy blood, n = 5 for COVID-19 blood, n = 10 for tonsillitis patients). Shown is the summary of 1 experiment out of at least 3 independent experiments.

CD8⁺ T cells acquire FcγRI through trogocytosis from myeloid cells. To investigate the origin of FcγRI on CD8⁺ T cells, we reanalyzed published single-cell RNA-seq (scRNA-seq) datasets of tumor-infiltrating T cells. Although FcγRI protein was readily detected, FCGR1A transcripts were absent from CD8⁺ T cell clusters, suggesting a non-transcriptional mechanism (Figure 3, A–C, and Supplemental Figure 2, A and B).

Figure 3

CD8⁺ T cells acquire FcγRI through trogocytosis from myeloid cells. (A) UMAP plot of tumor-infiltrating CD8⁺ T cell subclusters (n = 316). (B) Relative intensity of FCGR1A superimposed on UMAP projection. (C) Percentage of FCGR1A-expressing CD8⁺ T cells across T cell clusters (n = 316). (D) Mean percentages of FcγRI expression on CD8⁺ T cells co-cultured with monocytes and various cytokines after 1 and 6 days (n = 4). BF, bright-field. (E) Representative FACS analysis of FcγRI-expressing CD8⁺ T cells co-cultured with monocytes. (F) Representative structured illumination microscopy (SIM) of CD8⁺ T cells co-cultured with monocytes. Scale bars: 2 μm. Original magnification, ×630. Shown is the summary of 1 experiment out of at least 3 independent experiments. *P < 0.05; **P < 0.001; ****P < 0.00001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

Because FcγRI⁺CD8⁺ T cells were consistently observed near FcγRI⁺TCR^– cells, we tested whether FcγRI could be acquired through trogocytosis. Activated human CD8⁺ T cells were co-cultured with autologous CD14⁺ monocytes. FcγRI became detectable on approximately 1% of T cells after 1 day and increased to approximately 2% after 6 days or following IFN-γ exposure, whereas cytokines alone did not induce FcγRI expression (Figure 3, D and E, and Supplemental Figure 2, C and D). Lattice structured illumination microscopy (SIM) imaging demonstrated acquisition of large membrane fragments from monocytes by CD8⁺ T cells (Figure 3F and Supplemental Figure 2E). These findings indicate that FcγRI is acquired by CD8⁺ T cells through trogocytosis.

FcγRI acquired by CD8⁺ T cells remains functional for several hours. We next tested whether trogocytosed FcγRI remains functional. To this end, CD8⁺ T cells were co-cultured with autologous CD14⁺ cells and later exposed to HER2-containing lipid bilayers in the presence of trastuzumab. Approximately one-third of FcγRI⁺ cells formed immune synapses within 20 minutes and exhibited phosphorylated ERK1/2 (Figure 4, A–C). FcγRI accumulation at the synapse was also observed upon interaction with HER2-expressing tumor cells (Figure 4D and Supplemental Figure 2F). Importantly, FcγRI expression was lost within 4–6 hours after sorting, indicating rapid receptor turnover rather than a stable FcγRI⁺ T cell subset. Thus, CD8⁺ T cells can transiently acquire functional FcγRI from myeloid cells through trogocytosis.

Figure 4

FcγRI acquired by CD8⁺ T cells remains functional for several hours. (A) Representative TIRF imaging of sorted FcγRI-expressing CD8⁺ T cells interaction through intermediate antibodies with Her2 on SLB. (B and C) Mean fluorescence intensity of HER2 (B) and p-ERK (C) during interaction of FcγRI-expressing CD8⁺ T cells with HER2 on SLBs with and without a relevant antibody intermediate (n = 35). (D) Representative SIM-sorted FcγRI-expressing CD8⁺ T cells forming immune synapses with NCI-N87 cancer cells in the presence of trastuzumab. Shown is the summary of 1 experiment out of at least 3 independent experiments. Original magnification, ×1000 (A and D). **P < 0.001, ***P < 0.0001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

FcγRI signaling is controlled by antigen density. To define the signaling requirements of FcγRI-mediated ADCC, human CD8⁺ T cells were engineered to express FcγRI and its ITAM-containing FcRγ chain (AG2G) (Figure 5A). On HER2-containing lipid bilayers, FcγRI formed dense clusters with HER2-trastuzumab complexes while excluding TCRβ and CD45. FcγRI clustering required a minimum HER2 density of approximately 50 molecules/μm², whereas TCR signaling occurred at substantially lower densities (10–20 molecules/μm²) (Figure 5, B and C, and Supplemental Figure 3A).

