Antigen Specificity and Cell Engineering Determine CAR T Cell Efficacy in Group 3 Medulloblastoma
Giedre Krenciute, Meghan Ward, Justine Fouliard, Michaela Meehl, Diana Dinh, Jorge Ibanez-Vega, jingjing liu, Martine Roussel, Jiyang Yu, Stephen Gottschalk
TL;DR
This study compares CAR T cell therapies targeting EphA2 and B7-H3 in group 3 medulloblastoma, finding EphA2-CAR T cells more effective in most models.
Contribution
The study identifies EphA2 as a promising antigen target for CAR T cell therapy in group 3 medulloblastoma.
Findings
EphA2-CAR T cells showed greater cytolytic activity and TH1 cytokine production than B7-H3-CAR T cells in vitro.
EphA2-CAR T cells demonstrated superior tumor control in two of three orthotopic G3MB models.
Genetic modifications like DNMT3A deletion or IL-18 receptor expression improved EphA2-CAR T cell functionality.
Abstract
Group 3 medulloblastoma (G3MB) is a devastating disease of the central nervous system (CNS) that primarily affects infants and children. Chimeric antigen receptor (CAR) T cell therapy holds the promise to improve outcomes for CNS malignancies, but few studies have focused specifically on G3MB. We used publicly available datasets to demonstrate EphA2 and B7-H3 expression in primary G3MB and validated expression in patient-derived cell lines. EphA2-CAR T cells had greater cytolytic activity, persistence, and TH1 cytokine production than B7-H3-CAR T cells in coculture assays with MYC-driven G3MB cell lines in vitro. In vivo, EphA2-CAR T cells demonstrated superior tumor control and improved survival compared to B7-H3-CAR T cells in 2 of 3 orthotopic G3MB models. B7-H3-CAR T cells outperformed EphA2-CAR T cells in one model in which the antigen density of EphA2 was 5-fold lower than for…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsCAR-T cell therapy research · CNS Lymphoma Diagnosis and Treatment · Glioma Diagnosis and Treatment
Introduction
Brain tumors are a leading cause of cancer-related death in children. Advancements in surgical techniques, radiation technology, and targeted chemotherapeutics have greatly improved survivability of some pediatric brain tumors (PBT), but many tumors and subgroups of tumors remain lethal.^1^ Medulloblastoma (MB) is the most commonly diagnosed malignant PBT and is broadly classified into four molecular subgroups: Wingless-activated (WNT), Sonic Hedgehog-activated (SHH), Group 3 (G3MB), and Group 4 (G4MB). Among these, G3MB and G4MB have the poorest prognoses, particularly for high-risk patients defined by metastatic potential.^2^ New therapies are desperately needed to improve outcomes for these patients.
Chimeric antigen receptor (CAR) T cell therapy has transformed outcomes for patients with hematological malignancies, encouraging translation to solid and brain tumors. Multiple CAR T cell clinical trials for PBTs have shown that CAR T cells are safe, and promising signals of antitumor activity have been detected.^3, 4^ However, to-date, there are no reported MB patient outcomes.^5^
GPC2^6^, B7-H3^7, 8^, HER2^9^, and GD2^10^-directed CAR T cells have shown promising preclinical activity against MB. MB patients have been and are eligible for basket-trials investigating HER2 (NCT03500991) and B7-H3 (NCT05835687) CAR T cells and one trial has been initiated specifically for MB investigating GPC2-directed CAR T cells (NCT07087002). Recently, EphA2 was identified as a candidate for CAR T cell therapy for G3MB, and preliminary work showed promising in vivo results with locoregional delivery.^11, 12^ However, thorough preclinical characterization of EphA2-targeted CAR T cells against MB has not been performed.
We therefore sought to thoroughly characterize the therapeutic potential of EphA2-targeted CAR T cell therapy for MB. To contextualize our results, we compared second generation EphA2-CAR T cells against a clinically relevant, second generation B7-H3-CAR, both with CD28ζ stimulation domains. EphA2 expression and activity were restricted to WNT and G3MB, whereas B7-H3 was expressed by all MB subtypes. Patient-derived MYC-driven G3MB cell lines indeed expressed both EphA2 and B7-H3 on the cell surface. EphA2-CAR T cells had superior in vitro antitumor activity against G3MB that was associated with maintained phenotypic diversity upon tumor cell exposure. Further, B7-H3 antigen expression on B7-H3 CAR T cells alone could not explain these differences. We further show that EphA2-CAR T cells had potent in vivo antitumor activity against G3MB xenograft models and was further improved by deleting DNMT3A or constitutively activating IL-18 signaling. Together, our results support further preclinical and clinical development of EphA2-targeted CAR T cell therapy for G3MB.
Results
EphA2 is a promising CAR target for G3MB
We analyzed a previously curated microarray dataset that contains transcriptomic profiles of a total of 407 primary MB samples and 4 normal cerebellum controls compiled from multiple published sources.^13–15^ Consistent with previous findings,^12^ only WNT and G3MB subtypes had expression of EphA2 at higher levels than normal cerebellum (Fig. 1A), whereas B7-H3 expression was upregulated in all subgroups (Fig. 1B). We next inferred driver activity of EphA2 and B7-H3 for each subgroup, where positive activity scores (> 0) indicate activated regulatory programs.^16^ For WNT and G3MB, EphA2 driver activity was positive and significantly higher than normal cerebellum controls (Fig. 1C). B7-H3 driver activity was also significantly elevated compared to normal cerebellum controls for both WNT and G3MB, but B7-H3 regulatory programs were not activated in G3MB (Fig. 1D). We validated these findings by performing identical analyses in publicly available microarray^17^ and RNAseq^18^ datasets (Supplementary Fig. 1). Elevated expression and inferred driver activity for EphA2 in G3MB compared to B7-H3 and normal cerebellum suggests it is a biologically important, promising target for CAR T cell therapy.
