Clinical and molecular dissection of CAR T cell resistance in pancreatic cancer

Abstract

Patients with advanced pancreatic ductal adenocarcinoma (PDAC) have a median survival of less than a year, highlighting the urgent need for treatment advancements. We report on a phase 1 clinical trial assessing the safety and feasibility of intravenous and local administration of anti-mesothelin CAR T cells in patients with advanced PDAC. While therapy is well tolerated, it demonstrates limited clinical efficacy. Analyses of patient samples provide insights into mechanisms of treatment resistance. Single-cell genomic approaches reveal that post-infusion CAR T cells express exhaustion signatures, including previously identified transcription factors ID3 and SOX4, and display enrichment for a GZMK⁺ phenotype. Single knockout of ID3 or SOX4 enhances efficacy in xenograft models, though with donor-dependent variability. However, single-knockout cells eventually fail. Conversely, ID3 and SOX4 double-knockout CAR T cells exhibit prolonged relapse-free survival, demonstrating a sustained therapeutic effect and a potential avenue for engineering more potent CAR T cells in PDAC. This study was registered at ClinicalTrials.gov (NCT03323944).

Keywords: CAR T cells; ID3; SOX4; T cell dysfunction; T cell exhaustion; cancer; clinical trial; immunotherapy; pancreatic cancer; single-cell RNA-seq.

Graphical abstract

Figure 1

huCART-meso cells are tolerated and safe in a patient population with minimal cell expansion and cytokine response (A) Protocol schema for screening, apheresis, T cell manufacture, treatment with huCART-meso cells, and follow-up. i.v., intravenous; i.p., intraperitoneal, i.h.; intrahepatic. (B and C) (B) Overall survival and (C) progression-free survival of patients enrolled on the trial. (D) huCART-meso cell engraftment in peripheral blood by qPCR-based detection of CAR-specific sequences in genomic DNA. (E) huCART-meso cell expansion in pre- and post-infusion biopsy tissue samples by qPCR. Wilcoxon matched-pairs signed rank test, n = 8 patients. (F) Heatmap grouped by cohort depicting fold changes in proinflammatory cytokine production from baseline/pre-CAR T cell infusion to the peak cytokine level post infusion. IL-2 levels for patient 07 were below the range of detection of the assay and are indicated as “X” in the figure. See also Figure S1; Table S1.

Figure 2

Locally delivered huCART-meso cells are enriched for GzmK⁺ Tem phenotypes and exhibit a dysfunctional T cell phenotype including upregulation of SOX4 and ID3 (A) Summary of the samples collected for flow cytometry and scRNA-seq analyses. (B) Representative flow cytometry plot of patient 04 showing detection of peritoneal fluid infiltrating CD3⁺ cells with surface expression of the huCART-meso. (C and D) CAR detection performed by qPCR on peripheral blood (left) and peritoneal fluid (right) of patient 04 (C) and patient 05 (D). (E) UMAP projection of scRNA-seq data from sorted CD3⁺CD45⁺ cells infiltrating the peritoneal fluid 7 days post intraperitoneal infusion (patient 04) or 26 days post infusion (patient 05). UMAP plot is labeled with T cell subsets. (F) huCART-meso expression associated with (E). (G) Distribution of huCAR-meso+ (CARpos) and endogenous CAR-negative T cells (CARneg) across cell type clusters defined in (E). Unknown cluster from (E) was excluded. The y axis represents the percent of cells found within each cell type cluster, calculated by taking the number of cells per condition (condition being CARpos or CARneg) within each cell type cluster and dividing by the total number of cells in that condition, calculated for each patient individually. Dots represent the individual patient values, and the bar plot represents the average of the patient values; moderated t test via the propeller method (n = 2 patients). p values denoted as ∗p < 0.05 and ∗∗p < 0.01. (H) Expression of the 30-gene CAR T dysfunction signature in CD3⁺ CARpos and CARneg T cells of patients 04 and 05 post infusion. (I) Gene expression levels of huCART-meso, SOX4, ID3, SRGAP3, NDFIP2, CTLA4, TNFRSF9, PDCD1, HAVCR2, and TIGIT in CARpos (left, red) and CARneg cells (right, green). CARneg cells were randomly downsampled to have an equal number of CARpos and CARneg cells for analyses in (G)–(I). See also Figure S2.

Figure 3

Single KO of ID3KO or SOX4KO in huCART-meso cells induces variable antitumor primary responses (A) Experimental design to evaluate the efficacy of ID3KO and SOX4KO huCART-meso cells in AsPC-1 tumor-bearing mice. Suboptimal doses of huCART-meso cells were injected i.v. when tumors reached ∼250–300 mm³ (day 0, n = 7–12 mice per group). (B) Tumor growth volumes of AsPC-1 tumor-bearing mice treated with 1 × 10⁶ WT, ID3KO, or SOX4KO ND561 huCART-meso cells or with UTD (untransduced T cells) at equivalent amounts of total cells; two-Way ANOVA with Geisser-Greenhouse correction and Dunnett’s post hoc test; mean ± SEM (n = 8–11 mice per group). Dashed line indicates the day of infusion. This result is representative of three independent experiments performed with CAR T cells derived from three different donors. (C) Percentage of weight loss of mice infused with UTD, WT, ID3KO, or SOX4KO ND561 huCART-meso cells; mean ± SEM. (D) Tumor growth profiles of a representative good (ND539), intermediate (ND582), and non-responder (ND602) compared to mice treated with WT (black) or SOX4KO huCART-meso cells (green). 1 × 10⁶ and 5 × 10⁵ CAR T cells/mouse administered; two-way ANOVA with Geisser-Greenhouse correction; mean ± SEM (n = 5–11 mice per group). (E) Tumor volumes of representative WT and SOX4KO huCART-meso-treated mice at day 41–42 after infusion. Two donors were used per group. Good responders: n = 7–11 animals per donor, totaling 16–21 animals per group and CAR T dose; intermediate responders: n = 5–9 animals per donor, totaling 13–14 animals per group and CAR T dose; non-responders: n = 3–5 animals per donor, totaling 5–8 animals per group and CAR T dose; t test or Mann-Whitney test (for non-normally distributed data); mean ± SEM. (F) Post-expansion phenotype huCART-meso cells derived from healthy donors. ∗∗∗∗p < 0.0001, ∗∗∗p < 0.001,∗p < 0.05, ns: not significant. See also Figure S3; Table S2.

