Novel GABA_AR antagonists target networked gene hubs at the leading edge in high-grade gliomas

Abstract

Background

Ion channel activity underlying biological processes that drive high-grade gliomas (HGG) is largely unknown. We aimed to determine the networking of ion channel genes and validate their expression within HGG patient tumors, to identify ion channel-targeting drugs that would inhibit tumor-promoting processes.

Methods

We used weighted gene co-expression network analysis (WGCNA) of RNAseq data to identify ion channel gene hubs in diffuse midline glioma (DMG) and glioblastoma. Using scRNA-seq, spatial transcriptomics, and immunohistochemistry, we characterized the expression of identified hubs within patient tumors, validating their role by testing the efficacy of ion channel inhibitors alone or in combination with radiation and temozolomide on the growth and invasion of patient-derived glioblastoma explant organoids (GBOs).

Results

Network analysis revealed a preserved HGG “neuronal regulation” module, containing the greatest number of ion channels, with its corresponding genes concentrated at the tumor's leading edge. Hubs within this module included γ-Aminobutyric-acid type A receptor (GABA_AR) genes GABRA1 (α1) and GABRG2 (γ2), which immunohistochemically colocalized with GABAergic synaptic markers at the leading edge. GBOs failed to retain this synaptic architecture but expressed a glioblastoma hub GABRA5 (α5), a component of extrasynaptic GABA_ARs. S44819, an α5-GABA_AR antagonist strongly inhibited GBO invasion, with GABA(A)-compound 1b, a partial antagonist of GABA_ARs, robustly inhibiting GBO proliferation and invasion. Moreover, combined with standard-of-care (SOC) regimens, the anti-invasive properties of both compounds were enhanced in GBOs.

Conclusions

Our co-expression network analysis identified key ion channels at the leading edge in HGGs, which can be targeted by GABA_AR-acting drugs to disrupt tumor progression.

Key Points.

• A neuronal regulation module, enriched with ion channel and synaptic genes, is embedded in the leading edge of HGG tumors.

• GABRA1 and GABRG2 are top hubs and are expressed in neurogliomal GABAergic synapses.

• Antagonism of GABA_ARs in GBOs inhibits their growth and invasion.

Importance of the Study.

High-grade gliomas (HGGs) exploit ion channels to hijack neurodevelopmental mechanisms for progression, but their role in HGG network topology is unclear. Using weighted gene co-expression network analysis of RNA sequencing data from diffuse midline glioma (DMG) and glioblastoma patients, we identified ion channel hub genes within a preserved “neuronal regulation” module with synaptic protein transcripts concentrated at the leading edge of DMG and glioblastoma tumors. GABA_AR subunit genes, GABRA1 and GABRG2, were top hubs, with GABRA5 prominent in glioblastoma. Immunohistochemistry confirmed that GABRA1/GABRG2 protein products appose GABAergic synaptic markers in patient tumors. In patient-derived glioblastoma organoids, GABA_AR-antagonists blocked proliferation and invasion, with α5βγ2 receptor antagonism disproportionately suppressing invasion. Standard treatment enhanced the anti-invasive effects of GABA_AR inhibitors. Thus, GABA_AR-inhibiting drugs, by potentially disrupting neuron-glioma connections and blocking GABA-mediated oncogenic signaling pathways, are promising HGG adjuvant therapeutics.

High-grade gliomas (HGGs) are lethal brain and spinal cord tumors affecting both adults and children.¹ Glioblastoma, a highly aggressive HGG in adults, has a 95% 5-year fatality rate, while diffuse midline glioma (DMG), a pediatric HGG, is uniformly fatal.¹ Although HGGs share features with other advanced tumors, they uniquely hijack neurobiological systems to progress. Therefore, pharmacological targets for HGG must align with the tumor’s specific gene regulatory networks. Ion channel proteins meet this criterion and are targeted by ~20% of approved drugs, with many passing the blood–brain barrier (BBB).² These proteins selectively pass metal ions,³ generating electrical currents that mediate HGG-specific disease mechanisms, including neuron-glioma synaptic communication⁴ , and other cancer hallmarks.⁵

Recent studies demonstrate the importance of synaptic physiology in HGG etiology. Glutamatergic neurogliomal synapses support a highly connected microenvironment populated by synaptogenic HGG cells.⁶ In glioma mouse models, neuronal activity initiates⁷ and drives⁸ glioma growth through secreted factors like BDNF and NLGN3 in DMG,⁹ in part by promoting the formation of glutamatergic neurogliomal synapses,⁸ whose activity stimulates Ca²⁺-permeable AMPA receptors, leading to HGG cell membrane depolarization and increased proliferation and invasiveness.⁸ Moreover, in DMG and glioblastoma patients, highly cellular tumors enriched with neural epigenetic signatures upregulate synapse-associated proteins, further promoting neurogliomal synaptic connectivity, tumor progression and mortality.¹⁰

Glioma membrane depolarization alone can drive tumor growth,¹¹ revealing the pathological importance of bioelectric signaling in HGG. However, identifying therapeutic ion channel targets is challenging given the > 500 ion channel transcripts expressed in the brain.³ Network Science is a tool that enables a systems-level approach to identify disruptions in complex regulatory networks within the tumor microenvironment that contribute to progression.^12,13 Glioblastoma-centric network studies have traditionally relied on whole genome expression data,¹⁴ solely differentially expressed,^15,16 or highly variable genes^17–20 biasing the results to a specific subset of highly correlated genes with variable expression. Consequently, this can exclude key components with a low or stable expression that may underlie crucial bioelectric and neuropathological processes.

Here, we employed weighted gene co-expression network analysis (WGCNA) to identify key ion channels driving HGG progression. Across HGG our analysis revealed a preserved “neuronal regulation” module comprising the most ion channels, with GABA_AR genes as central hubs. These ligand-gated Cl^- permeable ion channels, crucial for inhibitory synaptic neurotransmission, are primarily located in the central nervous system (CNS) but are also fundamental to systemic organ function²¹ and disrupted GABA_AR activity is associated with non-CNS cancers.²¹ Considering the pathological role of neuron-glioma synaptic communication in HGG, we hypothesized that GABA_AR signaling is critical to its regulatory networks and GABA_AR-modulating drugs would disrupt its progression. To validate our network analysis, we used single-cell and spatially resolved transcriptomics and immunohistochemistry in patient tumors. We administered GABA_AR-targeting drugs on patient-derived glioblastoma explant organoids (GBOs)²² alone or in combination with chemotherapy and radiation, to assess their anti-cancer efficacy on growth and invasion. Overall, our comprehensive approach demonstrates the effectiveness of Network Science in revealing new treatment avenues and GABA_ARs as promising therapeutic targets for HGG.