To confirm that EphA2 and B7-H3 are expressed on the cell surface of G3MB tumor cell lines, we performed quantitative flow cytometric analysis on D341, HDMB03, and D425 MYC-driven G3MB patientderived cell lines (Fig. 1E, F). On a molecule per cell basis, B7-H3 was expressed at higher levels than EphA2 in 2 of 3 G3MB cell lines (HDMB03, D425) with the following hierarchy of EphA2 (D425 < D341<HDMB03) and B7-H3 (D341 < < HDMB03 < D425) expression.
EphA2-CAR T cells have superior performance in vitro against G3MB cell lines
To generate EphA2- and B7-H3-CAR T cells we used previously published CARs with a CD28ζ signaling domain, one of which (B7-H3-CAR) is being evaluated in combination with 41BBL costimulation in an early phase clinical study for pediatric patients with recurrent/refractory CNS tumors at our institution (Fig. 2A).^19–21^ Both CARs were highly expressed on the surface of primary human T cells with a slight increase in transduction efficiency of the B7-H3-CAR over EphA2 (Fig. 2B, Supplementary Fig. 2A). CD3/CD28-activated but non-transduced (NT) T cells were used as a control. Both CAR T cell products exhibited similar expansion post-transduction (Fig. 2C), and B7-H3-CAR T cells had a slight drop in viability 7 days post-transduction that was not observed with the EphA2-CAR (Fig. 2D). We next compared baseline phenotypes of EphA2- and B7-H3-CAR T cells via flow cytometric analysis. EphA2-CAR T cells slightly but significantly enriched for CD4^+^ subsets. (Fig. 2E, Supplementary Fig. 2B). In the CD4^+^ subset, EphA2-CAR T cells had a significantly higher proportion of effector memory and terminally differentiated effector (TEMRA) cells with a significant decrease in central memory cells. These phenotypic differences were also observed in the CD8^+^ subset but did not reach statistical significance (Fig. 2F, Supplementary Fig. 2C). This suggests that, at baseline, EphA2-CAR T cells are more differentiated towards an effector phenotype.
To evaluate cytolytic activity, EphA2-CAR T cells and B7-H3-CAR T cells were co-cultured with firefly luciferase (ffluc) expressing G3MB cell lines (D341, HDMB03, and D425) at an effector:target (E:T) ratio of 1:4 for 24 hours. Against all lines, EphA2-CAR T cells had increased cytolytic activity compared to B7-H3-CAR T cells (Fig. 2G) that reached statistical significance for D341 and HDMB03. This difference in cytolytic activity cannot be explained by target antigen density since only one cell line expressed lower levels of B7-H3 than EphA2 (Fig. 1D, E).
We next determined the ability of EphA2-CAR T cells to repeatedly kill tumor cells and expand using a repeat stimulation assay. Briefly, CAR T cells were challenged with fresh D341 (Fig. 2H), HDMB03 (Fig. 2I), and D425 (Fig. 2J) tumor cells at an E:T ratio of 1:2 in the presence of IL-15 every 7 days until the CAR T cells failed to eliminate the tumor cells or expand. Persistence is defined by the number of stimulations until failure up to 10 successful stimulations. Against all 3 cell lines, EphA2-CAR T cells had enhanced persistence that did not reach statistical significance (Fig. 2K).
Finally, we analyzed the cytokine production profiles of EphA2- and B7-H3-CAR T cells upon interaction with tumor cells. We first performed single-analyte ELISA analysis of co-culture supernatants from the first stimulation of the repeat stimulation assay described above for TH1-asociated cytokines IFNγ, IL-2, and GM-CSF (Fig. 2L–N). Across all cell lines and cytokines, EphA2-CAR T cells had significantly higher secretion. We further determined differences in other TH1-associted cytokines as well as other classes of cytokines (e.g. TH2, TH17), via multiplex cytokine analysis of co-culture supernatants from the first stimulation of EphA2- and B7-H3-CAR T cells against HDMB03 (Supplementary Fig. 3). Alongside TH1-associated cytokines, EphA2-CAR T cells also produced higher levels of some TH2 (IL-4, IL-10) and TH17 (IL-17A, IL-6) associated cytokines; however, this did not reach statistical significance (Supplementary Fig. 3G-S).
Together, these findings suggest that EphA2-CAR T cells exhibit potent in vitro antitumor effector function against G3MB, superior to a clinically relevant B7-H3 construct with matching stimulation domains.
Phenotypic diversity is associated with superior EphA2-CAR T cell in vitro responses
We next sought to determine EphA2-CAR T cell characteristics that correlate with in vitro potency. We first performed standard phenotypic analyses to evaluate differences in CAR T cell characteristics (CD4, CD8, effector/memory phenotypes (CCR7, CD45RA), and exhaustion) upon tumor cell exposure. We stimulated CAR T cells with G3MB cell lines at an E:T ratio of 1:2 in the presence of IL-15 and evaluated effector/memory phenotypes after the first stimulation and exhaustion after the third. After the first stimulation, the composition of B7-H3-CAR T cells changed significantly with a decrease in CD4^+^ cells, whereas CD4^+^ populations were maintained in EphA2-CAR T cells (Fig. 3A). Within the CD4^+^ subsets, EphA2-CAR T cells had lower proportions of TEMRA cells and higher proportions of effector memory cells which was significant after exposure to D341 and D425 (Fig. 3B). CD8^+^ EphA2-CAR T cells also had lower proportions of TEMRA against all G3MB tumor lines and higher effector memory which was significant against HDMB03 (Fig. 3C). Taking into consideration the differences in the proportions of CD4^+^ and CD8^+^ cells between B7-H3- and EphA2-CAR T cells, CD8^+^ TEMRA cells emerged as the dominant T cell subset for B7-H3-CAR T cells (Fig. 3D). In contrast, EphA2-CAR T cells maintained a high degree of phenotypic diversity, with consistent decreases in CD8^+^ TEMRA balanced by increases in CD8^+^ effector memory, CD4^+^ effector memory, and CD4^+^ central memory compared to B7-H3 CAR-T cells (Fig. 3D).
