Superkine IL-2 and IL-33 Armored CAR T Cells Reshape the Tumor Microenvironment and Reduce Growth of Multiple Solid Tumors

In contrast to hematologic tumors (1–4), chimeric-antigen receptor (CAR) T-cell therapy remains largely ineffective against solid tumors due to several obstacles that limit CAR T-cell persistence and activity within the solid tumor microenvironment (TME). These obstacles include tumor-intrinsic expression of inhibitory ligands such as PD-L1 that induce T-cell exhaustion (5); the heterogenous expression of tumor antigens, which contributes to immune evasion (6); the absence of essential nutrients required for T-cell survival (7); and the presence of tumor-associated immunosuppressive cells including T regulatory (Treg) cells (8, 9), tumor-associated macrophages (TAM; ref. 10), and myeloid-derived suppressor cells (MDSC; ref. 11). One or a combination of these mechanisms results in tumor progression beyond control by the immune system. Methods that counter one or more of these obstacles are critically needed to increase the chance that CAR T-cell therapy will be efficacious against solid tumors.

To improve CAR T-cell therapy against solid cancers, several groups have engineered CAR T cells to express stimulatory cytokines. These fourth-generation CAR T cells, which are known as TRUCKs or armored CARs, aim to enhance the intrinsic activity of the CAR T cells through expression of T-cell stimulating cytokines such as IL-15 or IL-21 (12–14). This approach largely improves CAR T cell–intrinsic activity but is not designed to shift the immunosuppressive TME, although these cytokines can affect neighboring endogenous immune populations. Other groups have expressed cytokines such as IL-12 or IL-18 to aid in both CAR T-cell responses and activate local immune cells including recruitment and stimulation of macrophages (15–19). However, strategies to engage the endogenous immune response in the context of CAR T-cell therapy remain limited to a small number of type 1 cytokines. To date, type 2 cytokines such as IL-33 and IL-25, which can also regulate antitumor immunity (20, 21), have yet to be studied in the context of CAR T cells.

IL-33 is a type 2 cytokine that can promote the activation, expansion, and mobilization of type 2 innate lymphoid cells (ILC2; ref. 22). It is known for its role as an alarmin that is released after cell injury to alert the immune system to tissue damage, infection, or trauma (23, 24). It was recently reported that IL-33 expression in pancreatic tumor patients activates ILC2s to recruit CD103⁺ dendritic cells (DC) via CCL5, which, in turn, promotes CD8⁺ T-cell priming and expansion (25, 26). In this study, we paired IL-33 with superkine IL-2 (Super2), an engineered variant of IL-2 that binds to the IL-2 receptor beta and gamma chain complex at a higher affinity than wild-type IL-2 (27). Levin and colleagues showed that Super2 induces a superior expansion of cytotoxic T cells, leading to improved antitumor responses in vivo compared with IL-2, and that it elicits proportionally decreased expansion of Tregs (27).

Preclinical studies often evaluate human CAR T-cell activity against human tumor cell lines or primary tumor grafts using NSG mice to prevent xenogeneic rejection of human cells. However, the immunodeficient background of NSG mice cannot recapitulate the suppressive TME found in patients with cancer (28). Although these models are necessary to study human T-cell responses, engineering CAR T cells to overcome the immunosuppressive TME required us to use an immunoreplete syngeneic tumor model. To maintain the integrity of both positive and negative features of the host immune system, we determined the ability of CAR T cells to limit solid tumors without preconditioning with lymphodepleting chemotherapy or radiotherapy. CAR T cells were assessed for their ability to control B16F10 melanoma, which develops an immunosuppressive TME that prevents productive antitumor immunity, is poorly immunogenic and does not respond to checkpoint blockade immunotherapy (5). The ability of engineered CAR T cells to control tumor growth was also examined in the B16F10 lung metastatic tumor model, the immunogenic and checkpoint inhibitor–responsive MC38 colon cell carcinoma tumor model, and the B16F10-B7H6 tumor antigen model to evaluate CAR T-cell efficacy in tumors that have various TMEs and express different tumor antigens.

Here, we show that CAR T cells that coexpress Super2 and IL-33–stimulated endogenous immune cells to mount a broad antitumor immune response and shifted the TME from immune suppressive to immune stimulatory independent of CAR target or tumor type. Delivery of this combination of type 1 and type 2 cytokines via CAR T cells significantly delayed solid tumor growth across four tumor models and three target antigens, demonstrating its potential to induce a potent, universal antitumor response.

C57BL/6 (C57BL/6NCrl, strain 556), CD45.1 congenic (NCI B6-Ly5.1/Cr, strain 564), and C57BL/6 albino (strain 493) mice were purchased from Charles River Laboratories. CD8-deficient (CD8 KO, Jax stock # 002665), IFNγ-deficient (IFNγ KO, Jax stock # 002287), Perforin-deficient (Prf1 KO, Jax stock # 002407), and RORα-deficient (RORα KO, Jax stock # 005047) mice were purchased from The Jackson Laboratories and bred in the Dartmouth CCMR facility. Mice were 8 to 14 weeks old at the start of all experiments. Male and female mice were both used in these studies. All animal experiments were conducted with the approval of Dartmouth College's Institution Animal Care and Use Committee.

Depleting anti-CD4 (mAb clone GK1.5) was produced as bioreactor supernatants from hybridoma cell lines (ATCC) and administered at 250 μg/dose i.p. Depleting anti-NK1.1 (mAb clone PK136; Bio X Cell) was administered at 250 μg/dose i.p. Each dose was administered every 3 to 4 days beginning 1 day prior to tumor injection.

The pCigar retroviral backbone (gift from Dr. Bill Sha, UC Berkley) was used to generate the full-length and tailless TA99 CAR, and all cytokine-expressing constructs. Combination cytokine constructs expressed the Super2 sequence (27), a T2A self-cleaving peptide, and the type 2 cytokine. Mouse cytokine sequences included the full-length coding sequence followed by a (GGS)₂ linker, tandem V5-(His)₆ tag, and an internal ribosomal entry sequence (IRES) and GFP to determine transduction efficiency. The TA99 CAR construct contained the TA99 scFv sequence (gift from Dr. Darrell Irvine, Massachusetts Institute of Technology) fused to the CD8 transmembrane domain, the CD28 cytoplasmic domain followed by the CD3 zeta cytoplasmic domain. The tailless TA99 CAR lacked the CD28 and CD3ζ domains. The pSFG backbone was used for the NKG2D CAR (described in ref. 29) and the TZ47 CAR (described in ref. 30) construct, which included the CAR sequence, a T2A, followed by a truncated mouse CD19 sequence in which all signaling components were removed to act as a marker for transduction.