Figure 5

FcγRI signaling is controlled by antigen density. (A) Illustration of FcγRI (A2G) and AG2G constructs. (B) Representative SIM-TIRF imaging of transduced T cell interaction through intermediate antibodies with HER2 on SLBs. (C) Integrated HER2 density on SLBs during interaction with transduced T cells (3 donors, n = 9–11 each). (D) Representative TIRF imaging of transduced T cell interaction through intermediate antibodies with HER2 on SLBs.Original magnification, ×1000 (B and D). (E–G). Integrated density of p-Syk (E), p-ERK (F), and p-LAT (G) during interaction of transduced T cells with HER2 on SLBs (n = 10). (H) Mean levels of IFN-γ in the supernatants of transduced human T cells incubated 48 hours with fixed antibody (n = 4). (I) Principal components analysis of RNA-seq from transduced T cells activated with human antigens (n = 4). Experiments were repeated at least 3 independent times. *P < 0.05; **P < 0.001; ***P < 0.0001; ****P < 0.00001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

FcγRI signaling depended on Syk rather than ZAP70 and induced phosphorylation of LAT and ERK1/2. Addition of the FcRγ chain enhanced phosphorylation of Syk, LAT, and ERK1/2 but did not reduce the antigen-density threshold, with AG2G activation also beginning at approximately 50 HER2 molecules/μm² (Figure 5, B and C, and Supplemental Figure 3B). Increasing ITAM copy number further resulted in ligand-independent tonic signaling (Supplemental Figure 3, C and D). Consistent with enhanced signaling, AG2G T cells exhibited greater tumor-cell killing and IFN-γ production than cells expressing FcγRI alone (Figure 5, D–H).

Transcriptomic analysis further demonstrated that AG2G T cells remained minimally activated below the antigen-density threshold but underwent marked activation above that threshold, characterized by increased TNF-α– and IFN-γ–associated programs and reduced TGF-β signaling (Figure 5I and Supplemental Figure 3, E–G).

T cell ADCC selectively targets cells exceeding a defined antigen-density threshold. We next tested whether antigen-density-dependent signaling enables selective tumor targeting. AG2G T cells were exposed to tumor cells expressing above-threshold EGFR, borderline HER2, or below-threshold MUC16. Cetuximab induced robust tumor-cell killing, whereas trastuzumab produced limited killing and anti-MUC16 antibody produced no detectable cytotoxicity (Figure 6, A and B). FcγRI and cortical actin accumulated at the tumor-cell interface only in the presence of the activating antibody, forming stable synapses lasting 30 minutes or longer (Figure 6, C and D, and Supplemental Figure 3H).

Figure 6

T cell ADCC selectively targets cells exceeding a defined antigen-density threshold. (A) FACS absolute quantification of antigen expression on HT29 tumor cells (n = 6). (B) Normalized HT29 cell confluence following incubation with transduced human T cells in combination with cetuximab or trastuzumab or sofituzumab (n = 4 for each combination). (C) Mean synapse length between AG2G-transduced T cells and tumor cells in the presence of relevant antibody (cetuximab), irrelevant antibody (trastuzumab), and no antibody (n = 18). (D) Representative SIM of HT29 tumor cells incubated overnight with AG2G-transduced T cells. Original magnification, ×630. (E) Integrated density of HER2 on SLBs during interaction with transduced T cells (3 donors, n = 8 cells each). (F) Normalized epithelial cell count following incubation with AG2G-, CAR HER2–, or sham-transduced T cells in combination with trastuzumab (n = 4). (G) Tumor size following treatment with AG2G and relevant antibody (anti–mouse HER2), irrelevant antibody (rituximab), or sham and relevant antibody (n = 4). Experiments were repeated at least 3 independent times. *P < 0.05; ***P < 0.0001; ****P < 0.00001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

Compared with donor-matched HER2 CAR T cells, which responded to HER2 densities as low as 10 molecules/μm², AG2G T cells displayed a substantially higher activation threshold (Figure 6E and Supplemental Figure 3I). Consequently, HER2 CAR T cells efficiently killed both normal epithelial cells and HER2-overexpressing tumor cells, whereas AG2G T cells selectively killed tumor cells while sparing normal cells (Figure 6F, Supplemental Figure 3J, and Supplemental Videos 1–6). In vivo, AG2G T cells significantly reduced established HER2-overexpressing tumors only when administered with a tumor-binding antibody and caused no detectable inflammation in major organs (Figure 6G). These findings demonstrate that FcγRI-mediated T cell activation is strongly dependent on antigen density and can discriminate between malignant and normal tissues expressing the same antigen.

Soluble IgG does not activate AG2G T cells. We next compared AG2G with a conventional FcR-based CAR containing the FcγRI extracellular domain linked to CD28 and CD3ζ signaling domains. Unlike AG2G, the FcR-based CAR induced cytotoxicity and cytokine production in response to both tumor-bound and irrelevant soluble antibodies (Figure 7, A–C, Supplemental Figure 4A, and Supplemental Videos 7 and 8). In contrast, AG2G T cells remained inactive in the presence of soluble trastuzumab unless antibody concentrations exceeded approximately 1.375 mg/mL (Figure 7D and Supplemental Figure 4, B and C). Fluorescence analysis demonstrated rapid removal of surface-bound IgG from AG2G T cells, with clearance within approximately 2 hours. High-resolution imaging showed that soluble IgG was internalized and degraded in endosomal compartments (Figure 7, E and F, Supplemental Figure 4D, and Supplemental Videos 9 and 10). Thus, soluble antibody is efficiently cleared without inducing productive FcγRI signaling.