We next hypothesized that the improved persistence of EphA2-CAR T cells (Fig. 1H–K) might be due to reduced surface expression of co-inhibitory receptors, or exhaustion markers.^22, 23^ We analyzed the surface expression of PD-1, LAG-3, TIM-3, and TIGIT on EphA2- and B7-H3-CAR T cells at baseline (Fig. 3E). There were no significant differences in expression of exhaustion markers in the CD4^+^ subset. Surprisingly, EphA2-CAR T cells had significantly higher expression of PD-1, LAG-3, and TIM-3 in the CD8^+^ subset, which most likely is an indicator of activation given their improved effector function (Fig. 2G–N).^24^ After three stimulations with D341, HDMB03, and D425, we did not observe significant differences in the expression of co-inhibitory receptors between EphA2- and B7-H3-CAR T cells (Fig. 3F–H). Together, this data suggests that expression of surface markers canonically associated with exhaustion did not significantly differ between EphA2 and B7-H3-CAR T cells and was not driving functional differences observed in in vitro.
B7-H3-CAR T cells are susceptible to fratricide from B7-H3 expression on the T cell surface
Previous work from our group and others showed that B7-H3 is expressed upon CAR T cell activation and might render B7-H3-CAR T cells susceptible to fratricide.^25–27^ Indeed, we observed increased expression of B7-H3 on the cell surface of T cells when B7-H3- and EphA2-CAR T cells were cultured with D341, HDMB03, and D425 for 7 days (Fig. 4A–C). The percentage of CAR T cells expressing B7-H3 was lower for B7-H3-CAR T cells compared to EphA2-CAR T cells after 7 days, suggesting that B7-H3^+^ B7-H3-CAR T cells might be eliminated through fratricide.
We next evaluated the potential for fratricide amongst B7-H3-CAR T cells. As a model system, we activated EphA2-CAR T cells with recombinant human (rh) EphA2 protein and found that B7-H3 expression was induced after 24 hours (Supplementary Fig. 4). We stained non-activated or activated EphA2-CAR T cells with CellTracker^™^ Red CMTPX and co-cultured them with non-activated B7-H3-CAR T cells (stained with CellTrace^™^ Violet) and performed live-cell fluorescence imaging. DRAG7 fluorescent dye was added to the culture to indicate apoptosis (Fig. 4D). We observed that over 50% of interactions between activated EphA2-CAR T cells and B7-H3-CAR T cells resulted in EphA2-CAR T cell apoptosis. In contrast, apoptosis of non-activated EphA2-CAR T cells was only observed following 10% of interactions with B7-H3-CAR T cells (Fig. 4E).
To further validate our finding, we activated EphA2-CAR T cells for 24 hours with increasing doses of rhEphA2 and cultured them with B7-H3-CAR T cells at 1:1 ratio. After 24 hours of co-culture, we determined the proportion of live B7-H3:EphA2-CAR T cells by flow cytometric analysis (Supplementary Fig. 5). At baseline (0 ng rhEphA2), the live B7-H3:EphA2-CAR T cell ratio remained approximately 1:1. Activation with increasing doses of rhEphA2 resulted in a stepwise decrease in the proportion of live EphA2-CAR T cells (Fig. 4F). Further, the percentage of B7-H3^+^ EphA2-CAR T cells only increased post rhEphA2 activation in a dose-dependent manner if B7-H3-CAR T cells were not present (Fig. 4G). Thus, activation induces B7-H3 expression on the cell surface of CAR T cells, rendering them sensitive to killing by B7-H3-CAR T cells.
Thus, we sought to determine if deleting B7-H3 would increase CAR T cell efficacy. We optimized CRISPR/Cas9 knockout (KO) of CD276 (B7-H3) and used AAVS1-KO as a control (Ctrl) (Supplementary Fig. 6A). KO of B7-H3 did not impact expression of either CAR (Supplementary Fig. 6B). To confirm functional deletion of B7-H3, we evaluated B7-H3 surface expression at the end of CAR T cell production. B7-H3 was expressed by 6–10% of EphA2- and B7-H3 CAR T cells post Ctrl KO, and the percentage of B7-H3^+^ CAR T cells was significantly reduced post B7-H3 KO (Supplementary Fig. 6C).
To assess the functionality of B7-H3-KO CAR T cells, we first performed cytotoxicity assays against G3MB cell lines at a 1:4 E:T ratio. B7-H3-KO did not significantly alter the cytolytic activity for either B7-H3- or EphA2-CAR T cells (Fig. 4H). After 7 days of stimulation, CD4^+^ CAR T cell percentages were also not significantly altered for either CAR T cell population (Fig. 4I). Additionally, we did not observe significant changes in secretion of TH1-associated cytokines (IFNγ or IL-2) or cytotoxic molecule granzyme B (GZMB) (Fig. 4J–L). Thus, preventing B7-H3 expression by KO in B7-H3-CAR T cells post activation is not sufficient to improve their effector function, suggesting that fratricide alone cannot explain the observed differences in effector function between EphA2- and B7-H3-CAR T cells against G3MB.
EphA2-CAR T cells have potent in vivo antitumor efficacy against G3MB
To evaluate the efficacy of EphA2- and B7-H3-CAR T cells against G3MB in vivo xenograft models, G3MB cell lines were modified to express firefly luciferase for bioluminescence imaging. Either 100K (D341, HDMB03) or 50K (D425) tumor cells were implanted intracranially into the cortices of NSG mice. A single dose of 2×10^6^ CAR T cells was injected intracranially 7 (D341, D425) or 14 (HDMB03) days later after confirmation of tumor engraftment (Fig. 5A, E, I). For all three cell lines, 2 different biological donors (D1 and D2) of CAR T cells were tested.
EphA2-CAR T cells delayed the rate of D341 tumor growth (Fig. 5B–C), resulting in an increase in median overall survival (OS) from 20 days for non-treated to 39 days for EphA2-CAR T-cell-treated mice (Fig. 5D). B7-H3-CAR T cells were unable to control D341 tumors, and extended OS by 7 days. This resulted in significantly improved survival of mice treated with EphA2-CAR T cells in comparison to mice treated with B7-H3-CAR-T cells (Fig. 5D).