B16F10 (CRL-6475), HEK 293T (CRL-1168), and K562 (CCL-243) cell lines were purchased from ATCC. The MC38 (ENH204-FP) cell line was purchased from Kerafast. The B16F10-B7H6 cell line was generated as described previously (30). All cell lines were frozen at early cell passages (5–8), thawed immediately prior to the experiment, and cultured for 1 or 2 passages prior to tumor inoculation. Cell lines were purchased prior to 2010 and checked annually for mycoplasma contamination or as needed when cell growth deviated from normal.

B16F10 and B16F10-B7H6 melanoma cell lines and primary T cells were cultured in Complete RPMI-1640 media [RPMI-1640 (Cytiva SH30027FS) supplemented with 10% heat-inactive FBS (HyClone SH30910), 10 mmol/L HEPES (Gibco 15630080), 100 μmol/L nonessential amino acids (Gibco 11140050), 1 mmol/L sodium pyruvate (Gibco 11360070), 100 U/mL penicillin (Gibco 15140122), 100 μg/mL streptomycin (Gibco 15140122), and 50 μmol/L of 2-mercaptoethanol (Gibco 21985023)]. Primary T cells were also cultured in 25 U/mL recombinant human IL-2 (NCI-Frederick BRB Repository). HEK293T, K562, and MC38 cells were cultured in complete Dulbecco's modified Eagle media (DMEM) with a high glucose concentration (4.5 g/L) (Cytiva SH30022). Complete DMEM was DMEM supplemented as for Complete RPMI-1640 media without 2-mercaptoethanol.

To package retrovirus, 2.5 × 10⁶ HEK-293T cells were plated on a 10-cm dish in complete DMEM 18 hours before transfection to produce ecotropic virus. The cells were transfected using the calcium phosphate transfection method by premixing 10 μg of CAR or cytokine plasmid and 10 μg of pCL-Eco packaging plasmid (Addgene 12371) with 250 mmol/L CaCl₂ in 510 μL water prior to mixing with 510 μL 2X HEPE-buffered saline (HBS; Alfa Aesar J62623) and dropwise addition to 4×10⁶ HEK 293T cells. Media were replaced with fresh media approximately 8 hours after calcium phosphate treatment. Viral supernatants were harvested 48 hours after transfection and filtered through a 0.45-μm filter before immediate use or flash freezing and storing at −80°C.

Splenocytes from C57BL/6 mice or congenically marked CD45.1 C57BL/6 mice were transduced 18 to 24 hours after concanavalin A (1 μg/mL; Sigma C5275) stimulation and cultured in Complete RPMI media plus 25 U/mL of human IL-2. Mouse T cells were transduced with empty control or retrovirally encoded cytokine supernatant by resuspending activated T cells in viral supernatant and polybrene (1 μg/mL; Sigma 107689) at 8 million cells per well in 24-well plates followed by spinoculation at 1,500 × g for 90 minutes at 37°C. Approximately 24 hours later, indicated cell populations were transduced a second time to express the CAR receptor using the same protocol as for the retrovirally encoded cytokine. Two days after infection, cells were analyzed by flow cytometry to detect CAR expression before experimental set-up or CAR T-cell administration to mice (see Flow cytometry).

For human CAR T cells, deidentified PBMCs (DartLab) were stimulated for 2 days starting with anti-CD3 (OKT3, BioLegend 317326) in RPMI complete media with 100 U/mL human IL-2 for 2 days prior to transduction with TZ47 CAR and Super2 + IL-33 viral supernatants on sequential days as described above. Human CAR T cells were expanded for a total of 8 days before cells were analyzed by single-cell RNA sequencing (scRNA-seq) or cocultured with K562-luciferase tumors at effector-to-target ratio of 0.3:1, 1:1, 3:1, and 6:1 for 24 hours followed by the addition of 50 μL of luciferin (200 μg/mL; GoldBio LUCK), incubation at 37°C for 30 minutes, and detection of luminescence using a Centro LB960 Berthold Technologies luminometer.

1×10⁵ TZ47 CAR only or Super2 + IL-4, IL-5, IL-25, IL-33 or thymic stromal lymphopoietin (TSLP) T cells were cocultured with recombinant B7H6 (BioLegend 794006) or B16F10-B7H6 tumor cells for 24 hours at 37°C. Tumor cell lines were plated at a 1:1 and 0.5:1 effector:target ratio. The cell-free medium was collected and analyzed for IL-4, IL-5, IL-25, IL-33, and TSLP using ELISAs or LEGENDplex assays (BioLegend, 740055, 740056, 447104, 434104; Invitrogen 88-7333-22) as described in the manufacturers’ protocols using an Epoch BioTek plate reader with Gen5 1.11 2005 software or BD Accuri C6 Flow Cytometer and Software.

B16F10 tumor cells expressing PyRE9 luciferase (31) were plated at 5×10⁴ cells per well in a white 96-well flat-bottom plate. TA99 CAR only or Super2 + IL-4, IL-5, IL-25, IL-33 or TSLP T cells were added at various T-cell effector-to-target ratios (E:T) of 5:1, 1:1, 0.5:1. Cells were cocultured in Complete RPMI at 37°C for 24 hours followed by addition of 50 μL of luciferin (200 μg/mL) and incubated at 37°C for 30 minutes before detecting luminescence via a Centro LB960 Berthold Technologies luminometer and analyzed using MikroWin 2000 software.

Male and female mice were used for in vivo experiments. Tumor volume was calculated by multiplying the tumor length, width, and depth three times a week, and mice reaching maximum tumor burden (15 mm in diameter) or exhibiting moribund signs were euthanized. Each in vivo experiment was independently repeated at least twice.

For the B16F10 primary tumor model, 2×10⁵ B16F10 cells in 50 μL of HBSS were injected via intradermal (i.d.) injection into the right flank of shaved C57BL/6 mice. Tumors were established in mice for 6, 11, or 14 days and then treated with 7×10⁶ mouse TA99 CAR T, Super2 TA99 CAR T, IL-33 TA99 CAR T or Super2 + IL-4, IL-5, IL-25, IL-33, or TSLP CAR T cells via intravenous (i.v.) tail-vein injection. For the metastatic tumor model, 2 × 10⁵ B16F10 cells in 200 μL of HBSS were injected i.v. through the tail vein of C57BL/6 mice. After the tumors established for 6 days, 7 × 10⁶ TA99 or Super2 + IL-33 CAR T cells were injected through the tail vein. For luminescence imaging, TA99 CAR T-luciferase or Super2 + IL-33 CAR T-luciferase cells were visualized for 1-, 2- or 5-minute exposures using a Xenogen VivoVision IVIS bioluminescent and fluorescent imager (PerkinElmer) following i.p. injection of 200 μL of 15 mg/mL of luciferin. Total luminescence signal acquired at the 5-minute exposure time were quantified using Living Image software (PerkinElmer).