Figure 7

Soluble IgG does not activate AG2G T cells. (A) Normalized H1975 cancer cell confluence following incubation with AG2G- or CAR-based FcγR-transduced T cells in combination with trastuzumab (n = 4). (B) Normalized kidney epithelial cell confluence following incubation with AG2G- or CAR-based FcγR-transduced T cells in combination with trastuzumab (n = 4). (C) Mean levels of IFN-γ in the supernatants of AG2G- and CAR-based FcγR-transduced T cells incubated with various concentrations of free antibody (n = 3). (D) Normalized HT29 cell confluence following incubation with AG2G-transduced T cells in combination with various concentrations of trastuzumab (n = 4) (E) Mean percentages of stained transduced human T cells following wash of FITC-IgG1 (n = 3). (F) Representative SIM of AG2G-transduced human T cells following addition of FITC-IgG1. Scale bars: 1 μm. Original magnification, ×630. Arrowheads indicate positively stained organelles. Experiments were repeated at least 3 independent times. *P < 0.05; ****P < 0.00001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

Soluble IgG does not interfere with FcγRI-mediated tumor-cell killing. To define the mechanism underlying this selectivity, we compared FcγRI signaling following exposure to immobilized versus soluble IgG. Immobilized trastuzumab induced FcγRI and phosphorylated-Syk clustering, and cortical actin accumulation, whereas soluble IgG induced none of these responses (Figure 8, A and B, and Supplemental Figure 4E). These findings indicate that productive AG2G signaling requires spatially restricted receptor clustering at an antibody-coated target–T cell interface.

Figure 8

Soluble IgG does not interfere with FcγRI-mediated tumor-cell killing. (A) Representative SIM of AG2G-transduced T cells incubated for 20 minutes with trastuzumab. Scale bars: 1 μm. Original magnification, ×630. (B) Mean percentages of p-Syk in AG2G-transduced T cells incubated for 20 minutes with trastuzumab. (C) Mean counts of HER2⁺ HT29 tumor cells following incubation with AG2G-transduced human T cells and in the presence of trastuzumab and IVIg (n = 4). (D) Mean percentages of stained transduced human T cells blocked with 10 mg/mL trastuzumab and incubated with FITC-IgG1 (n = 3). Experiments were repeated at least 3 independent times. *P < 0.05; **P < 0.001; ***P < 0.0001; ****P < 0.00001 by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

Importantly, high concentrations of irrelevant human IgG (IVIg) did not substantially inhibit AG2G-mediated killing of HER2⁺ tumor cells in the presence of trastuzumab (Figure 8C). Although soluble IgG rapidly occupied most FcγRI molecules, subsequent antibody exchange demonstrated continuous receptor turnover (Figure 8D, Supplemental Figure 4F, and Supplemental Videos 11 and 12). Thus, soluble IgG neither activates AG2G T cells nor prevents their response to antibody-coated tumor cells.

Repeated AG2G T cell administration promotes tumor eradication and effector memory T cell formation. Next we tested the in vivo effect of AG2G therapy. A single infusion of 5 × 10⁶ AG2G T cells combined with trastuzumab controlled but did not eradicate established NCI-N87 gastric tumors in NSG mice. Yet, residual tumors retained HER2 expression, arguing against antigen loss as a mechanism of resistance (Figure 9A). AG2G T cells preferentially accumulated in tumors, peaking approximately 14 days after transfer and declining substantially by day 30 (Figure 9B). Weekly trastuzumab injection combined with 2 infusions of 5 × 10⁶ AG2G T cells, administered 21 days apart, resulted in complete tumor eradication, whereas a single infusion produced only partial control (Figure 9, C–E, and Supplemental Figure 5, A–D). Most persisting AG2G T cells in the bone marrow and spleen of tumor-free mice displayed an effector memory phenotype (Figure 9, F and G).

Figure 9

Repeated AG2G T cell administration promotes tumor eradication and effector memory formation. (A) Representative histology of HER2 in NCI-N87 tumor core from NSG mice treated with AG2G- or sham-transduced T cells in combination with trastuzumab. Scale bar: 100 μm. (B) AG2G-transduced T cell relative quantification (RQ) in various organs of NCI-N87 tumor–bearing NSG mice (n = 6). (C) NCI-N87 tumor volume in NSG mice following treatment with AG2G- and sham-transduced T cells in combination with trastuzumab (n = 6). (D) NCI-N87 tumor necrosis score in NSG mice following treatment with AG2G- and sham-transduced T cells in combination with trastuzumab (n = 6). (E) NCI-N87 tumor weight in NSG mice following treatment with AG2G- and sham-transduced T cells in combination with trastuzumab (n = 6). (F and G) Mean percentages of central memory (CM) and effector memory T cells from bone marrow (F) and spleen (G) in NSG mice challenged with NCI-N87 tumors following treatment with AG2G-transduced T cells in combination with trastuzumab (n = 4). Experiments were repeated at least 3 independent times. Statistical significance assessed by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

AG2G T cells produce memory subsets in vivo and provide long-term protection. To determine whether tumor eradication reflected cumulative T cell dose, mice received escalating numbers of AG2G T cells. Doses of 1 × 10⁷ or 2 × 10⁷ cells induced complete tumor regression, whereas 1 × 10⁶ cells had limited activity (Figure 10A). Persisting cells again predominantly displayed an effector memory phenotype (Figure 10, B and C).