In HDMB03, the durability of EphA2-CAR T-cell-mediated tumor control was donor-dependent. For D1, EphA2-CAR T cells eradicated tumors in 4/4 mice, and no tumors relapsed (Fig. 5F), whereas D2 EphA2-CAR T cells initially eradicated 3/4 tumors, but all tumors relapsed (Fig. 5G). B7-H3-CAR T cells generated from these two donors were less variable, but ineffective in inducing long-term tumor control, resulting in a significant decrease in OS of mice treated with B7-H3-CAR T cells compared to mice treated with EphA2-CAR T cells (OS of 55 days vs. undefined at 150 days, respectively) (Fig. 5H).
Against D425, EphA2-CAR T cells generated from both donors had short-term antitumor activity (Fig. 5J–K) which still resulted in a significant survival advantage in comparison to non-treated mice (OS: 40 days vs. 20 days, respectively) (Fig. 5L). The antitumor activity of B7-H3-CAR T cells was donor-dependent with D2 having significant antitumor activity, which was superior in comparison to EphA2-CAR T cells. Combined, mice treated with B7-H3-CAR T cells had a slight survival advantage over those treated with EphA2-CAR T cells, but the difference did not reach statistical significance (MS: 55 days vs. 40 days, respectively) (Fig. 5L).
DNMT3A KO or constitutive IL-18 signaling improves EphA2-CAR T cell in vivo antitumor efficacy
Finally, we explored if the antitumor activity of EphA2-CAR T cells could be further improved by modifications that have been previously tested by our group: EphA2-CAR T cells with MyD88.CD40 costimulation instead of CD28 (EA2.MyD88),^28^ EphA2-CAR T cells with CD28 co-stimulation with KO of the negative regulator DNMT3A (EA2.D3A)^29^, or EphA2-CAR T cells with transgenic expression of leucine zipper-based chimeric cytokine receptors (ZipRs) with IL-2 (EA2.Zip2R)^30^ or IL-18 (EA2.Zip18R)^31^ receptor signaling domains. We generated a panel of modified EphA2-CAR T cells from two different healthy donors (D1 and D2), determined transduction efficiency (Supplementary Fig. 7A) and knockout efficiency (Supplementary Fig. 7B) when appropriate, and compared them head-to-head against D425 (Fig. 5I) using unmodified EphA2-CAR T cells (EA2.std) as a control.
Surprisingly, we found that mice treated with EA2.MyD88-CAR T cells experienced acute toxicity described by rapid weight loss within 1–2 weeks of CAR T cell treatment that was not observed for any other EphA2-CAR product (Supplementary Fig. 8). Within this time frame, antitumor activity was not improved compared to EA2.std (Fig. 6A–B), resulting in significantly worsened survival (OS 16 days vs. 41 days, respectively) (Fig. 6C). Of the other modifications tested, transgenic expression of Zip2R had no meaningful benefit (Fig. 6G–I), whereas DNMT3A-KO (Fig. 6D–F) or transgenic expression of Zip18R (Fig. 6J–L) significantly improved OS from 41 days to 58 and 62 days, respectively, in comparison to unmodified EphA2-CAR T cells. Notably, only EphA2-CAR T cells with Zip18R expression induced long-term remission (> 100 days) in 2 of 10 treated mice (Fig. 6L)
Discussion
Collectively, our results demonstrate that EphA2-CAR T cells have potent antitumor activity against G3MB cell lines in vitro and in vivo. EphA2 overexpression has been noted on several adult solid cancers of the prostate, lung, colon, cervix, ovaries, breast, and skin and pediatric osteosarcoma and Ewing’s sarcoma.^32^ EphA2 is also highly expressed by adult glioblastoma multiforme,^33–35^ and an ongoing clinical trial is assessing the therapeutic potential of synNotch receptor-induced anti-EphA2 and IL-13Ra2-CAR T cells for this patient population (NCT06186401). Our group and others previously showed EphA2 is also expressed on diffuse intrinsic pontine gliomas and ependymomas.^7, 36^ Thus, the development of effective EphA2-CAR T cells has the potential to transform not only outcomes for G3MB patients, but also many pediatric and adult solid and brain tumors.
Importantly, EphA2 is expressed on a number of healthy tissues, including low levels in the brain.^37^ While low levels of target antigen in the brain does not preclude their targeting via intracranial administration of CAR T cells, best exemplified by the use of GD2-CAR T cells for GD2-positive brain tumors,^3^ concerns regarding on target/off cancer toxicity remain. To reduce the risk of on target/off cancer toxicity for EphA2-targeted CAR T cell therapy, we have employed the 4H5 anti-EphA2 single chain variable fragment (scFv) as a CAR antigen binding domain since it specifically recognizes a conformational epitope of EphA2 that is unique to malignant cells and not when EphA2 is expressed at lower levels in healthy tissue.^38^ However, since 4H5 does not recognize murine EphA2, formal safety studies in immune competent models cannot be conducted. Nevertheless, EphA2-CAR T cells are being evaluated in early phase clinical studies, and data from three adult patients with high-grade glioma suggest a promising safety profile.^39^
Our results showed that EphA2-CAR T cells have superior in vitro antitumor efficacy compared to B7-H3-CAR T cells that is associated with lower proportions of terminally differentiated CD8^+^ cells and higher proportions of CD4^+^ effector cells after tumor cell exposure. Terminally differentiated CD8^+^ cells are associated with high, short-term cytotoxic potential, but not long-term efficacy.^40, 41^ The ability of EphA2-CAR T cells to resist rapid enrichment of this population is consistent with observed prolongation of efficacy in chronic stimulation assays. Broadly, CD4^+^ cells provide critical support to CD8^+^ cytotoxic cells,^42^ and maintenance of this population may be crucial for long-term antitumor efficacy.^43, 44^ Here, we observed overall maintenance of CD4^+^ proportions as well as significant enrichment of CD4^+^ effector cells which are poised to secrete cytokines and even participate in target cell elimination.^45^ We found that EphA2-CAR T cells had high secretion of TH1-associated cytokines IFNγ, IL-2, GM-CSF, and TNFα; but did not consistently upregulate TH2-associated cytokines IL-4, IL-5, and IL-13. In many cancers, tipping the endogenous immune balance towards TH1 enrichment has been associated with better prognosis due to direct impacts on tumor cells as well as activating other antitumor immune cells.^46^ This suggests that EphA2-CAR T cells may be better poised for both cytotoxicity and recruitment of other immune components to effectively eliminate tumors.
