CHD2 Regulates Neuron-glioma Interactions in Pediatric Glioma

Abstract

High-grade gliomas (HGG) are deadly diseases for both adult and pediatric patients. Recently, it has been shown that neuronal activity promotes progression of multiple subgroups of HGG. However, epigenetic mechanisms that govern this process remain elusive. Here we report that the chromatin remodeler CHD2 regulates neuron-glioma interactions in diffuse midline glioma (DMG) characterized by onco-histone H3.1K27M. Depletion of CHD2 in H3.1K27M DMG cells compromises cell viability and neuron-to-glioma synaptic connections in vitro, neuron-induced proliferation of H3.1K27M DMG cells in vitro and in vivo, activity-dependent calcium transients in vivo, and extends the survival of H3.1K27M DMG-bearing mice. Mechanistically, CHD2 coordinates with the transcription factor FOSL1 to control the expression of axon-guidance and synaptic genes in H3.1K27M DMG cells. Together, our study reveals a mechanism whereby CHD2 controls the intrinsic gene program of the H3.1K27M DMG subtype, which in turn regulates the tumor growth-promoting interactions of glioma cells with neurons.

Introduction

High-grade gliomas (HGG) are classified into different subtypes partly based on their genetic mutations and gene expression profiles (1). For instance, diffuse midline glioma H3K27-altered (DMG) is a deadly pediatric glioma subtype with few treatment options available and has a median survival of less than one year after diagnosis (2). Most DMG cases contain mutations at a histone H3 gene (H3K27M) (3–6). Of the H3K27M DMG tumors, about 80% carry the heterozygous mutations at the H3F3A gene, one of the two genes encoding histone H3.3, with the remaining 20% harboring heterozygous mutations at HIST1H3B or HIST1H3C, two of 13 genes encoding histone H3.1 (3,5,7). We and others have previously shown that the expression of H3.3K27M or H3.1K27M mutant proteins leads to a global reduction of H3K27me2 and H3K27me3 levels on wild-type histone H3, even though the H3K27M mutant proteins account for only 5–10% of the total histone H3 proteins in cancer cells (8–12). Furthermore, H3.1K27M and H3.3K27M DMG tumors represent two subtypes of pediatric HGG based on analyses of DNA methylation, transcriptional profiles, associated co-driver mutations, as well as potential cell of origin (7,13–21). Therefore, it is imperative to elucidate the epigenetic mechanisms governing the distinct gene expression profiles and to identify potential therapeutic targets for each HGG subtype.

Recently, it has been shown that neuronal activity promotes the proliferation and invasion of multiple subgroups of HGG (including DMG, adult and pediatric glioblastoma and anaplastic oligodendroglioma) via activity-dependent paracrine signaling and/or electrochemical signaling through the neuron-to-glioma synapses (22–30). For example, neuronal activity-regulated paracrine factors NLGN3 and BDNF boost DMG proliferation and their integration into neural network (22,25–27). Furthermore, electrochemical signaling through functional synapses between presynaptic neurons and postsynaptic tumor cells enhance DMG progression (22,27). All these studies utilized human HGG cells that were cultured alone in vitro prior to co-culture with neurons or implantation into the mouse brain, suggesting that the intrinsic gene expression program of these HGG cells is well preserved in vitro for their ability to interact with neurons (22–28). However, it is largely unknown whether and how epigenetic regulators control the intrinsic gene expression program of DMG cells for their interactions with neurons.

Chromodomain Helicase DNA Binding Protein 2 (CHD2) is a member of ATP-dependent CHD chromatin remodeling family (31,32). The canonical function of CHD2, like other chromatin remodelers, is to regulate chromatin accessibility (31,33). In addition, CHD2 is also involved in nucleosome assembly of histone H3.3 during gene transcription (34–36) and DNA repair (37). Heterozygous mutation of Chd2 in mice results in reduced proliferation of neural progenitors, deficits in synaptic transmission and impaired long-term memory (38), and increased susceptibility to lymphomas (39). Furthermore, CHD2 haploinsufficiency has been associated with a broad range of neurodevelopmental disorders in patients, such as intellectual disability, childhood epilepsies and autism spectrum disorder (40–46). These studies indicate that CHD2 plays important roles in neural development, likely through regulating the expression of genes involved in these processes. However, it was not known whether CHD2 has any roles in DMG tumors.

