Abstract Glioblastoma (GBM) remains uniformly lethal, driven by profound intratumoral heterogeneity, and an immunosuppressive tumor microenvironment (TME) dominated by tumor-associated macrophages (TAMs). The identification of novel therapeutic targets is crucial for advancing promising therapies that address both tumor and TME. Here, we identify glycoprotein non-metastatic melanoma protein B (GPNMB) as a clinically relevant surface marker expressed on both GBM cells and TAMs. Genetic loss of GPNMB in GBM models reduced tumor cell proliferation, delayed intracranial tumor growth, and altered transcriptional programs linked to immune regulation and leukocyte activation, indicating GPNMB sustains both tumor fitness and an immunosuppressive niche. We generated both anti-human and anti-mouse GPNMB CAR-T cells showing potent activity: In orthotopic patient-derived xenografts, GPNMB CAR-T therapy induced complete tumor remission; murinized GPNMB CAR-Ts similarly produced sustained tumor control in syngeneic GBM models. In a humanized mouse model of recurrent GBM, anti-human GPNMB CAR-T therapy drove tumor regression in most animals despite a highly suppressive myeloid-rich TME. An integrated central nervous system myeloid single-cell RNA-sequencing atlas revealed GPNMB+ macrophages are enriched for lipid transport, and GPNMB expression correlates with immunosuppressive gene signatures. Consistent with this, GPNMB was preferentially expressed by immunosuppressive macrophages, and GPNMB CAR-Ts selectively eliminated immunosuppressive over pro-inflammatory macrophages in vitro and in vivo. Collectively, these data establish GPNMB CAR-T therapy as an extremely promising therapy inducing complete GBM remission in PDXs and immunocompetent models by simultaneously targeting GPNMB+ tumor cells and immunosuppressive macrophages. Citation Format: Sheila Kumari Singh, Neil Savage, Muhammad Vaseem Shaikh, Shan Grewal, Franz J. Zemp, Nick Mikolajewicz, Joanna Pyczek, Hinda Najem, Jeffrey Wei, Shawn C. Chafe, Kui Zhai, William Maich, CHIRAYU CHOKSHI, Nazanin Tatari, Dillon McKenna, Mohamed Taleb, Lucas Asselstine, Hong Han, Kevin Brown, Chitra Venugopal, Thomas Kislinger, Amy B. Heimberger, Jennifer A. Chan, Jason Moffat, Douglas J. Mahoney. GPNMB CAR-T cells target both glioblastoma and the tumor microenvironment to relieve immune suppression [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4044.

Supplementary Fig. 1. Single-cell RNA sequencing of human meningiomas with loss of chromosome 22q. A, Single-cell counts across UMAP clusters from human meningioma samples with loss of chromosome 22q analyzed using single-cell RNA sequencing. B, UMAP showing single-cell RNA sequencing of human meningioma samples shaded by sample of origin. C, Stacked bar plots showing the distribution of chromosome 22q loss (red) across single-cell RNA sequencing clusters from human meningioma samples. D, UMAP and stacked bar plot showing single-cell RNA sequencing cell cycle analysis across human meningioma samples. E-G, Heatmap and feature plots showing differentially expressed genes across UMAP clusters from human meningioma samples analyzed using single-cell RNA sequencing. Colors as in A. See also Supplementary Table 1. G, Feature plots showing expression of meningioma (SSTR2, PTGDS) or cancer stem-cell markers (NOTCH3, THY1) across UMAP clusters from human meningioma samples analyzed using single-cell RNA sequencing.