For the MC38 model, 1 × 10⁶ MC38 cells in 200 μL of HBSS were injected via subcutaneous injection (s.c.) into the right flank of shaved C57BL/6 mice. Tumors were established in mice for 6 days and then treated with 7 × 10⁶ to 8 × 10⁶ mouse NKG2D or Super2 + IL-33 NKG2D CAR T cells via i.v. tail vein injections.

For the B16F10-B7H6 tumor antigen model, 2×10⁵ B16F10-B7H6 cells in 50 μL of HBSS were injected i.d. into the right flank of shaved C57BL/6 mice. Tumors were established in mice for 6 days and then treated with 4×10⁶ mouse TZ47 CAR T or Super2 + IL-33 TZ47 CAR T cells via i.v. tail vein injections.

Final tumor weights were determined following tumor removal on day 17 or 19 after tumor inoculation, when control tumors reached a maximum diameter of 15 mm. B16F10 tumors that were analyzed by flow cytometry or scRNA-seq were harvested on day 15 after tumor inoculation and weighed. Tumors were then filtered through a wet 70-μm cell strainer (Falcon) and resuspended in RPMI media at 4°C prior to further flow cytometry or scRNA-seq analysis.

To isolate the tumor-infiltrating leukocytes (TIL), tumors were digested and resuspended in 80% percoll (Cytiva 17-0891) and overlayed by 40% percoll to create a separation gradient. Samples were spun at 500 × g for 25 minutes with no break. Buffy coat layer between 40% and 80% percoll was harvested and washed in PBS before further analysis and staining for flow cytometry (see Flow cytometry).

CAR T cells were characterized prior to adoptive transfer by flow cytometry following staining with antibodies specific for CD3ε (145-2c11, BioLegend), and CD4 (GK1.5, BioLegend) and biotinylated protein L (Pierce 29997) followed by streptavidin–PE (BioLegend 740452) to detect scFv CAR expression. TILs were isolated and characterized by flow cytometry following staining with antibodies specific for B7H6 (875001; R&D Systems FAB7144P), CD279 (PD-1) (29F.1A12, BioLegend 135231), CD3 (17A2, BioLegend 100204), CD366 (Tim-3; RMT3-23, BioLegend 119715), CD4 (RM4-5, BioLegend 100509 or GK1.5, BD Biosciences 563790), CD45.1 (A20; BioLegend 110727, 110736), CD45.2 (104; BioLegend 109839, 109851), CD64 (X54-5/7.1, BioLegend 139305), CD8β (H35-17.2, BD Biosciences 751609; 53–5.8; BioLegend 140415), F4/80 (BM8, BioLegend 123131), Foxp3 (MF-14, BioLegend 126403), I-A/I-E (MHC II; M5/114.15.2, BioLegend 107622), LIVE/DEAD Fixable Blue Dead Cell Stain (Invitrogen L34962), NK1.1 (PK136; BioLegend 108709), protein L (Pierce, Thermo Scientific, 29997), Thy1 (30-H12; BioLegend 105341), and TRP2 [Flex-T Biotin H-2 K(b) TRP2 Monomer (SVYDFFVWL); BioLegend 280061]. Surface proteins were stained in staining buffer (PBS, 2% FBS, 1 mmol/L EDTA) on ice for 20 minutes and washed twice in staining buffer. Intracellular proteins were stained by incubating surface-stained cells in fixation buffer for 20 minutes at room temperature (BioLegend 420801), followed by two washes in Perm/Wash buffer (BioLegend 421002), incubation with anti-Foxp3 for 30 minutes at room temperature, two washes in Perm/Wash buffer and then resuspended in staining buffer prior to analysis. Samples were analyzed on an Accuri C6 Cytometer (BD Biosciences) or Aurora Spectral Cytometer (Cytek) using SpectroFlo Version 3.03. Flow cytometry analysis was performed using FlowJo software v10.6.1.

Tumor-infiltrating cells were subjected to scRNA-seq using the 10x Genomics platform according to the manufacturer's directions. Briefly, single-cell suspensions were counted, and viability was confirmed using a K2 automated fluorescence cell counter (Nexcellom Bioscience). Cells were loaded onto a Chromium NextGEM Chip G, targeting 10,000 cells for capture. Reverse transcription, cDNA amplification, and library preparation were performed following the user guide (PN for the Chromium NextGEM 3′ Single Cell v3.1 kit; 10× Genomics, 1000121). Gene-expression libraries were uniquely indexed (10× Genomics, 1000079) and pooled for sequencing on an Illumina NextSeq500 instrument with the settings: Read1 = 28 bp, Read2 = 56 bp, Index1 = 8 bp, Index2 = 0 bp, targeting 30,000 reads/cell. Raw bcl files were processed using Cellranger v6.0.0 to produce gene-expression matrices used for downstream analysis. Transcripts were aligned to the mm10 mouse reference genome; barcode processing and transcript quantification were performed on the Cell Ranger Single-Cell Software Suite (10× Genomics). A total of 9,767 cells were analyzed: 3,151 cells from control B16F10 tumor–bearing mice, 5,569 cells from B16F10 tumor–bearing mice treated with TA99 CAR T cells, and 1,047 cells from B16F10 tumor–bearing mice treated with Super2 + IL-33 TA99 CAR T cells, for a total of 15,295 genes. The transcript counts for these genes were then analyzed using Seurat v.4.0.1 in R v.4.0.2 (32). Cells where mitochondrial genes made up more than 5% of total reads were removed, as were cells that had positive read counts fewer than 100 genes or greater than 3,000 genes. Genes present in fewer than 10 cells were removed from the analysis. The cells and genes that remained were then normalized and transformed using SCTransform (33). Following transformation, nearest neighbor unsupervised clustering was performed using Seurat with a cluster resolution of 0.25 (32). Visualization of these clusters was carried out using dimensionality reduction with Uniform Manifold Approximation and Projection (UMAP; ref. 34) on the first 30 principal components of the gene-expression data. Differential gene expression was performed across the Seurat-determined clusters using a Wilcoxon rank-sum test (32). The most upregulated genes in each cluster were then leveraged to manually determine cluster cell types.