Through CRISPR/Cas9 screens, we found that H3.1K27M, but not H3.3K27M, DMG cells depend on chromatin remodeler CHD2 for cell fitness. Further exploration of CHD2’s role in DMG cells surprisingly revealed that CHD2 governs the gene expression and epigenome of H3.1K27M DMG cells for their interactions with neurons and for the neuron-induced proliferation of H3.1K27M DMG cells.

Results

CHD2 is essential for the viability of H3.1K27M DMG cells in vitro

To uncover the epigenetic mechanisms that govern the unique gene expression program of H3.1K27M DMG tumors, we first identified the epigenetic vulnerabilities of H3.1K27M DMG cells. Briefly, we performed CRISPR screens using a pooled library of single-guide RNAs (sgRNAs) targeting the functional domains of 180 chromatin regulators (47) (Fig. 1A; Supplementary Table S1) in seven human patient-derived glioma cell culture models, including one H3.1K27M DMG culture (SU-DIPG4), three H3.3K27M DMG cultures (SF8628, SU-DIPG6 and SU-DIPG17) and three glioblastoma (GBM) cultures without histone mutations (GBM6, GBM22 and GBM28). These screens revealed KAT2A, CHD2 and KDM4A as top hits for H3.1K27M DMG cells (Fig. 1A and 1B). Using GFP-based competition assays in which the sgRNA targeting a specific gene and GFP were co-expressed from the same plasmid, we confirmed that depleting CHD2, KAT2A and KDM4A via two independent sgRNAs for each reduced the fitness of two H3.1K27M DMG cultures (SU-DIPG4 and SU-DIPG36) (Supplementary Fig. S1A–S1D). As controls, cells expressing sgRNAs targeting the essential gene PCNA were outcompeted by those not expressing the sgRNA targeting PCNA in all four DMG cultures and two other glioma cultures (KNS42 and GBM22) analyzed, whereas cells expressing sgRNAs targeting the ROSA26 locus (sgNeg) had no apparent effects on all 6 glioma cultures. Because CHD2 depletion exhibited growth-inhibitory effects only in H3.1K27M DMG cells, but not other glioma cultures including two H3.3K27M DMG cells (Supplementary Fig. S1A–S1D), we focused our subsequent studies on CHD2, a member of the CHD family chromatin remodelers (32).

Figure 1. H3.1K27M DMG cells depend on CHD2 for cell viability in vitro.

A, Heatmap of the average log₂ fold-change of sgRNA abundance between T0 (3 days post infection) and T1 (10 population doublings after T0) in 7 glioma cultures from the CRISPR screens using an sgRNA library targeting the domains of 180 chromatin regulators. WT, wild type. Candidate genes were ranked by the specificity in H3.1K27M SU-DIPG4 cells compared to the rest 6 glioma cultures. B, Volcano plot showing normalized beta-score and negative log₁₀ transformed P-value based on MAGeCKFlute analysis of the CRISPR screens, with essential genes RPA3 and CDK1 and a top hit CHD2 highlighted. C-D, Effects of CHD2 depletion on the viability of two H3.1K27M (C) and two H3.3K27M (D) DMG cultures based on MTT cell viability assays (N = 4). E-F, Effects of CHD2 depletion on the ability of DMG cells to form colonies, with representative images (E) and quantification (F) shown (N = 6). G, Quantification of the ratio of apoptotic cells based on the annexin V apoptosis assays (N = 3). Error bars represent the mean ± SD. P values were calculated using two-tailed Student’s t-test. n.s., not significant; * P < 0.05, ** P < 0.01, **** P < 0.0001.