Figure 10

AG2G T cells produce memory subsets in vivo and provide long-term protection. (A) NCI-N87 tumor volume in NSG mice following treatment with 250 μg trastuzumab and i.v. injection of AG2G-transduced T cells (n = 6). (B and C) Representative FACS analysis (B) and mean percentages (C) of memory T cells in mice challenged with NCI-N87 cells on day 60 (n = 6). (D) Illustration of experimental design. (E) Image of NCI-N87 tumor after tumor rechallenge with and without treatment with trastuzumab (n = 4). (F) NCI-N87 tumor volume in NSG mice following treatment with AG2G-transduced T cells and 250 μg trastuzumab. (n = 6). Experiments were repeated at least 3 independent times. Statistical significance assessed by ANOVA with Tukey’s correction for multiple comparisons. Error bars represent SEM.

To test the functionality of this memory subset, mice were rechallenged with NCI-N87 cells 2 months after tumor eradication. Following trastuzumab administration, all previously cured mice rapidly eliminated the rechallenged tumors, whereas untreated mice developed progressive disease (Figure 10, D–F). These results indicate that AG2G therapy can generate persistent effector memory T cells capable of antibody-dependent tumor control upon rechallenge.

AG2G T cells display enhanced tumor infiltration compared with CAR T cells. AG2G T cells’ ability to eradicate tumors suggested overcoming the infiltration limitation posed by solid tumors. To test that, we compared AG2G T cell tumor-infiltrating capacities to that of HER2 CAR– and sham-transfected T cells, all from the same donors. In all 3 tested donors, the levels of infiltrating AG2G-transduced T cells were significantly higher compared with CAR- and sham-transduced T cells, composing up to 7% of the total lymphocytes, and AG2G T cells could be found in the tumor core (Figure 11, A and B).

Figure 11

AG2G T cells display enhanced tumor infiltration compared with CAR T cells. (A) Mean percentage of transduced T cells out of non-tumor cells in NCI-N87 tumor core taken from NSG mice after treatment (n = 6) (B) Representative histology of core tumor from NSG mice treated with AG2G-transduced T cells in combination with trastuzumab. (C) Mean migration distance of transduced T cells on Matrigel (n = 3). (D) Mean cell area of transduced T cells on glass and Matrigel (n = 18). (E) Representative SIM of AG2G-, HER2 CAR–, and sham-transduced T cells on Matrigel. (F) Representative SEM of AG2G- and sham-transduced T cells on Matrigel. Experiments were repeated at least 3 independent times. Scale bars: 50 μm (B), 1 μm (E) (original magnification, ×630), and 7 μm (F). Statistical significance was calculated using ANOVA with Tukey’s correction for multiple comparisons. For cumulative distribution function statistical analysis, a Kolmogorov-Smirnov test was used. For interaction classification analysis, a Mann-Whitney U test was used. *P < 0.05; **P < 0.001; ***P < 0.0001; ****P < 0.00001. Error bars represent SEM.

Since there are no relevant suppressive cues from the tumor microenvironment of NSG mice, the main factor limiting T cell infiltration is the basement membrane that forms a mechanical barrier. Hence, we coated migration slides with artificial basement membrane (Matrigel) and recorded T cell motility and migration distance towards CCL5 by confocal microscopy. Consistent with their in vivo infiltration capacities, AG2G-transduced T cells moved rapidly on the slides and for a longer distance compared with CAR- and sham-transfected T cells from matched donors (Figure 11C and Supplemental Videos 13 and 14). Interestingly, we noticed that when AG2G-transduced T cells are placed on Matrigel, they acquire polarity characterized by actin polymerization at the moving front and a spreading morphology (Figure 11, D and E, and Supplemental Figure 6A). Analysis by electron microscopy further corroborated the flattened spreading morphology that AG2G T cells acquire once placed on Matrigel (Figure 11F).

Reduced basement membrane adhesion promotes AG2G T cell infiltration. Reanalysis of tumor-infiltrating T cell datasets identified a distinct integrin profile associated with FcγRIII⁺CD8⁺ T cells, characterized by increased CD11b, CD11a, and CD11c and reduced CD29 and CD49a, the latter being prominent mediators of collagen IV interactions (Figure 12A and Supplemental Figure 6B). AG2G T cells similarly displayed reduced CD49a and increased CD11b expression (Figure 12, B and C).

Figure 12

Reduced basement membrane adhesion promotes AG2G T cell infiltration. (A) Pearson correlation scores between the mean log-normalized expression of various integrin and FcR genes across T cell clusters from human single-cell published data (n = 316). (B) Mean fluorescence intensity (MFI) of CD11b on transduced T cells (n = 3). (C) MFI of CD49a on transduced T cells (n = 3). (D) Force spectroscopy setup illustration. P_asp, aspiration pressure. (E) Representative fluorescence images of AG2G, CAR HER2, and sham cells aspirated with a micropipette (green, CD8; red, CD64\HER2; magenta, F-actin). Scale bars: 2 μm. (F) Cumulative distribution function (CDF) of transduced T cell interaction with Matrigel-coated beads. (G) Interaction classification for transduced T cells. The data represent 5 independent experiments, with up to 3 measured interactions per cell (AG2G – 20 cells, n = 59; CAR HER2 – 21 cells, n = 61; sham – 22 cells, n = 62). Boxes represent the fraction of each interaction type, and whiskers indicate 5%–95% CI, shown only for “bound.” Supplemental Figure 7J shows CI’s for all interaction types. Experiments were repeated at least 3 independent times. Statistical significance was calculated using ANOVA with Tukey’s correction for multiple comparisons. For CDF statistical analysis, a Kolmogorov-Smirnov (K-S) test was used. For interaction classification analysis, a Mann-Whitney U (MWU) test was used.*P < 0.05; **P < 0.001; ****P < 0.00001. Error bars represent SEM.

