Novel CAR T cell blend targeting PDPN and GD2 to overcome glioblastoma heterogeneity

Abstract

Background

While chimeric antigen receptor (CAR) T cells have achieved encouraging remission rates in hematological malignancies, they have demonstrated limited success in treating glioblastoma (GBM), particularly due to high intratumoral and intertumorous heterogeneity. In this study, we identified a consistently expressed target antigen, podoplanin (PDPN), and evaluated the potential of a PDPN-CAR T cell and GD2-CAR T cell blend to target heterogeneous GBM.

Methods

Target antigen screening included clinical samples, cell lines and healthy tissues, as well as public RNA sequencing datasets. The anti-tumor function of CAR T cells was examined in co-culture experiments with GBM cell lines and patient-derived organoids (PDOs), and in vivo after locoregional delivery in orthotopic xenograft models.

Results

CAR T cells demonstrated strong anti-tumor activity against several cell lines and PDOs from multiple patients. PDPN and GD2 were expressed in all PDOs, and regardless of the antigen expression pattern, the CAR T cell blend induced significantly higher apoptosis levels in organoids compared with single-antigen targeting CAR T cells. In vivo, we observed efficient tumor regression after locoregional administration of monospecific CAR T cells. While heterogeneous orthotopic tumors eventually relapsed in these groups, therapy with the CAR T cell blend significantly increased overall survival and even achieved a cure in the majority of mice.

Conclusion

This novel PDPN/GD2 CAR T cell blend demonstrated robust efficacy in advanced preclinical GBM models, suggesting its potential to treat heterogeneous GBM and address limitations associated with single-antigen CAR T cell therapies.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Glioblastoma (GBM) is the most prevalent and lethal primary malignant brain tumor and remains incurable despite aggressive multimodal standard therapy. Although novel treatment strategies, such as chimeric antigen receptor (CAR) T cell therapy, have shown encouraging results, the highly heterogeneous tumor antigen landscape in GBM presents a major challenge for monovalent treatment strategies.

WHAT THIS STUDY ADDS

• In this study, podoplanin (PDPN) and GD2 were identified as the most consistently expressed target antigens through a comprehensive screening campaign, with expression levels notably higher in recurrent compared with primary GBM. The combination of PDPN-directed and GD2-directed CAR T cells into a blend markedly enhanced efficacy against patient-derived organoids and induced long-term complete remission in an orthotopic xenograft model, outperforming single-antigen targeting approaches.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The identification of PDPN and GD2 as a viable target antigen combination, together with the depth of response achieved using the applied CAR T cell blend, supports the use of this axis for treating heterogeneous GBM and may inform future research and clinical trial design.

Introduction

Glioblastoma (GBM) is the most common and most aggressive primary malignant brain tumor in adults and is considered incurable despite extensive treatment.^{1 2} The median survival rate after standard therapy, comprising surgical resection followed by radiotherapy with concomitant and adjuvant chemotherapy with temozolomide and tumor treating fields,³ remains limited to 21 months, with a 5-year survival rate of only 13%.4,6

Chimeric antigen receptor (CAR) T cells targeting CD19 have shown remarkable clinical effectiveness against various hematologic neoplasms.⁷ However, effective CAR T cell therapy has not yet been demonstrated in cases of solid tumors, particularly GBM.^{8 9} The immunosuppressive tumor microenvironment (TME), CAR T cell trafficking to tumor sites and especially the high degree of intertumorous and intratumoral heterogeneity present major challenges for CAR T cell therapy.^{10 11} Accordingly, phase I clinical trials demonstrated partial responses in patients with recurrent GBM (rGBM) following therapy with, for example, EGFRvIII-CAR T cells, IL13Rα2-CAR T cells, and HER2-CAR T cells, but these treatments failed to induce durable remission.12,14

The objectives of this study were to identify and validate antigens that are consistent among patients and to establish a dual CAR T cell approach capable of targeting heterogeneous GBM. For this purpose, we characterized antigen expression levels in GBM and identified podoplanin (PDPN), which is a mucin-like type-I integral membrane glycoprotein involved in tumorigenesis and metastasis, as a lead candidate. Increased PDPN expression levels have been detected in numerous cancers, particularly those that originate from immune-privileged organs such as the testis and brain. This overexpression is associated with tumor malignancy, treatment resistance and a dismal prognosis.15,17 The second most promising candidate emerging from our screening process was GD2, which is a tumor-specific disialoganglioside that is highly expressed by tumors of neuroectodermal origin and has been validated as a potent and tolerable target antigen for CAR T cell therapy.18,20

We evaluated the effectiveness of PDPN-CAR T cells and GD2-CAR T cells in experiments with multiple GBM cell lines (GCLs), patient-derived organoids (PDOs) and xenograft models. Generated directly from tumor tissue obtained intraoperatively, PDOs preserve the complexity and heterogeneity in the paternal tumor and constitute models that more closely replicate the conditions within patients than monolayer cultures.²¹ In contrast, orthotopic xenograft models, including those involving the locoregional delivery of CAR T cells into the lateral ventricle, represent a physiologically suitable environment for the assessment of CAR T cell functionality. In these preclinical GBM models, dual targeting with a PDPN-CAR T cell/GD2-CAR T cell blend demonstrated superior antitumor function, achieving durable tumor regression and curing in most mice.

Results

PDPN is a consistently expressed target antigen in GBM

We screened the expression levels of potential target surface antigens in GBM on the basis of freshly resected tissue obtained intraoperatively from patients with a confirmed GBM diagnosis (n=15). Flow cytometric analyses revealed that only PDPN was consistently expressed in all the samples, followed by GD2 with greater variance (figure 1A).

Figure 1. Antigen screening experiments. (A) ΔMFIs of primary tissue samples (n=15). Dark bars represent the median, and light bars represent the quartiles. For patient characteristics, refer to online supplemental Table 2. (B) Antigen expression in primary (light) and recurrent (dark) GBM specimens (n=33). Medians and quartiles are shown. Whiskers represent the 10th–90th percentiles, and the points represent statistical outliers. Significant differences in antigen expression (GD2 and PDPN) are marked with an asterisk (*p<0.05; **p<0.01). For patient characteristics, refer to online supplemental Table 2. (C) Kaplan-Meier survival curve of the TCGA-GBM patient cohort stratified on the basis of PDPN expression (p=log rank test p value). (D) Results of the analysis of the association between overall survival and the expression of selected genes (optimal cut-off point: expression cut-off achieving the greatest difference between survival curves; p-value: log-rank test p-value; HR: hazard ratio; HR p-value: p-value corresponding to the HR; # high/low: number of patients stratified into high-expression and low-expression groups based on the calculated optimal cutpoint). EGFRvIII was excluded from this analysis because of the lack of an underlying variant-specific gene signature. (E) ΔMFIs of GBM antigens are displayed logarithmically as violin plots for (n=6) GCLs (DKMG, U87, U138, U251, U343, and U373). Dark bars represent the median, and light bars represent the quartiles. GBM, glioblastoma; MFI, mean fluorescence intensity; PDPN, podoplanin; TCGA, The Cancer Genome Atlas.