It is known that activation induces B7-H3 expression on T cells^26^ which may leave B7-H3-CAR T cells susceptible to fratricide and contribute to inferior in vitro responses compared to EphA2-CAR T cells. It is important to note that other clinically-relevant B7-H3 scFvs have shown greater antitumor efficacy than MGA271 that is related to the avidity of CAR T/tumor cell interactions.^47^ It is possible that B7-H3-CAR T cells with alternate scFvs differentially regulate B7-H3 expression, and they may have antitumor efficacy that is comparable to the EphA2-CAR presented here. However, KO of B7-H3 was not sufficient to improve acute CAR T cell functionality. While investigating the impact of scFv/protein avidity on B7-H3 expression is outside of the scope of this work, it could significantly add to general knowledge about CAR design in the context of self-antigen expression.
When we evaluated EphA2-CAR T cells in vivo, we found that they were superior to B7-H3-CAR T cells against D341 and HDMB03 but inferior against D425 G3MB tumor models. Our results could be easily explained by low EphA2 antigen density on D425, but we hypothesized that further optimization may overcome this limitation. We tested an alternate stimulation domain, MyD88.CD40, genetic deletion of a known negative regulator DNMT3A, and transgenic expression of constitutively active IL-2 and IL-18 Zip receptors. Of the modifications, deletion of DNMT3A and overexpression of IL-18 signaling successfully prolonged mouse survival. Importantly, activating the MyD88 pathway through Zip18R overexpression did not result in the lethal toxicities that were observed when MyD88 was built into the co-stimulatory domain of the CAR. To date, no clinical trials have been initiated to study safety or efficacy of DNMT3A KO in CAR T cells. However, IL-18 armored CAR T cells are being actively clinically investigated due to their ability to broadly improve CAR T cell therapy for numerous CAR targets and diseases.^48^ Thus, we propose that boosting IL-18 signaling in EphA2-CAR T cells may be a promising clinical approach for G3MB.
Overall, we report in vitro and in vivo characterization of EphA2 targeted CAR T cells for G3MB. Compared to a clinically relevant B7-H3-CAR, EphA2-CAR T cells have enhanced cytotoxicity, persistence, and cytokine secretion despite lower antigen density. We emphasize the importance of phenotypic diversity and resistance of self-expression of target antigens for CAR T cell efficacy. Our results support further development of EphA2-targeted CAR T cells with IL-18 signaling support for clinical testing against G3MB.
Material and Methods
EphA2 activity analysis
A previously curated microarray dataset (DKFZ_SJ), which aggregates transcriptomic profiles from multiple published sources, was utilized.^13–15^ This dataset comprises 407 primary medulloblastoma samples and 4 normal cerebellum controls. To reconstruct medulloblastoma-specific gene-gene interaction network (MBi), we applied the SJARACNe algorithm^49, 50^ to the expression profiles of the MB tumor samples. An initial list of 11,239 candidate regulators, comprising transcription factors (TFs) and signaling molecules (SIGs) based on Gene Ontology (GO) classifications, served as potential hub genes. Using SJARACNe, we generated separate TF and SIG interaction networks, where edges between hub genes and their downstream targets were inferred based on pairwise gene-gene mutual information. The final consolidated MBi comprised 38,621 probe nodes and 1,567,381 interactions, encompassing 12,495 hub probes (representing 7,155 unique genes) and their predicted target genes.
The functional activity of each hub gene defined in the MBi was inferred for individual microarray or RNA-seq samples using the NetBID2 software package.^16^ Specifically, we applied the cal.Activity function (method=‘weightedmean’), which calculates a sample-specific activity score for a hub gene based on the aggregated, network-informed expression of its predicted target genes. For a given hub gene (i) in sample (s), its activity (HUB_si_) was computed as the weighted mean expression of its n target genes (j = 1…n):
Where:
EXP_sj_ is the z-normalized expression value of target gene j in sample s.MI_ij_ is the mutual information between driver i and target gene j, representing the strength of their regulatory interaction, as inferred by SJARACNe.SIGN_ij_ is the sign (+ 1 or −1) of the Spearman correlation between driver i and target gene j, indicating an activating or repressive relationship.n is the total number of target genes for hub gene i.
The network parameters—MI_ij_, SIGN_ij_, and the list of targets for each hub gene—are intrinsic properties of the pre-computed MBi network generated by SJARACNe. Consequently, this method transforms a sample’s gene expression profile into a quantitative signature of inferred functional activity for each transcriptional and signaling regulator within the MBi network.
Cell lines
D341, HDMB03 and D425 G3MB cell lines were obtained from the laboratory of Dr. Martine F. Roussel at St. Jude Children’s Research Hospital. D341 cells^51^ were maintained as neurospheres in EMEM (ThermoFisher, CAT:21103049) supplemented with 20% FBS (Gibco, CAT:A56708–01) and 1% GlutaMAX (ThermoFisher, CAT:35050079) in NunclonSphera^™^ Flasks (FisherSci CAT:174952). HDMB03^52^ and D425^53^ cells were maintained in Neurobasal media (Gibco CAT:21103049) supplemented with 0.02% heparin (StemCell Technologies CAT:07980), 2% B27 (ThermoFisher CAT:12587010) and 1% GlutaMAX (Gibco, CAT:35050079) with 0.1 ug of each EGF (PeproTech, CAT:AF-100–15) and FGF (Peprotech, CAT:100–18B) added every 2–3 days.