scRNA-seq analysis of human TZ47 CAR and Super2 + IL-33 TZ47 CAR T cells was performed as described above. Transcripts were aligned to a modified GRCh38 reference genome to include mCD19 and IRES-GFP for the detection of TZ47 CAR and Super2 + IL-33 expression, respectively. Count matrices for each library were filtered to exclude outliers. Cells with greater than 20% mitochondrial gene percentage, 7,500 genes, or 30,000 transcript reads were excluded, as were cells with fewer than 350 genes, or fewer than 1 transcript read. After filtering, 12,196 cells remained for analysis: 5,896 cells from TZ47 CAR only and 6,300 from Super2 + IL-33 TZ47 CAR. Clustering and visualization were performed as stated above, with a clustering resolution of 0.22 and with 20 principal components. Reference-based cluster annotation was performed using the SingleR package with MonacoImmuneData as a reference for immune cell types (35, 36).

The scRNA-seq data generated in this study are publicly available in the Gene-Expression Omnibus database under accession numbers GSE199401 and GSE199491. All additional data associated with this study are available within the article and its supplementary data files or are available from the corresponding author upon reasonable request.

In in vitro assays, experimental conditions were run as biological duplicates and experimental triplicates. All repeat experiments were combined for statistical analysis unless otherwise stated in the figure legend as a representative. For in vivo mouse efficacy studies, each independent experiment included 3 to 5 mice per experimental group. Data from two independent in vivo mouse experiments were combined for analyses. To quantify the synergistic impact of the Super2 and IL-33 treatments, we fit a linear regression model on the day 17 tumor volume for all 47 tumors in the no treatment, CAR only, Super2 CAR, IL-33 CAR, and Super2 + IL-33 CAR treatment groups. This statistical model took the form “volume ∼ CAR + Super2 + Il-33 + Super2*Il-33” where the dependent variable “volume” represents day 17 tumor volume and the predictor variables “CAR,” “Super2,” and “Il-33” are indicator variables capturing the status of the respective treatments. When fit to the experimental data, the coefficient estimates (and unadjusted Wald test P values) were 1306 (1.55e−12) for the intercept, 111.6 (0.56) for CAR only, −309.8 (0.12) for Super2 CAR, −425 (0.037) for IL-33 CAR, and −511.2 (0.07) for the interaction between Super2 and IL-33. Survival was graphed on Kaplan–Meier plots and the log-rank Mantel–Cox test was used to assess statistical significance. Two-way analysis of variance (ANOVA) was used to analyze differences between CAR T-cell experimental groups with appropriate post hoc analysis (Tukey multiple comparison unless otherwise stated). All data are represented as means, and error bars as SD. All statistical analyses were run for two-sided comparisons under the assumption of a normal distribution when groups have similar variance and nonparametric if variance among groups differs. Statistical analysis was assessed through GraphPad Prism software.

We generated C57BL/6-derived mouse T cells expressing a second-generation CAR, TA99 scFv-CD8TM-CD28-CD3zeta (hereafter referred to as TA99 CAR; Fig. 1A). The TA99 CAR recognizes tyrosinase-related protein 1, TRP1, which is expressed on melanocytes and melanoma cells. To assess the antitumor effects of the TA99 CAR T cells, C57BL/6 mice were inoculated i.d. with 2 × 10⁵ B16F10 melanoma cells, an aggressive tumor in which Tregs suppress infiltration and effector activity of cytotoxic lymphocytes (37). After B16F10 tumors were established for 6 days, mice were i.v. administered 7 × 10⁶ TA99 CAR T cells (Fig. 1B). Tumor growth was monitored every other day. As previously reported (38), TA99 CAR T cells had no effect on tumor growth kinetics in vivo compared with nontreated mice (Fig. 1C and D). In contrast, TA99 CAR T cells induced a dose-dependent decrease in luminescence when cocultured in vitro with B16F10-luciferase cells (Supplementary Fig. S1). This indicates that TA99 CAR T cells are capable of recognizing and killing B16F10 cells in vitro but are ineffective in controlling solid tumors in immunoreplete mice as a single-agent, single-dose therapy, thus highlighting a need for engineering improvements to allow effective responses against solid tumors.

We hypothesized that arming CAR T cells with cytokines that stimulate immune activation would reverse TME immunosuppression. We first evaluated cytokines with reported antitumor activity. Super2 is an engineered variant of human IL-2 that binds to the intermediate IL-2 receptor complex expressed on naïve T cells and NK cells with 200-fold higher affinity than native IL-2 (27). IL-33 is a type 2 cytokine that can activate ILC2 cells to promote tissue-specific cancer immunity (26). TA99 CAR T cells engineered to express either Super2 or IL-33 had little to no effect on tumor growth as compared with no treatment (NT) or TA99 CAR T cell–treated tumors (Fig 1C and D). However, TA99 CAR T cells expressing both Super2 and IL-33 (hereafter referred to as Super2 + IL-33) acted synergistically to significantly delay tumor growth. Super2 + IL-33 TA99 CAR T cells remained effective when administered on day 11 after tumor inoculation against medium-size tumors (∼7.5 mm × 6 mm; Fig. 1E; Supplementary Fig. S2). However, delaying Super2 + IL-33 TA99 CAR T-cell treatment until day 14 after tumor inoculation resulted in variable responses against large tumors (∼10.5 mm × 8 mm), suggesting that the TME of large tumors may be partially resistant to Super2 + IL-33 CAR T-cell therapy and require additional combination therapies.

To determine whether different type 2 cytokines could cooperate with Super2 to reduce tumor growth, TA99 CAR T cells were engineered to express Super2 in combination with IL-25 or TSLP, type 2 cytokines that also activate ILC2 cells, or in combination with IL-4 or IL-5, traditional type 2 cytokines that activate other type 2 immune populations. Retroviral transduction of TA99 and cytokine constructs as determined by GFP coexpression was comparable across samples (Supplementary Fig. S3a). When administered to B16F10 tumor–bearing mice, TA99 CAR T cells expressing Super2 with TSLP, IL-4, or IL-5 had no effect on tumor growth compared with nontreated mice despite considerable in vitro tumor lysis and cytokine expression (Fig. 1F and G; Supplementary Fig. S3b–d). TA99 CAR T cells expressing Super2 in combination with IL-25 did significantly decrease B16F10 tumor growth and increase survival; however, the Super2 and IL-33 combination led to the most effective antitumor response and was superior to all the other combinations (Fig. 1F and G).