To further validate the results, we depleted CHD2 using two independent sgRNAs in two H3.1K27M DMG cultures (SU-DIPG4 and SU-DIPG36) and two H3.3K27M DMG cultures (SU-DIPG6 and SU-DIPG17) (Supplementary Fig. S1E) and performed cell viability and colony formation assays (Fig. 1C–1F). CHD2 depletion significantly impaired the viability and colony formation ability of two H3.1K27M, but not two H3.3K27M, DMG cultures (Fig. 1C–1F). Furthermore, depletion of CHD2 markedly increased apoptosis in H3.1K27M, but not H3.3K27M, DMG cells (Fig. 1G; Supplementary Fig. S1F and S1G). These cell-intrinsic effects in glioma monoculture likely do not stem from the impairment of DNA synthesis, as CHD2 depletion in H3.1K27M DMG cells did not affect the incorporation of a thymidine analog, EdU, in vitro (Supplementary Fig. S1H).

CHD2 expression is high in H3.1K27M DMG tumors compared to H3.3K27M DMG tumors and is required for the growth of H3.1K27M DMG xenografts in vivo

During course of the experiments, we observed that the level of CHD2 in each of two H3.1K27M DMG cultures was higher than that of two H3.3K27M DMG cultures (Fig. 2A). Therefore, we analyzed single-cell RNA-seq datasets generated from primary biopsy samples of pediatric HGG tumors (20) and found that the primary H3.1K27M DMG tumors also expressed significantly higher levels of CHD2 compared to H3.3K27M DMG tumors or pediatric HGG with wild type histone H3 (Fig. 2B). The higher levels of CHD2 were unlikely due to H3K27M mutation or ACVR1 mutation in H3.1K27M DMG cells (Supplementary Fig. S2A and S2B), suggesting that the high levels of CHD2 in H3.1K27M DMG cells compared to H3.3K27M DMG cells may be due to other factors such as cell of origin of these two DMG subtypes, which are thought to arise from oligodendroglial precursor cells at differing stages of immaturity (48). Nonetheless, these results indicate that the differential expression of CHD2 in H3.1K27M and H3.3K27M DMG may contribute to the differential impact of CHD2 depletion on cell fitness of these two DMG subtypes. Indeed, exogeneous expression of FLAG-tagged CHD2 in SU-DIPG4 cells fully rescued the defects in viability and colony formation caused by depletion of endogenous CHD2 (Fig. 2C–2E; Supplementary Fig. S2C and S2D), ruling out the off-target effects of sgCHD2–1 sgRNA.

Figure 2. H3.1K27M DMG cells depend on CHD2 for tumorigenicity in vivo.

A, Western blot analysis of CHD2 in the four DMG cultures used in this study. B, CHD2 expression in H3.1K27M DMG tumors (green, n = 11,644 cells, 7 patients) is higher than that in H3.3K27M DMG tumors (blue, n = 76,112 cells, 11 patients) or pediatric HGG with wild type histone H3 (pink, n = 9,581 cells, 4 participant patients) based on published single-cell RNA-seq data (20). C, Western blot analysis of CHD2 in SU-DIPG4 cells harboring empty vector or sgRNA-resistant CHD2-FLAG cDNA without (sgNeg) and with depletion of endogenous CHD2 (sgCHD2–1). D-E, Effects of depleting endogenous CHD2 on the viability of SU-DIPG4 cells harboring empty vector (D) or sgRNA-resistant CHD2-FLAG cDNA (E) (N = 3). F, Normalized luminescence levels from xenograft tumors over time in mice xenografted with luciferase-expressing SU-DIPG36 cells with CHD2 depletion (sgCHD2–1, n = 8 mice) and without CHD2 depletion (sgNeg, n = 7 mice). G, Normalized luminescence levels from xenograft tumor mass over time in mice xenografted with luciferase-expressing HSJD-DIPG-007 cells with CHD2 depletion (sgCHD2–1, n = 8 mice) and without CHD2 depletion (sgNeg, n = 8 mice). H-I, Kaplan-Meier survival curves of mice xenografted with SU-DIPG36 cells (H) or HSJD-DIPG-007 cells (I) without (sgNeg) and with CHD2 depletion (sgCHD2–1). Error bars represent the mean ± SD. P values in B were calculated by Wilcoxon test, with P values in D-G derived using two-tailed Student’s t-test and P value in H and I derived by log-rank test. n.s., not significant; *** P < 0.001.