To determine whether target antigens are maintained during disease progression, we performed immunohistochemistry (IHC) analysis of tumor samples from patients with primary GBM and matched rGBM samples from 33 patients who experienced disease relapse. Unlike the levels of other antigens that remained unchanged, the levels of GD2 and PDPN were significantly elevated in recurrent tumors, highlighting their potential as target molecules for treating rGBM (figure 1B).

To assess the impact of GBM antigen expression on overall survival (OS), we analyzed data from The Cancer Genome Atlas (TCGA)-GBM dataset and identified a clear association between elevated PDPN expression and reduced OS levels (figure 1C). Notably, compared with other antigens, PDPN exhibited the strongest inverse correlation with patient survival (figure 1D).

To serve as a predictive model system, GCLs should ideally recapitulate the antigen expression patterns observed in primary tissue. Accordingly, we performed flow cytometric analyses of all the markers across a panel of GCLs (DKMG, U87, U138, U251, U343, and U373) and detected varying levels of surface expression for each antigen, with PDPN exhibiting the highest overall median expression (figure 1E). Interestingly, several antigens expressed on GCLs (eg, CD70 and IL13Rα2) were not consistently detected in primary tissue, in contrast to GD2 and PDPN, which showed consistent expression in both (figure 1A and E).

To evaluate potential on-target off-tumor expression, we performed quantitative PCR (qPCR) analysis of pooled complementary DNA (cDNA) from healthy organs. Among the tissues tested, ovarian, placental, and lung tissues presented the highest PDPN expression; however, compared with those in GBM, the overall PDPN expression levels remained low (online supplemental figure S1A). Although PDPN expression was negligible in pooled pan-brain cDNA, we conducted additional qPCR analysis on pooled cDNA from 24 distinct healthy brain regions to further examine its low expression in healthy tissue (online supplemental figure S1B). While some brain regions showed minimal PDPN expression at the transcript level, IHC analysis using both D2-40 and NZ-1 PDPN antibodies revealed no detectable protein expression in healthy brain tissue. Furthermore, tumor sections from various locations displayed variable expression patterns of all the tested antigens, ranging from low to high levels, which is consistent with the pronounced intratumoral heterogeneity characteristic of GBM (online supplemental figure S1C).

Notably, when the D2-40 PDPN antibody was used, lung tissue, including type 1 alveolar cells and the membrane lining of the epithelium, was identified as strongly positive. In ovarian tissue, D2-40 staining revealed PDPN expression on lymph vessels and stromal tissue. In contrast, both parenchyma were negative for PDPN when the NZ-1 antibody was used (online supplemental figure S1D).

These findings demonstrate that PDPN is a consistently expressed target antigen in GBM and that its expression is elevated in recurrent tumors and strongly inversely correlated with patient survival; thus, PDPN is an attractive target for CAR T cell therapy in GBM.

PDPN-specific CAR T cells efficiently target GCLs in vitro and in vivo

On the basis of our findings, we equipped T cells with a second-generation CAR comprising a single-chain fragment variable (scFv) derived from the PDPN-specific monoclonal antibody (mAb) NZ-1 (figure 2A). Both CD4⁺ and CD8⁺ CAR T cells underwent efficient purification, gene transfer, and overall expansion (figure 2B–D), resulting in a well-balanced T cell product predominantly composed of central memory-like (T_CM), effector memory-like (T_EM), and stem cell memory-like (T_SCM) phenotypes (figure 2E–G). Co-culture of PDPN-CAR T cells with GCLs led to specific target cell lysis in a 24-hour bioluminescence-based cytotoxicity assay (figure 2H–J). Similarly, the surface expression of PDPN promoted proliferation and significantly enhanced the secretion of interferon-gamma (IFNγ) in response to the PDPN⁺ cell lines (figure 2K,L).

Figure 2. PDPN-CAR T cells are functional against tumor cell lines in vitro. (A) Structural features of the PDPN-specific CAR construct. Created with BioRender.com. (B, C) Representative purities of CD4⁺ and CD8⁺ CAR T cells. Freq. of parent (cells/single cells/live cells) is displayed. (D) Enrichment of CAR⁺ T cells in a 10-day expansion protocol. Data are shown as mean+SEM for n=3 independent donors. (E, F) Representative memory-like phenotypes of expanded CD4⁺ and CD8⁺ CAR T cells. (G) Quantification of the memory-like phenotypes of expanded CD4⁺ and CD8⁺ CAR T cells. Data are shown as mean+SEM for n=3 independent donors. (H) Percentages of PDPN-positive cells in GCLs (DKMG, U87, U138, U251, U343, and U373) determined by flow cytometry. Freq. of parent (cells/single cells/live cells) is displayed for n=3 technical replicates. (I, J) Tumor cell lysis after incubation with utd T cells and PDPN-CAR T cells at an E:T ratio of 10:1 after 24 hours. Data are shown as mean+SEM for n=3 independent donors. Asterisks indicate a significant increase in lysis compared with the respective utd control (*p<0.05; **p<0.01, ****p<0.0001; p>0.05 not depicted). (K) A representative CFSE assay shows the proliferation of CD8⁺ CAR T cells after 72 hours of co-culture with GCLs. IL-2 (250 U/µl) was used as a positive control. (L) ELISA-based quantification of IFNγ from supernatants collected after 24 hours of coincubation of CD8⁺ PDPN-CAR T cells with GCLs at an E:T ratio of 4:1. Data are shown as mean+SEM for n=3 independent donors (****p<0.0001; p>0.05 not depicted). CAR, chimeric antigen receptor; CFSE, carboxyfluorescein succinimidyl ester; E:T, effector-to-target; GCLs, glioblastoma cell lines; IFNγ, interferon-gamma; PDPN, podoplanin; scFv, single-chain fragment variable; T_CM, central memory-like T cell; T_EFF, effector-like T cell; T_EM, effector memory-like T cell; T_SCM, stem cell memory-like T cell; utd, untransduced.

To assess the effectiveness of PDPN-CAR T cells, we conducted a pilot in vivo experiment with the U343 cell line, which exhibits robust PDPN expression (online supplemental figure S2A; figure 2H). Notably, U343 tumors displayed low tumorigenicity in NSG mice and lacked aggressive growth characteristics, as indicated by the relatively slow increase in fold-change radiance in control mice up to day 63. Nevertheless, a treatment effect was still observed in PDPN-CAR–treated mice (online supplemental figure S2B and C). In contrast, long-term follow-up revealed that U343 tumors in control groups eventually progressed, whereas PDPN-CAR T cells—but not untransduced T cells—effectively eliminated orthotopic U343 tumors, resulting in durable complete remission throughout the 4-month observation period (online supplemental figure S2D). Importantly, peripheral blood analysis on day 10 revealed only sporadic detection of both untransduced and PDPN-CAR T cells in the circulation, suggesting limited migration of T cells from the brain (online supplemental figure S2E).