T cell–mediated killing of tumor cells is largely dependent on the perforin and granzyme B–dependent pathway with some contribution from the IFNγ and FasL pathways (39). To determine whether the CAR T-cell–intrinsic expression of perforin or IFNγ is required for tumor control, Super2 + IL-33 TA99 CAR T cells were engineered from both perforin- and IFNγ-deficient donor mice. Surprisingly, mice treated with either Prf1^–/– or IFNg^–/– Super2 + IL-33 TA99 CAR T cells were similarly effective at controlling tumor growth as compared with wild-type (WT) Super2 + IL-33 TA99 CAR T cells (Fig. 2A and B). In addition, Super2 + IL-33 CAR T cells expressing a truncated, signaling defective (tailless) TA99 CAR or without a CAR retained partial ability to control tumor growth, albeit at reduced efficiency compared with cells that also expressed the TA99 CAR together with Super2 + IL-33 (Fig. 2C–E). Together, these data demonstrate that CAR expression and antigen signaling contribute only in part to antitumor responses. Additionally, CAR T cell–mediated tumor cytotoxicity through perforin/granzyme- and IFNγ-dependent mechanisms are dispensable for Super2 + IL-33 TA99 CAR T cell–mediated tumor control. This suggests that the CAR T cells may act indirectly, perhaps as a delivery vehicle for Super2 and IL-33 to stimulate endogenous immune responses that are then largely responsible for antitumor activity.

To distinguish the effects of Super2 and IL-33 on CAR T cells from the effects of these cytokines on endogenous immune cells at the tumor site, Super2 + IL-33 TA99 CAR T cells were generated from CD45.1 congenically marked cells and administered 6 days after tumor inoculation of CD45.2 hosts. TILs were isolated from nontreated tumors or tumors treated with TA99 CAR T cells or Super2 + IL-33 TA99 CAR T cells on day 15 and analyzed by flow cytometry. TA99 CAR T cells were 3-fold more abundant in tumors when they expressed Super2 and IL-33 (Fig. 3A). To track CAR T-cell abundancy and localization over time, TA99 CAR T cells and Super2 + IL-33 CAR TA99 T cells were engineered to express luciferase (Fig. 3B–D). Whole-body luminescence imaging showed similar CAR T-cell abundance 1 to 3 days after transfer whether or not the cells expressed Super2 and IL-33 (days 7 and 9). However, a significant increase in Super2 + IL-33 TA99 CAR T cells compared with TA99 CAR T cells was observed 5 to 7 days after transfer (days 11 and 13). Tumor infiltration of Super2 + IL-33 TA99 CAR T cells but not TA99 CAR T cells could be clearly observed by luminescence imaging on days 13 and 15 (Fig. 3D). Together, these data indicate that Super2 and IL-33 expression promotes CAR T-cell expansion and/or accumulation within the tumor.

As for the CAR T cells, endogenous CD45.2-marked CD8⁺ T cells, CD4⁺ T cells, NK cells, and ILC2 cells were also significantly increased in total number in TILs isolated from B16F10 tumor–bearing mice treated with Super2 + IL-33 TA99 CAR T cells compared with those treated with TA99 CAR T-cell or not treated (Fig. 3E; Supplementary Fig. S4a). Endogenous T and NK cells outnumbered Super2 + IL-33 TA99 CAR T cells by 3- to 4-fold. Thus, Super2 + IL-33 TA99 CAR T cells can also promote increased tumor infiltration of endogenous lymphocytes, something that was not observed when T cells expressing only the CAR were administered.

Next, we sought to determine whether endogenous cell types contribute to Super2 + IL-33 TA99 CAR T cell–induced antitumor responses. Individual populations of endogenous immune cells were depleted with antibodies or genetically disrupted to determine their requirement for antitumor immunity in this model (Fig. 4A). NK cells and CD4⁺ T cells were depleted by administering depleting antibodies specific for NK1.1 and CD4, respectively. CD8-deficient and RORα-deficient mice were used as models of CD8⁺ T-cell and ILC2-cell deficiency, respectively. Consistent with the ability of Tregs to suppress antitumor immunity in the B16F10 tumor model, CD4⁺ T-cell depletion led to reduced tumor growth in nontreated mice and had no effect on tumor control in Super2 + IL-33 TA99 CAR T cell–treated mice (Fig 4B and C). In contrast, tumors in nontreated mice tended to grow more quickly in NK cell–depleted and CD8-deficient mice, indicating that both cytotoxic cell types contribute to baseline antitumor responses (Fig. 4D–G). However, Super2 + IL-33 TA99 CAR T cells retained their ability to induce efficient antitumor immunity and decrease tumor growth compared with WT mice (Fig. 4D–G). Antitumor responses in RORα-deficient mice were also not statistically different from responses in WT mice (Supplementary Fig. S4b, c). Together, these data suggest that while endogenous CD8⁺ T cells and NK cells promote and CD4⁺ T cells dampen antitumor immunity, Super2 + IL-33 CAR T cell–induced responses do not rely on any one immune cell type but rather activate compensatory populations to decrease tumor burden.

To identify global changes in TIL populations, we performed scRNA-seq analysis on all tumor-infiltrating cells purified from nontreated B16F10 tumor–bearing mice, as well as B16F10 tumor–bearing mice treated with TA99 CAR T cells or Super2 + IL-33 TA99 CAR T cells. Cells from all samples were aggregated, and quality control and normalization were performed using the Seurat framework and SCTransform method. The normalized scRNA-seq data were clustered using the Seurat implementation of shared nearest neighbor unsupervised clustering and visualized using the UMAP method (40, 41). Cluster cell identities were determined using the top upregulated genes in each cluster as computed by the Wilcoxon rank-sum test on the normalized transcript counts (Supplementary Fig. S5).