Next, we asked whether CHD2 is specifically required for H3.1K27M DMG tumorigenicity in vivo. We depleted CHD2 in luciferase-expressing H3.1K27M SU-DIPG36 cells and H3.3K27M HSJD-DIPG-007 cells (Supplementary Fig. S2E), and then orthotopically injected them into the brainstem of mice, respectively. Bioluminescence imaging and subsequent tumor mass analysis revealed that CHD2 depletion specifically decelerated H3.1K27M, but not H3.3K27M DMG, tumor growth in vivo compared to controls (sgNeg) (Fig. 2F–2G; Supplementary Fig. S2F–S2I). Furthermore, relative to mice transplanted with control SU-DIPG36 cells, mice bearing CHD2-depleted SU-DIPG36 xenografts exhibited significantly extended survival compared to controls (Fig. 2H), but mice bearing CHD2-depleted HSJD-DIPG-007 xenografts did not show such effects (Fig. 2I). Thus, CHD2 is specifically required for the growth of H3.1K27M DMG xenografts in vivo.

CHD2 regulates the expression of axon-guidance and synaptic genes in H3.1K27M DMG cells

To investigate how CHD2 depletion affects the fitness of H3.1K27M DMG cells, we analyzed the effects of CHD2 depletion on transcriptomes using RNA-sequencing (RNA-seq) in two H3.1K27M (SU-DIPG4 and SU-DIPG36) and one H3.3K27M (SU-DIPG17) DMG cultures (Fig. 3A; Supplementary Fig. S3A and S3B). We identified 2,350 (1,119 upregulated and 1,231 downregulated genes) and 1,181 (413 upregulated and 768 downregulated genes) differentially expressed genes (DEGs) in CHD2-depleted SU-DIPG4 and SU-DIPG36 cells, respectively (Fig. 3A). However, we identified only 159 DEGs in CHD2-depleted H3.3K27M SU-DIPG17 cells (Supplementary Fig. S3B), providing a potential explanation for the minimal impact of CHD2 depletion on the viability of this DMG subtype (Fig. 1C–1F). Furthermore, 252 downregulated DEGs and 161 upregulated DEGs were shared between two H3.1K27M DMG cultures (Supplementary Fig. S3C). Gene ontology (GO) analysis showed that the 252 downregulated genes were enriched in axonogenesis, neuron projection and neuronal development (e.g., regulation of neural projection development and regulation of axonogenesis) (Fig. 3B; Supplementary Table S2). Gene set enrichment analysis (GSEA) also revealed that axon-guidance and synaptic genes (49) were downregulated in CHD2-depleted H3.1K27M DMG cells (Fig. 3C and 3D; Supplementary Fig. S3D), such as GABRA5 and GABRG3, genes encoding Gamma-aminobutyric acid (GABA) receptor subunits (Supplementary Fig. S3E), concordant with the recent description of GABAergic synapses in H3K27M DMG (bioRxiv 2022.11.08.515720). Importantly, the expression of 44 of these downregulated axon-guidance and synaptic genes showed significant correlations with CHD2 expression in 37 patient-derived primary H3.1K27M DMG tumors (Fig. 3E and 3F). However, the correlation between the expression levels of these 44 genes and that of CHD2 was much weaker in the 116 primary H3.3K27M DMGs and 97 primary H3 wild-type pediatric HGG cases (Fig. 3F; Supplementary Fig. S3F). Finally, using RT-qPCR, we validated the effects of CHD2 depletion on the expression of seven randomly selected genes in both H3.1K27M and H3.3K27M DMG cultures (Supplementary Fig. S3G). These findings indicate that CHD2 specifically regulates the expression of axon-guidance and synaptic genes in H3.1K27M DMG cells, and likely in H3.1K27M DMG primary tumors as well.

Figure 3. CHD2 binds and regulates axon-guidance and synaptic genes in H3.1K27M DMG cells.