Materials and Methods

Ethics Statement

Studies using patient tissues were approved by (i) the Royal Adelaide Hospital and Southern Adelaide Clinical Human Research Ethics Committees (HREC/17/RAH/358, HREC/18/SAC/16, HREC/17/RAH/372 and SAC (286.10)), (ii) the University of South Australia Human Research Ethics Committee (App ID 202306), and (iii) the Government of Western Australia Child and Adolescent Health Human Research Ethics Committee (RGS0000005103, RGS0000005298). Participants gave informed consent.

Bulk and scRNAseq Datasets

Primary adult glioblastoma multiform RNA-sequencing (RNA-seq) data (n = 151) were accessed using TCGAbiolinks,²³ summarized as gene-level counts from The Cancer Genome Atlas (TCGA). RNA-seq data for pediatric primary HGG tumor samples (n = 75; NBS-HGG n = 44, DMG n = 31) was obtained from the St. Jude Cloud^19,24 (https://www.stjude.cloud) and pre-processed as detailed in the Supplementary Material. Metadata for pediatric HGG is detailed in Supplementary Table 2. The scRNA-seq data used in this study are included in our previously published dataset25 and those available through the Gene Expression Omnibus (GEO) repository (accession numbers: GSE173280,²⁶ GSE182109,²⁷ GSE141383,²⁸ GSE184357,²⁹ and GSE102130³⁰). Spatial transcriptomics data from DMG samples were obtained from Ren et al. (2023)³¹ (GSE194329). scRNAseq from Ebert et al. (2020)²⁵ is available upon request from the authors to G.A.G. and spatial transcriptomic glioblastoma data are available upon request from the authors from the European Genome-Phenome (EGA) archive. All R and MATLAB scripts for single cell and spatial transcriptomics data analysis are available at: https://github.com/chloe-shard/Shard_and_Jones_J_Neurooncol_2025.

Quality control (QC)

Sequencing datasets were analyzed in the statistical computing environment R (version 4.1.1). Genes with an official HUGO Gene Nomenclature Committee symbol were retained. Exploratory data analysis was carried out with EDASeq and data was adjusted for unwanted variation using RUVSeq.³² The data was filtered to retain genes with at least 10 reads total across samples and normalized (log₂ transformation) using the function varianceStabilizingTransformation from DESeq2. Boxplots were used to visualize the log₂ gene expression of GABA_AR channels.

Network Analysis

Prior to network analysis, unwanted variation in normalized data was removed using removeBatchEffect from limma, and lowly expressed genes were removed using DCGL package in R,³³ resulting in 6005 and 6596 genes in DMG/NBS-HGG and glioblastoma, respectively. Network analysis was performed using weighted gene co-expression network analysis (WGCNA), employing the following parameters (DMG, NBS-HGG, glioblastoma): signed network, softPower = 7/10, minModuleSize = 50, deepsplit = 0. Computation of module preservation statistics to identify module preservation, network reconstruction, and functional enrichment are detailed in the Supplementary Material.

Spatial Transcriptomics

Glioblastoma patient tissue specimens were obtained from four patients through the South Australian Neurological Tumor Bank (SANTB). Spatial RNA sequence analysis was performed on 5 µm thick FFPE sections using the Visium CytAssist Spatial Gene Expression kits for FFPE, Human Transcriptome (11 × 11 mm, 10X Genomics) following Visium CytAssist Spatial Gene Expression User Guide (CG000495, Rev C, 10x Genomics). Libraries were prepared and pooled to a final library concentration of 1.8 nM. The samples were loaded on a NovaSeq 6000 System (Illumina) using NovaSeq 6000 SP Reagent kit (200 cycles, 20040326, Illumina) and sequenced at a depth of approximately 150 M reads per sample. The read protocol was set as read 1 at 28 cycles, i7 index read at 10 cycles, i5 index read at 10 cycles, and read 2 at 94 cycles. FASTQ files were processed using Space Ranger 1.0.0 using GRCh38-3.0.0 as reference, and outputs were imported and processed into Seurat v5 (https://satijalab.org/seurat/articles/spatial_vignette). The count data were log normalized with a scale factor of 10,000. Publicly available spatial transcriptomic data of 5 DMG tissues,³¹ was processed using the same pre-processing workflow and normalization.

GBO Proliferation and Invasion Assays

GBOs derived from three patients (Supplementary Table 10) were generated and cultured as described.²² Protocols assessing proliferation and invasion in the presence of 50 µM GABA(A)-Compound 1b, 10 µM S44819 (also named Afizagabar), or 0.1% DMSO (vehicle), alone, or with SOC conditions, are detailed in the Supplementary Methods. Our screen for the cytotoxic efficacy of 65 small-molecule oncology compounds on GBO viability showed they have an effective IC₅₀ of 50 µM.²² Owing to the solubility properties of S44819 in DMSO, the maximum working concentration attainable for this was 10 µM. Proliferation was assessed by monitoring the relative change (from day zero) of GBO 2D-projected area (µm²) during a 20-day drug incubation, testing the effects of drug withdrawal at day six. Invasion assays were performed in parallel: on day six, GBOs were transferred in Matrigel and monitored for peak radial displacement (µm) relative to “day zero in Matrigel” measurements, with or without drug incubation, for a further 14 days. To test the efficacy of GABA_AR antagonism with the SOC Stupp protocol,³⁴ GBOs were treated with 50 µM temozolomide (TMZ) and radiation (10 Gy), with GABA_AR-drugs/0.1% DMSO for 10 days,²² where SOC ceased. Proliferation on day 10 was measured as above. Invasion was assessed after an additional GBO transfer step into Matrigel at day 10, with DMSO or GABA_AR-drug treatment continued for six days. Peak radial displacement relative to “day zero in Matrigel” measurements was calculated. Five to eight GBOs per patient were used for each treatment.

Bioinformatics Analysis

Detailed scRNA-seq and spatial transcriptomics bioinformatic analysis are included in the Supplementary Information.

Immunohistochemistry and Automated Image Analysis

Glioblastoma and DMG patient tissues for immunohistochemistry analysis were obtained from the Sydney Brain Tumour Bank, Australia (SBTB) and the Queensland Children's Tumour Bank, Australia (QCTB), respectively. Immunohistochemistry was performed using standard techniques (Supplementary Tables 3–4) detailed in the Supplementary Methods. CellProfiler automated image analysis of confocal microscope images was used to independently estimate marker co-expression. Methodology to determine the relative abundance of protein co-expression is described in the Supplementary Methods.

Statistical Analysis

The Shapiro-Wilk test was used to determine data normality and Grubbs’ test to detect outliers. Depending on normality, statistical analyses (GraphPad Prism 9.0) used were parametric (t-test; one-way ANOVA Tukey’s multiple comparisons test, or two-way ANOVA with Tukey’s multiple comparisons test) or non-parametric (Mann–Whitney) tests with P<.05 considered significant.

