Diffuse intrinsic pontine glioma (DIPG) and other histone H3 K27M (H3K27M) mutated diffuse midline gliomas (DMGs) 1–5 are aggressive and universally fatal pediatric brain cancers 6 . Chimeric antigen receptor (CAR)-expressing T-cells have mediated impressive clinical activity in B-cell malignancies 7–10 , and recent results suggest benefit in CNS malignancies 11–13 . Here, we report that patient-derived H3K27M-mutant glioma cell cultures exhibit uniform, high expression of the disialoganglioside GD2. Anti-GD2 CAR T-cells incorporating a 4-1BBz costimulatory domain 14 demonstrated robust antigen-dependent cytokine generation and DMG cell killing in vitro. In five independent patient-derived H3K27M+ DMG orthotopic xenograft models, systemic administration of GD2-CAR T-cells cleared engrafted tumors, save a small number of residual GD2-low glioma cells. To date, GD2-targeting CAR T-cells have been well-tolerated in clinical trials 15–17 . While GD2-CAR T-cell administration was tolerated in the majority of animals, peri-tumoral neuroinflammation during the acute phase of antitumor activity resulted in hydrocephalus that was lethal in a fraction of animals. Given the precarious neuroanatomical location of midline gliomas, careful monitoring and aggressive neurointensive care management will be required for human translation. With a cautious multidisciplinary clinical approach, GD2-CAR T-cell therapy for H3K27M+ diffuse gliomas of the pons, thalamus and spinal cord could prove transformative for these lethal childhood cancers. Immune checkpoint inhibitors mediate substantial benefit in adult cancers refractory to traditional therapies, but have not yet demonstrated widespread benefit in sporadic childhood cancers, possibly due to the paucity of neoantigens in these diseases 18 . In contrast, CAR T cell therapies have mediated highly potent effects in childhood B-cell acute lymphoblastic leukemia (B-ALL). To identify potential targets for CAR T-cell immunotherapy in DIPG, we screened cell surface antigens using an antibody array in patient-derived DIPG cultures (Fig. 1a, Supplementary Table 1). Significant overlap between independent patient-derived cultures (Fig. 1b), suggests conservation of a core group of surface markers across DIPG patients. From these common targets, we observed that the disialoganglioside GD2 was expressed at high levels on each of the four patient-derived DIPG cultures screened (Fig. 1a). Hit validation by flow cytometry in six H3K27M+ DIPG cultures confirmed uniform, high GD2 expression in all H3K27M+ DIPG cultures examined, including those with the H3F3A K27M mutation (SU-DIPG-6, 13, 17, 25, 29) and the less-common HIST1H3B K27M mutation (SU-DIPG-21) 2,19,20 (Fig. 1c). GD2 expression was far lower in two histone-3 WT pediatric high-grade gliomas (pHGG), including a case of H3WT DIPG (Fig. 1c). To assess whether transcriptional perturbations resulting from the H3K27M mutation might be linked to GD2 overexpression, we profiled gene expression of ganglioside synthesis enzymes in patient-derived DIPG and pHGG cultures and found higher expression of upstream ganglioside synthesis enzymes in H3K27M+ cultures (Supplementary Figure 1). Double immunostaining of primary human DIPG tissue for H3K27M to identify infiltrating malignant cells and GD2 confirmed local expression of GD2 in the native tumor context (Fig. 1d). GD2-targeting immunotherapies are currently under clinical and preclinical investigation in several diseases, including neuroblastoma, osteosarcoma, and melanoma 14–17,21–24 . Unlike monoclonal antibodies which do not efficiently cross the blood-brain barrier, activated T-cells can infiltrate the CNS following adoptive transfer 7,25 . We generated human GD2-targeting CAR T-cells incorporating a 4-1BBz costimulatory domain (GD2-CAR) 14 (Fig. 1e) and observed significant GD2-dependent killing (Fig. 1f) and cytokine generation (Fig. 1g) upon exposure to patient-derived DIPG cultures relative to control CD19-CAR T-cells incorporating 4-1BBz (CD19-CAR). Notably, GD2-CAR T-cells do not produce significant