Glioblastoma, IDH-wildtype, is characterized by diffuse invasion, profound immunosuppression, treatment resistance, and near-inevitable recurrence. Although canonical genetic alterations establish malignant capacity, they do not fully explain how glioblastoma cells adapt to hypoxic, perivascular, invasive, immunosuppressive, metabolically constrained, and treatment-injured microenvironments. We examined how chemokine signaling contributes to these adaptive behaviors. We used the hallmarks of cancer as an organizing framework to synthesize preclinical, translational, and clinical evidence on chemokine circuits in glioblastoma. We evaluated recurrent mechanistic pathways, distinguished causal functions from context-dependent biomarker associations, and assessed their therapeutic relevance. Chemokines act primarily as spatial and stress-responsive regulators rather than initiating oncogenic drivers. Recurrent circuits include CXCL12–CXCR4 in vascular repair, invasion, and stem-like persistence; CCL2 and CCL7 signaling through CCR2 in suppressive myeloid recruitment and metabolic–immune remodeling; CCL5–CCR5 in perivascular protection, invasion, and DNA-damage tolerance; and CXCL8 signaling through CXCR1 and CXCR2 in angiogenesis, immune evasion, and therapy-induced plasticity. Most chemokine-directed strategies remain preclinical or early translational. Therapeutic development should prioritize biomarker-defined dependencies and appropriately timed combinations that disrupt selected chemokine-dependent interactions within specific biological and treatment contexts. Clinical translation will require verification of target engagement, disruption of the relevant cellular interactions, and evidence that chemokine modulation improves treatment response or delays recurrence.

Glioblastoma is a lethal brain tumour for which current multimodal treatment rarely prevents recurrence1. Therapeutic failure is driven by extensive intratumoural cellular heterogeneity2 with a microenvironment dominated by tumour-associated macrophages that sustain tumour growth and immunosuppression3. Although chimeric antigen receptor (CAR)-T cell therapies are being developed for glioblastoma, sustained response has been undermined by non-uniform antigen expression, antigen loss and microenvironmental barriers that are not directly engaged by tumour-targeting designs4. These limitations motivate new strategies that address the disease as a coupled tumour-immune system rather than a single malignant compartment. Here we use a multi-omic target discovery platform to identify GPNMB as a dual-compartment antigen in glioblastoma. Anti-GPNMB CAR-T cells showed potent anti-tumour activity, with long-term disease control in orthotopic patient-derived xenografts and syngeneic glioma models through concomitant depletion of GPNMB+ tumour and immunosuppressive myeloid populations. By collapsing tumour control and microenvironmental reprogramming, these findings provide a new strategy for antigen selection and targeting in heterogenous, myeloid-rich solid cancers.

Oligodendroglioma is a primary central nervous system tumor classified by the presence of isocitrate dehydrogenase (IDH) mutations and codeletion of 1p/19q. Here we describe the generation of an IDH-mutant 1p/19q-codeleted oligodendroglioma mouse model using in utero electroporation. We identified IDH1R132H, PIK3CAE545K, Cic KO, Fubp1 KO and Cdkn2a KO as the optimal combination (termed OligoCdkn2a) to drive fully penetrant tumors that histologically resemble human grade II/III IDH-mutant, 1p/19q-codeleted oligodendroglioma. Replacing Cdkn2a with Trp53 loss in this mouse model shifted tumor histology towards high grade astrocytoma. OligoCdkn2a tumors displayed metabolic and transcriptional changes associated with IDH and CIC mutations, and single cell sequencing identified a bias towards oligodendrocyte differentiation compared to an IDH wild-type glioblastoma mouse model. OligoCdkn2a tumors represent the first mouse model system to recapitulate the genetic, histological and transcriptional features of human IDH-mutant 1p/19q-codeleted oligodendrogliomas, offering a platform to further dissect tumor biology and test new therapeutic strategies.

Supplementary Fig. 16. NOTCH3 inhibition blocks meningioma xenograft growth without causing toxicity. A, A NOTCH1 negative regulatory region neutralizing antibody (αNRR1), but not αNRR3, causes diarrhea and rash resulting in weight loss in mice harboring meningioma xenografts. Antibodies were delivered using biweekly IP injection. Student’s t tests. B, The -secretase inhibitor LY-411575 (GSI) delivered using daily IP injection causes diarrhea and rash resulting in weight loss in mice harboring meningioma xenografts. Student’s t tests. C, LY-411575 attenuates the growth of CH-175MN (left) or IOMM-Lee (right) meningioma xenografts in mice. Student’s t tests. Lines represent means and error bars represent standard error of means.

