GABAergic neuronal lineage development determines clinically actionable targets in diffuse hemispheric glioma, H3G34-mutant

Abstract

Diffuse hemispheric gliomas, H3G34R/V-mutant (DHG-H3G34), are lethal brain tumors lacking targeted therapies. They originate from interneuronal precursors; however, leveraging this origin for therapeutic insights remains unexplored. Here, we delineate a cellular hierarchy along the interneuron lineage development continuum, revealing that DHG-H3G34 mirror spatial patterns of progenitor streams surrounding interneuron nests, as seen during human brain development. Integrating these findings with genome-wide CRISPR-Cas9 screens identifies genes upregulated in interneuron lineage progenitors as major dependencies. Among these, CDK6 emerges as a targetable vulnerability: DHG-H3G34 tumor cells show enhanced sensitivity to CDK4/6 inhibitors and a CDK6-specific degrader, promoting a shift toward more mature interneuron-like states, reducing tumor growth, and prolonging xenograft survival. Notably, a patient with progressive DHG-H3G34 treated with a CDK4/6 inhibitor achieved 17 months of stable disease. This study underscores interneuronal progenitor-like states, organized in characteristic niches, as a distinct vulnerability in DHG-H3G34, highlighting CDK6 as a promising clinically actionable target.

Figure 1.. Developmental spectrum of interneuronal lineage tumor cells in DHG-H3G34.

(A) Differentially expressed genes (rows) between DHG-H3G34 (blue), DMG-H3K27 (green), and H3-wildtype HGG (grey), in n=191 pediatric patient HGG bulk RNA-seq profiles. (B) Boxplots of select differentially expressed genes. The thick line within the box represents the median, the lower and upper limits of the boxes first and third quartiles, the whiskers 1.5x the interquartile range. All comparisons DHG-H3G34 versus DMG-H3K27: p-value <0.0001, save for CDK6, p-value=0.016, t-test. (C) Uniform manifold approximation and projection (UMAP) of 2,587 DHG-H3G34 glioma cells/nuclei profiled by single-cell (2 samples)/single-nucleus RNA-seq (7 samples). (D) Relative expression of NMF metaprogram marker genes (rows) in whole-cell transcriptomes (columns) from fresh sample MUV94. (E) Projection of fresh single-cell metaprograms (y-axis) onto a single-cell atlas of the developing human cortex. Color scale: scores of normal cell signatures in tumor cells. Dot sizes: scores of tumor cell signatures in normal cells. (F)-(G) Joint UMAP of normal human cortex cells and DHG-H3G34 fresh tumor cells after data integration. Cells are colored by (F) data source and (G) cell type as well as fresh/frozen tumor metaprograms. (H) Sankey plot depicting distribution of single-nuclei (columns), scored for either metaprograms derived from frozen tumors (left) or for all unique, granularly resolved metaprograms derived from fresh and frozen tumors (right). (I) Log-normalized expressions (y-axes) of outer radial glia marker genes across fresh single-cell metaprograms (x-axes) in MUV94. (J) Diffusion map of MUV94 single-cell transcriptomes colored by fresh metaprograms (left), and pseudotime analysis with cells colored by relative pseudotime (right). (K) Z-scored expressions (color scale) of temporally dynamic genes (rows) in MUV94 single-cells ordered along pseudotime (columns). (L) Proportion (y-axis) of cycling vs. non-cycling (color-scale) single-cells/nuclei in combined fresh and frozen metaprograms (x-axis). (M) 2D-representation of the stem-like RG/NPC-like scores (y-axis) and neuronal (INPC/eIN-like) vs. AC/MES-like scores (xaxis) for all single-nuclei, colored by their cycling status. See also Figure S1 and Tables S1–S2.

Figure 2.. Interneuronal lineage programs are preferentially expressed and epigenetically maintained in DHG-H3G34.