cytokines or induce cell killing when exposed to the H3WT, GD2-negative VUMC-DIPG10 patient-derived DIPG culture, providing evidence of therapeutic specificity of GD2-CAR T-cells toward H3K27M DIPG. To further confirm the targeting specificity of GD2-CAR T-cells, we used CRISPR-Cas9-mediated deletion of GD2 synthase (B4GALNT1) in patient-derived DIPG cells to generate GD2 knockout DIPG cells (Supplementary Figure 2). Loss of GD2 antigen expression eliminated cytokine production by the GD2-CAR T-cells in comparison to untreated or DIPG cells electroporated with a control guide sequence targeting the AAVS1 locus (Fig. 1h). While GD2 expression in normal brain has been previously reported 26 , GD2-CAR T-cell therapy in human trials in which the targeting domain is derived from the same monoclonal antibody (14g2a) have reported no cases of significant neurotoxicity 15–17 . Moreover, intrathecal and intraventricular administration of anti-GD2 antibody radioconjugates in patients has been well-tolerated in clinical trials 27,28 . Taken together, these findings indicate specific reactivity of GD2-CAR T-cells to H3K27M+ glioma cells. To evaluate in vivo efficacy of GD2-CAR T-cells against DIPG, we prepared orthotopic mouse xenografts of DIPG cultures derived from post-mortem patient tissue. DIPG cultures were transduced with a luciferase-expressing construct to enable longitudinal monitoring of tumor burden. These xenograft models faithfully recapitulate the diffusely infiltrating histology of DIPG 29,30 . Mice were distributed by tumor burden into equivalent treatment and control groups before receiving 1×107 GD2-CAR or CD19-CAR T-cells by a single intravenous injection 7-8 weeks after establishment of pontine xenografts. Within 40 days post-treatment (DPT), marked reductions in tumor burden were observed across two independent GD2-CAR T-cell treated cohorts of mice bearing SU-DIPG6 xenografts 31 (Fig. 2a). Similar results were observed in a second patient-derived xenograft model, SU-DIPG13FL 30 (Fig. 2e). All GD2-CAR treated animals demonstrated complete tumor clearance by bioluminescence imaging (Supplementary Figure 3). By contrast, no mice in the CD19-CAR T-cell control groups exhibited significant tumor regression. At 50 DPT brains were harvested, and immunostaining for the mutant histone H3K27M – present in all engrafted tumor cells – revealed that GD2-CAR treated tumors had been largely eradicated (Fig. 2c,d,g,h,i). The small number of H3K27M+ tumor cells that remain after treatment are negative for GD2 by immunostaining (Supplementary Figure 4). We hypothesize that the potency of the GD2-CAR in this model is driven by very high expression of the target antigen in H3K27M mutant DIPG, which was consistently higher than that present on GD2+ neuroblastoma and sarcoma cell lines (Supplementary Figure 5). This study adds to the growing evidence suggesting that CAR T-cells are more efficacious at high levels of target antigen expression 32,33 . Most patient-derived orthotopic DIPG xenograft models require many months for lethality, limiting the ability to monitor survival benefit due to development of xenogeneic graft versus host disease (GVHD) after treatment with human T-cells 34 . We therefore used SU-DIPG-13P*, a model that exhibits a dense pattern of growth histologically 35 , and is consistently lethal within one month. Substantial improvement in survival was seen in GD2-CAR treated animals compared with CD19-CAR treated controls (Fig. 3a). However, in one out of three independent cohorts, lethal toxicity occurred in several GD2-CAR T-cell treated animals, while all GD2-CAR T-cell treated animals in the other cohorts survived to endpoint (Supplementary Figure 6). GD2-CAR treated animals that survived the initial phase of glioma clearance returned to a visibly healthy state indistinguishable from untreated immunodeficient mice until the onset of GVHD symptoms 4+ weeks after CAR administration that invariably triggered endpoint criteria (Supplementary Figure 7). Histologic analysis of the brains of endpoint GD2-CAR treated animals