To determine whether this profile alters T cell–ECM interactions, we developed an optical tweezers/micropipette force spectroscopy assay using Matrigel-coated beads (Figure 12D and Supplemental Figure 6, C–G). Integrin-dependent binding was confirmed by EDTA-mediated reduction of unbinding forces (Supplemental Figure 6, H and I). Although AG2G, sham, and HER2 CAR T cells exhibited similar probabilities of forming initial interactions with Matrigel, AG2G T cells showed substantially fewer strong interactions (~7% vs. ~35% for sham and CAR T cells). Consistent with this finding, most AG2G interactions were weak, with 52% occurring between 10 and 20 pN, whereas sham and CAR T cells displayed distributions shifted toward stronger interactions (>28 pN) (Figure 12, E–G, Supplemental Figure 6J, and Supplemental Videos 15–18). Jointly, our data suggest that the integrin profile expressed on AG2G-transduced T cells results in lower interactions with the basement membrane and subsequently improves their capacity to infiltrate cancers.

Activating the host T cells against cancer cells has been the focus of intensive research and the clinical success of several therapeutic strategies. Among them, cancer vaccines, checkpoint blockade, and adoptive cell transfer have demonstrated that efficient activation of tumor-reactive T cells can result in tumor eradication, at least temporarily (26–30). Nonetheless, most patients do not experience long-term benefits from these treatments due to reasons that are not fully clear (31–33). One of the problems underlying these strategies is that they rely on activation of the host T cell repertoire, which may not be sufficient to cover and target the appropriate tumor antigens (34, 35). Additionally, the inherent low affinity of TCRs for most tumor antigens (36) and the low expression patterns of these antigens on tumor cells (37) make its activation insufficient to kill tumors. Attempts to increase TCR affinity through point mutations often result in cross reactivity and off-target toxicities (38–40). To overcome these limitations, Gross and Eschar linked the high-affinity binding properties of an antibody to T cell signaling chains. Indeed, revolutionary clinical trials based on this fundamental design have demonstrated complete regression of refractory B cell lymphoma in patients (41, 42). Nonetheless, these therapies have proved effective only with hematological malignancies. One of the major limiting factors is the lack of identified tumor-specific antigens that limits these therapies and often results in on-target, off-tumor toxicity (43).

Here, we identify a rare subset of CD8⁺ T cells at inflammatory sites that transiently acquires functional FcγRI through trogocytosis from adjacent myeloid cells. Over the past decade, a growing body of evidence has demonstrated the transfer of functional surface receptors and membrane components from myeloid and antigen-presenting cells to CD8⁺ T cells (44). Huang and colleagues were the first to show that T cells, similarly to B cells, can extract MHC-peptide complexes from the immunological synapse with antigen-presenting cells and subsequently present them to neighboring T cells (45). Subsequent studies identified additional transferred molecules, including CD86 (46), CD14 (47), and recently even synthetic chimeric receptors (48), highlighting trogocytosis as a broader mechanism of intercellular communication within the immune system (44). Although several mechanisms have been proposed to explain T cell trogocytosis, the precise processes governing membrane acquisition and retention remain incompletely understood. While we did not directly investigate these mechanisms experimentally, we observed that large membrane fragments derived from myeloid cells can become transiently associated with the T cell surface before being internalized within approximately 1 hour.

We observed an increase in the rates of trogocytosis of FcγRI by T cells under inflammatory cytokines. This suggests that T cell–mediated ADCC could participate in inflammatory conditions where self-reactive IgG1 antibodies are present, such as autoimmunity or following allogeneic hematopoietic cell transplantation (49). Clinical observations demonstrate lower relapse rates in patients who develop mild graft-versus-host disease, as well as the efficacy of donor lymphocyte infusion in relapsed disease. Nakasone et al. have suggested that elevation in tumor-binding antibodies predicts host antitumor immunity (50). While several subsets of T cells have been shown to express FcγRs enabling ADCC, their role in graft-versus-tumor has mainly been attributed to direct killing through TCR-MHC mismatch, or activation of immune cells (49).