A, Volcano plots of differentially expressed genes (DEGs) in two H3.1K27M DMG cultures upon CHD2 depletion based on RNA-seq (N = 2). B, GO analysis of the 252 downregulated genes in both SU-DIPG4 and SU-DIPG36 cells upon CHD2 depletion (sgCHD2), with axonogenesis and neural development-related terms highlighted. C-D, GSEA analysis for the effects of CHD2 depletion on the expression of synaptic genes in SU-DIPG4 cells (C) and SU-DIPG36 cells (D). NES, normalized enrichment score. E, Correlation heatmap illustrating the Pearson coefficients of expression levels among CHD2 and 44 CHD2-regulated axon-guidance and synaptic genes in the 37 patient-derived primary H3.1K27M DMG tumors. The bulk RNA-seq data used in this analysis are obtained from Suzy Baker, Alan Mackay (2017) (18), and Selin Jessa (2022) (20). The CHD2-regulated axon-guidance and synaptic genes whose expression exhibit a statistically significant correlation with CHD2 expression (p < 0.01) in H3.1K27M DMG tumors, have been selected for this analysis. F, Boxplots depicting the distribution of Pearson coefficients between CHD2 expression and the expression of the 44 CHD2-regulated axon-guidance and synaptic genes in E. G, Genomic distribution of CHD2 CUT&RUN peaks in SU-DIPG4 and SU-DIPG17 cells. H, Heatmap of CHD2 CUT&RUN signals at each CHD2 CUT&RUN peak identified from two biological repeats (rep 1 and rep 2) in SU-DIPG4 cells. CHD2 CUT&RUN peaks were categorized as promoters, enhancers, or other regions based on their colocalizations with H3K4me3 and H3K27ac CUT&RUN peaks, with the average CUT&RUN signals at each group shown at right. I, Representative genome tracks showing CHD2 CUT&RUN signals and RNA-seq signals at EPHB3 (an axon guidance gene) and NGEF and NEFL (two synaptic genes) in SU-DIPG4 cells. Y-axis indicates reads per million reads (RPM). J, Effects of EPHB3 depletion (sgEPHB3) on the viability of H3.1K27M SU-DIPG4 cells (N = 3). K-L, Effects of EPHB3 depletion in SU-DIPG4 cells on the ability to form colonies, with representative images (K) and quantification (L) shown (N = 3). Error bars represent the mean ± SD. P values were calculated using two-tailed Student’s t-test. ** P < 0.01, *** P < 0.001.

CHD2 directly binds to axon-guidance and synaptic genes

To explore how CHD2 regulates the expression of axon-guidance and synaptic genes, we first performed CHD2 CUT&RUN (50) in SU-DIPG4 (H3.1K27M) and SU-DIPG17 (H3.3K27M) cells and analyzed CHD2 chromatin occupancy genome-wide. We identified 13,960 and 16,257 CHD2 CUT&RUN peaks in SU-DIPG4 and SU-DIPG17 cells, respectively (Fig. 3G), with only 4,259 peaks shared between these two DMG cultures. The majority of the CHD2 CUT&RUN peaks in SU-DIPG4 or SU-DIPG17 cells were localized at promoters, introns, and distal intergenic regions which are likely enhancers (Fig. 3H and Supplementary Fig. S3H). Indeed, based upon their overlap with H3K4me3 (a mark for promoters) and H3K27ac (a mark for active enhancers), we estimated that 63.8% (8,912/13,960) and 30.2% (4,209/13,960) of the CHD2 CUT&RUN peaks in SU-DIPG4 cells were located at promoters and enhancers, respectively (Fig. 3H). Similar analysis on CHD2 CUT&RUN peaks in SU-DIPG17 cells also revealed its predominant distributions at promoters and enhancers (Supplementary Fig. S3H). Because CHD2 depletion led to a broad transcriptional dysregulation only in H3.1K27M DMG cells (Fig. 3A; Supplementary Fig. S3B), CHD2 likely regulates gene expression in a context-dependent manner.