Since NZ-1 does not exhibit cross-reaction with mouse tissue, we performed ex vivo co-cultures with type I alveolar cells to extend our toxicity screening (online supplemental figure S3A). Human microvascular endothelial cells (HMVECs) contain a small fraction of highly PDPN-positive cells, whereas small airway epithelial cells (SAECs) exhibit a uniform shift, although with overall low expression (online supplemental figure S3B). After 24 hours of coincubation with PDPN-CAR T cells, SAECs did not induce T cell activation in either CD4⁺ or CD8⁺ CAR T cells (as indicated by a fold change of 1 in CD25⁺, CD69⁺, and CD25⁺CD69⁺ expression). In contrast, HMVECs induced a modest increase in CD25⁺ and CD25⁺CD69⁺ expression; however, this response was considerably weaker than the tumor-specific activation observed following co-culture with U87 cells (with fold changes approximately twice as high in CD8⁺ CAR T cells) and occurred only at a high effector-to-target (E:T) ratio of 10:1 (online supplemental figure S3C and D).

Dual-antigen CAR T cell strategies targeting PDPN and GD2

Consistent with its expression in cell lines as well as in primary and recurrent tumor tissues, the expression pattern of GD2 suggests its suitability as a combinatorial antigen in a dual targeting CAR T cell approach (figure 1A–C). Therefore, we generated CAR T cells based on an scFv fragment derived from the GD2-specific antibody 14.18. Purification, gene transfer rates and T cell expansion were equally efficient for GD2-CAR T cells and PDPN-CAR T cells (figure 2B–D and online supplemental figure S4A-D). Notably, antigen staining, cytotoxicity, and cytokine analyses indicated the complementary functionality of GD2-CAR T cells and PDPN-CAR T cells. In 24-hour assays, five of the six GCLs were efficiently lysed by GD2-CAR T cells or PDPN-CAR T cells, and significant IFNγ production was detected (online supplemental figure S4E-H and figure 2J,L).

We employed and compared multiple methods of dual-antigen targeting in a systematic in vitro study: coadministration of PDPN-CAR T cells and GD2-CAR T cells at a 1:1 ratio (“blend”; figure 2A; online supplemental figure S4A), as well as cotransduction, tandem CAR and bicistronic vector strategies (online supplemental figure S5A). T cells could be efficiently expanded using all strategies except for the cotransduction approach. To achieve robust coexpression of PDPN and GD2, multiple rounds of magnetic enrichment were needed, ultimately resulting in reduced overall cell numbers by the end of the 10-day expansion protocol (online supplemental figure S5B).

The different formats only slightly influenced the CAR T cell memory phenotype, resulting in minimal changes within the memory T cell subpopulations. However, the tandem CAR format resulted in the highest proportion of CD4⁺ T_CM cells, while the blend format resulted in the lowest proportion of effector-like (T_EFF) CD8⁺ T cells (online supplemental figure S5C). In contrast to the relatively consistent phenotypic characteristics, functional evaluation using GCLs revealed more pronounced differences. Notably, the tandem CAR T cells failed to induce substantial target cell lysis. All other formats eradicated the DKMG, U87, U343, and U373 cells and lysed the majority of the U138 cells after 24 hours of co-culture. Importantly, treatment with the blend resulted in unabated cytotoxicity despite containing only half as many GD2-specific or PDPN-specific CAR T cells as single-antigen treatments and still achieved complete lysis of DKMG, U343, and U373 cells (online supplemental figure S5D).

A significant increase in CAR-specific proliferation in response to GCLs was observed only with the blended and cotransduced CAR T cells. In contrast, the tandem and bicistronic CAR T cells did not exhibit meaningful proliferation (online supplemental figure S5E). Furthermore, analysis of cytokines and cytotoxic mediators in the supernatants collected after 24 hours of co-culture revealed robust levels of IFNγ, granzyme B, and Fas ligand (FasL) in the blended and cotransduced CAR T cells. The other formats, however, failed to induce substantial cytokine or effector molecule release (online supplemental figure S5F).

Collectively, these findings led us to select the CAR T cell blend as the preferred format for dual-antigen targeting based on its favorable functional and phenotypic characteristics, as well as its modular design.

CAR T cell blend consistently demonstrates superior effectiveness in PDOs

To investigate the potential synergy in a heterogeneous model representing primary tumor characteristics, we conducted dual targeting experiments with PDOs. CAR T cells derived from healthy donors were co-cultured with PDOs generated from resected GBM tumor tissue at an inverse E:T ratio of 1:4. After 20 hours of co-culture, the organoids were either directly processed for immunofluorescence (IF) microscopy or treated with the Caspase-Glo 3/7 3D Reagent to quantify luminescence signals proportional to caspase activity within the PDOs. The supernatants were collected prior to reagent addition for cytokine profiling using cytometric bead arrays (LEGENDplex) (figure 3A). For the blended CAR T cell therapy, PDPN-targeting and GD2-targeting CAR T cells were combined at a 1:1 ratio, each comprising CD4⁺ and CD8⁺ T cells (figure 3B). In these co-culture experiments, complete organoid degradation was observed after 48 hours, hindering evaluation by IF microscopy. Consequently, an interim analysis time point at 20 hours was selected for PDO harvest and subsequent processing (figure 3C,D).

Figure 3. Dual targeting PDPN-CAR T cells/GD2-CAR T cells significantly increase caspase activity in PDOs and induce favorable cytokine and cytotoxic mediator profiles. (A) CAR T cells (CD4⁺:CD8⁺ 1:1) were co-cultured with PDOs at an inverse E:T ratio of 1:4 for 20 hours. PDOs were fixed and processed for IF microscopy, or the supernatants were collected for LEGENDplex analysis prior to the addition of the luminogenic Caspase-Glo 3/7 3D reagent. Following incubation for 30 min to 2 hours at 500 rpm, caspase activity was quantified by measuring the luminescence signal proportionally. Created with BioRender.com. (B) The independently generated PDPN-CAR T cells and GD2-CAR T cells were mixed at a 1:1 ratio for dual targeting therapy of heterogeneous tumor cells. (C) Representative light microscopy images of PDOs without (only) and with (PDPN) CAR T cell treatment at different time points. Bright-field mode with standard settings was used. Scale bar: 50 µm. (D) Representative IF images of tumor cell apoptosis (DAPI⁺/TAA⁺/CC3⁺). Magnification: 40×. Scale bar: 50 µm. (E) PDPN and GD2 expression in various PDOs was analyzed by flow cytometry. Freq. of parent (cells/single cells/live cells) is depicted. (F) Caspase 3/7 activity in PDOs from n=4 independent patients after 20 hours of co-culture with CAR T cells from n=3 independent donors. All experiments were performed in triplicates (three organoids per patient per T cell donor). (G) LEGENDplex-based quantification of cytotoxic mediators in the supernatants of PDOs from n=4 independent patients co-cultured with CAR T cells from n=3 independent donors at an E:T ratio of 1:4. Data are shown as mean+SEM (*p<0.05; **p<0.01; ****p<0.0001). CAR, chimeric antigen receptor; DAPI, 4′,6-diamidino-2-phenylindole; E:T, effector-to-target; FasL, Fas ligand; GBM, glioblastoma; IF, immunofluorescence; IFNγ, interferon-gamma; ns, not significant; PDOs, patient-derived organoids; PDPN, podoplanin; RLU, relative light units; TAA, tumor-associated antigen; utd, untransduced.