(A) & (B) Markers of interneuron lineage development/specification (y-axis), compared for their expression across (A) DHG-H3G34, DMG-H3K27, and H3-wildtype GBM single-cell populations, and (B) DHG-H3G34 and H3-wildtype GBM pathway-based single-cell populations (x-axis). PPR: proliferative/progenitor, NEU: neuronal, GPM: glycolytic/plurimetabolic, MTC: mitochondrial. Dot size: percentage of cells expressing the gene in the given population. Color scale: scaled average relative expression. (C)-(E) Volcano plots depicting differentially expressed genes with -log10 of the BH-adjusted Wilcoxon rank-sum test-derived p-value (y-axis) plotted against log2 fold change (x-axis) in DHG-H3G34 NPC-like vs. GBM PPR cells (C), DHG-H3G34 INPC-like vs. GBM NEU cells (D), and DHG-H3G34 eIN-like vs. GBM NEU cells (E). (F) Input-normalized, rank-ordered H3K27ac signals (x-axis) against H3K27ac occupancy (y-axis) in DHG-H3G34. Super-enhancer-associated genes are colored blue. (G) Gene ontology analysis of DHG-H3G34 super-enhancer-associated genes, plotted as dots according to their enrichment score (x-axis), scaled by the number of genes in the annotated ontology. (H) Ranked cell state-specific TF regulatory networks (regulon, x-axis) derived from 7 frozen tumors by SCENIC, plotted against their normalized specificity score (y-axis). (I) Overrepresentation analysis of SCENIC-derived cell state-specific TF regulons (x-axis), overlayed with H3K27ac ChIP-seq derived super-enhancers. Dot size: -log10 of FDR-adjusted p-values calculated using a Fisher’s exact test. Color scale: number of intersecting genes. See also Figure S2 and Tables S3–S4.

Figure 3.. In situ validations of scRNA-seq derived DHG-H3G34 interneuronal lineage programs.

(A) Markers of RG/NPC-like, INPC/eIN-like and AC-like DHG-H3G34 tumor cell populations used for immunofluorescence (IF) validations. Dot sizes: percentage of cells expressing the gene in the given cluster. Color scale: scaled average relative expression. (B) Schematic of nests of DCX⁺ neuroblasts surrounded by different Nestin⁺ SOX2⁺ progenitor cells detected in human developing MGE during 2^nd trimester, as described by Paredes et al., 2022. i/oSVZ: inner/outer subventricular zone. (C)-(E) Nest-like structures of DCX⁺ (green) G34R⁺ (blue) tumor cells, surrounded by Nestin⁺ (red) G34R⁺ tumor cells in patient DHG-H3G34 G34PT1. Scales: 200 μm. (F) IF representation of HOPX⁺ RG/NPC-like tumor (G34R+) cells, along STMN2⁺ INPC/eIN-like tumor cells and AQP4⁺ AC-like tumor cells in patient DHG-H3G34 G34PTE9. Scale: 200 μm. (G) IF of HOPX⁺ RG/NPC-like tumor (G34R⁺) cells, along TUJ1⁺ INPC/eIN-like tumor cells and AQP4⁺ AC-like tumor cells in patient DHG-H3G34 G34PTN16. Scale: 100 μm. (H) IF of SOX2⁺ RG/NPC-like tumor cells alongside rare STMN2⁺ INPC/eIN-like cell populations in patient DHG-H3G34 G34PT1. Scale: 200 μm. See also Figure S3 and Table S5.

Figure 4.. DHG-H3G34 display features of GABAergic interneuron development and exert neuron-glioma interactions distinct from other HGG types.

(A) Log-normalized expression levels of GAD1 splice isoforms (top middle and right) and GAD2 transcripts (bottom) in fresh DHG-H3G34 eIN-like cells. (B)&(C) IF showing GABA⁺ (red) malignant cells (G34R⁺) in two patient DHG-H3G34, G34RPN31 (B) & G34RPE93 (C), scales: 100 μm (B) & 200 μm (C). (D) IF of GABA (green) and VGAT (purple) in HSJD-GBM-002 (H3.3G34R) and normal GABA neuron controls. Scales: 100 μm (top), 50 μm (bottom). (E) Schematic of experimental design for panel (F). (F) Neuronal hyperexcitability in patient-derived DMG-H3K27 (SU-DIPGVI, green) but not in two DHG-H3G34 (HSJD-GBM-002, dark blue; OPBG-GBM-001, light blue) xenografts in the mouse hippocampus. Plot of presynaptic fiber volley versus amplitude of fEPSP at increasing axonal stimulation intensities (10, 20, 30, 50, 75, 100, 150 and 200 μA) in glioma-bearing or needlestick control hippocampus (needlestick control: n = 8 slices from 4 mice; HSJD-GBM-002: n=9 slices from 4 mice; OPBG-GBM-001 : n=7 slices from 3 mice; SU-DIPGVI: n=7 slices from 3 mice at each data point). Data fit to a nonlinear regression and compared to needlestick by extra-sum-of-squares F-test (p-values for HSJD-GBM-002: 0.0727; OPBG-GBM-001: 0.77; SU-DIPGVI: 0.0001). Data are mean ± SEM. See also Figure S4.

Figure 5.. Progenitor-like states along interneuronal lineage development present a distinct vulnerability in DHG-H3G34.