Results

WGCNA Analysis Reveals Modules Associated with Neuronal Regulation and ECM in HGG that Contain Co-expressed Ion Channels

To identify the gene network architecture in pediatric DMG, NBS-HGG, and adult glioblastoma tumors, we constructed independent co-expression networks for each cancer using WGCNA. The network analysis revealed the underlying modular architecture of HGGs (Figure 1A,B; Supplementary Figure 1A), clustering into nine, nine, and eight highly co-expressed modules, respectively, for DMG, NBS-HGG, and glioblastoma (Supplementary Tables 5-7).

Figure 1.

Network analysis of DMG and glioblastoma identifies neuronal regulation associated network module that is expressed at the tumor’s leading edge. A, B, Network analysis was carried out independently for DMG and glioblastoma using WGCNA. The modular architecture was visualized employing a topological overlap matrix (i.e., heatmap and dendrogram) in A, DMG, and B, glioblastoma coexpression networks. Color bands signify identified functional modules (rows and columns) of highly connected genes. Red intensity across the diagonal signifies a high correlation. Black boxes represent the identified modules in DMG (nine modules) and glioblastoma networks (eight modules). C, D, Heatmaps representing the relative expression (min-max normalized by cell type) of neuronal regulation genes (columns) across cell types in C, two DMG and D, four glioblastoma scRNA-seq datasets. E, F, The spatial expression pattern of four tumor regions defined by Puchalski et al., (2018) in a representative E, DMG and F, glioblastoma patient tumor alongside the neuronal regulation module gene signature in the same G, DMG and H, glioblastoma tumors. I, J, Heatmaps representing the relative expression (min-max normalized by region) of neuronal regulation module genes (columns) in tumor regions (rows) for I, five DMG and J, four glioblastoma tissues profiled by spatial transcriptomics.

Pathways analysis demonstrated that the modules were enriched with coherent biological functions, particularly neuronal regulation, ECM organization, and cell cycle modules were present across all HGGs (Figure 1A, B; Supplementary Figure 1A). The neuronal regulation module had the most significant number of ion channel genes, including 31, 71, and 114 ion channel genes in DMG, NBS-HGG, and glioblastoma, respectively. Mainly, the ion channels are derived from four families: voltage-gated calcium, potassium, sodium channels, and GABA_ARs, demonstrating a common wiring of ion channel genes in pediatric and adult networks that potentially contribute to HGG progression.

Also, by comparing module preservation (Supplementary Table 8), as well as examining the topological changes of the networks, we found that modules related to metabolism and MAPK signaling/VGKC were unique to DMG (Z < 10) compared to NBS-HGG, while all modules in NBS-HGG were preserved in DMG. Comparing the NBS-HGG network to glioblastoma showed that ribosomal translation, immunoglobulins/complement, and ErbB receptor signaling were moderately preserved, and ECM organization (Z = 0.85) was unique to NBS-HGG. The axon guidance module in glioblastoma was moderately preserved in the pediatric networks (Z < 6). The remaining HGG modules were highly preserved (Z > 10).

GABA_ARs Are Ion Channel Network Hubs in HGG

Next, we aimed to identify ion channel hub genes within the network structure since these can play essential roles in glioma progression and be targeted pharmacologically. Therefore, we reconstructed the underlying wiring diagram of the modules dominated by co-expressed ion channels, using Cytoscape to identify highly connected hub genes (Supplementary Figure 1C–E).

The reconstructed neuronal regulation modules showed that amongst the highest-ranking ion channel hub genes (based on intramodular connectivity and degree, i.e., number of edges; Supplementary Tables 5–7) were CACNG3 in NBS-HGG and glioblastoma, GABA_AR genes GABRA1 (number of edges for DMG/NBS-HGG/glioblastoma:19/11/42) and GABRG2 (37/7/37 edges) in the pediatric and adult networks (Supplementary Table 1). Unique to glioblastoma, the reconstructed axon guidance module contained two ion channel hubs, SCNN1D (34 edges) and ANO8 (7 edges; data not shown). The ECM module in DMG contained three ion channel hubs, cholinergic receptors CHRNG (36 edges), CHRND (23 edges), and CHRNA1 (5 edges), whilst the adult ECM module only contained one ion channel hub, TRPM2 (10 edges). There were no ion channel hubs in NBS-HGG (data not shown). There were no ion channel hubs in the reconstructed MAPK signaling/VGKC module in DMG nor the immune system module in NBS-HGG (data not shown).

Tumor Cells at the Leading Edge Express Neuronal Regulation Module Genes

Since the neuronal regulation module exhibited the highest density of ion channels, we examined its expression in malignant tumor cells and spatial distribution across tumor regions. First, we performed scRNA-seq analysis of 36 DMG,^29,30 and 31 glioblastoma^26–28,25 patient samples from multiple datasets to determine cell type-specific expression of neuronal regulation module genes. We confirmed that most neuronal regulation genes are highly expressed in tumor cells compared to nonmalignant cells (Figure 1C,D), with 66 shared genes showing high expression (calculated as mean z-score > 1) in both glioblastoma and DMG malignant cells (Supplementary Table 9; Supplementary Figure 1B).

A smaller proportion of the neuronal regulation module genes were also highly expressed in oligodendrocytes in both DMG and glioblastoma, in vascular (endothelial and pericyte), and immune cells (myeloid and T/B cells) in glioblastoma. To map spatial expression of the module across anatomically distinct tumor regions—leading edge, cellular tumor, microvascular proliferation, and pseudopalisading necrosis (Figure 1E,F)—we performed spatial transcriptomics on glioblastoma tissues from four patients (Visium10x) and obtained published spatial transcriptomics data from five DMG patients.³¹ Tumour region annotations were assigned by averaging the expression of region-specific gene signatures obtained from micro-dissected glioblastoma RNA-seq³⁵ (Figure 1F). These signatures also successfully demarcated analogous regions in DMG tissues (Figure 1E). Neuronal regulation genes were highly expressed in the leading edge, a zone rich in neurogliomal synapses but showed low expression in the other regions (Figure 1G–J). This pattern was consistent in glioblastoma and DMG tumor sections that adequately captured the leading edge (3/4 glioblastoma and 1/5 DMG tumor tissues).

HGG Tumors Express GABA_Aα1 and GABA_Aγ2

The GABA_AR gene family codes for nineteen subunits: α (1-6), ß (1-3), γ (1-3), δ, θ, ε, π and ρ (1-3), that assemble into pentamers, typically arranged 2α,2β,1γ.³⁶ Since GABA_ARs were hubs amongst the three neuronal regulation modules, gene expression of the 19 subunits were verified in RNA-Seq DMG (n = 31)/NBS-HGG (n = 44) and glioblastoma (n = 151) (Figure 2A,B; Supplementary Figure 2A; < 5.0 log₂ gene expression is negligible), showing that the most highly connected GABA_AR genes (Supplementary Table 1) were not the most abundant within tumors.

Figure 2.