Flow cytometry
All staining was performed on 0.3×106 cells in PBS supplemented with 2% FBS (Gibco, CAT:A56708–01) for 45 minutes in the dark at 4C. For live/dead cell discrimination, either Aqua Fluorescent Reactive Dye (Invitrogen, CAT: L34966A) or Near-IR Fluorescent Reactive Dye (Invitrogen, CAT:L34976A) was used at 0.2 μL/test. Antigen density on tumor cell lines was quantified using the Quantum^™^ MESF kit with anti-B7-H3 AF647 (R&D Systems, CAT:FAB1027R, 3 μL/test) and anti-EphA2 PE (BioLegend, CAT:356804, 5 μL/test). Transduction efficiency was evaluated using anti-G4S linker AF647 (Cell Signaling, CAT:69782S, 0.5 μL/test) or anti-CD19 PeCy7 (BioLegend, CAT:502538412, 3 μL/test) for the EphA2-CAR only. For immunophenotyping, T cells were stained with anti-CD4 BUV395 (BD Biosciences, CAT:563550, 3 μL/test) or anti-CD4 Pe-Cy7 (BD Biosciences, CAT:557852, 3 μL/test); anti-CD8 PerCP (BD Biosciences, CAT: 347314, 3 μL/test) or anti-CD8 FITC (BD Biosciences, CAT: 347313, 5 μL/test); anti-CCR7 FITC (BD Biosciences, CAT: 561271, 3 μL/test); anti-CD45RA APC (BioLegend, CAT: 304112, 3 μL/test). B7-H3 expression was assessed using anti-CD276 BV421 (BD Biosciences, CAT:565829, 3 μL/test). To distinguish the B7-H3 and EphA2-CAR T cell populations when mixed, B7-H3-CAR T cells were labelled with CellTrace Violet (Invitrogen, CAT:C34557A, 1 μL/mL for 2×106 cells), while EphA2-CAR T cells were identified using anti-CD19 PE-Cy7 (BioLegend, CAT:302216, 3 μL/test). Samples were acquired on a BD FACSCanto or BD LSRFortessa and files were analyzed using FlowJo version 10.10.0.
Generation of CAR T cells
B7-H3 and EphA2-CAR T cells: Retroviral particles containing B7-H3 and EphA2-CAR molecules were generated by transient transfection of 293Vec.RD114 cells as previously described.^19, 26^ Human peripheral blood mononuclear cells (PBMCs) were isolated from healthy donors. One day prior to thawing PBMCs, 24-well non-tissue culture treated plates were coated with anti-human CD3 (Miltenyi Biotec, CAT:130–093-387) and anti-human CD28 (Miltenyi Biotec, CAT:130–093-375) at 0.5 μg/mL in 500 μL of sterile water per well and incubated at 4C overnight. The next day, PBMCs were thawed and plated at a density of 1×10^6^ cells/well in 2 mL of complete RPMI. After 24 hours, 1 mL of RPMI was removed and replaced by 1 mL of complete RPMI supplemented with IL-7 (Peprotech, CAT:200–7) and IL-15 (Peprotech, CAT:200 − 15) for a final concentration of 10ng/mL and 5 ng/mL, respectively. On the same day, each well of a non-TC treated 24 well plate was coated with 10 μg of RetroNectin (Takara, CAT:T100B) in 500 μL PBS (Gibco, CAT:14190144) and incubated overnight at 4°C. The following day, RetroNectin was removed and 500 μL of appropriate viral supernatant was added to each RetroNectin coated well. Viral supernatants were centrifuged at 2,000xg for 90 minutes, After centrifugation, supernatants were removed and activated T cells were plated at a density of 2.5×10^5^ cells/well in 2 mL of complete RPMI supplemented with IL-7 and IL-15 at 10 ng/mL and 5 ng/mL, respectively. After 72 hours, CAR T cells were transferred to a clean, TC-treated 24 well plate and maintained in complete RPMI supplemented with IL-7 and IL-15 at 10 ng/mL and 5 ng/mL, respectively. CAR T cells were split 1:2 and media replenished every 2–3 days.
AAVS1 and B7-H3 knockout (KO) CAR T cells: For CRISPR/Cas9 KO of AAVS1 and B7-H3, RNP complexes were prepared separately as previously described (AAVS1 guide RNA sequence: GGGAACCCAGCGAGTGAAGA; B7-H3 guide 1 RNA sequence: TCTCCAGCACACGAAAGCCANGG; B7-H3 guide 2 RNA sequence: GTGCTGCAGCAGGATGCGCANGG).^29^ Briefly, 3 μL of 60 μM guide RNA stock (2 nmol guide RNA suspended in 13.5 μL TE buffer + 20 μL RNase-free water) was combined with 1 μL of 40 μM Cas9 (MacroLab, University of California, Berkeley) per desired electroporation reaction. RNP complexes were incubated at room temperature for 10 minutes prior to storage at −20C until use. For B7-H3 KO, 1×10^6^ activated T cells (activated 48 hours prior) were suspended in 14 μL of P3 Primary cell buffer (Lonza; V4XP-3032) and combined with 3 μL of B7-H3 RNP.1 (guide 1) and 3 μL of B7-H3 RNP.2 (guide 2) per reaction. For AAVS1 knockout, 1×10^6^ activated T cells (activated 48 hours prior) were suspended in 17 μL of P3 Primary cell buffer (Lonza; V4XP-3032) and combined with 3 μL of AAVS1 RNP per reaction. Each reaction was electroporated using the EH-115 program on a Lonza 4D-nucleofector. Immediately following electroporation, 80 μL of recovery media (RPMI complete at 20% FBS supplemented with IL-7 and IL-15 at 10ng/mL and 5 ng/mL, respectively) was gently added to each electroporation reaction. After 30 minutes, 100 μL of electroporated cells were transferred to 650 μL of recovery media in TC-treated 48-well plates. After 24 hours, electroporated T cells were transduced (two transduction wells for every one electroporation reaction).