(A) Schematic of the genome-wide CRISPR-Cas9 screen protocols. (B) Top 100 gene dependencies from both CRISPR-Cas9 screens scored for their expression (color scale) in snRNA-seq data, plotted as 2D-representation of RG/NPC-like scores (y-axis) and INPC/eIN-like vs. AC/MES-like scores (x-axis). (C) Left: Ranked gene effect for the KNS42 screen. Blue points indicate genes with dependency scores <0. Right: Ranked gene dependencies for the DepMap CRISPR-Cas9 screening panel of pan-cancer cell lines. Blue point indicates the ranked position of KNS42. (D) Left: Ranked gene effect for the OPBG-GBM-001 screen. Turquoise points indicate genes with dependency scores <0. Right: Log2 read counts of multiple gRNAs targeting each gene at the beginning (T0) and after 10 cell doublings (T10). The thick line within the boxplot represents the median, the lower and upper box limits the first and third quartiles, the whiskers the 1.5x IQR. (E) Cell viability after CRISPR-Cas9-directed single-gene knockouts compared to AAVS1 negative gRNA control in 3 DHG-H3G34 cell lines. ***p-values <0.001 for all targets relative to AAVS1 negative control by unpaired two-sided t-test. Data shown as mean ± SEM. (F) Bioluminescent (BLI) signal quantification of animals orthotopically xenografted with HSJD-GBM-002 cells following CRISPR-Cas9-mediated single-gene knockouts (n=6 animals per group), shown as mean ± SEM. P-values derived from a two-sided unpaired Student’s t-test. Representative BLI images from each group are shown on the right. (G) Kaplan-Meier analysis of mice harboring HSJD-GBM-002 DHG-H3G34 tumors with CRISPR-Cas9-mediated single-gene knockouts vs. AAVS1 safe-harbor locus control (n=6 animals per group). P-values derived from log-rank test compared with AAVS1 control animals: for SOX2: 0.0014; DLX2: 0.11; DLX5: 0.24. (H) Scaled gene effect calculated by CHRONOS for the genome-scale CRISPR-Cas9 screens in OPBG-GBM-001 (x-axis) vs. KNS42 (y-axis). (I) Methylation beta-values plotted as ribbons across the CDK6 genomic locus, with samples separated by H3.3-mutation status (color legend). Ribbon size: degree of variation across samples. (J) Log-normalized expression of CDK6 in DHG-H3G34 single-nucleus transcriptomes, plotted as 2D-representation of RG/NPC-like score (y-axis) and INPC/eIN-like vs. AC/MES-like scores (x-axis). (K) Relative cell viability of DHG-H3G34 cells after CRISPR-Cas9 knockout of CDK6 using two different sgRNAs, compared to AAVS1 negative gRNA control (***p-value <0.001, unpaired two-sided t-test). Data shown as mean ± SEM. (L) BLI quantification of animals with orthotopic BT690 DHG-H3G34 tumors, followed by shRNA doxycycline-inducible knockdown of CDK6 compared to scramble shRNA (n=5 animals per group), shown as mean ± SEM. P-values calculated by two-sided unpaired Student’s t-test. (M) Kaplan-Meier analysis of mice from (L), with shRNA doxycycline-inducible knockdown of CDK6 after tumor formation compared to shRNA Scramble control (n=5 animals per group). P-value <0.05, log-rank test. See also Figure S5 and Table S6.

Figure 6.. Enhanced efficacy of CDK4/6 pharmacological inhibition in DHG-H3G34.