Supplementary Fig. 2. Single-cell RNA sequencing of dog meningiomas. A, Single-cell counts across UMAP clusters from dog meningioma samples analyzed using single-cell RNA sequencing. B, UMAP showing single-cell RNA sequencing of dog meningioma samples shaded by sample of origin. C, UMAP showing single-cell RNA sequencing cell cycle analysis across dog meningioma samples. D-E, Heatmapt and feature plots showing differentially expressed genes across UMAP clusters from dog meningioma samples analyzed using single-cell RNA sequencing. Colors as in A. F, Magnetic resonance imaging (MRI, left) or H&E images (right) of spontaneous dog meningiomas. Scale bars, 100µm. See also Supplementary Table 2.

Background Chordomas are locally aggressive notochordal tumors with no systemic therapy options. As an ultra-rare cancer type, our understanding of its immune landscape is limited. While tumor-associated macrophages (TAMs) and T cells are critical components of the immune landscape, their functional states and interactions remain poorly understood.Methods We conducted an integrative analysis of 35 chordoma samples and six paired tumor-PBMC samples using single-cell RNA sequencing (scRNA-seq), T-cell receptor (TCR) profiling, and multiplex immunofluorescence. Immune cell phenotypes, spatial distribution, TCR motif diversity, and functional states were assessed using unbiased co-expression network analysis and predictive modeling.Results Chordomas exhibited remarkable immune cell heterogeneity, ranging from highly infiltrated to immune-desert tumors. Tumor-associated macrophages dominated the tumor microenvironment (TME) and were enriched for antigen-processing pathways. T-cell receptor profiling revealed clonal overlap between tumor-infiltrating and peripheral T cells, suggesting systemic anti-tumor responses. Exhausted CD8+ T cells exhibited restricted clonality and tumor-specific amino acid motifs. Weighted gene co-expression network analysis (WGCNA) identified gene modules associated with immune activation and suppression, underscoring the dual roles of immune cells in the TME. Spatial analysis revealed fibrous septa as immune interaction hubs, where immune cell clustering was significantly higher than in tumor regions.Conclusions This study advances understanding of the chordoma immune landscape by integrating spatial, transcriptomic, and TCR data. The findings highlight systemic and local immune dynamics, reveal tumor-specific TCR motifs, and identify potential therapeutic targets. These insights provide a foundation for developing personalized immunotherapies to overcome immune suppression and enhance anti-tumor immunity in chordomas.

The glioblastoma tumor immune microenvironment (TIME) is an immunosuppressive barrier to therapy that encumbers glioblastoma responses to immune checkpoint inhibition (ICI). Immunosuppressive cytokines, pro-tumor myeloid cells, and exhausted T-cells are hallmarks of the glioblastoma TIME. Here we integrate spatial and single-cell analyses of patient-matched human glioblastoma samples before and after ICI with genetic, immunologic, single-cell, and pharmacologic studies in preclinical models to reveal that interleukin-6 (IL-6) inhibition reprograms the glioblastoma TIME to sensitize mouse glioblastoma to ICI and radiotherapy. Rare human glioblastoma patients who achieve clinical responses to ICI have lower pre-treatment IL-6 levels compared to glioblastomas who do not respond to ICI. Immune stimulatory gene therapy suppresses IL-6 tumor levels in preclinical murine models of glioblastoma. Furthermore, survival was longer in Il-6 knockout mice with orthotopic SB28 glioblastoma relative to wild-type mice. IL-6 blockade with a neutralizing antibody transiently sensitizes mouse glioblastoma to anti-PD-1 by increasing MHCII+ monocytes, CD103+ migratory dendritic cells (DCs), CD11b+ conventional DCs, and effector CD8+ T cells, and decreasing immunosuppressive Tregs. To translate these findings to a combination treatment strategy for recurrent glioblastoma patients, we show that IL-6 blockade plus ICI durably sensitizes mouse glioblastoma to high-dose radiotherapy.

Supplementary Fig. 7. WHO grade 1 meningioma immunofluorescence microscopy for NOTCH3 and the endothelial cell marker VWF. NOTCH3 expression is restricted to the perivascular niche adjacent to endothelial cells in meningiomas with WHO grade 1 histology. SF numbers indicate individual meningiomas. Representative of n=10 WHO grade 1 meningiomas. WHO grade defined using histological criteria. Scale bars, 10µm.