DNMT3A KO CAR T cells: For CRISPR/Cas9 KO of DNMT3A, RNP complexes were prepared as previously described,^29^ and activated T cells were electroporated with DNMT3A (guide RNA sequence: CCTGCATGATGCGCGGCCCANGG, IDT) RNP complexes 48 hours activation. 1×10^6^ activated T cells were suspended in 17 μL of P3 buffer and electroporated with 3 μL of RNP per reaction a Lonza 4D-nucleofector. Immediately following electroporation, 80 μL of recovery media (RPMI complete at 20% FBS supplemented with IL-7 and IL-15 at 10ng/mL and 5 ng/mL, respectively) was gently added to each electroporation reaction. After 30 minutes, 100 μL of electroporated cells were transferred to 650 μL of recovery media in TC-treated 48-well plates. After 24 hours, electroporated T cells were transduced (two transduction wells for every one electroporation reaction).
Zip2R and Zip18R EphA2-CAR T cells: For double transduction with Zip2R and Zip18R, CAR T cells underwent sequential transduction. Briefly, Zip2R and Zip18R retroviral supernatants were generated by transient transfection of 293Vec.RD114 cells. For double transduction, the EphA2-CAR construct was transduced as described above. One day later, 1 mL of Zip2R or Zip18R retroviral supernatants were centrifuged onto RetroNectin coated plates, and EphA2 transduced T cells were transferred directly from EphA2 wells to Zip2R or Zip18R wells. Zip2R and Zip18R plasmids also contain mClover for detection by flow cytometry.
Cytotoxicity Assays
Firefly luciferase-expressing medulloblastoma cells (50K/well) were plated on Geltrex-coated (ThermoFisher, CAT:A1413302) 96-well TC-treated plates (white wall, clear bottom) in their respective culture media. CAR T cells, suspended in 100 μL complete RPMI, were then added at specified effectorto-target (T cell:tumor cell) ratios. After 24 hours, D-luciferin (Revvity, CAT:122799) was added to a final concentration of 1 μg/μL, and luminescence was measured using a TECAN plate reader. Wells containing media only were used to subtract background luminescence from all wells. Wells containing only tumor cells were arbitrarily set to 100% live tumor cells, and all wells containing CAR T cells + tumor cells were normalized appropriately to approximate % tumor cell killing.
Repeat stimulation assays
G3MB cells (5×10^5^) were plated on Geltrex-coated, TC-treated 24-well plates in 1 mL of respective media and allowed to adhere for 1 hour at 37°C. B7-H3 or EphA2-CAR T cells (2.5×10^5^) were then added to the tumor cells in 1 mL of complete RPMI supplemented with IL-15 (5 ng/mL final concentration in 2 mL). After seven days, T cells were counted manually using a hemocytometer and replated with fresh tumor cells at the same E:T ratio with IL-15. The assay was repeated every 7 days until CAR T cell failed to eliminate > 80% of tumor cells or failed to expand past the plating density of 2.5×10^5^.
Cytokine secretion analysis
IFNγ, IL-2, GM-CSF, and granzyme B (GZMB) were analyzed using either standard single-cytokine analysis assays (R&D Systems IFNG:SIF50 CAT: IL-2: CAT:S2050 GM-CSF: CAT:DGM00) or the automated ELLA immunoassay platform, with the Simple Plex Cell Activation Panel 1 (Bio-Techne, CAT:SPCKC-CS-003222). Multiplex cytokine quantification was performed using the HCYTA-60K-PX48 MILLIPLEX kit (Millipore Sigma). Analysis was performed using a Luminex FlexMap 3D instrument and Belysa software (Millipore Sigma).
Live cell imaging
One day prior to EphA2 CAR T cell activation, 24-well non TC-coated plates were coated with rhEphA2 at 880 ng/well in 500 μL of PBS (Sigma-Aldrich, CAT:P8920) and incubated at 4C overnight. EphA2 CAR T cells were plated at 2×10^6^ cells/well in 2 mL of complete RPMI. After 24 hours of activation, EphA2 CAR T cells were labeled with CellTracker^™^ Red CMTPX at 1:1000 (ThermoFisher, CAT:C34552T) and non-activated B7-H3 CAR T cells were labeled with CellTrace^™^ Violet at 1:1000 (Invitrogen, CAT:C34557A) in PBS (for 2×10^6^ cells/mL), for 30 minutes at 37°C. Labeled 200K EphA2 and 200K B7-H3-CAR T cells were combined in μ-slide 8 well slides (Ibidi, CAT:80826). To assess apoptosis, 10 μL of the fluorescent apoptosis marker DRAG7 (ThermoFisher, CAT: D15106) was added to each chamber. Imaging was performed every 10 minutes for 24 hours using a widefield microscope (Nikon Eclipse Ti2 inverted microscope) with a 20x objective.
Images were imported into ImageJ (FIJI) for analysis. Trackmate^54^ with default settings (LoG=10μm, max linking-distance=20μm, Gap = 2 frames) was used to track target cells (Epha2-CAR T cells) over time. The resulting matrix was imported into Rstudio (Version 4.5) where a custom-made script was used to filter target and effector interacting cells, synchronize tracks, normalize, and quantify effective killing events.
In vivo models
NOD scid gamma (NSG) mice were obtained from breeding colonies maintained by the St. Jude Animal Resource Center. All animal experiments were performed on protocols approved by the St. Jude Institutional Animal Care and Use Committee (623–100650). Mice were 8–10 weeks old at the time of tumor implantation. For D341 and HDMB03, 100K cells labeled with firefly luciferase were implanted into the cortex at a depth of 2 mm. For D425, 50K firefly luciferase labeled cells were implanted into the cortex. The tumor cells were suspended in an equal mix of PBS and Matrigel (Corning, CAT:356234), and 5 μL of cell suspension was injected per mouse. Tumor cell growth was tracked via bioluminescence imaging. Briefly, mice were injected intraperitoneally with 15 mg/mL D-Luciferin (Revvity, CAT:122799) in PBS at a dose of 150 mg/kg (10 μL/g). After luciferin administration, mice were anesthetized and imaged 7 min after luciferin injection using the In vivo Imaging System (IVIS). Bioluminescence was quantified using Living Image software (Living Image). Bioluminescence imaging continued weekly until endpoint, which was defined by loss of weight (20% from baseline), neurological symptoms, or other reduction in health (lethargic, hunched, scruffy, paralysis, etc) according to IACUC protocols. Humane tumor flux endpoint was defined as 1E10 photons/second.