(A) Heatmap of GI50 values [μM] assessed for CDK4/6-inhibitors (ribociclib, abemaciclib, palbociclib) and CDK6 degrader (BSJ-03–123) at day 14 following 4 doses across DHG-H3G34 (BT690 [assessed independently at DFCI and ICR, plotted separately], OPBG-GBM-001, HSJD-GBM-002 [assessed independently at DFCI and ICR, plotted separately], CHOP-GBM-001, ICR-CXJ046), DMG-H3K27 (BT869, SUDIPG45), H3-wildtype (SU-pcGBM2, HSJD-GBM-001, ICR-CXJ001, ICR-CXJ008, ICR-CXJ015) cell lines, non-malignant human astrocytes (NHA) and mice OPCs (columns). GI50 values were inferred for each cell line and drug from at least two biological replicates. (B) Relative viabilities of DHG-H3G34, DMG-H3K27, H3-wildtype cell lines, non-malignant human astrocytes (NHA) and mice OPCs, at day 14 following 4 doses of treatment compared to DMSO control, shown as mean ± SEM of at least two biological replicates per cell line and drug. (C) Boxplot representation of CDK6 expression across 5 ribociclib sensitive vs. 6 non-sensitive cell lines. P-value was calculated using a Wald-test and BH-adjusted for multiple testing by DESeq2. The thick line within the box represents the median, the lower and upper limits the first and third quartiles, the whiskers 1.5x the interquartile range. (D) Representative gating strategy for cell cycle analysis in OPBG-GBM-001 following treatment with 0.5 μM BSJ-03–123 or DMSO. (E)-(F) Venn diagrams of commonly upregulated (E) and downregulated (F) genes between ribociclib, CDK6 degrader BSJ-03–123, and shRNA-mediated CDK6 knockdown. P-values calculated by Fisher’s exact test. (G)-(H) Gene ontology analyses of (G) upregulated and (H) downregulated genes shared between ribociclib, CDK6 degradation, and CDK6 shRNA knockdown. Color scale: -log10 FDR/adjusted p-values, size: number of genes within the respective GO term. (I) Relative expression of select cell cycle, stemness, adhesion, and interneuron differentiation-associated marker genes down- or upregulated following 14 days of ribociclib compared to DMSO treatment in HSJD-GBM-002. (J) Histogram of flow cytometry analysis in BT690 DHG-H3G34 line treated with 1 μM ribociclib over 10 days for SOX2⁺ (top) and TUBB3⁺ (bottom) cells. (K) Genes down- (‘DMSO gene score’, left) or upregulated (‘ribociclib gene score’, right) in BT690 and HSJD-GBM-002 following ribociclib treatment scored in patient DHG-H3G34 snRNA-seq data, plotted as 2D-representation of RG/NPC-like score, INPC/eIN-like vs. AC/MES-like scores. (L) Dotplot of gene set enrichment analysis for DHG-H3G34 scRNA-seq tumor signatures in differentially expressed genes after ribociclib treatment in BT690 and HSJD-GBM-002. Color scale: normalized enrichment score, dot size: -log10 of the BH-adjusted p-value. See also Figure S6.

Figure 7.. Efficacy of CDK4/6 pharmacological inhibition in DHG-H3G34 in vivo and in a patient with progressive DHG-H3G34.

(A) BLI quantification of animals with DHG-H3G34R BT690 tumors treated orally with 75 mg/kg abemaciclib (n=7) or vehicle (n=8) (*p-value <0.05, two-sided unpaired Student’s t-test). (B) BLI quantification of animals with DHG-H3G34R HSJD-GBM-002 tumors treated with 75 mg/kg abemaciclib or vehicle. Last datapoint represents 75 mg/kg abemaciclib (n=5) or vehicle (n=6), groups started with n=10 mice each. (C) BLI quantification of animals with DHG-H3G34R BT690 tumors treated orally with 200 mg/kg ribociclib (n=7) or vehicle (n=9). In (A)-(C), the y-axis denotes mean ± SEM of BLI signal measurements, the x-axis denotes the days post-treatment; *p-values <0.05; **p-values <0.01, two-sided unpaired t-test. (D) Kaplan-Meier analysis BT690 xenografts from (C). Treatment window is colored in grey. P-value=0.0002 for comparison between ribociclib (n=7) and vehicle (n=9), log-rank test. (E) Representative IF images following treatment with ribociclib or vehicle, stained for H3.3G34R (green), DCX (magenta), DAPI (blue), Nestin (red). Scales: top: 3 mm; bottom: 200 μm. (F) Quantification of DCX⁺ G34R⁺ tumor cells and Nestin⁺ G34R⁺ tumor cells in top: vehicle (n=4) vs. ribociclib (n=4)-treated tumors. Bottom: Quantification stratified by tumor rim vs. tumor core. P-values derived from an unpaired two-sided t-test. Data shown as mean ± SEM. (G) Timeline of disease course and treatment history in a 14-year-old female patient with recurrent DHG-H3G34. (H) MRI scans (top: axial T2 w/coronal T2-FLAIR, bottom: axial T1w post-Gadolinium) of DHG-H3G34 patient at initial diagnosis (i, ii), after complete resection (iii, iv), after first-line radiochemotherapy (v, vi),first recurrence (vii, viii), after two cycles of PCV (ix, x). The patient then received a third cycle of procarbazine/CCNU as a holding measure while securing access to ribociclib. The scans after two (xi, xii), six (xiii, xiv) and nine cycles (xv, xvi) of ribociclib showed disease stabilization, which was sustained until the scan after 18 cycles of ribociclib showed disease progression (xvii-xx). EOT: end of treatment; PCV: procarbazine, CCNU (lomustine), vincristine; RT: radiotherapy; TMZ: temozolomide. See also Figure S7.