GABA _A R gene expression in HGG patient tumors. Log₂ gene expression in A, DMG (n = 31), B, glioblastoma (n = 151). The dotted line signifies a log₂ gene expression of >5 where the gene is considered expressed. Red lettering denotes neuronal regulation gene hubs. C, D, Heatmaps representing the normalized expression (min-max normalized by cell type) of GABA_AR genes (rows) across cell types (columns) C, two DMG and D, four glioblastoma single-cell RNA-seq datasets. Red lettering denotes gene hubs. E, GABA_Aα1 and GABA_Aγ2 expression within primary HGG tumor samples. Quadruple immunofluorescence for GABA_Aα1 (red) and GABA_Aγ2 (green), OLIG2 (gray; tumor marker), and DAPI (blue; nuclear stain) in DMG (E_1-3) and glioblastoma (E_4-6) tumors shows GABA_Aα1 and GABA_Aγ2 expression within patient tumor samples. Images are maximum intensity z-projections of single plane scans taken at 0.5 µm increments through 4-5 µm of tumor using a x40 objective. Top right inserts are taken from single plane images, note GABA_Aα1 and GABA_Aγ2 colocalization in OLIG2+/- cells. F, Quantification of the percentage of OLIG2+ cells co-expressing either GABA_Aα1 or GABA_Aγ2 in DMG and glioblastoma patient tissue. *** P<.001, Mann-Whitney.

Single-cell analysis across individual glioblastoma and DMG datasets (Figure 2C,D; Supplementary Figure 1B) revealed that GABA_AR subunit genes, including top hubs GABRA1, GABRG2, (and GABRA5 in glioblastoma), are consistently enriched in tumor (Figure 2C, D) cells, specifically in cells with a neural precursor-like (NPC-like) expression³⁷ profile in glioblastoma (Supplementary Figure 2B). GABA_AR subunit genes were also enriched in immune cells in DMG, and endothelial cells in both diseases (eg, GABRD in glioblastoma).

To confirm protein expression in tumor cells of the top ion channel hubs, GABRA1 (GABA_Aα1) and GABRG2 (GABA_Aγ2), we used immunohistochemistry on frozen tissue sections from primary glioblastoma and DMG tumors (n = 3 each; Supplementary Table 4). Using OLIG2 as a tumor marker, we observed the expression of both GABA_AR subunits in tumor cells in DMG and glioblastoma tissues (Figure 2E_1-6). Quantification of GABA_AR-subunit expressing OLIG2+ cells revealed that compared to DMG, both GABA_AR proteins were detected at an increased frequency of ~2-3 fold higher with OLIG2 in glioblastoma (DMG vs glioblastoma both α1 and γ2, P<.001, Mann–Whitney), with a mean of 67% and 63% of OLIG2+ glioblastoma cells co-expressing GABA_Aα1 or GABA_Aγ2 respectively (Figure 2F). Both proteins also were highly colocalized in OLIG2+/- cells (inserts Figure 2E_1-6), which suggests an abundance of α1βγ2 channels in HGG patient tumors.

GABA_Aα1 and GABA_Aγ2 Appose Presynaptic Proteins in HGG Tumors

Using gene signatures identified by Puchalski et al., (2018)³⁵ to delineate four distinct HGG tumor regions (Figure 1E,F), we performed Manders’ Overlap coefficient analysis to determine the spatial overlap of these regions with the expression of the GABA_AR subunits genes in DMG and glioblastoma tumors. This enabled us to observe the relative abundance of gene expression within different tumor regions. We confirmed that GABRA1 and GABRG2 expression overlapped most strongly with the leading edge signature in DMG (Figure 3A,B) and glioblastoma (Figure 4A,B) supporting a potential role for GABRA1 and GABRG2 in a high functional connectivity microenvironment defined by neurogliomal synapses. Additionally, common to DMG and glioblastoma, we observed GABRB2 (β2), a less connected hub GABA_AR gene within the same region (Supplementary Table 1; Figures 3B and 4B). In glioblastoma, a fourth hub gene GABRA5 (α5), was also concentrated at the leading edge (Supplementary Table 1; Figure 4B). Consistent with our glioblastoma network analysis, which showed a high expression of GABA_AR genes within the neuronal regulation module (13/19 subunits), most GABA_AR genes were enriched at the leading edge in glioblastoma tissues (Figure 4B), while demonstrating lower expression in other tumor regions.

Figure 3.

GABRA1 and GABRG2 are expressed at the leading edge with GABAergic synaptic markers in DMG tumors A, The spatial expression pattern of gene signatures for the leading edge region defined by Puchalski et al., (2018) within a representative DMG tumor with spatial plots of GABRA1 and GABRG2 expression. B, Manders’ Overlap coefficient across the four regions for each GABA_AR subunit transcript in five DMG tumors. C, Spatial plots of tumor (OLIG2), neuronal (NFH) and GABAergic synapse marker genes (GAD1/GAD67, GAD2/GAD65, VGAT and GPHN/Gephyrin) in DMG. D, Manders’ Overlap coefficient across the four regions for each marker gene in five DMG tumors. E_1-3, Quadruple immunofluorescence for GABA_Aα1 (red), the presynaptic marker GAD65/67 (green), the neuronal marker NFH (blue) and OLIG2 (gray) in three DMG tumors. Note close apposition of GABA_Aα1 and GAD65/67 (arrows). F_1-3, Quadruple immunofluorescence for GABA_Aγ2 (red), the presynaptic marker GAD67 (green), NFH (blue) and OLIG2 (gray) in three DMG tumors. Note close apposition of GABA_Aγ2 and GAD67 (arrows). G_1-5, Immunofluorescence images of DMG patient tissue showing the postsynaptic marker, gephyrin (red) in close apposition to presynaptic markers GAD65/67 (green) and VGAT (blue) adjacent to OLIG2 + cells (arrows). H_1-5,  Immunofluorescence images of DMG patient tissue showing GABA_Aα1 (red) and GABA_Aγ2 (green) apposing VGAT (blue; arrows). All confocal images are single plane taken with a x40 objective.

Figure 4.

GABRA1 and GABRG2 are expressed at the leading edge with GABAergic synaptic markers in glioblastoma tumors A, The spatial expression pattern of gene signatures for the leading edge region defined by Puchalski et al., (2018) within a representative glioblastoma tumor with spatial plots of GABRA1 and GABRG2 expression. B, Manders’ Overlap coefficient across the four regions for each GABA_AR subunit transcript in four glioblastoma tumors. C, Spatial plots of tumor (OLIG2), neuronal (NFH) and GABAergic synapse marker genes (GAD1/GAD67, GAD2/GAD65, VGAT and GPHN/Gephyrin) in glioblastoma. D, Manders’ Overlap coefficient across the four regions for each marker gene in four glioblastoma tumors. E_1-3, Quadruple immunofluorescence for GABA_Aα1 (red), the presynaptic marker GAD65/67 (green), the neuronal marker NFH (blue) and OLIG2 (gray) in three glioblastoma tumors. Note close apposition of GABA_Aα1 and GAD65/67 (arrows). F_1-3, Quadruple immunofluorescence for GABA_Aγ2 (red), the presynaptic marker GAD67 (green), NFH (blue) and OLIG2 (gray) in three glioblastoma tumors. Note close apposition of GABA_Aγ2 and GAD67 (arrows). G_1-5, Immunofluorescence images of glioblastoma patient tissue showing the postsynaptic marker, gephyrin (red) in close apposition to presynaptic markers GAD65/67 (green) and VGAT (blue) adjacent to OLIG2+ cells (arrows). H_1-5, Immunofluorescence images of glioblastoma patient tissue showing GABA_Aα1 (red) and GABA_Aγ2 (green) apposing VGAT (blue; arrows). All confocal images are single plane taken with a x40 objective.