OBJECTIVE:Ventricular entry (VE) during resection can maximize high-grade glioma (HGG) resection, but it remains unclear whether tumor contiguity to the ventricles or VE increases the risk of leptomeningeal metastasis (LM) and/or worsens overall survival (OS). METHODS:To clarify the role of VE and tumor location in LM incidence and OS, the authors retrospectively reviewed the charts of patients who underwent their first resection of supratentorial HGG at The University of Texas MD Anderson Cancer Center between 1993 and 2021. OS and time to LM diagnosis were estimated using the Kaplan-Meier method; their associations with patient and treatment variables, including the tumor proximity to the ventricle, were assessed via Cox regression analysis. RESULTS:The authors identified 884 patients: 390 (44%) had VE and 444 (50%) had ependymal contact (EC) tumors. Eighty-two percent of patients with VE had EC, while only 25% of those without VE had EC (p < 0.0001). On multivariate analysis, VE did not significantly predict LM (hazard ratio [HR] [95% CI] 1.32 [0.57-3.04], p = 0.520) or OS (HR 1.03 [0.87-1.22], p = 0.744). However, EC significantly increased LM risk (HR 3.97 [1.43-11.01], p = 0.008) and worsened OS (HR 1.33 [1.1-1.6], p = 0.003). Although patients with VE had an overall lower complete resection rate compared to those without VE (63% vs 72%, p = 0.005), VE improved the extent of resection among EC tumors with 58% having complete resection (vs 47% of EC tumors without VE). CONCLUSIONS:Tumors with EC predict higher LM risk and shorter OS, while VE during resection does not increase LM risk or worsen OS. Surgeons can use VE to maximize resection of supratentorial HGGs without increasing the risk of subsequent LM.

Meningiomas are common tumors of the central nervous system that are typically treated with surgery or radiation, but lack established systemic therapies. Activation of the stimulator of interferon genes pathway with an agonist such as 8803 can trigger anti-tumor immune responses. Using integrated molecular approaches, here we show that this pathway is targetable in both neoplastic and immune populations within the meningioma microenvironment. Meningioma tumor cells exhibit promoter hypomethylation and increased chromatin accessibility of the STING genomic locus, associated with robust expression of this gene. Treatment of diverse patient meningiomas ex vivo with 8803 induces direct tumor cytotoxicity through inflammatory cell death pathways, including induction of gasdermin D membrane pore formation. Release of necrotic tumor debris triggered by 8803 activates macrophages and upregulates matrix metalloproteinase production, facilitating degradation of extra-cellular collagen. Injection of preclinical meningiomas with 8803 induces survival benefits, including in an immunocompetent orthotopic setting, through remodeling of the tumor microenvironment, immune infiltration, and downregulation of tumor-mediated immune suppression, thereby nominating 8803 for treatment consideration in meningiomas.