reveals clearance of this high-burden tumor while surrounding neural tissues appear grossly normal (Fig. 3b). To better understand the etiology of treatment-related toxicity in these DIPG xenograft models, we examined the brains of treated SU-DIPG6 xenograft-bearing mice acutely at DPT14 (Fig. 3c). GD2-CAR treatment was accompanied by a widespread inflammatory infiltrate involving brain parenchyma, meninges and ventricles that was most prominent in the brainstem. Ventriculomegaly was observed, consistent with hydrocephalus. We observed histologically normal-appearing neurons present throughout the pons, hippocampus, and cortex of GD2-CAR T-cell-treated animals with no evidence of neuronal cell killing nor other tissue destruction in this model (Fig. 3c). Thus, neuropathological evaluation indicates that the toxicity described above results from brainstem inflammation and hydrocephalus due to fourth ventricular compression during the tumor-clearing interval and not on-target, off-tumor toxicity of GD2-CAR T-cells. To visualize CAR T-cell infiltration into the parenchyma and tumor, we generated GD2-4-1BBz-mCherry and CD19-4-1BBz-mCherry fusion constructs (Fig. 3d). By DPT7, GD2-CAR T-cells are extensively distributed throughout the leptomeninges of treated animals, leptomeningeal tumor has been largely eradicated, and few mCherry+ cells are present within the brain parenchyma (Fig. 3h, Supplemental Figure 8). By DPT14, mCherry+ GD2-CAR T-cells had widely infiltrated throughout the parenchyma and numerous foci of Iba1+ macrophages (Fig. 3e) are present in the xenografted site, along with extensive apoptotic cleaved caspase 3+ cells (Fig. 3f). Notably, very few cleaved caspase 3+ apoptotic cells are neurons as identified by NeuN double immunostaining (10 total apoptotic neurons identified across 4 mice; Fig. 3g, Supplementary Figure 9). By DPT21, mCherry+ GD2-CAR T-cells remain present throughout the CNS; whereas few CD19-CAR T-cells infiltrate the parenchyma (Supplementary Figure 8). This supports a model in which intravenously administered GD2-CAR T-cells enter through the meningeal lymphatic system 36 , then subsequently infiltrate brain parenchyma, although the mechanism of CAR T-cell trafficking to the tumor remains to be defined. Given that resolution of tumor clearance and ventriculomegaly temporally coincide in treated animals, it is likely that antigen-specific antitumor activity, rather than on-target, off-tumor cell killing, precipitates neuroinflammation and edema during active tumoricidal activity that results in hydrocephalus. Understanding systemic and microenvironmental mechanisms that contribute to the resolution of acute CAR T-induced neuroinflammation is a critical area of future study that may be best addressed in immunocompetent models. Recent WHO criteria place DIPG within a larger classification of diffuse midline gliomas (DMG) expressing the H3K27M mutation (Fig. 4a) 37 . In patient-derived cultures of pediatric H3K27M thalamic (QCTB-R059, derived from resection) 4 , and spinal cord (SU-pSCG1, derived post-mortem) DMGs, GD2 is also highly and uniformly expressed (Fig. 4b) and triggers IFNγ and IL-2 production by GD2-CAR T-cells (Fig. 4c). While these data might reasonably predict similar efficacy of GD2-CAR T-cells in these midline H3K27M gliomas, we reasoned that the neuroanatomical site of disease could impact outcomes of CAR T-cell therapy. To explore in vivo GD2-CAR T-cell efficacy in these H3K27M DMGs, we generated patient-derived orthotopic xenograft models of spinal cord (SU-pSCG1) and thalamic (QCTB-R059) glioma. When engrafted in the medulla to avoid the paralysis induced by injection into the spinal cord, widespread SU-pSCG1 growth is observed throughout the CNS (Fig. 4d). Systemic administration of GD2-CAR T-cells achieves potent and lasting tumor-clearing in this spinal cord glioma xenograft model, assessed both by longitudinal bioluminescence imaging (Fig. 4d,e) and endpoint histology, where approximately 16 residual H3K27M+ cells per animal remained across the sampled volume of 3 GD2-CAR