These findings encouraged investigators to harness T cell–mediated ADCC into therapeutic strategies against cancer. Clémenceau and colleagues were the first to apply that discovery to a therapeutic strategy in cancer by expressing CD16^158V fused to the FcεRIγ signaling chain in CD8⁺ T cells (16). While transduced T cells show in vitro killing of antibody-coated B cells, the effector/target ratios were very high (30:1) and no in vivo models were subsequently reported. The premise behind that construct is the generation of heterodimers with CD3ζ. Since it occurs specifically in NK cells, subsequent translation of this strategy was carried out using NK92 cells. An alternative strategy has been proposed by Kudo and Caratelli, who fused CD16^158V or CD32A^131R to 4-1BB (or CD28) followed by the CD3ζ chain (21, 25). While these constructs showed promising results in preclinical models, they have failed in the clinic, inducing the death of cancer patients in 2 cohorts. A similar design was later suggested by Ochi, who simply fused CD16^158V to the CD3ζ transmembrane (TM) and signaling domain (22). One issue with this design is that the CD3ζ TM, much like the CD8 and CD28 TM, contains signals for dimerization (positions D6, Y12, and T17) and dictate its association with the TCR-CD3 complex, hence making it hard to understand how this construct is organized. Yet, our results, in this work as well in our previous study (24), strongly suggest that maintaining the original FcγR scaffold is required to avoid response to freely circulating antibodies. This design maintains the natural regulation machinery and ensures that T cells will augment cytotoxic responses only against cells expressing antigens with densities above the activation threshold (approximately 15 nM) (51).

Following systemic administration, a substantial proportion of circulating T cells accumulates within the hepatic vasculature, where they are retained through interactions with liver sinusoidal endothelial cells, Kupffer cells, and other resident myeloid populations (52–54). Likewise, intravenously administered antibodies often exhibit prominent hepatic uptake due to FcR-mediated clearance and filtration by the reticuloendothelial system. This liver sequestration markedly reduces the fraction of therapeutic agents that ultimately reaches tumor sites and therefore represents a major obstacle to effective cancer immunotherapy (55, 56). In our mouse models, we found that increasing the dose of injected antibodies and T cells could partially compensate for this hepatic “sink” effect and restore therapeutic activity. Although elucidating the mechanisms underlying liver retention and clearance will be important for improving the pharmacokinetics and efficacy of T cell–based therapies, such investigations are beyond the scope of the present study. Overall, this study highlights the advantages of harnessing T cell–mediated ADCC as a possible therapeutic strategy for the treatment of a broad range of solid cancers.

Sex as a biological variable. Both male and female mice were used throughout the experiments in this study. The experiments were not specifically designed to detect sex-dependent differences, and the data were therefore not analyzed separately by sex. We expect the findings to be broadly applicable to both male and female mice.

Mice. WT C57BL/6 mice were purchased from Envigo and from The Jackson Laboratory, and NSG mice were purchased from The Jackson Laboratory. All mice were housed in an American Association for the Accreditation of Laboratory Animal Care–accredited animal facility and maintained in specific pathogen–free conditions. Male and female 8- to 12-week old mice were used in all experiments.

Cell lines. B16F10, YUMM1.7, HEK-293FT, and HT-29 cells were cultured in DMEM supplemented with 10% heat-inactivated FBS, L-glutamine, and penicillin/streptomycin (DMEM-Full). Cells were routinely tested for mycoplasma. B16F10 and HT-29 cells expressing H2B-tdTomato were generated by lentiviral transduction and isolated by FACS. Human PBMCs were obtained commercially, activated with anti-CD3 and IL-2 (Miltenyi Biotec) for 48 hours, and subsequently transduced. Primary human epithelial and endothelial cells were cultured according to the manufacturers’ recommendations. CD4⁺ and CD8⁺ T cells were isolated by magnetic selection (Miltenyi Biotec).

Construct design. Mouse and human FcγRI-based constructs, including AG2G, FcγRI variants, and control CAR constructs, were synthesized by GeneART and cloned into the pMIGII or pMSVG.1 vector. Constructs were verified by sequencing. HER2, EGFR, and H2B-tdTomato expression vectors were generated in pLVX. Lentiviral and retroviral particles were produced in HEK293-derived cells and used to transduce tumor cells or primary T cells, respectively, following standard protocols.

Primary T cell isolation. Human PBMCs were separated using Histopaque-1077 HYBRI-MAX (H8889, Sigma-Aldrich), activated for 48 hours with 50 ng/mL Ultra-LEAF Purified anti-human CD3 antibody (OKT3, BioLegend), together with 300 U IL-2 (Peprotech), followed by retrovirus transduction. T cells were cultured in RPMI 1640 supplemented with 1% pen-strep, 10% heat-inactivated FBS, 1% sodium pyruvate, 1% MEM-Eagle non-essential amino acids, 1% insulin-transferrin-selenium (PeproTech), and 50 μM β-mercaptoethanol (Sigma-Aldrich).

Primary human cells. Human PBMCs were purchased from STEMCELL Technologies and CellGeneration, activated for 48 hours with 50 ng/mL Ultra-LEAF Purified anti-human CD3 antibody (OKT3, BioLegend) and 300 U IL-2 (Peprotech), followed by retrovirus transduction. Human primary alveolar epithelial cells (H-6053, CellBiologics) and human primary kidney epithelial cells (H-6034, CellBiologics) were cultured in Complete Human Epithelial Cell Medium (H6621, CellBiologics) containing supplements and 100 U/mL penicillin and 100 μg/mL streptomycin on gelatin-coated cell culture flasks (0.2% solution on Greiner Bio-one flasks). Human pulmonary microvascular endothelial cells (3000, ScienCell/Provitro AG) were cultured in ECM (ScienCell, Provitro AG) containing supplements and 100 U/mL penicillin and 100 μg/mL streptomycin on fibronectin-coated cell culture flasks (Greiner Bio-one flasks). CD4⁺ and CD8⁺ cells were separated by Mojosort negative magnetic selection kits (BioLegend, 480010 and 480012).