Both antigens could be detected in PDOs derived from four different patients at varying levels, demonstrating conserved yet heterogeneous expression. However, three of these four PDOs contained a GD2 mono-positive population, whereas no PDPN mono-positive population was observed. The double expression of both markers varied widely, ranging from low levels to nearly 100% (figure 3E). Despite the predominance of GD2 expression across these organoids, dual-antigen targeting using our CAR T cell blend resulted in significant functional enhancement. Specifically, caspase 3/7 levels were significantly elevated following coincubation with the CAR T cell blend compared with single-antigen targeting CAR T cells (figure 3F). Furthermore, the blend showed unabated IFNγ, granzyme B, and FasL production, producing substantial quantities comparable to those released by GD2-CAR T cells, which is consistent with the observed GD2 expression (figure 3G).

We subjected an additional set of PDOs (n=8) to IF analysis and detected both antigens in all the samples (online supplemental figure S6A and B). In this cohort, PDPN was predominantly expressed, and compared with GD2-CAR T cells, PDPN-CAR T cells induced significantly higher levels of apoptosis. Importantly, dual treatment with the CAR T cell blend consistently proved to be the most effective therapeutic modality across all individual PDOs, irrespective of antigen expression levels (online supplemental figure S6C). When combined as a therapeutic blend, PDPN-CAR T cells and GD2-CAR T cells acted synergistically, inducing significantly higher levels of tumor cell apoptosis than single-antigen targeting approaches did (online supplemental figure S6D).

Finally, co-culture experiments were repeated with PDOs and CAR T cells generated from the same patients in an autologous setting (online supplemental figure S7A). Peripheral blood mononuclear cells (PBMCs) were isolated following surgery and histopathological confirmation of the GBM diagnosis, prior to the initiation of radiochemotherapeutic treatment. PDOs were harvested after 20 hours of co-culture with either autologous or allogeneic PDPN-CAR T cells and GD2-CAR T cells and subsequently processed for IF analysis. Representative IF images demonstrated that CAR T cells induced robust levels of apoptosis, with no statistically significant differences observed between autologous PDPN-CAR T cells or GD2-CAR T cells and those derived from unrelated healthy donors (online supplemental figure S7B and C).

These findings not only demonstrate the potential of this organoid model for patient-specific preclinical assessment of CAR T cell effectiveness prior to therapy but also demonstrate that the CAR T cell blend effectively targets heterogeneous PDOs by significantly increasing cytotoxic activity and unabated key effector molecule production through dual-antigen targeting.

Curative dual targeting in an orthotopic xenograft model

To evaluate whether a dual targeting CAR T cell blend confers a gain of function in vivo, we first established a heterotopic xenograft model in NSG mice. The GCLs DKMG and U343 were coinjected subcutaneously within an extracellular matrix to promote tumor formation. This approach aimed, on the one hand, to generate tumors expressing both PDPN and GD2 and, on the other hand, to increase tumor malignancy by situating the cells in close proximity, thereby facilitating high cellular density and intercellular crosstalk.^{22 23} 7 days after tumor implantation, T cells were administered intravenously, and tumor volumes were continuously monitored using calipers (online supplemental figure S8A).

This strategy yielded a heterogeneous and highly aggressive tumor model with rapid growth kinetics. In vivo passaging in NSG mice altered the antigen expression profile of the injected cell mixture, resulting in tumors comprising four distinct subpopulations: cells expressing PDPN only, GD2 only, both antigens, or neither (online supplemental figure S8B and C). Despite the aggressive expansion and high cellular density of the tumors, we observed a significant therapeutic response to the CAR T cell blend on days 10 and 14 (online supplemental figure S8D). The initial treatment response correlated with OS outcomes, with the CAR T cell blend conferring the most substantial survival benefit (online supplemental figure S8D).

Following the confirmation of the enhanced effectiveness of the CAR T cell blend compared with that of single-antigen targeting CAR T cells in this short-term heterotopic xenograft model, we proceeded to establish a heterogeneous orthotopic NSG model to evaluate its therapeutic potential in a more physiologically relevant context. U87 firefly luciferase⁺/green fluorescent protein⁺ (ffluc⁺/GFP⁺) cells were injected into the caudate nucleus of the right hemisphere on day 0, and tumor engraftment was verified by bioluminescence imaging (BLI) on day 6 (figure 4A). We selected U87 cells for our xenografts because of their heterogeneous antigen expression pattern according to flow cytometric analysis, with populations expressing GD2, PDPN, both or neither of these markers (figure 4B). Notably, U87 tumors isolated from the brains of untreated mice exhibited variable expression patterns, including elevated GD2 expression, elevated PDPN expression, elevated expression of both markers, and the absence of both markers, indicating a certain degree of plasticity in vivo (figure 4C,D). This heterogeneity consequently constituted a challenging environment for our dual targeting approach.

Figure 4. Synergistic effects of a PDPN-CAR T cell and GD2-CAR T cell blend in orthotopic mouse models. (A) 6–8 weeks old female NSG mice received an i.c. injection of 2×10⁵ U87 ffluc/GFP⁺ cells into the caudate nucleus of the right hemisphere on day 0 using a stereotactic frame. After verifying tumor engraftment on day 6 using BLI, 2×10⁶ CAR T cells or untransduced T cells (1:1 mixture of CD4⁺ and CD8⁺ T cells) were administered via i.c.v. injection into the lateral ventricle of the left hemisphere. The tumor reduction and overall survival of the mice were then monitored. Created with BioRender.com. (B) Representative PDPN/GD2 expression pattern of U87 cells after standard in vitro culture analyzed by flow cytometry. Freq. of parent (cells/single cells/live cells) is depicted. (C, D) Analysis of the antigen expression patterns of untreated U87 tumors. Freq. of parent (cells/single cells/live cells) is depicted. Data are shown as mean. (E, F) Monitoring of the fold change in average radiance over time. Data are shown as medians and individual curves for the mice. An asterisk indicates time points at which the tumor load significantly decreased between the dual targeting group and the utd control group (*p<0.05; p>0.05 not depicted). (G) Analysis of the initial response depth on day 7 after T cell administration (corresponding to day 14 after tumor engraftment). Medians and quartiles are shown (*p<0.05; p>0.05 not depicted). Whiskers represent the 10th–90th percentiles. (H) Analysis of PFS, defined in our model as the time interval of tumor recurrence beyond the initial BLI signal on day 6. Data are shown as mean+SEM (*p<0.05; p>0.05 not depicted). (I) Kaplan-Meier survival analysis of untreated (n=4), utd T cell-treated (n=3) and CAR T cell-treated (n=4) mice (*p<0.5; **p<0.01). (J) Percentage of surviving mice and curative rates after treatment with GD2-CAR T cells, PDPN-CAR T cells, or the PDPN-targeting/GD2-targeting blend at the end of the 125-day observation period. BLI, bioluminescence imaging; CAR, chimeric antigen receptor; FC, fold change; ffluc, firefly luciferase; GFP⁺, green fluorescent protein⁺; i.c., intracerebral; i.c.v., intracerebroventricular; PDPN, podoplanin; PFS, progression-free survival; utd, untransduced.