For each model, CAR T cells were generated from two different healthy PBMC donors. Bioluminescence images were used to redistribute mice before CAR T cell injection to ensure equal tumor burden across conditions. A single dose of 2E6 CAR T cells suspended in 3 μL of PBS was injected intracranially on day 7 for D341 and D425; and day 14 for HDMB03 at a depth of 2 mm.
Statistical Analysis
Statistical analyses were performed using GraphPad Prism 10.6 (GraphPad Software Inc, La Jolla, CA). Statistical analyses were performed only when the number of replicates was at least three. Biological replicates were generated by performing identical experiments with multiple T cell donors. For all analyses, Shapiro Wilk test was performed to check normality assumption, and Bartlett’s test or Brown-Forsyth tests were used to determine equal variance where appropriate. For comparisons of two groups, two-tailed paired or unpaired T tests were used. For multiple paired T tests, a false discovery rate (FDR) of 5% was used. For comparisons of 2 or more groups, one or two-way ANOVA tests were used with Dunnett’s, Sidak’s, or Tukey’s multiple comparison tests. Survival data were analyzed using the log-rank Mantel Cox test. For all experiments, p-values of less than 0.05 were considered significant. All details of assumption verification and statistical tests performed for each figure panel are detailed in Supplementary Table 1.
Supplementary Material
Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Kulubya ES, Kercher MJ, Phillips HW, Antony R, Edwards MSB (2022) Advances in the Treatment of Pediatric Brain Tumors. Children (Basel). ;10(1). Epub 20221227. 10.3390/children 10010062. · doi ↗
- 2Shi X, Liu S, Sun X, Li Y, Jiang M (2025) Treatment Outcomes and Prognostic Factors for Patients With Medulloblastoma Having Defined Molecular Subtypes. Adv Radiat Oncol 10(9):101796 Epub 20250627. 10.1016/j.adro.2025.10179640837387 PMC 12361598 · doi ↗ · pubmed ↗
- 3Monje M, Mahdi J, Majzner R, Yeom KW, Schultz LM, Richards RM, Barsan V, Song K-W, Kamens J, Baggott C, Kunicki M, Rietberg SP, Lim AS, Reschke A, Mavroukakis S, Egeler E, Moon J, Patel S, Chinnasamy H, Erickson C, Jacobs A, Duh AK, Tunuguntla R, Klysz DD, Fowler C, Green S, Beebe B, Carr C, Fujimoto M, Brown AK, Petersen A-LG, Mc Intyre C, Siddiqui A, Lepori-Bui N, Villar K, Pham K, Bove R, Musa E, Reynolds WD, Kuo A, Prabhu S, Rasmussen L, Cornell TT, Partap S, Fisher PG, Campen CJ, Grant G, Prolo L, Ye X, Sahaf B, Davis KL, Feldman SA, Ramakris · doi ↗ · pubmed ↗
- 4Vitanza NA, Ronsley R, Choe M, Seidel K, Huang W, Rawlings-Rhea SD, Beam M, Steinmetzer L, Wilson AL, Brown C, Beebe A, Lindgren C, Gustafson JA, Wein A, Holtzclaw S, Hoeppner C, Goldstein HE, Browd SR, Hauptman JS, Lee A, Ojemann JG, Crotty EE, Leary SES, Perez FA, Wright JN, Alonso MM, Dun MD, Foster JB, Hurst D, Kong A, Thomsen A, Orentas RJ, Albert CM, Pinto N, Annesley C, Gardner RA, Ho O, Pattabhi S, Gust J, Wendler JP, Park JR, Jensen MC (2025) Intracerebroventricular B 7-H 3-targeting CAR T cells for diffuse intrinsic pontine glio · doi ↗ · pubmed ↗
- 5Eksteen C, Riedemann J, Johnson H, Engelbrecht AM (2025) Literature review: CAR T-cell therapy as a promising immunotherapeutic approach for medulloblastoma. Rev Neurol 181(8):703–712. 10.1016/j.neurol.2025.07.00540764212 · doi ↗ · pubmed ↗
- 6Okada R, Fanuzzi M, Rodriguez C, Oh J, Sreenivasan A, Stack HG, Puebla M, Phadke I, Cole AP, Bradford S, Kelly MC, Lei H, Ho M, Cotter JA, Thiele CJ, Zhang X, Nellan A, Nguyen R (2025) GPC 2-CAR T cells have potent preclinical activity against orthotopic medulloblastoma xenografts. Mol Ther Oncol 33(4):201067 Epub 20250930. 10.1016/j.omton.2025.20106741159102 PMC 12554907 · doi ↗ · pubmed ↗
- 7Haydar D, Houke H, Chiang J, Yi Z, OdéZ, Caldwell K, Zhu X, Mercer KS, Stripay JL, Shaw TI, Vogel P, De Renzo C, Baker SJ, Roussel MF, Gottschalk S, Krenciute G (2021) Cell-surface antigen profiling of pediatric brain tumors: B 7-H 3 is consistently expressed and can be targeted via local or systemic CAR T-cell delivery. Neuro Oncol 23(6):999–1011. 10.1093/neuonc/noaa 27833320196 PMC 8168826 · doi ↗ · pubmed ↗
- 8Majzner RG, Theruvath JL, Nellan A, Heitzeneder S, Cui Y, Mount CW, Rietberg SP, Linde MH, Xu P, Rota C, Sotillo E, Labanieh L, Lee DW, Orentas RJ, Dimitrov DS, Zhu Z, Croix BS, Delaidelli A, Sekunova A, Bonvini E, Mitra SS, Quezado MM, Majeti R, Monje M, Sorensen PHB, Maris JM, Mackall CL (2019) CAR T Cells Targeting B 7-H 3, a Pan-Cancer Antigen, Demonstrate Potent Preclinical Activity Against Pediatric Solid Tumors and Brain Tumors. Clin Cancer Res 25(8):2560–2574 Epub 20190117. 10.1158/1078-0432.Ccr-18-043230655315 PMC 8 · doi ↗ · pubmed ↗