Across HGG, a significant proportion of genes within the neuronal regulation module were associated with synaptic transmission (Supplementary Tables 5–7). Therefore, we hypothesized that GABA_Aα1 and GABA_Aγ2 subunits present at the leading edge, formed GABA_ARs within neurogliomal synapses, where neurons form the presynaptic component, and HGG cells the postsynaptic¹¹ and used immunohistochemistry analysis on HGG patient tumors to determine if GABA_Aα1 and GABA_Aγ2 colocalized with synaptic markers. Presynaptic structures of GABAergic synapses were identified by GAD67 (GAD1) and GAD65 (GAD2), which catalyze the decarboxylation of glutamic acid into GABA, and the GABA vesicular reporter VGAT (VGAT). The scaffold protein gephyrin (GPHN), and the GABA_Aα1 receptor subunit itself are well-established postsynaptic markers.³⁸ As many neuronal markers are also expressed by glioma cells we used neurofilament protein heavy chain (NFH), to demarcate mature neurons, as there are no reports that glioma cells express the heavier isoform.^39–41

Immunohistochemistry confirmed that the structural components of GABAergic synapses were present within HGG patient tumors (Figures 3G and 4G), where colocalized VGAT and GAD65/67 apposing gephyrin were observed. To test our central hypothesis, we continued our immunohistochemistry on three primary DMG and glioblastoma patient tumors, co-staining with GAD65/67 and NFH and marking HGG cells with OLIG2. GAD+ bouton-like structures apposing GABA_Aα1 and GABA_Aγ2 were observed in DMG and glioblastoma (α1: Figure 3E_1-2 & Figure 4E₂; γ2: Figure 3F₂ & Figure 4F_1-3); with instances of neuronal colocalization (Figure 3F₃ & Figure 4F_2-3). Corroborating our findings, we also observed GABA_Aα1 and GABA_Aγ2 apposing VGAT in HGG tumors (Figures 3H and 4H). Furthermore, GAD65/67 cytoplasmic staining was also shown in DMG and glioblastoma in α1+ and γ2+ OLIG2+/- cells (Figures 3E₃, F₁ & 4E_1,3), suggesting that GABA is actively synthesized outside of synaptic structures in different cell types within the tumors. Collectively our results support our network analysis that GABA_Aα1 and GABA_Aγ2 are localized within synapses as postsynaptic components of GABAergic synapses within DMG and glioblastoma tumors.

To further confirm these results, we performed Manders’ Overlap coefficient analysis on the spatial transcriptomics data (Figures 3C–D, 4C-D) and to measure overlap between neuronal (NFH) and pre- (GAD1/2, VGAT) and postsynaptic (GPHN) markers across different tumor regions, which revealed that these synapses are highly concentrated at the leading edge, in DMG and glioblastoma tumors.⁶

GABAergic Signaling Is Prominent in GBOs that Express Neuronal Regulation Hub Genes

Depending on the chloride concentration gradient across the cell membrane, which is regulated by the cation-chloride co-transporters NKCC1 and KCC2, GABA_AR activation can be either depolarizing or hyperpolarizing.⁴² Our RNAseq analysis showed significant upregulation (Figure 5A) and diffuse enrichment of NKCC1 in four regions of glioblastoma and two regions in DMG, while KCC2, was more concentrated at the leading edge (Figure 5B,C; Supplementary Figure 3B,C). We hypothesized that overall GABA_AR currents are depolarizing, but at the leading edge with higher KCC2 expression, chloride currents may reverse direction. We used GBOs derived from resected glioblastomas to better recapitulate the genetics, intratumoral heterogeneity, and nonmalignant cellular composition of tumors compared to cell lines.²² However, caveats of GBOs are the uncertainty of specific tumor regions resected and the susceptibility of neuronal damage during the culturing process. Indeed, neuronal (NFH) expression within GBOs was infrequently detected by immunocytochemistry (Supplementary Figure 3A). Given the poor preservation of the leading edge in GBO culture, we hypothesized that blocking GABA_AR currents in GBOs would inhibit HGG growth and invasion, as glioma cell membrane depolarization drives HGG progression.¹¹

Figure 5.

GABA _A R antagonism inhibits GBO proliferation and invasion A, Boxplots of NKCC1 and KCC2 expression in HGG tumors. Statistical analysis was performed using t-test: ****P<.0001. B, The spatial expression pattern of NKCC1 and KCC2 in glioblastoma. C, Manders’ Overlap coefficient of NKCC1 and KCC2 with studied GABA_AR gene hubs D, E, GABA_Aα1 (D₂) and GABA_Aα5 (E₂) are expressed with GABA_Aγ2 (D₃ and E₃) in GBOs and are concentrated at the periphery with VGAT (D₄, E₄). F, G, GBOs strongly express genes related to GABA_AR signaling (gephyrin F₂, G_2; VGAT F₄, G₄) and metabolism (GAD65/67, F₃, G₃). H, I, Representative images of GBOs derived from two patients H, in proliferation at days 0 and 20 (scale bar: 250 µm) and I, invading Matrigel at day 14 (scale bar: 1000 µm) following incubation with either 0.1% DMSO (vehicle control), 50 µM GABA(A) Compound 1b, a partial GABA_AR antagonist, or 10 µM S44819, a GABA_A α5βγ2 antagonist. J, Assay protocols to test the effects of GABA_AR antagonism on two patient-derived GBOs’ proliferation (J₁) and invasive potential (J₂). K, L, Quantification of GBO proliferation calculated as the fold-change (FC) of spheroid area (µm²) normalized to day 0 measurements derived from two patients during a 20-day drug incubation period with GABA_AR-drugs and vehicle control. All data represented as means±SEM. Statistical analysis was performed using Two-way ANOVA with Tukey’s multiple comparisons: *P<.05; **P<.01; ***P<.001; ****P<.0001. M, N, Quantification of GBO spheroid Matrigel invasion calculated as the FC in peak radial displacement (µm) normalized to day 0 in Matrigel measurements (see J₂). Between days 0-6 GBOs were treated with vehicle (0.1% DMSO), 50 µM Compound GABA(A)1b or 10 µM S44819. At day 6, organoids were embedded in Matrigel and were treated for an additional 14 days. All data represented as means±SEM. Statistical analysis was performed using Two-way ANOVA with Tukey’s multiple comparisons: *P<.05; ***P<.001; ****P<.0001.