As the intrathecal injection of CAR T cells via an Ommaya reservoir appears to be superior to intravenous application,24,26 we administered our CAR T cells directly into the brain on day 7, thereby bypassing the blood–brain barrier. Injection into the lateral ventricle rather than into the tumor cavity added an additional layer of complexity to CAR T cell trafficking. The reduction in the tumor BLI signal and OS of the mice was subsequently monitored (figure 4A).

Stereotactic injections were successful, as evidenced by consistent tumor engraftment in all experimental groups (online supplemental figure S9A). 6 weeks after treatment, all mice in the control group met the termination criteria. CAR T cell-treated mice exhibited tumor regression and remission following intracerebroventricular (i.c.v.) delivery. However, tumor relapse occurred in mice treated with single-antigen targeting CAR T cells, whereas durable responses were observed only in mice receiving dual-antigen treatment with the PDPN-targeting/GD2-targeting blend at the same total CAR T cell dose (figure 4E,F). Assessment of the initial depth of response on day 14 after T cell administration revealed that the number of established tumors was moderately reduced in the PDPN-CAR T cell and GD2-CAR T cell groups; however, only treatment with the blend resulted in a consistent and significant reduction in tumor size (figure 4G). Consistent with these findings, progression-free survival, which is defined in our model as the time interval until tumor recurrence beyond the initial BLI signal on day 6, was significantly prolonged only in mice receiving dual-antigen therapy (figure 4H).

With a median survival time of 55 days, even a single injection of untransduced T cells conferred a survival advantage over no treatment in mice, which is consistent with the in vitro cytotoxicity data (figure 2I). Treatment with either PDPN-CAR T cells or GD2-CAR T cells significantly extended the median survival to 103 and 118 days, respectively. Notably, the combination of both CAR T cell products in a dual targeting regimen resulted in 100% survival of the treated mice (figure 4I and online supplemental figure S9A and B). Flow cytometric analysis of the brain from surviving mice confirmed the absence of GFP⁺ tumor cells in those with durable remission (online supplemental figure S9C). Thus, while single-antigen targeting therapy led to curing in only one mouse (PDPN group), the CAR T cell blend demonstrated synergistic activity, resulting in an encouraging curative effect in the majority of treated animals (figure 4J).

Collectively, these data demonstrate that i.c.v. delivery of a dual targeting PDPN-CAR T cell/GD2-CAR T cell blend efficiently reduces the number of heterogeneous tumors within a physiological environment.

Discussion

Given the success of CAR T cells in the treatment of several hematological malignancies, high expectations exist for the treatment of solid tumors, including GBM. However, single-antigen targeting CAR T cell therapies have encountered significant challenges in GBM, including a highly heterogeneous tumor antigen landscape.

In this study, we aimed to target two antigens, both of which are ideally involved in tumor progression and maintenance, to effectively address the heterogeneity in GBM. To this end, we analyzed cell lines, paraffin-embedded GBM specimens and primary tissue samples using IHC, qPCR, and flow cytometry. Additionally, we investigated the association between antigen expression and patient survival through analysis of a publicly available dataset. Notably, established GBM target antigens such as EGFRvIII and IL13Rα2^{12 14} were expressed across all GCLs but were detected at low levels in our primary tissue cohort. In contrast, the type I integral membrane glycoprotein PDPN was highly expressed in all tissue samples, and survival analysis revealed a clear correlation between PDPN expression and a poor prognosis. A potential limitation of our antigen screening is that, at the time of data collection, malignant cells were not distinguished from surrounding stromal cells, should they also express the respective markers, in the flow cytometric analyses. Accordingly, normalization to tumor content was not performed, which may have influenced the observed expression patterns of the investigated markers depending on their expression in non-malignant components of the TME.

PDPN is a marker of tumor lymphatic vessels, and its expression increases with increasing glioma WHO grade.²⁷ Functionally implicated in tumorigenesis, PDPN overexpression promotes the phosphorylation of ERM proteins and activates RhoA, leading to enhanced cell motility and the induction of epithelial-to-mesenchymal transition.^{28 29} An aggressive subset of glioma stem cells, characterized by elevated PDPN expression, exhibits increased treatment resistance, and PDPN depletion sensitizes glioma cells to radiotherapy.³⁰ Additionally, glioma cells secrete PDPN via exosomes, which induces PDPN surface expression on macrophages and M2 polarization within the TME following phagocytosis.³¹ Together with its surface expression on cancer-associated fibroblasts within the TME,³² these findings highlight the abundant expression of PDPN in GBM. Moreover, the ability to target both tumor cells and components of the TME via PDPN further supports its potential as a promising therapeutic target. This is particularly relevant given the immunosuppressive nature of the GBM TME, which drives CAR T cell exhaustion and severely limits their trafficking to and infiltration into the tumor core.³³

Derived from the mAb NZ-1, our PDPN-CAR T cells demonstrated substantial effectiveness both in vitro and in vivo. Although U343 cells exhibit relatively low tumorigenicity in NSG mouse brains and slow growth kinetics, this model provides valuable insights.^{34 35} Following locoregional delivery, CAR T cells effectively migrated toward and lysed PDPN-positive U343 tumors, resulting in long-term remission. Only a small fraction of CAR T cells entered the peripheral circulation; therefore on-target off-tumor effects in healthy organs are likely to remain within a manageable range, especially considering the systemic administration of other CAR T cell products currently under clinical consideration that target antigens with similar profiles.³⁶ Additionally, T cell activation following co-culture with HMVECs was minimal and occurred only at a high E:T ratio (10:1), indicating that the model was highly stringent and intentionally exaggerated.

Previous studies in which NZ-1-derived PDPN-CAR T cells were used either did not address GBM heterogeneity or failed to achieve curative outcomes in mouse models.^{37 38} Here, we combined CAR T cells targeting PDPN and GD2, a tumor-specific disialoganglioside that is highly expressed in various cancers, particularly in central nervous system tumors.39,41 Mainly absent in normal tissues, GD2 is linked to invasion, motility, and proliferation and plays a role in oncogenesis through several signaling pathways.^{18 42 43} GD2 expression was consistently high in GCLs and fresh GBM tissue, and in addition to PDPN, it was the only antigen among the candidates that exhibited significantly elevated expression levels in rGBM. Additionally, GD2 is a well-established CAR T cell target and has been evaluated in multiple clinical trials, demonstrating its feasibility and safety.^{19 20} Given this favorable profile, we selected GD2 for incorporation into a dual-antigen targeting strategy alongside PDPN.