T-cell treated animals (Fig. 4f,g). No mice from this cohort died during the tumor-clearing phase. To evaluate efficacy in H3K27M thalamic glioma, we engrafted QCTB-R059 cells orthotopically in the thalamus (Fig. 4h). Tumor clearance was observed in this model (Fig. 4h,i, Supplementary Figure 10) on a similar time scale as observed for DIPG and spinal cord tumors discussed above. However substantial toxicity occurred in GD2-CAR T-cell treated animals during the period of maximal therapeutic effect (Fig. 4k). The results are reminiscent of “pseudoprogression”, well described following immunotherapy with checkpoint inhibitors 38 , and highlight the danger of a robust immunotherapeutic response and subsequent neuroinflammation in neuroanatomical locations intolerant of swelling. The thalamus – located just above the cerebellum tentorial notch – is a precarious location for edema, particularly when already expanded by tumor, and swelling in this location can precipitate hydrocephalus from third ventricular compression, increased intracranial pressure and lethal transtentorial herniation (Fig. 4j). Meticulous clinical monitoring and neurointensive management of edema will be required for successful clinical translation of this powerful immunotherapy, particularly for thalamic tumors, but may be insufficient to mitigate the peril of neuroinflammation at midline brain locations susceptible to herniation. Although stark tumor clearance occurs in these xenograft models, the persistence of small numbers of tumor cells that are negative for GD2 expression by immunofluorescence staining suggests that multimodal therapy could be required to circumvent potential antigen escape (Supplementary Figure 4). While GD2 antibody-associated neuropathy has been observed in clinical trials, it is critical to note that in GD2-CAR T-cell human trials to date in which the targeting domain is derived from the same antibody as used here, treatment-associated neuropathy has not been observed 15–17 . The effects of robust T-cell infiltration and potent killing with associated inflammation within the tumor site represents a CNS tumor-specific element of the larger category of possible neurotoxic complications of CAR T-cell therapy, distinct from the observed CAR T-cell therapy-associated encephalopathy syndrome (CRES) that occurs in the context of cytokine release and appears to result from endothelial dysfunction 39 . The importance of careful monitoring for hydrocephalus and signs of increased intracranial pressure cannot be overemphasized and will require careful inpatient monitoring with frequent neurological and fundoscopic exams, and neuroimaging as indicated. Neurosurgical interventions, such as intraventricular shunt placement for relief of hydrocephalus or even craniectomy for decompression, may be required to support children through the phase of tumoricidal neuroinflammation. In summary, we have identified GD2 as a novel immunotherapy target in H3K27M mutant DMGs, and demonstrated potent antitumor efficacy of GD2-CAR T-cells delivered systemically in five independent patient-derived orthotopic xenograft models representing three distinct H3K27M mutant glioma subtypes of the pons, thalamus and spinal cord. Upregulation of ganglioside synthesis pathway component genes in H3K27M mutant pediatric high-grade gliomas relative to H3 WT pediatric high-grade gliomas suggests this mutation drives GD2 overexpression. Tumor associated inflammation generated by CAR T-cell activity is tolerated by the majority of mice in this setting, suggesting the possibility for a therapeutic window with neurointensive management despite the precarious anatomic sites of these midline tumors, but putative tolerability of this approach remains to be demonstrated in children with H3K27M+ DMG. If the results in mouse models are predictive of results in humans, this immunotherapeutic strategy could be transformative for the outcomes of children with H3K27M-mutant diffuse midline gliomas. Methods DMG cultures Patient-derived glioma cell cultures were generated as previously described 