Figure 4

Mice injected with single KO of SOX4 or ID3 eventually experience tumor recurrence (A) Experimental design for recurrent tumor model. (B) KO efficiency in sorted T cells from relapsing tumors. Each dot represents a technical replicate (independent sequencing for KO assessment from one mouse per group); mean ± SEM. (C) CD3⁺CD45⁺ T cell infiltration in ND561 WT and SOX4KO relapsing tumors (left) and percent of CD3⁺CD45⁺ surface huCART-meso in relapsing tumors and infused products (right). Each dot represents a technical replicate (n = 1 mouse per group); Mann-Whitney U test (left) and two-Way ANOVA and Sidak post hoc test (right); mean ± SEM. (D) Total number of MSLN⁺Epcam⁺ cells (#) per gram of tumor (left) and representative flow cytometry plots (right) of MSLN expression in ND561 WT and SOX4KO-treated relapsing tumors (n = 1 mouse per group). AsPC-1 cells were used as positive control and MSLN minus as negative control for MSLN staining; Mann-Whitney U test. Error bars represent variation between technical replicates; mean ± SEM. (E) UMAP projection of scRNA-seq data from recurrent TILs harvested from mice injected with WT CAR T cells (ND561, n = 1 donor) and SOX4-KO CAR T cells (ND561 and ND539, n = 2 donors). (F) UMAP projection with broad T cell subsets labeled. NK-like T cells clustered on the left (cluster 2) and all other clusters were defined as GZMK⁺ exhausted T cells (Tex), right. (G) huCART-meso expression in recurrent tumor TILs. (H) Cluster distribution depicting the frequency that WT and SOX4KO-injected recurrent tumor TILs were found within the broad clusters defined in (F). y axis represents the percent of cells found within each cell type cluster for each condition (WT or SOX4KO). The percent was calculated by taking the number of cells per condition within each cell type cluster and dividing by the total number of cells in that condition. SOX4KO condition was averaged for the two mice and visualized via bar plot, and circles depict the individual values for each mouse; moderated t test using the propeller method. (I) Gene expression levels of TNFRSF9, TIGIT, LAG3, and HAVCR2 in recurrent tumor CARpos TILs extracted from mice injected with WT and SOX4KO CAR T cells. ∗∗∗∗p < 0.0001, ∗∗∗p < 0.001, ns: not significant. See also Figure S4.

Figure 5

Double KO of SOX4 and ID3 in huCART-meso cells provides sustained protection against tumor recurrence (A) Population doublings, viability, and average cell volume of ND584 WT, ID3KO, SOX4KO, and DKO huCART-meso (solid lines) and un-transduced cells (UTD) (dashed lines). (B) Post-expansion phenotype characterization using flow cytometry of huCART-meso DKO cells derived from healthy donors; mean ± SEM. Each dot represents one donor, n = 7. (C) Tumor growth volumes of AsPC-1 tumor-bearing mice treated with suboptimal doses of UTD (n = 8 mice), WT, ID3KO, SOX4KO, or DKO ND584 huCART-meso cells (5 × 10⁵ CAR T cells/mouse, n = 12 mice per group); two-Way ANOVA with Geisser-Greenhouse correction and Dunnett’s post hoc test; mean ± SEM. Dashed line indicates the day of huCART infusion (day 0). (D) Percentage of weight loss of mice shown in (C), mean ± SEM. (E) Experimental design for relapse analyses in AsPC-1 tumor-bearing xenografts treated with curative doses of WT, ID3KO, SOX4KO, and DKO huCART-meso cells (1 × 10⁶ CAR T cells/mouse). (F) Individual tumor volumes of mice cured with ND528 UTD, WT, ID3KO, SOX4KO, and DKO huCART-meso cells until 4 months after infusion (day 123). (G) Individual tumor volumes of mice treated with curative doses of ND528 and ND584 WT, ID3KO, SOX4KO, and DKO huCART-meso cells at the end of the study or last seen, whatever comes first; Kruskal-Wallis plus Dunn’s test (mean ± SEM). (H) Relapse-free survival (RFS) plot for mice treated with WT, ID3KO, SOX4KO, or DKO huCART-meso cells. Plot represents RFS of all groups combined (ND528 + ND584, n = 170 mice). Multiple independent experiments were performed. (I) RFS analysis by treatment group. Based on Kaplan-Meier method, median RFS for the WT group (n = 42) was 111 days (95% confidence interval [CI]: 86, NE), and ID3KO group (n = 42) had a median RFS of 115 (95% CI: 100, NE) days. The median survival time for the SOX4KO (n = 43) and DKO (n = 43) group was not reached (the estimated survival curve did not fall below 0.5 by the data cutoff time). Subjects at risk refer to the number of mice in the study being monitored at each time point and that have not yet experienced relapse, recurrence, or death. Latest endpoint day for some of the studies in the combined analysis of (H) and (I) was day 141 post infusion. NE, not estimable. ∗∗p < 0.01,∗p < 0.05, ns: not significant. See also Figure S5.