Limited numbers of CAR T cells were identified by either flow cytometry or scRNA-seq in TILs isolated following TA99 CAR T-cell treatment with or without Super2 and IL-33 expression compared with endogenous leukocytes (Fig. 3A; Supplementary Fig. S6). As previously reported, B16F10 tumors were infiltrated with endogenous NK cells, T cells, and TAMs (Fig. 5C–F). Comparison of endogenous TIL- populations across samples showed that similar immune cell types were present in all settings; however, proportional shifts in CD8⁺ T cells, TAMs, and MDSCs were observed (Fig. 5A–C). Compared with NT, TA99 CAR T-cell treatment led to modest differences in macrophage and T-cell populations (Fig. 5B–H). However, Super2 + IL-33 TA99 CAR T-cell treatment led to a reduction in M2-like macrophages, which are associated with immunosuppression, and the appearance of an M1-like macrophage population that expressed transcripts associated with antigen presentation including CIITA and H-2Ab (Fig. 5C–F).

scRNA-seq analysis showed that some T-cell subpopulations also shifted in proportion. Although CD8⁺ effector memory T cells, NK cells, and Tregs did not change appreciably in proportion across samples, naïve CD8⁺ T cells decreased and PD-1⁺CXCR6⁺CD8⁺ effector T cells that expresses IFNγ and perforin effectors increased in proportion (Fig. 5G–I). Flow-cytometric characterization of TIL- populations identified a proportional increase in CD8⁺ T cells from Super2 + IL-33 TA99 CAR T-cell–treated mice in comparison with nontreated or TA99 CAR T cell only–treated mice (Fig. 6A–D; Supplementary Fig. S7). Further analysis to distinguish differentially exhausted CD8⁺ effector T cells revealed a general increase in CD8⁺ T cells of all phenotypes, terminally exhausted PD-1⁺Tim-3⁺, and less exhausted PD-1⁺Tim-3^– and PD-1^–Tim-3^– cells, in Super2 + IL-33 TA99 CAR T cell–treated mice compared with TILs isolated from nontreated or TA99 CAR T cell–treated mice (Fig. 6E and F). Although Tregs did not decrease in frequency, the ratio of CD8⁺ T cells to Tregs increased significantly following Super2 + IL-33 TA99 CAR T-cell treatment (Fig. 6G and H). Characterization of TAMs also revealed a proportional decrease in macrophages lacking MHC class II expression, consistent with a decrease in M2-like macrophages following Super2 + IL-33 TA99 CAR T-cell treatment (Fig. 6I–K). In contrast, treatment with TA99 CAR T cells lacking Super2 and IL-33 expression had little to no effect on TIL populations (Fig. 6).

These data indicate that Super2 and IL-33 delivery by TA99 CAR T cells altered TIL populations. This included a proportional shift from M2-like to M1-like macrophages that express high levels of MHC class II. The infiltration of CD8⁺ effector T cells increased substantially, especially in proportion to immunosuppressive Tregs. Together, these changes appear to promote enhanced tumor control, although depletion of any one endogenous population had little effect on Super2 + IL-33 TA99 CAR T cell–induced antitumor immunity (Fig. 4), suggesting that Super2 and IL-33 have pleiotropic effects, each of which contributes to shifting the balance toward tumor elimination.

Our data suggest that Super2 + IL-33 TA99 CAR T cells shift TIL populations from immune suppressive to immune stimulatory in the primary B16F10 melanoma tumor model. As immunosuppression is a general mechanism limiting antitumor immunity against many tumor types, we next sought to determine whether our CAR platform could be universally applied to improve antitumor responses to other solid tumors irrespective of CAR target or tumor type. To test this, we examined three additional tumor models. First, we tested TA99 CAR T cells in a B16F10 metastatic tumor model in which the B16F10 cells are administered i.v. through the tail vein and tumors establish in the lungs. After tumors were established for 6 days, we administered TA99 CAR T cells with or without Super2 and IL-33 and quantified lung metastases 15 days after tumor inoculation. Mice treated with Super2 + IL-33 TA99 CAR T cells had few tumor colonies in the lungs, and in several animals, complete tumor clearance was observed even after a single dose of Super2 + IL-33 TA99 CAR T cells (Fig. 7A and B). Second, we established the MC38 colon cell carcinoma model in which we administered 1×10⁶ MC38 tumor cells subcutaneously into C57BL/6 mice. In this model, we used NKG2D CAR to target Rae1, a stress ligand that is expressed endogenously on MC38 tumors (42). A significant decrease in MC38 tumor growth was observed in mice treated with Super2 + IL-33 NKG2D CAR T cells compared with mice treated with NKG2D CAR T cells or nontreated mice (Fig. 7C and D).

In a third model, mice were inoculated with B16F10 tumors with ectopic expression of the human tumor antigen B7H6. B16F10 tumor cells express heterogenous levels of B7H6 and lose B7H6 tumor antigen in immunoreplete but not immunodeficient mice over time (Supplementary Fig. S8a, b). B16F10-B7H6 tumors were injected i.d. and established for 6 days prior to treatment with 4×10⁶ CAR T cells expressing the TZ47 CAR, a CAR containing an scFv specific to human B7H6 (30). Even with a heterogeneous tumor population and continual tumor antigen loss, Super2 + IL-33 TZ47 CAR T cells were able to significantly control tumor growth as compared with other treatment groups (Fig. 7E and F). On day 19, when nontreated tumors reached 15 mm in diameter, we characterized TIL populations by flow cytometry to examine endogenous antitumor responses by using an MHC class I tetramer loaded with a peptide derived from TRP2, a shared melanocyte and tumor antigen. We observed a significant increase in TRP2 tetramer⁺ T cells in tumors from mice treated with Super2 + IL-33 TZ47 CAR T cells compared with tumors from mice treated TZ47 CAR T cells or NT (Fig. 7G).

Together, these data demonstrate that a single dose of CAR T cells that express Super2 and IL-33 delayed tumor growth and even led to complete remission in various models irrespective of the CAR target and tumor type. Additionally, CAR T cells expressing Super2 and IL-33 effectively controlled heterogeneous tumors with the propensity to downregulate tumor antigen, consistent with the ability of Super2 and IL-33 to induce endogenous immune responses and shift immunosuppressive cells within the TME to a tumor-specific, immunostimulatory phenotype.

To move toward clinical therapy, we next examined the effect of Super2 and IL-33 on the ability of human TZ47 CAR T cells to mediate cytotoxicity against K562 tumor targets that naturally express B7H6 antigen. As for mouse CAR T cells, Super2 and IL-33 expression did not alter the in vitro cytolytic activity of human TZ47 CAR T cells against tumor targets across multiple effector-to-target ratios (Supplementary Fig. S9a). In addition, we conducted scRNA-seq analysis on human TZ47 CAR T cells with or without Super2 and IL-33 expression 8 days after in vitro manufacturing. Six T-cell subsets were identified by reference-based SingleR cluster analysis as four CD8⁺ effector memory T-cell subsets and two CD4⁺ Th1 cell subsets (Supplementary Fig. S9b–c). However, there were no observable differences in cluster proportions or cytotoxic activity between human CAR T cells with or without Super2 and IL-33 expression when characterized in vitro (Supplementary Fig. S9a–e). It remains to be determined whether Super2 and IL-33 expression grants human CAR T cells the ability to promote endogenous antitumor immunity in vivo in the setting of an immunosuppressive TME.