Next, we integrated RNA-seq datasets upon CHD2 depletion with CHD2 CUT&RUN datasets in unperturbed SU-DIPG4 cells (H3.1K27M) and SU-DIPG17 cells (H3.3K27M), and found 607 genes specifically affected by CHD2 depletion in SU-DIPG4 cells, including 266 downregulated genes and 341 upregulated genes, with at least one adjacent CHD2 binding peak specifically in SU-DIPG4 cells (Supplementary Fig. S3I). GO analysis for these CHD2-bound and downregulated genes upon CHD2 depletion prioritized the terms related to axon guidance (e.g., axon development) and synapse (e.g., GABA-ergic synapse) (Supplementary Fig. S3J and S3K). For instance, EPHB3 (a gene involved in axon guidance and excitatory synapse formation, and previously found to be associated with super-enhancers in H3K27M DMG) (48,51–53) had CHD2 CUT&RUN peaks nearby and were downregulated upon CHD2 depletion (Fig. 3I). CHD2 CUT&RUN-qPCR also confirmed the CHD2 binding at five synaptic genes specifically in H3.1K27M DMG (SU-DIPG4 and SU-DIPG36) cells (Supplementary Fig. S3L). Importantly, EPHB3-depleted H3.1K27M DMG cells displayed significantly impaired viability and colony formation ability in vitro (Fig. 3J–3L; Supplementary Fig. S3M). Together, these results indicate that CHD2 directly controls the expression of axon-guidance and synaptic genes in H3.1K27M DMG cells and some of these genes most likely also play an important role in the fitness of gliomas cells cultured alone.

CHD2 is critical for neuron-induced proliferation of H3.1K27M DMG cells in co-culture

Recently, it has been shown that genes involved in axon guidance such as SEMA4F regulate neuron-glioma interactions and promote glioma invasion (54). To further probe the functional implications of CHD2 in controlling the expression of axon-guidance and synaptic genes, we asked whether CHD2 regulates neuron-glioma interactions. First, we assessed the effects of CHD2 depletion on neuron-induced glioma proliferation using the established neuron-glioma co-culture system. Consistent with the previous studies (22,27), the proliferation of both H3.1K27M (SU-DIPG4, SU-DIPG36) and H3.3K27M (SU-DIPG6) DMG cells increased dramatically when co-cultured with neurons compared to the corresponding DMG cells cultured alone, with the ratio of EdU-positive DMG cells increasing from the baseline ~15% to 40–60% upon co-culture (Fig. 4A and 4B; Supplementary Fig. S4A–S4C). CHD2 depletion in DMG cells cultured alone did not affect the ratio of proliferating cells (Fig. 4B; Supplementary Fig. S4B and S4C). Surprisingly, CHD2 depletion in H3.1K27M cells co-cultured with neurons, but not H3.3K27M cells co-cultured with neurons, resulted in a drastic reduction in neuron-induced proliferation, decreasing the EdU-positive DMG cells from ~50% to ~20% (Fig. 4B). As expected (22), treatment with NBQX, an antagonist against AMPA receptors that mediates neuron-to-glioma synaptic signaling, dramatically compromised neuron-induced proliferation of H3.3K27M DMG cells in co-culture (Fig. 4B). Thus, CHD2 is required for neuron-induced proliferation of H3.1K27M DMG cells in the in vitro neuron-glioma co-culture system.

Figure 4. CHD2 regulates neuron-induced proliferation of H3.1K27M DMG cells in co-culture.

A-B, Representative image (A) and proliferation index (B) for the impact of CHD2 depletion on the neuron-induced proliferation of H3.1K27M SU-DIPG4 cells (B, left), H3.1K27M SU-DIPG36 cells (B, middle) and H3.3K27M SU-DIPG6 cells (B, right). Red denotes proliferating cells in EdU Click-iT reaction; green denotes DMG cells in H3K27M staining. Proliferation index was calculated as the percentage of EdU positive DMG cells (identified by EdU+ & H3K27M+) over total number of DMG cells (identified by H3K27M+) (n > 1,600 cells, 4 coverslips, per group). Scale bar, 50 μm. Treatment with 10 μM NBQX was used as the control. C, Representative confocal image of PSD95-RFP labeled wild-type (sgNeg) SU-DIPG4 cells co-cultured with neurons. White box and magnified view (right) highlight the region of synaptic puncta colocalization. All SU-DIPG4 cells expressed GFP. Red and pink denote PSD95-RFP staining (postsynaptic puncta) from DMG cells and Synapsin (presynaptic puncta) from neurons, respectively. Arrow depicts the colocalization between Synapsin and PSD95-RFP in a SU-DIPG4 cell. Scale bars, 10 μm (left) and 2 μm (right). D, Quantification of the colocalizations between presynaptic Synapsin from neurons and postsynaptic PSD95-RFP from wild-type (sgNeg) or CHD2-depleted (sgCHD2) SU-DIPG4 cells (left), SU-DIPG36 (middle) and SU-DIPG6 cells (right) co-cultured with neurons (n > 210 cells, 3–4 coverslips, per group). E, Representative GCaMP6s fluorescence images from calcium imaging in SU-DIPG36 cells co-cultured with neurons. Scale bar, 50 μm. F, Comparison of the log₁₀ value of calcium peaks amplitude in SU-DIPG36 cells cultured alone and co-cultured with neurons (n > 1,500 peaks, > 1,000 cells, 22 recordings from 3 biological repeats per group). G, Frequency of calcium peaks in SU-DIPG36 cells cultured alone and co-cultured with neurons. Calcium peaks with the amplitude (ΔF_max/F0) more than 0.1 are used in this analysis (n = 22 recordings from 3 biological repeats per group). H, Representative confocal images showing the tumor microtubes of SU-DIPG4 cells co-cultured with neurons. Scale bars, 50 μm. I-J, Numbers (I) and length (J) of membrane protrusions in SU-DIPG4 cells co-cultured with neurons (n > 400 cells, 4 coverslips per group in I; n > 800 protrusions, 4 coverslips per group in J). Error bars in all panels represent the mean ± SD. P values were calculated using two-tailed Student’s t-test. n.s., not significant; * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001.