First, using immunocytochemistry analysis, we confirmed GABA_Aα1 and GABA_Aγ2 expression in GBOs (Figure 5D; Supplementary Figure 3D). Interestingly, both subunits showed a high degree of colocalization, and with VGAT, localized toward the periphery of the GBO together with OLIG2+ cells (Figure 5A; Supplementary Figure 3D), suggesting that the gradient of GABRA1 and GABRG2 expression, with GABAergic synapse genes, is preserved in culture. However, the in vivo synaptic architecture in GBOs was not preserved with pre- and post-synaptic proteins overlapping (Figure 5G, G₅ yellow arrows). Nonetheless, gephyrin and GAD65/67 expression was uniform in OLIG+/- cells, indicative of widespread GABA_AR expression and GABA synthesis in glioma cells (Figure 5F,G). Therefore, we decided to focus on α5, the α-subunit component of non-synaptic (extrasynaptic) GABA_ARs and the product of another gene hub, GABRA5, in the glioblastoma neuronal regulation module (Supplementary Table 1). Using immunohistochemistry on patient tumors we confirmed that GABA_Aα5 was expressed in HGG patient tumors (Supplementary Figure 3F-H). In our three glioblastoma tumors GABA_Aα5 was detected at a significantly increased frequency in OLIG2+ cells compared to GABA_Aα1 and GABA_Aγ2, although this was not observed in our RNA-seq analyses with larger datasets (Supplementary Figure 3H; Figure 2B,D). Spatially in our GBOs GABA_Aα5 was evenly distributed, and colocalized with GABA_Aγ2 in VGAT/OLIG2+ cells (Figure 5E; Supplementary Figure 3E). Collectively, our observations suggest that our GBOs possess a GABAergic phenotype and even in the absence of the healthy brain environment, GABRA1, GABRG2, and VGAT expression is spatially regulated.

Inhibition of GABAergic Signaling Blocks the Growth and Invasiveness of GBOs

Since GABRA1, GABRG2, GABRA5, and GABRB2 were hubs in the glioblastoma neuronal regulation module and GABA_Aα1/5 and GABA_Aγ2 colocalized in our GBOs (Figure 5D,E), we hypothesized that α1β2γ2 and α5β2γ2 receptors were expressed and that modulating their activity would have a greater impact than targeting other GABA_AR isoforms. We were limited in our choice of antagonists because few GABA_AR drugs specifically target certain isoforms, cross the BBB, and have manageable pro-convulsive effects. We therefore examined the anti-cancer efficacy of two specific α5βγ antagonists that fulfilled our criteria: GABA(A)-Compound 1b and S44819 (also named Afizagabar). GABA(A)-Compound 1b is a selective α5βγ antagonist at nanomolar concentrations, and a broad partial antagonist of GABA_AR isoforms, including pentamers containing GABA_Aα1, at micromolar concentrations, significantly limiting its pro-convulsive effects in rodents⁴³ and S44819, an antagonist of α5βγ2 receptors, is a drug specifically designed to inhibit α5-receptors, e.g., demonstrating an IC₅₀ ~600 nM for α5β2γ2,⁴⁴ which also showed no adverse effects in a Phase II stroke trial.⁴⁵

In preliminary experiments using three patient-derived GBOs (Supplementary Figure 4A,B, Supplementary Table 10) and 50 µM  TMZ as a positive control, we observed that a short incubation (6 days) with 50 µM GABA(A) Compound 1b broadly inhibited invasion and proliferation, with 10 µM S44819 inhibiting invasion in GBOs from patient 3. We next considered whether GABA_AR antagonism would confer additional therapeutic benefit in clinically relevant scenarios, determining whether GBOs developed resistance to GABA_AR drugs, mimicking a longer dosage regimen, and assessing their efficacy in combination with SOC protocols. We used GBOs from patients 1 and 2 (Supplementary Table 10) that poorly responded to S44819, to determine if sustained α5-GABA_AR antagonism would more broadly suppress the carcinogenesis of glioblastoma cells (protocol in Figure 5J_1,2;  Supplementary Figure 4C). Also, considering current SOC, we aimed to assess outcomes based on tumor cell MGMT promoter methylation status (i.e., patient 1, methylated and patient 2 unmethylated), to determine if patients with varying sensitivity to TMZ would differentially benefit from GABA_AR-antagonism (see protocol in Figure 6A).

Figure 6.

The effects of GABA _A R antagonism on GBO proliferation and invasion under clinically relevant treatment regimens A, Assay protocol to test the effects of GABA_AR antagonism with the Stupp protocol (as indicated in red) on two patient-derived GBOs’ proliferation and invasion. B, C, Quantification of B, methylated and C, unmethylated MGMT patient GBO proliferation calculated as the relative fold-change (FC) in spheroid area (µm²) normalized to day 0 measurements in vehicle (0.1% DMSO), 50 µM Compound GABA(A)1b and 10 µM S44819 conditions with or without the Stupp regimen (see A for protocol). All data represented as means±SEM. Statistical analysis was performed using Two-way ANOVA with Tukey’s multiple comparisons: ns=not significant; ****P<.0001. D, E, Representative images of D, methylated and E, unmethylated MGMT populations of GBOs in proliferation at day 0 and day 10 during radiation and drug administration protocols depicted in A. Scale bar: 250 µm. F, G, Quantification of GBO Matrigel invasion of F, methylated MGMT and G, unmethylated patient GBOs calculated as the peak radial displacement (µm) normalized to day 10 measurements (see A for protocol). All data represented as means±SEM. Statistical analysis was performed using Two-way ANOVA with Tukey’s multiple comparisons: *P<.05; ****P<.0001. H, Representative images of methylated (top panel) and unmethylated (bottom panel) MGMT populations of patient GBOs invading Matrigel at day 16 following vehicle (0.1% DMSO), 50 µM Compound GABA(A)1b and 10 µM S44819 administration with or without TMZ and radiation pretreatment (see A; scale bar: 1000 µm).

To test prolonged 20-day dosing regimens on GBO proliferation, with either 50 µM GABA(A) Compound 1b, 10 µM S44819 or vehicle controls, we monitored fold-changes in 2D spheroid-projected area (µm²) relative to day zero, up to day 20, during continuous treatment or washout at day six (Figure 5H,J₁,K,L; Supplementary Figure 4C,D-G). For invasion, at day six GBOs were transferred into Matrigel, and monitored for fold-changes in peak radial displacement (µm) over 14 days, with treatment continued or withdrawn (Figure 5I,J₂,M,N; Supplementary Figure 4C,H-K).