Multiple dual targeting approaches, including bicistronic or tandem CAR designs and the simultaneous administration of two separate CAR T cell products, are being investigated.44,46 The optimal approach for specific malignancies and antigen combinations remains unclear, as each strategy presents distinct advantages and disadvantages, with CAR design and intrinsic signaling further contributing to the complexity of multitargeting approaches. In our hands, a 1:1 blend of PDPN-CAR T cells and GD2-CAR T cells emerged as the most promising format in this study. The blend resulted in unabated specific target cell lysis in single-positive cell lines while demonstrating favorable proliferative capacity and effector molecule profiles. Additionally, this approach maintains modularity, allowing for the easy addition of or combination with other CARs without increasing engineering and manufacturing complexity or compromising gene transfer efficiency because of increased genetic cargo.⁴⁷

Cell lines inadequately represent paternal tumor characteristics because of monolayer cultivation, disruption of cell-to-cell contacts during passaging and the total duration of in vitro cultivation. In contrast, PDOs, generated directly from intraoperative primary material within 2–4 weeks, can partially retain the tumor vasculature, glioma stem cell markers, and hypoxic gradients. They represent the primary tumor at the histological, molecular, and transcriptional levels, even after weeks of culture, freezing, and thawing.²¹ Therefore, PDOs represent a heterogeneous and challenging model for evaluating the effectiveness of our CAR T cell blend. PDPN and GD2 were detected at varying levels in each patient examined, underscoring the conserved yet heterogeneous expression of both antigens. We chose to process the organoids for IF analysis within 20 hours of co-culture with CAR T cells to preserve tissue integrity and prevent degradation and loss of the organoids after longer incubation times. At this intermediate time point, our CAR T cell blend induced significantly higher levels of apoptosis in PDOs than single-antigen targeting CAR T cells did and resulted in favorable effector molecule release. Collectively, these findings suggest that dual PDPN-targeting/GD2-targeting is promising for the treatment of heterogeneous GBM.

To evaluate our CAR T cell blend in vivo, we first established a highly aggressive subcutaneous xenograft model for a pilot study and observed an increased depth of response compared with that of single-antigen targeting CAR T cells, which ultimately translated into significantly prolonged OS. To complete our preclinical assessment, we used an advanced orthotopic NSG mouse model that allows direct evaluation of the functionality of our human CAR T cells against human GBM within a physiologically relevant environment. In this model, we inoculated U87 cells, which exhibited considerable plasticity and heterogeneity in vivo, forming distinct GD2-expressing and PDPN-expressing subpopulations across different mice. Despite the absence of a complex TME in NSG mice, this model enabled investigation of the benefits of dual targeting in reducing heterogeneous tumors following locoregional administration.

Rather than systemic or intratumoral delivery, we administered CAR T cell products into the lateral ventricle of mice, as clinical studies suggest that direct delivery into the cerebrospinal fluid enhances therapeutic effectiveness.^{24 48 49} In a clinical setting, the placement of an Ommaya reservoir in the ventricle facilitates repeated administration, pharmacodynamic monitoring, and rapid management of elevated intracranial pressure and toxicity.¹⁹ Owing to the procedural complexity and to minimize the surgical burden in the murine model, we investigated the effect of a single injection administered on day 7. Nevertheless, the locoregional administration of single-antigen targeting CAR T cells resulted in remarkable tumor reduction and partial remission, outcomes that were not observed in previous studies using similar models.^{37 50}

Importantly, dual targeting therapy significantly extended OS and achieved curative effects and durable remission in the majority of treated mice; only one mouse treated with the CAR T cell blend eventually relapsed but survived symptom-free with a relatively low tumor burden until the end of the 125-day observation period. Given that a single injection elicited such sustained responses against heterogeneous U87 tumors, our findings demonstrate the synergistic and curative potential of pooled PDPN-CAR T cells and GD2-CAR T cells in vivo.

Overall, we present a novel dual targeting approach employing a PDPN-CAR T cell/GD2-CAR T cell blend that effectively targets diverse GCLs, heterogeneous PDOs, and heterotopic and orthotopic xenografts. Clinically, this innovative combination has the potential to address GBM heterogeneity and potentially overcome the limitations associated with single-antigen CAR T cell therapies.

Materials and methods

Glioblastoma tissue

Fresh GBM tissue for flow cytometry and PDO generation was obtained intraoperatively after patients provided informed consent to the study. Paraffin-embedded sections were provided by the Department of Neuropathology, University of Würzburg.

Cell lines

GCLs DKMG, U373, and U251 were purchased from the German Collection of Authenticated Cell Cultures (DSMZ, Leibniz, Germany). The U138, U343 and U87 cell lines were purchased from the European Collection of Authenticated Cell Cultures (Porton Down, England). GCLs were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 1 g/L glucose, sodium pyruvate, 3.7 g/L NaHCO₃ and 4 mM L-glutamine (Gibco, Carlsbad, California, USA), as well as 10% v/v heat-inactivated fetal calf serum, 2× non-essential amino acids (NEAA, 100×stock), and 1% penicillin/streptomycin (Invitrogen, Carlsbad, California, USA) as a monolayer in 75 cm³ flasks (Corning, New York, New York, USA) at 37°C, 5% CO₂ and 95% humidity.

Alveolar cells

HMVECs (CC-2527) and SAECs (CC-2547) were purchased from Lonza (Basel, Switzerland) and cultivated using the EGM −2 MV Microvascular Endothelial Cell Growth Medium-2 BulletKit and SAGM Small Airway Epithelial Cell Growth Medium BulletKit according to the manufacturer’s instructions.

Generation of patient-derived organoids

Organoids were prepared as previously described.⁵¹ In brief, acute tumor tissue was temporarily stored on ice in Hibernate A medium (Gibco). The tumor tissue was cleared from necrotic tissue and blood vessels, carefully cut into pieces with a McIlwain Tissue Chopper (Cavey Laboratory Engineering, Guildford, UK), treated with RBC lysis buffer (Invitrogen) for 10 min and washed two times with Hibernate A medium containing 1% GlutaMAX, 0.4% penicillin/streptomycin and 0.1% amphotericin (all from Gibco). The sections were transferred into GBO medium consisting of a 1:1 mixture of DMEM/F12 and neurobasal media supplemented with 2% B27 without vitamin A (50×), 1% GlutaMAX, 1% NEAA, 0.4% penicillin/streptomycin, 0.1% β-mercaptoethanol (all from Gibco) and 0.023% human insulin (Sigma-Aldrich, Merck KGaA, Darmstadt, Germany) in ultralow-attachment 6-well plates (Corning) and incubated at 37°C, 5% CO₂ and 95% humidity on an orbital shaker at 120 rpm. After 2 weeks of culture, organoids measuring approximately 100–200 µm had formed, adopted a spherical morphology, and could be used for experimental procedures.