OBJECTIVE:Stereotactic brain biopsy (SBB) is a widely used and generally safe diagnostic procedure. However, the utility of routine postoperative CT scans to screen for hemorrhage remains controversial, and reported postbiopsy hemorrhage rates vary widely (1%-60%). This study aimed to identify factors associated with postbiopsy hemorrhage and determine whether a selective, symptom-driven imaging strategy could safely replace routine imaging. METHODS:The data of 751 patients who underwent 753 SBBs between 1993 and 2021, all of whom received a postoperative CT within 48 hours, were retrospectively reviewed. The presence of hemorrhage of any size, the onset of new or worsening neurological symptoms within 30 days, and relevant clinical and radiographic characteristics were recorded. Neurological symptoms were categorized as early (present by the time of the first postoperative CT study) or delayed (developed after the initial CT study). Clinically significant hemorrhage was defined as bleeding on CT that prompted a change in management directly attributable to the hemorrhage. Associations between variables and hemorrhage were assessed using logistic regression and chi-square analysis. RESULTS:Blood was detected on postoperative CT imaging in 316 (42%) biopsies, most commonly at the biopsy site or along the trajectory (97%). On multivariable analysis, early postoperative symptoms (OR 3.82, 95% CI 1.34-10.9; p = 0.012), detecting blood through the biopsy needle intraoperatively (OR 2.88, 95% CI 1.37-6.06; p = 0.005), preoperative intralesional hemorrhage (OR 31.4, 95% CI 1.67-592; p = 0.021), and a platelet count of > 100 to 150 × 109/L (OR 1.7, 95% CI 1.00-2.89; p = 0.050) were associated with blood on the CT study. Patients with platelet counts ≤ 100 × 109/L conferred a fourfold increased risk that did not reach significance. New or worsening neurological symptoms were detected in 161 (21%) cases. Cases with altered mental status post-SBB were more likely to have blood on CT (69% vs 31%, p = 0.009). Four cases required intervention (2 hemorrhage evacuations, 2 ventriculostomies). The positive predictive value of postoperative CT in detecting a new or expanding hemorrhage was 17%, and the negative predictive value was 98%. Postoperative CT findings altered management in 5% of cases, predominantly in symptomatic cases. CONCLUSIONS:Routine postoperative CT after SBB may not be warranted in all patients. A symptom-driven imaging approach may reduce healthcare costs and unnecessary radiation exposure without compromising patient safety. The authors recommend selective imaging in patients with bleeding diathesis, intraoperative bleeding, existing intralesional hemorrhage, and/or new or worsening neurological symptoms. The final decision to perform a CT study is left to the treating physician's discretion.

Supplementary Fig. 14. Meningioma invasion of Virchow-Robin spaces. A, H&E low magnification (top) and high magnification (box, bottom) images of human meningioma invasion into perivascular fluid-filled cavities surrounding perforating vasculature of the brain. Densely cellular meningioma is shown at the top of each image, and islands of perivascular tumor within the brain parenchyma (dashed lines, bottom) are shown at the bottom of each image. Scale bars, 100µm. B, Low magnification (top) and high magnification (box, bottom) images of IHC for the brain parenchyma marker GFAP validates invasion of unlabeled meningioma cells into Virchow-Robin spaces without direct invasion of the brain parenchyma itself. C, IHC for the mural cell marker SMA (top) or the meningioma cell marker SSTR2A (bottom) validates meningioma invasion into Virchow-Robin spaces.

Supplementary Table S1. Cell cluster marker genes from single-cell RNA sequencing of human meningiomas with loss of chromosome 22q. Supplementary Table S2. Cell cluster marker genes from single-cell RNA sequencing of dog meningiomas. Supplementary Table S3. RNA sequencing differential expression analyses across recurrent versus primary human meningiomas. Supplementary Table S4. Antibodies for multiplexed sequential immunofluorescence. Supplementary Table S5. Cell cluster marker genes from single-cell RNA sequencing of human cells from meningioma xenografts. Supplementary Table S6. Cell cluster marker genes from single-cell RNA sequencing of mouse cells from meningioma xenografts.

Supplementary Fig. 9. UMAP markers for CyTOF gating of T cells from intracranial mouse glioblastoma allografts after IL6 blockade and immune checkpoint inhibition. a, UMAP cell density plots showing protein expression used for assigning cell types in Fig. 7d.

Supplementary Fig. 12. Body weights for immunocompetent mice treated with radiotherapy. a, Body weights for C57J/B6 WT mice harboring intracranial GL261 (left) or SB28 (right) intracranial allografts treated with radiotherapy (RT, 18Gy/1Fx) vs no-treatment control (n=10 mice/condition). See also Fig. 7a.

Supplementary Fig. 6. WHO grade 3 meningioma immunofluorescence microscopy for NOTCH3 and the mural cell marker SMA. NOTCH3 is expressed in and out of the perivascular niche, and colocalizes with mural cells in the perivascular niche, in meningiomas with WHO grade 3 histology. SF numbers indicate individual meningiomas. Representative of n=10 WHO grade 3 meningiomas. WHO grade defined using histological criteria. Scale bars, 10µm.