As before, GABA(A) Compound 1b significantly inhibited proliferation (Figure 5H,K,L) and invasion (Figure 5I,M,N), with less potency after drug withdrawal (Supplementary Figure 4C,D,H,F,J). Prolonged S44819 incubation showed late anti-proliferative activity at day 20 (Figure 5H,K,L), making drug withdrawal at day six irrelevant (Supplementary Figure 4E,G). S44819 was more potent against invasiveness, significantly slowing organoid invasion by day 14 (Figure 5I,M,N), with sustained inhibition remaining after washout in one GBO (Supplementary Figure 4K). Collectively, these data demonstrate the therapeutic potential of persistently targeting different GABA_AR isoforms.

Next, monitoring the same GBO proliferation and invasion parameters, we tested whether GABA_AR antagonism would provide additional therapeutic benefit to GBOs subjected to the Stupp regimen^22,34 (protocols: Figure 6A). GBOs were treated with 50 µM TMZ and 10 Gy radiation over 10 days, with concomitant GABA_AR-drug or vehicle treatment. In assessing invasion, the latter conditions were continued for six days on GBOs embedded in Matrigel. We observed that the Stupp protocol and GABA_AR antagonism had similar effects on GBO proliferation, which is not enhanced by their combination (Figure 6B-E). However, TMZ and radiation enhanced the inhibition of invasiveness of distinct GABA_AR-drug treated GBO populations (Figure 6F-H), where methylated and unmethylated GBOs responded better to combined therapies including S44819 (Figure 6F,H) and GABA(A) Compound 1b, respectively (Figure 6G,H), compared to single GABA_AR-drug treatment. Whilst determining if these observations reveal a specificity of GABA_AR isoforms acting on distinct invasive signaling pathways, and whether this is related to methylation status, is outside the objectives of this study, our data suggests that patients with varying TMZ sensitivity may benefit from SOC regimens combined with different subclasses of GABA_AR-antagonists.

Discussion

We are the first to characterize ion channel-related network topology in HGG using highly variable genes, identifying shared network wiring across DMG, NBS-HGG, and adult glioblastoma, including neuronal regulation, cell cycle regulation, ECM organization, immune responses, alongside glioblastoma-specific modules like axon guidance. Our results suggest that while common biological processes underpin HGG progression, network differences may offer HGG subtype-specific treatment opportunities.

Across all HGG networks, GABRA1 and GABRG2 emerged as top hub genes within the neuronal regulation module. Many preserved genes were linked to synaptic signaling and were expressed by malignant cells at the leading edge in DMG and glioblastoma. Immunohistochemistry confirmed postsynaptic GABA_Aα1 and γ2-containing receptors within neurogliomal GABAergic synapses, suggesting that within the leading edge microenvironment, highly connected GABRA1 and GABRG2 may play a prominent role in driving HGG progression through GABA_AR-regulated synaptic processes. Although our GBOs did not completely preserve the leading edge microenvironment they expressed GABA_Aα1/γ2, and GABA_Aα5 the product of GABRA5, a hub in glioblastoma. Pharmacologically, antagonizing α5βγ2 receptors in GBOs disproportionately suppressed their invasion compared to proliferation, and partial antagonism of GABA_ARs, significantly inhibited proliferation and invasion. Moreover, under SOC protocols the anti-invasive effects of GABA_AR-drugs were enhanced in distinct GBO populations. Collectively, our ion channel network analysis successfully identified a neuronal modulation gene network embedded within the leading edge tumor microenvironment that uses GABA_AR signaling to drive HGG progression, which can be disrupted by GABA_AR antagonists.

GABA, the ligand for GABA_ARs, may not be derived exclusively from GABAergic interneurons particularly as they are vulnerable to glutamate neurotoxicity near the tumor’s leading edge.⁴⁶ Glioblastoma cells release glutamate, a substrate for the synthesis of GABA by GAD65/67, as well as GABA itself into the extracellular environment.^47,48 In support, our results show GAD65/67 cytoplasmic staining in HGG cells, indicative of GABA synthesis, and in vivo studies demonstrate that within the brain, breast-to-brain metastases adapt to a GABAergic phenotype showing a proliferative advantage through GABA metabolism.⁴⁹ However, we know that subpopulations of HGG cells form functional GABAergic synapses with neurons. For example, in mice bearing H3K27M+ DMG xenografts, inhibitory interneurons and HGG cells formed depolarizing GABAergic synapses (attributed to increased NKCC1 activity), which upon repeated stimulation significantly increased tumor proliferation,⁵⁰ supporting our pharmacological findings.

Consistent with the depolarizing effects of GABA in HGG, our analysis showed that overall NKCC1 was significantly upregulated. Likely reflecting the neural progenitor origins of HGG cells and/or the increased instances of peritumoral ictal effects that can stimulate transcriptional changes in cation-chloride co-transporters.⁴⁶ During neuronal development and in oligodendrocytes/astrocytes,⁴² GABA_AR activation results in Cl^- efflux, because of higher NKCC1 activity, which in immature neurons surpasses KCC2 activity.⁴² In mature neurons, KCC2 activity is more prominent but around epileptic glioma peritumoral regions KCC2 expression decreases within neurons, contributing to their hyperexcitability, and can switch nearby neuronal GABA neurotransmission to excitatory in glioma patients.^46,51,52  KCC2 was concentrated at the leading edge but without detailed electrophysiological analysis, we cannot predict the chloride current direction within this microenvironment as in vitro glioma cells permeate endogenous depolarizing and hyperpolarizing GABA_AR currents.⁵³ Nonetheless, dynamic changes in cation-chloride co-transporter expression and resulting Cl^- gradients, and directional flow, may significantly impact HGG tumor progression and peritumoral neuronal network function.

Caveats of our study include that we did not consistently identify neurons as presynaptic cells nor confirm these “orphaned” GABA_Aα1/γ2-containing synapses were functional. Nonetheless, GABA_Aα1/γ2-containing synapses may reflect structural artifacts of predetermined developmental processes, with our network analysis possibly revealing epigenetic/genetic programs established during ontogeny,⁵⁴ which as evidenced by gene expression patterns in our GBOs, may also be spatially regulated. Indeed, GABA_AR-derived excitation mediates activity-dependent synaptogenesis during embryogenesis, where GABA acts as a trophic factor that depolarizes neuronal progenitors, regulating cell proliferation, survival, and growth.⁵⁵ Although incompletely understood mechanistically, in in vitro culture systems α1β2γ2-GABA_ARs can act as postsynaptic structural proteins that initiate the formation of GABAergic synapses,⁵⁶ raising the possibility that highly networked GABA_AR genes may be master regulators of synaptic protein recruitment. Particularly as GABA_AR transcripts were predominantly expressed in the proneural glioblastoma cohort, i.e., OPC/NPC-like cells that are synaptogenic,¹⁰ which are also enriched at the leading edge of glioblastoma tumors.⁵⁷