Altogether, our findings show that CAR T cell–induced engagement of both type 1 and type 2 immunity is an effective treatment strategy for combating solid cancers. Expression of Super2 and IL-33 with different CAR constructs enabled effective control of multiple tumor types, highlighting the potential of this approach as a universal CAR T-cell platform.

The therapeutic efficacy of CAR T cells is predominantly determined in preclinical studies using immunodeficient mouse models such as NSG mice to allow engraftment of both human CAR T cells and their tumor targets. Yet these models poorly recapitulate the immunosuppressive TME that determines whether CAR T cells ultimately control solid tumor growth. When immunocompetent models are used, host preconditioning with irradiation, immunodepleting regimens or immunosuppressive chemotherapies to generate space for CAR T cells and to reduce immunosuppression is often necessary to drive CAR T-cell expansion; however, therapeutic responsiveness remains limited (13, 38, 43–46). Thus, substantial opportunity remains to develop approaches to overcome the daunting hurdle presented by the TME for CAR T-cell therapy for solid cancers. We show that Super2 and IL-33 expressing CAR T cells mount a potent antitumor response in the absence of preconditioning and as a single-dose, single-agent therapy. In fact, general depletion of host immune cells could arguably impede the overall efficacy of Super2 + IL-33 CAR T cells given the ability of these cells to activate a robust antitumor response from endogenous immune cells. Lymphodepleting regimens aimed at generating immunologic space to allow the accumulation of excess homeostatic cytokines such as IL-2 may also not be required with Super2- and IL-33-engineered CAR T cells. Future studies that directly evaluate the effects of lymphodepletion on Super2 and IL-33 CAR T-cell therapy will help to further inform clinical translation.

Loss of tumor antigens due to tumor heterogeneity and immune editing is an additional hurdle limiting the efficacy of CAR T cells and other antigen-directed therapies, including T cells with ectopically expressed TCR- or Cas9-targeted insertion of tumor-specific TCR or CAR sequences in the native TCR locus. To examine the effect of antigen heterogeneity in the context of CAR T-cell therapy, we engineered T cells to express a B7H6-specific CAR, TZ47, and then treated mice that had been inoculated with B16F10 expressing heterogeneous levels of the human tumor antigen B7H6. Preferential loss of B7H6-expressing tumor cells following TZ47 CAR T-cell treatment effectively mimicked antigenic loss observed with human solid tumors. In addition to enhancing CAR T-cell numbers, Super2 and IL-33 expression increased tumor infiltration of endogenous melanoma antigen–specific T cells. The generation of a broad antitumor response is essential for the field of CAR T-cell therapy to effectively manage solid tumors characterized by antigen heterogeneity. To date, there are a limited number of tumor-specific antigens identified for solid tumors, and those that are tumor selective are rarely expressed uniformly throughout the entire tumor. Thus, the expression of Super2 and IL-33 in CAR T cells has the potential to address antigen loss by activating innate cytotoxic NK and myeloid cells and priming endogenous tumor-reactive T cells.

IL-33 is an IL-1 family cytokine with alarmin properties that impact both innate and adaptive immune cells including, but not limited to, macrophages, DCs, NK cells, T cells, and ILCs (47). Although IL-33 is a type 2 cytokine known for its ability to aid in type 2 immunity, its full activity remains unclear. Based on its activity in pancreatic cancer, we anticipated a role for IL-33 in activating, expanding, and mobilizing ILC2 cells, which then recruit CD103⁺ DCs to license and expand CD8⁺ T cells (26). However, the unperturbed antitumor responses we observed with CD8- and RORα-deficient mice, which lack CD8⁺ T cells and ILC2 cells, respectively, led us to consider a broader role for IL-33. It has been reported that IL-33 can skew Th1 and Th2 T-cell responses depending on the environmental conditions (48). Additionally, IL-33 has been shown to activate both macrophages and DC (49, 50). Moreover, NK and NKT cells have been shown to respond to IL-33 by increasing IFNγ production and release upon receptor engagement (51, 52). The modest reduction in tumor protection observed following administration of IFNγ-deficient CAR T cells supports further investigation of additional mechanisms downstream of IL-33. scRNA-seq and flow-cytometric analyses support a pleiotropic effect of IL-33 and Super2 in inducing proportional shifts in effector CD8⁺ T cells in relation to Tregs and skewing of macrophage phenotypes toward M1-like properties. Elimination of individual cell populations, such as NK cells, CD8⁺ T cells, or ILC2 cells, or disarming CAR T-cell effector molecules, perforin or IFNγ, had little disruptive effect on Super2 and IL-33 CAR T-cell efficacy. Future studies that selectively deplete macrophage subsets or multiple cell populations in combination will be required to fully understand the IL-33-dependent mechanisms for promoting antitumor responses. Given the multistep multicellular pathway that exists downstream of IL-33 in pancreatic ductal adenocarcinomas that involves the sequential activation of ILC2, DCs, and CD8⁺ T cells (26), identifying a single required mechanism for Super2 + IL-33 CAR T cells may prove technically challenging. The pleiotropic effects of Super2 and IL-33 on various immune subsets may also prove challenging to identify a necessary mechanism, although the engagement of compensatory mechanisms could be advantageous in providing a multiple offensive play strategy against the challenges presented by solid tumors.

The synergy that results from combining two diverse cytokines, Super2 and IL-33, is unexpectedly robust. Other type 2 cytokines were unable to synergize with Super2 except IL-25, which demonstrated partial efficacy. The partial overlap between IL-25 and IL-33 could perhaps be attributed to their shared ability to activate TRAF6 signaling, albeit downstream of distinct receptor-associated adapter proteins (53, 54). In contrast, Super2, TSLP, IL-4, and IL-5 activate JAK/STAT signaling (55). Whether or not Super2 and IL-33 synergy occurs by activating JAK and TRAF6 pathways in cis, within the same cellular target, or in trans, within distinct cells, remains to be determined. The ability of other type 1 cytokines such as IL-15 and IL-21 to synergize with IL-33 also requires future exploration.