Preparation of CAR T cells

CAR T cells were generated as previously described. In brief, lentiviral transduction of healthy donor bulk CD4⁺ and bulk CD8⁺ T cells isolated from PBMCs was performed with vectors encoding second-generation PDPN-CARs and/or GD2-CARs comprising a single-chain fragment derived from NZ-1 and/or hu14.18, respectively, an IgG4 spacer and a signaling module of 41BB_CD3ζ. After enrichment of CAR-positive cells via EGFRt or HER2t selection, both CD4⁺ and CD8⁺ CAR T cells were expanded using an antigen-independent rapid expansion protocol over 10 days⁵² and used for functional evaluation from day 21 on after initial T cell isolation.

Subcutaneous xenograft model

All animal experiments were approved by local authorities (Regierung von Unterfranken). 6–8 weeks old female NSG mice (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (JAX Stock number 005557)) were subcutaneously injected with a total volume of 100 µL, consisting of 50 µL containing 1×10⁶ DKMG and U343 cells in 0.9% sodium chloride injection solution (1:1 ratio) and 50 µL of extracellular matrix gel (Sigma-Aldrich, E6909). On day 7, 5×10⁶ CAR T cells or untransduced T cells (1:1 mixture of CD4⁺ and CD8⁺ T cells) were administered via intravenous injection. The tumor volumes were continuously monitored using calipers.

Orthotopic xenograft model

6–8 weeks old female NSG mice were stereotactically injected on day 0 with 2×10⁵ U87 ffluc/GFP⁺ or U343 ffluc/GFP⁺ GBM cells into the caudate nucleus of the right hemisphere at 1.8 mm lateral, 0.5 mm anterior to the bregma and 3.5 mm below the dura. At a rate of 0.1 mm/sec, the needle was advanced 3.8 mm before injection, paused there for 1 min to allow the tissue to bounce back, and then moved back 0.3 mm to create a pocket for the tumor cells. A total of 5 µL of cell suspension was injected at a rate of 1 µL/min before the needle was retracted at 1 mm/min to avoid efflux of the suspension. Tumor engraftment was verified under anesthesia with isoflurane on day 6 via BLI on an IVIS Lumina (PerkinElmer, Waltham, Massachusetts, USA) 10 min after the intraperitoneal injection of D-luciferin (Biosynth, Staad, Schweiz) at 0.3 mg per g body weight. Imaging data were analyzed using LivingImage software (PerkinElmer). On day 7, 2×10⁶ CAR T cells or untransduced T cells (1:1 mixture of CD4⁺ and CD8⁺ T cells) were administered via i.c.v. injection into the lateral ventricle of the left hemisphere at 0.95 mm medial, 0.22 mm posterior to the bregma and 2.3 mm below the dura. A total of 5 µL of cell suspension was injected at a rate of 1 µL/min before the needle was retracted at 1 mm/min. The degree of tumor reduction was subsequently analyzed via BLI, and the OS of the mice was monitored.

Immunohistochemistry

For IHC, tissue was cut into 2-µm-thick slices, deparaffinized in xylene and hydrated in a graded series of alcohol. Heat-induced retrieval was either performed for 20–40 min with citrate buffer at pH=6.0 (EGFRvIII, GD2, HER2, and PDPN) or for 5 min with Tris-EDTA buffer at pH=9.0 (CD70, CD133, CSPG4, EphA2, and IL13Rα2). After blocking with 0.7% hydrogen peroxide (Sigma-Aldrich), the slides were treated with 10% normal goat serum (Invitrogen), and the primary antibody diluted with antibody buffer (DCS Innovative Diagnostic Systeme, Hamburg, Germany) was applied overnight at 4°C (see online supplemental Table 1). The slides were then incubated with the secondary antibody for 30 min and labeled with streptavidin peroxidase (Super Sensitive Link-Label IHC Detection System, DCS/BioGenex, Hamburg, Germany). Diaminobenzidine (Liquid DAB+Substrate Chromogen System, Dako, Wiesentheid, Germany) was applied for 5 min, and after rinsing with water, the cell nuclei were counterstained using hemalum solution acid (Carl Roth GmbH, Karlsruhe, Germany) before the slides were mounted.

On primary tissue slides, five representative fields of view per slide were imaged with a Leica DMI3000 B light microscope, Leica DFC450 camera and LAS V.4.5 software (all Leica Microsystems, Wetzlar, Germany) with standardized settings at 40× amplification and analyzed for staining intensity via the batch processing function of the open-source program Fiji.⁵³ For IHC, the percentage of low-antigen-expressing cells, medium-antigen-expressing cells and high-antigen-expressing cells on each slide was quantified by the use of a custom macro (see online supplemental Data). The expression score was further weighted by multiplication by 1 (low), 2 (medium) or 3 (high expression). Finally, the median expression values of all the tumors combined for each antigen were calculated. All analyses were monitored, double-checked and approved by an experienced neuropathologist.

Flow cytometry

15 patients at the Department of Neurosurgery, University Hospital Würzburg, were included. The cell surface expression of target antigens was analyzed by flow cytometry. After acute tissue was removed from vessels, surrounding brain tissue and debris, homogenized mechanically, washed, resuspended in 100 µL of phosphate-buffered saline/0.5% bovine serum albumin (Thermo Fisher Scientific, Waltham, Massachusetts, USA) and treated with Fc receptor-blocking reagent (Human TruStain FcX, BioLegend, San Diego, California, USA) for 5 min, it was stained with 0.5 µL of the appropriate antibody for 20 min at 4°C (see online supplemental Table 1). If necessary, secondary staining was performed after the tissue was washed for 20 min at 4°C.

Mouse brains and subcutaneous tumors were harvested directly after euthanasia with CO₂, cut into small pieces and subsequently digested using a human tumor dissociation kit (Miltenyi, Bergisch Gladbach, Germany) and a gentleMACS Octo dissociator with heaters (Miltenyi, Program 37C_h_TDK1). Cell debris was removed from viable cells using debris removal solution (Miltenyi), and red blood cell lysis was performed using BD PharmLyse Buffer (BD, Franklin Lakes, New Jersey, USA). After live/dead staining with Zombie Aqua (mouse brains, 1:1000, BioLegend) or Ghost Dye Red 710 (subcutaneous tumors, 1:10000, Cytek Biosciences, Fremont, California, USA) and Fc receptor-blocking reagent treatment, the cells were stained with the appropriate antibodies (see online supplemental Table 1).

PDOs were digested using a tumor dissociation kit (Miltenyi) and a gentleMACS Octo dissociator with heaters (Miltenyi, Program 37C_h_TDK1). After live/dead staining with Ghost Dye Red 710 (1:10000, Cytek) and Fc receptor-blocking reagent treatment, the cells were stained with the appropriate antibodies (see online supplemental Table 1) and analyzed.

CAR T cells cultured in vitro were washed and processed via the use of the same staining protocol. To isolate CAR T cells from murine blood samples, red blood cell lysis was performed using BD PharmLyse Buffer (BD), and identical staining procedures were applied as described above.

All acquisitions were conducted using a FACSCanto II (BD) or Cytek Northern Lights (Cytek) flow cytometer. Subsequent data analysis was performed with FlowJo software V.10.7.1 (BD).