40 . Briefly, tumor tissue was dissociated mechanically and enzymatically (Liberase DH, Roche) prior to separation of myelin and debris by sucrose centrifugation. Neurosphere-generating cultures were maintained in serum-free media supplemented with B27 (ThermoFisher), EGF, FGF, PDGF-AA, PDGF-BB (Shenandoah Biotechnology), and Heparin (StemCell Technologies). All cultures were validated and monitored by STR-fingerprinting (Supplementary Table 2) and verified to be mycoplasma-free within the previous 6 months (MycoAlert Plus, Lonza). SU-DIPG6 and SU-DIPG13 have been previously referred to as and are identical to SU-DIPG-VI and SU-DIPG-XIII, respectively. Clinical characteristics and STR fingerprints of all DIPG and pSCG cultures 30 along with QCTB-R059 4 used here have been previously reported. For all studies using human tissue, informed consent was obtained per guidelines of the approved Stanford Institutional Review Board protocol. Cell surface screening Cell surface markers present on DIPG cell cultures were screened using a panel of monoclonal antibodies against human cell surface markers (Lyoplate, BD Biosciences). Low passage ( 60%) were then administered by intravenous injection into the tail vein of animals. Where transduction efficiencies varied between GD2-CARs and CD19-CARs, the concentration of CD19-CAR cells was adjusted to match the total dose of human T-cells present in the GD2-CAR infusion. Tumor burden was monitored longitudinally by in vivo luminescence imaging. Due to the obvious nature of GD2-CAR response, blinding in initial cohorts was deemed ineffective and subsequently not performed. All images were scaled to display minimum flux intensity as 5E4 and maximum as 5E6, then images of individual animals were arranged with like-treated animals in the cohort for display in figures. Trial endpoint at 50 days post treatment was determined in initial cohorts where substantial hair loss, reduced activity, and weight loss in both GD2-CAR and CD19-CAR groups that triggered morbidity criteria for euthanasia. CRISPR/Cas9-mediated deletion of GD2 synthase Deletion of GD2 synthase (B4GALNT1) in SU-DIPG13 cells was accomplished by electroporation of DIPG13 with Cas9:sgRNA ribonucleoprotein complexes as previously described 45 . Briefly, guide RNAs targeting exon1 of B4GALNT1 (CGUCCCGGGUGCUCGCGUAC and CCGGCUACCUCUUGCGCCGU, Synthego) were incubated with Cas9 nuclease to form ribonucleoprotein complexes and electroporated with an Amaxa 4-D nucleofector (SE Buffer, program DS-112). In parallel, a control gRNA targeting the AAVS1 locus 46 (GGGGCCACUAGGGACAGGAU) was electroporated with Cas9 nuclease as a ribonucleoprotein complex using identical parameters. GD2-negative cells electroporated with B4GALNT1-targeting gRNAs were isolated by FACS sorting, and deletion was confirmed by Sanger sequencing and TIDE analysis 47 . Statistics and Reproducibility Statistical tests were conducted using Prism (GraphPad) software unless otherwise indicated. Gaussian distribution was confirmed by the Shapiro–Wilk normality test. For parametric data, unpaired, two-tailed Student’s t-tests and one-way ANOVA with Tukey’s post hoc tests to further examine pairwise differences were used. For survival analysis, a log-rank (Mantel-Cox) test was used. A level of P < 0.05 was used to designate significant differences. On the basis of the variance of xenograft growth in control mice, we used at least 3 mice per treatment group to give 80% power to detect an effect size of 20% with a significance level of 0.05. For all animal experiments, the number of independent mice used is listed in figure legend. For each of the five patient-derived xenograft models used, at least two independent cohorts were tested (i.e. independent litters of mice on different days with independent batches of cells.) For cytokine and in vitro cell killing experiments, n=3 and experiments were repeated twice. Data Availability Statement The datasets generated during the current study are available from the corresponding author upon reasonable request. Supplementary Material 1 2