In the clinic, systemically delivered high-dose IL-2 can effectively activate the immune system and drive tumor clearance (28). Although it has affected cures in a small subset of metastatic melanoma patients, toxicity remains a general concern, and coupled with a lack of predictive markers to identify responsive patients, it is not typically used as a single-agent therapy today (56). However, it is common practice for CAR T cells to be administered with systemic IL-2 to boost CAR T-cell expansion, which often correlates with improved patient responsiveness and outcomes, but the toxicity concerns associated with high-dose systemic IL-2 remain. By engineering CAR T cells to express IL-2, systemic toxicity may be ameliorated by the direct delivery of IL-2 to the TME by CAR T cells, which utilize the CXCR3–CXCL9/10 axis to infiltrate solid tumors. Although toxicity was not observed in our mouse tumor models, the more limited toxicity observed in mice compared with humans in response to systemic IL-2 suggests that careful pharmacodynamic studies need to be conducted unless additional safety measures, such as engineering kill switches or inducible expression, are used to derisk the use of IL-2 and its variants.

The high efficacy that Super2 + IL-33 CAR T cells exhibited in four tumor models suggests that Super2 and IL-33-based cell therapies can universally promote clearance of solid tumors independent of tumor type and tumor antigen. Additionally, the ability of the Super2 and IL-33 combination to induce antitumor responses in the absence of CAR expression or CAR signaling offers hope that it could be applied to allogeneic therapies, which drastically reduces manufacturing time and costs and can be used off-the-shelf.

Given the increase in tumor-infiltrating CD8⁺ T cells expressing PD-1, we predict that combination therapy with checkpoint blockade, which has dramatically affected the cancer field due to its universality and efficacy, will further increase Super2 + IL-33 CAR T cell–induced antitumor immunity toward cures. This prediction is supported by scRNA-seq and flow-cytometric analyses, which identified increased endogenous CD8⁺ effector T cells that are PD-1⁺ but Tim-3^–, which include progenitor-exhausted cells predicted to remain responsive to PD-1 blockade (57). Combining Super2 + IL-33 CAR T cells with checkpoint blockade immunotherapy has the potential to revitalize endogenous CD8⁺ T cells to drive complete tumor remission. A requirement for additional immunotherapies is particularly important in the context of large established tumors, for which we observed variable efficacy following Super2 + IL-33 CAR T-cell treatment.

The next step in advancing Super2 + IL-33 CAR T cells toward clinical application will require additional human CAR T-cell studies. Although we found no detectable effect of Super2 and IL-33 expression on either in vitro human CAR T-cell cytotoxic activity or T-cell subsets in the manufactured CAR T-cell product, our mouse studies implicate the activation of endogenous antitumor immunity as a key mechanism linked to Super2 + IL-33 CAR T-cell efficacy. Because patient-derived xenograft mouse NSG models lack an endogenous human immune system, further studies will require the use of immunoreplete humanized mouse models to appropriately determine the ability of Super2 + IL-33 CAR T cells to control human solid tumors. A variety of human tumor cell lines grow in NSG mice reconstituted with CD34⁺ hematopoietic stem cell grafts (58). Tumors including A375, a human melanoma cell line expressing B7H6 tumor antigen (59, 60), become infiltrated by a variety of innate and adaptive human immune cells including NK cells, macrophages, CD4⁺ and CD8⁺ T cells and Tregs (58). Future studies to determine whether human TZ47 CAR T cells with or without Super2 and IL-33 expression can limit A375 tumor growth through CAR T cell–mediated or endogenous antitumor immunity may help support the translational use of this novel type 1 and 2 cytokine combination in adoptive T cell therapies for patients with solid cancers.

R.A. Brog reports grants from NIH grant T32-AI007363 to RAB during the conduct of the study; in addition, R.A. Brog has a patent for immunostimulatory cytokine combination and therapeutic use thereof pending. L. Abdullah reports other support from Burroughs Wellcome Fund outside the submitted work. J. Zou reports a patent for immunostimulatory cytokine combination and therapeutic use thereof pending to PCT/US22/14017. P. Kumar reports a patent for immunostimulatory cytokine combination and therapeutic use thereof pending to PCT/US22/14017. C.L. Sentman reports personal fees from Bellicum Pharmaceuticals, CytomX Therapeutics, Javelin Oncology, and Celdara Medical outside the submitted work; in addition, C.L. Sentman has a patent for NKG2D CAR pending, issued, licensed, and with royalties paid from Celyad, a patent for B7H6 CAR pending, issued, licensed, and with royalties paid from Celyad, a patent for high-affinity B7H6-specific Abs and scFv pending and issued, a patent for novel signaling platforms for CAR T cells pending, a patent for coexpression of CARs and transcription factors pending, and a patent for TCR-deficient T cells pending, issued, licensed, and with royalties paid from Celyad. Y.H. Huang reports grants from NIH UT2-GM130176, NIH P30-CA023108, and Munck-Pfefferkorn Novel and Interactive Grant during the conduct of the study; in addition, Y.H. Huang has a patent for PCT/US22/14017, immunostimulatory cytokine combination and therapeutic use thereof pending. No disclosures were reported by the other authors.

R.A. Brog: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S.L. Ferry: Investigation, writing–review and editing. C.T. Schiebout: Formal analysis, visualization, writing–review and editing. C.M. Messier: Formal analysis, investigation. W.J. Cook: Investigation. L. Abdullah: Investigation, writing–review and editing. J. Zou: Conceptualization, methodology. P. Kumar: Investigation, writing–review and editing. C.L. Sentman: Supervision, methodology, project administration, writing–review and editing. H.R. Frost: Formal analysis, supervision, visualization, writing–review and editing. Y.H. Huang: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, visualization, writing–original draft, project administration, writing–review and editing.

This work was supported by NIH grant UT2-GM130176, a Norris Cotton Cancer Center Prouty Developmental Grant, a Munck-Pfefferkorn Novel and Interactive Grant, and a gift from Tom and Susan Stepp to Y.H. Huang, NIH grants R21-CA253408 and P20-GM130454 to H.R. Frost, NIH grant T32-AI007363 to R.A. Brog, Big Data in the Life Sciences Training Program to L. Abdullah, and NIH grant P30-CA023108 to Steve Leach, which supports the Norris Cotton Cancer Center's genomics, microscopy and flow cytometry cores. We thank Darrell Irvine and Leyuan Ma at MIT for sharing the TA99 CAR sequence, Gregory Ho for his early contributions to cloning the IL-33 construct, Fred Kolling IV for scRNA sequencing and analysis.

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