Lentiviral infection. HEK-293FT cells were transfected with pLVX plasmids containing H2B-tdTomato, HER2, or EGFR under the EF1a promoter together with psPAX2 (Addgene plasmid 12260) and pCMV-VSV-G (Addgene plasmid 8454). Media containing viruses were collected after 24 and 48 hours. For transduction, cells were incubated with viruses and 100 μg/mL polybrene (Sigma-Aldrich) for 30 minutes followed by a 30-minute centrifugation before medium was replaced. After 3 days, cells that expressed tdTomato were sorted by FACSAria II (BD Biosciences).

Retroviral infection. For infection of human cells, 1.2 × 10⁶ 293GP cells (Takara Bio) were cultured in poly-L-lysine–coated 6-well plates and transfected using JetOptimus (Polyplus Transfections) with 1.4 μg RD114 envelope plasmid and 2 μg pMIGII or pMSVG.1 plasmid. Forty-eight and 72 hours later, supernatants were collected, filtered through 0.45 μm PVDF filters, and 1.5 × 10⁵ cells were infected in 24-well suspension plates coated with Retronectin (Takara) for 24 hours, then transferred to 12-well culture plates with additional supernatant from retroviral collection.

Tumor models. For the immunocompetent melanoma model, 2 × 10⁵ B16F10 cells were injected subcutaneously into C57BL/6 mice, and tumor growth was monitored twice weekly. Tumors were harvested at the predefined endpoint and enzymatically dissociated for downstream analysis.

For the xenograft model, 2.5 × 10⁶ NCI-N87 cells were injected subcutaneously into NSG mice. Treatment was initiated at a tumor volume of 70–120 mm³ and included saline, trastuzumab (250 μg/mouse/dose), or trastuzumab plus AG2G T cells (250 μg trastuzumab + 5 × 10⁶ cells/mouse/dose).

Immunohistochemistry. Tumors and healthy organs were paraffin embedded and sectioned at 5 μm thick. The sections were stained with hematoxylin-eosin (H&E) and studied using an Olympus BX51 microscope equipped with the 12MP SPOT InSight CMOS camera. HER2 staining was performed using an anti-HER2 antibody (Cell Signaling Technology, catalog 76799, lot 2). CD3 staining was performed using a rabbit monoclonal antibody against human CD3ε (Abcam, catalog ab25959, lot GR3273911-4).

Flow cytometry. Purified T cells were analyzed using flow cytometry (CytoFLEX, Beckman Coulter) for GFP expression or by staining with Alexa Fluor 647 anti-mouse CD64 (FcγRI) antibody (clone X54-5/7.1, BioLegend), Brilliant Violet 421 or APC anti-human CD64 antibody (clone 10.1, BioLegend), and Alexa Fluor 488 anti-human CD3 (clone HIT3a, BioLegend). Datasets were analyzed using FlowJo software (Tree Star).

CyTOF. Purified antibodies were obtained from commercial vendors and labeled with metal isotopes using the MAXPAR X8 chelating polymer kit (Fluidigm Corporation). A full list of antibodies, their elemental isotope tags, clone, and source are listed in Supplemental Table 1. Approximately 3 million cells per sample were used for staining and incubated with isotope-labeled antibodies. For viability staining, cells were washed with CyTOF Staining Media (composed of PBS with 2% FBS and 0.5% sodium azide) and incubated for 15 minutes in 500 μL of 250 nM rhodium solution (Fluidigm Corporation) at room temperature. For preliminary data filtration, acquired data were normalized and concatenated using Helios Software (Fluidigm Corporation) and uploaded into Cytobank (Cytobank Inc.) for further QC data processing. Gaussian parameters of the Helios system were used for doublet exclusion, and Ir⁺ and Rh^– were used for gating out dead cells and normalization beads. Data were transformed using an arcsinh(x/5) transformation.

CyTOF analysis. Cell population gating was performed using a custom hierarchical strategy implemented via the flowWorkspace (4.10.1) and openCyto (2.10.1) packages (https://bioconductor.org/packages/openCyto/). For downstream analysis, CD8⁺ T cells were isolated and clustered using a Self-Organizing Map (FlowSOM) based on the expression of manually selected phenotypic and functional markers. Differential abundance (DA) testing between tissue groups was performed using the diffcyt-DA-GLMM method from the diffcyt package (https://bioconductor.org/packages/release/bioc/html/diffcyt.html). Cluster-level counts were modeled using generalized linear mixed models with sample ID as a random effect.

Cytotoxicity and apoptosis assays. Tumor-cell killing was assessed using IncuCyte imaging and target-cell confluence. Mouse B16F10 cells were tested with anti-TRP1 or control antibodies, whereas human HT-29 cells were tested with trastuzumab or cetuximab. Apoptosis was independently assessed by flow cytometric measurement of caspase-3/7 activity.