LEGENDplex

Cytokines in the supernatants of co-cultures of CAR T cells and GCLs as well as CAR T cells and PDOs were analyzed using the LEGENDplex Human CD8/NK Panel (BioLegend) according to the manufacturer’s instructions. Acquisitions were conducted using a Cytek Northern Lights system (Cytek). Subsequent data analysis was performed with the Data Analysis Software Suite for LEGENDplex.

qPCR

Pooled cDNA from healthy organs (Human MTC Panel I and II, Takara Bio, Kusatsu, Japan) and healthy brain tissue (TissueScan, Human Brain cDNA Array, HBRT501, OriGene, Rockyville, Maryland, USA) was analyzed for PDPN expression. Each sample was analyzed with the StepOnePlus Real-time PCR System using TaqMan PDPN Gene Expression Assays (Thermo Fisher Scientific, FAM-MGB_PL, Hs00366766_m1) with GAPDH as an endogenous control (Thermo Fisher Scientific, VIC-MGB_PL, Hs99999905_m1). Three technical replicates were performed according to the manufacturer’s instructions.

To rank the expression in the obtained healthy tissue, PDPN expression in 29 patients with GBM was analyzed. Total mRNA was extracted from frozen tissue using the PureLink RNA Mini Kit (Thermo Fisher Scientific) and subsequently reverse transcribed using the High Capacity RNA to cDNA Kit (Thermo Fisher Scientific). cDNA concentrations were adjusted to 5 ng/µL with ultrapure water for storage at −80°C. qPCR analysis was performed as described above, and the mean ΔCt value of the patients was used as a calibrator for healthy tissue expression according to the ΔΔCt method.

Survival analysis based on the TCGA GBM dataset

Survival analyses relied on a publicly available cohort of human GBMs made available from TCGA (https://tcga-data.nci.nih.gov/tcga/). In particular, clinical annotations and gene expression data of the TCGA GBM dataset were downloaded by relying on the TCGAbiolinks package (V.2.28.4, https://bioconductor.org/packages/TCGAbiolinks) within R V.4.3.1, applying the following custom filters: experimental.strategy = “RNA-Seq”, workflow.type = “STAR - Counts”, and sample.type = “Primary Tumor”. Raw gene counts were then normalized and transformed using the variance-stabilizing transformation (VST) method implemented in the Bioconductor package DESeq2 (V.1.40.0).⁵⁴ The obtained VST counts were used as expression values in the following survival analyses.

The ganglioside synthase genes ST8SIA1 and B4GALNT1 were used as GD2-phenotype-predicting signatures⁵⁵, and EGFRvIII had to be excluded from this analysis because of the lack of an underlying variant-specific gene signature. To standardize the time measurements, the time to death or last follow-up was converted into months for both deceased and living patients. For each target gene, patients were stratified into high-expression and low-expression groups on the basis of the surv_cutpoint function within the survminer package in R (V.0.4.9, https://CRAN.R-project.org/package=survminer). Unlike a fixed median split, this method determines the optimal cutpoint by maximizing the log-rank statistic across all possible thresholds, effectively identifying the point of greatest separation in clinical outcome. This data-driven approach accounts for the biological distribution of target genes expression while adjusting for multiple testing to provide a robust prognostic stratification. For each gene, the difference in survival rates between the two groups was assessed using the log-rank test. In addition, the HR and its associated p value were computed using the survival package in R (V.3.5.7 https://CRAN.R-project.org/package=survival).

Cytotoxicity assays

GCLs stably transduced with the ffluc/GFP fusion gene were incubated in triplicate wells at 1×10⁴ cells/well. T cells were added to the respective wells at an E:T ratio of 10:1. D-luciferin substrate (Biosynth) was added to the co-culture to a final concentration of 0.15 mg/mL, and the decrease in the luminescence signal in wells that contained target cells and T cells compared with target cells only was measured using a luminometer (Tecan, Crailsheim, Germany).

In PDOs, all apoptotic tumor cells were identified after 20 hours of incubation with CAR T cells at an E:T ratio of 1:4. Triplicate samples were embedded in paraffin and subsequently subjected to costaining and analysis as described above with primary antibodies against CC3 (1:400, clone: Asp175:400, Cell Signaling, Leiden, Netherlands) and GD2 (1:50, clone: 53831, Santa Cruz, Dallas, Texas, USA) or PDPN (ready-to-use, clone: D2-40, Zytomed Systems GmbH, Berlin, Germany). After secondary antibody staining with goat anti-mouse IgG (H+L), AlexaFluor Plus 488 (1:1000, # A32723, Invitrogen), goat anti-rabbit IgG (H+L), and Alexa Fluor Plus 555 (1:1000, # A32732, Invitrogen), the slices were mounted with Fluoroshield mounting medium containing 4′,6-diamidino-2-phenylindole (DAPI; Abcam, Cambridge, UK) and analyzed as described above.

Quantitative measurement of caspase 3/7 activity in PDOs was performed after 20 hours of incubation with CAR T cells at an E:T ratio of 1:4 using the Caspase-Glo 3/7 3D Assay (Promega, Walldorf, Germany) according to the manufacturer’s instructions. Luminescence signals were measured using a luminometer (Tecan).

Cytokine secretion

T cells (5×10⁴) were plated in triplicate wells with target cells at an E:T of 4:1, and the secretion of IFNγ was measured by ELISA (ELISA MAX, BioLegend) in the supernatant after 24 hours of incubation. Analysis was performed using a luminometer (Tecan).

Proliferation assays

T cells were labeled with 0.2 µM carboxyfluorescein succinimidyl ester (Invitrogen), washed, and plated in triplicate wells with irradiated (80 Gy) target cells at an E:T ratio of 4:1. No exogenous cytokines were added to the culture medium. After 72 hours of incubation, the cells were analyzed by flow cytometry to assess the division of T cells.

Activation assays

On day 0, HMVECs, SAECs, and U87 cells were seeded in flat-bottom 96-well plates at a density of 10,000 cells/cm². After 24 hours, T cells were added at an E:T ratio of 10:1, comprising a 1:1 mixture of CD4⁺ and CD8⁺ T cells. After 24 hours, the T cells were harvested, washed, and processed for flow cytometry. Acquisitions were conducted using a Cytek Northern Lights system (Cytek). Subsequent data analysis was performed with FlowJo software V.10.7.1 (BD).

Statistical analyses

Analyses were performed using GraphPad Prism V.9 (GraphPad Software, San Diego, California, USA). The data were examined for a Gaussian distribution by the Kolmogorov-Smirnov test before significance testing was conducted. Data from experiments with cell lines, primary tissue or organoids were analyzed by either paired t-test or two-way analysis of variance (ANOVA) with Sidak’s multiple comparison test. In vivo data were analyzed by one-way ANOVA followed by Dunn’s multiple comparison test and the log-rank (Mantel-Cox) test for Kaplan-Meier survival analysis.

Data availability statement

Data are available upon reasonable request.