When considering the clinical implications of our study, treating HGG patients with GABA_AR-inhibiting drugs may cause seizures and requires caution. Within GABA_AR pentamers, α-subunits are critical for drug interactions, as binding sites are typically located where α-subunits interface with other subunits.³⁶ α1, most commonly associated with β2γ2, is targeted by several GABA_AR-drugs,⁵⁸ but as the most abundant subunit in the brain, complete inhibition of α1 is risky. Therefore, we tested GABA(A)-compound 1b, a partial antagonist of α1-4, βγ receptors, which has shown minimal CNS side-effects in rodents, although is yet to be tested in humans.⁴³ Our results indicate that sustained partial antagonism of GABA_AR activity, with GABA(A) Compound 1b, would help to significantly control disease burden, potentially improving outcomes in glioblastoma patients less responsive to TMZ therapy. Further, whether the cancer-suppressive effects of GABA(A) Compound 1b results from glioma cell hyperpolarization or the modulation of oncogenic signaling pathways remains unclear. Indeed, in systemic cancers like pancreatic, gastric, ovarian, and breast, enhanced GABA receptor activation stimulates several cancer pathways like cAMP, EGFR, AKT, and MAPK/ERK, and ECM breakdown through matrix metalloproteinase activation.⁵⁹ Further study is needed to assess the efficacy of GABA(A)-compound 1b in DMG, but it holds promise due to a similar GABA_AR expression, and with NBS-HGG, gene network profile to glioblastoma.

Sustained administration of S44819, a specific α5βγ2 antagonist, more weakly inhibited the proliferation of glioblastoma cells, better-suppressing invasion, revealing for the first time a more defined role of specific GABA_AR pentamers in HGG pathophysiology. Therapeutically, this is promising as theoretically, the various combinations of GABA_AR subunit can create 400,000+ receptor subtypes.⁶⁰ S44819 has an excellent safety profile⁴⁵ and as an abundant α-subunit in HGG, α5-pentamers are an attractive target as they comprise <5% of GABA_ARs in the CNS.⁶¹ Since combinational treatments including S44819 demonstrated robust anti-invasive effects on methylated GBOs—in patients likely to be more responsive to SOC—S44819 may offer a safer adjuvant therapy to more broadly acting GABA_AR-inhibitors. However, in a broader disease context, it remains to be determined whether α5-GABA_AR antagonism is more effective in glioblastoma compared to pediatric HGG, where GABRA5 is a hub within the invasive microenvironment. More study is required to determine how GABA_A-α5-containing receptors may contribute invasive potential in cancer, but the oncogenic role of α5 has been studied in medulloblastoma, where GABRA5 expression is enhanced in the Group 3 subtype. In vitro α5-GABA_AR agonism causes membrane depolarization, blocks cellular proliferation, and induces apoptosis.⁶² The opposing effects of α5-GABA_AR activity in cancer progression reveal its complex oncogenic role and emphasize the need for context-specific therapeutics in brain cancer.

Indeed, in vivo, GABA_AR expression in other cell types within other regions may convey changes in drug sensitivity between HGG subtypes. For example, another α-subunit gene, GABRA3, was identified as a hub involved in immune cell response, specifically in NBS-HGG. Many immune cell subtypes, B and T cells, dendritic cells, macrophages, and neutrophils, express GABA_ARs,²¹ and a recent study demonstrated that GABA signaling significantly modulates immune responses against tumors. GABA secreted from B cells promoted anti-inflammatory macrophages and inhibited the anti-tumor responses of CD8 T cells via their GABA_ARs in a mouse model of colon carcinoma.⁶³ Whether a similar mechanism occurs in HGG is yet to be tested, but antagonizing specific GABA_ARs may have unwanted anti-tumor immune responses supporting the need for isoform-specific drug interventions.

In summary, our findings show that GABA_ARs are critical in the progression of HGG and that a conserved neuronal regulation network within the tumor’s leading edge uses neurogliomal GABAergic synapses to drive HGG growth and invasion. We have delineated GABA_AR subtypes associated with specific hallmarks, highlighting their potential as targets for precision therapies. However, based on the complex interactions between GABA_ARs, the tumor microenvironment, and immune modulation, further studies are required to optimize their therapeutic potential while minimizing side effects.

Funding

This study was conducted with support from The Kids Research Institute Australia Ascend fellowship [to E.V.F]; The Cure Starts Now [to T.J.], Perth Children’s Hospital Foundation [to T.J. and S.V.], the Western Australian Future Health Research and Innovation Fund [to T.J. and E.V.F.], the Robert Connor Dawes Foundation [to T.J.], the Children’s Leukemia & Cancer Research Foundation, Western Australia [to T.J. and E.V.F]. This work was carried out with the support of a collaborative cancer research grant provided by the Cancer Research Trust “Enabling advanced single-cell cancer genomics in Western Australia,” an enabling grant from the Cancer Council of Western Australia [to A.R.R.F and E.V.F.]. A McCleary Murchland Fellowship [to G.A.G.]; grants from the NHMRC (Ideas grant 2021/GNT2013180 and 2023/GNT2030541 to G.A.G., A.P., R.J.O. ad S.P.); the Cure Brain Cancer Foundation [to G.A.G., and S.P.]; the Neurosurgical Research Foundation [to G.A.G., H.M.P., C.S.]; the Cancer Council S.A. Beat Cancer Project [to G.A.G.]; the Charlie Teo Foundation [Rebel Grant to G.A.G.]; The MAWA Trust [to G.A.G and H.M.P.]; Tour de Cure Grants to [G.A.G., and H.M.P]; and a Neurosurgical Research Foundation Chris Adams award [to C.S.].

Conflict of Interest Statement

A.P. is the founder and CEO of Combinatics. A.B. is the founder and equity holder of the startup company INSiGENe Pty Ltd that is related to this work. A.J. was an employee of INSiGENe. A.B. is a co-founder, equity holder, and director of the startup company Respiradigm Pty Ltd that is unrelated to this work. A.B. has grants or contracts with NIH R21 AI176305-01A1, NIH R01AI099108-11A1, Sanofi-Aventis, and the Bill & Melinda Gates Foundation. All other authors have declared no competing interests.

Author Contributions

Conceptualization: CS, ACJ, TGJ, GAG, EVF. Data collection/analysis/interpretation: CS, ACJ, AF, HMP, YB, BD, YY, IM, AF, MB, IK, MEJ, AP, RJO, SR, ZYN, EVF. Manuscript writing: CS, ACJ, AF, HMP, RJO, GAG, EVF. Funding acquisition: ARRF, TGJ, GAG, EVF. Scientific and infrastructural support: AB, RJO, SIP, SV, ARRF, TGJ, GAG, EVF. Provision of samples: RJO, SIP. Manuscript Editing/Critical Revision of the Manuscript: All authors.

Data Availability

Data will be made available upon reasonable request.