RNA-seq and scRNA-seq analysis. RNA-seq differential expression was analyzed using DESeq2 (https://bioconductor.org/packages/DESeq2/), with pathway enrichment performed using clusterProfiler, GAGE (https://bioconductor.org/packages/gage/), and Pathview. Published tumor-infiltrating CD8⁺ T cell scRNA-seq data (GEO GSE156728) were reanalyzed using Scanpy (https://scanpy.readthedocs.io/en/1.9.8/), BBKNN (https://github.com/Teichlab/bbknn), UMAP (https://umap-learn.readthedocs.io/), and Leiden clustering (https://github.com/vtraag/leidenalg).

TIRF-SIM. Structural illumination and total internal reflection fluorescence–SEM (TIRF-SIM) were used to analyze T cell localization, signaling, migration, and synapse formation. Supported lipid bilayers (SLBs) containing 12.5 mol% DOGS-NTA and 87.5 mol% DOPC were prepared and functionalized with recombinant HER2 at defined densities (10 or 1,000 molecules/μm²). T cell synapse formation and phosphorylation of signaling molecules were assessed by fluorescence microscopy.

Confocal microscopy. Tumor tissues were fixed in 4% paraformaldehyde for 2 hours and equilibrated in a 30% sucrose solution for 48 hours. Tissues were then embedded in frozen tissue matrix (Scigen O.C.T. Compound Cryostat Embedding Medium, Thermo Fisher Scientific) and frizzed at –80°C. Infected T cells were plated on glass-bottom confocal plates, fixed, and stained overnight at 4°C with anti-human CD3 (clone HIT3a, BioLegend), APC–anti-human CD64 (clone 10.1, BioLegend). Mouse cells were stained with anti-CD3 (clone 17A2, BioLegend), anti-TCRβ (clone H57- 597), anti–FcγRI (clone X54-5/7.1, BioLegend). For migration assays, 15m Slide I^0.4 luer (80176, Ibidi) were covered with 1:2 Matrigel matrix (356231, Corning) in PBS with 5 × 10⁴ transduced T cells and videos were acquired after 1-hour incubation with 5-minute intervals between frames. All specimens were imaged using a Zeiss LSM 800 and analyzed by ZEN 2.3 (Zeiss) and IMARIS software 8.4 (Bitplan, RRID: SCR_007370).

SLB preparation. SLBs were generated from small unilamellar vesicles prepared by mechanical extrusion containing 12.5 mol% DOGS-NTA (Ni²⁺) and 87.5 mol% DOPC (all lipid reagents obtained from Avanti Polar Lipids) used at a final lipid concentration of 0.4 mM. Plasma-cleaned glass cover slips (Schott UK) were fixed to 6-channel flow chambers 489 (Sticky-Slide VI 0.4, Ibidi, Thistle Scientific LTD). Each channel was filled with liposome solution and incubated at room temperature for 30 minutes. SLBs were blocked with 3% BSA and subsequently incubated with fluorescently labeled recombinant HER2 for 60 minutes.

Bead coating with Matrigel matrix. Silica non-porous particles (beads; Spherotech Inc.) of 3.15 μm diameter were introduced into a 600 μL solution of cold Matrigel matrix (1:100, unless specified otherwise; 356231, Corning). Only beads with uniform Matrigel coating were used. The experiments were performed using a C-trap confocal fluorescence optical tweezers setup (Lumicks) and 60× water-immersion objective (NA 1.2). The optical trap is generated by a 10 W 1064 nm laser. T cells and Matrigel-coated beads were added to the chamber with DMEM-Full at room temperature. A micropipette aspiration setup, including a micromanipulator (uMp-3, Sensapex) holding a capillary of 5 μm diameter (VICbl-5-10-1-55, BioMedical Instruments) connected to a microfluidic flow controller (EZ-25, Fluigent), was integrated into our optical tweezers instrument.

Matrigel unbinding force measurement. Following cell introduction into the optical tweezers experimental chamber, a single cell and a Matrigel-coated bead were captured using the setup described above. The bead was held at a 4% laser power (~0.4 W), and its force was calibrated using a passive power–spectrum method. The trapped bead was brought into contact with the cell at a pushing force of approximately 10 pN and was maintained in contact for 30 seconds. Next, the bead was pulled at a constant speed of 0.8 μm/s, and its unbinding force was measured.

Statistics. Each experiment was performed at least 3 times and each experimental group consisted of 4–5 mice. Statistical tests were performed in Prism (GraphPad Software, Inc.). In cases where 2 conditions were compared across groups, statistical significance was determined by 2-way ANOVA with Tukey’s or Bonferroni-Šídák post hoc test. Statistical significance of survival curves was assessed using the log-rank Mantel-Cox test. In all cases, multiple hypothesis testing–adjusted P values were used to assess significance. A P value of less than 0.05 was considered significant.

Study approval. All animal experiments were approved by the Tel-Aviv University Institutional Animal Care and Use Committees, protocols TAU-MD-IL-2310-163-5 and TAU-MD-IL-2401-102-5.

Data availability. The scRNA-seq data we reanalyzed are publicly available in the NCBI GEO database under accession GSE156728 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE156728).