Neurons regulate the proliferation of multiple subgroups of HGG through paracrine signaling and direct neuron-to-glioma synapses (22–30). To understand the mechanism by which CHD2 controls neuron-induced proliferation of H3.1K27M DMG cells, we examined neuron-to-glioma synaptic connections in co-culture using DMG cells expressing RFP-tagged postsynaptic density protein-95 (PSD95). Consistent with the previous studies (22,27), we observed the co-localization of neuronal presynaptic puncta (Synapsin) with DMG postsynaptic puncta (PSD95-RFP) (Fig. 4C and 4D), which indicates the formation of neuron-to-glioma synapses (22,27). Importantly, CHD2 depletion in H3.1K27M DMG cells (SU-DIPG4 and SU-DIPG36) co-cultured with neurons dramatically reduced the number of neuron-to-glioma synaptic structures (Fig. 4D). In contrast, CHD2 depletion in H3.3K27M SU-DIPG6 cells had minimal impact on the number of neuron-to-glioma synaptic structures formed between these cells and neurons in co-culture (Fig. 4D), revealing an important role of CHD2 in regulating neuron-glioma synapse formation specifically in H3.1K27M DMG cells in vitro.

We further examined the neuron-glioma interaction by monitoring calcium transients in DMG cells expressing the calcium indicator GCaMP6s (55). SU-DIPG36 cells co-cultured with neurons exhibited increased amplitude and frequency of spontaneous calcium transients than these cells cultured alone (Fig. 4E–4G). Importantly, CHD2 depletion significantly impaired the amplitude and frequency of spontaneous calcium transients in H3.1K27M DMG cells co-cultured with neurons, whereas it had no apparent effects on the basal level of calcium transients in these cells cultured alone (Fig. 4F and 4G; Supplementary Fig. S4D and S4E). These results suggest the important role of CHD2 in regulating the neuron-glioma interaction for H3.1K27M DMG cells. We also observed periodic calcium transients in a small subpopulation of H3.1K27M DMG cells co-cultured with neurons (Supplementary Fig. S4F), consistent with previous observations of “pacemaker” glioma cells exhibiting autonomous, periodic calcium transients in glioblastoma (56) and in H3K27M DMG (22). It has been shown that glioblastoma subpopulations with periodic calcium transients are extensively connected to other tumor cells through tumor microtubes, long cell processes that couple glioma cells through gap junctions (56). Like glioblastoma, tumor microtubes also connect H3K27M DMG cells (22) and tumor microtubes-associated genes (26) and process extension (24) are upregulated by interactions with neurons. Indeed, we found that both SU-DIPG4 and SU-DIPG36 DMG cells co-cultured with neurons, but not cultured alone, exhibited tumor microtubes with various length (Fig. 4H and 4I; Supplementary Fig. S4G–S4I). Importantly, CHD2 depletion in both SU-DIPG4 and SU-DIPG36 cells significantly reduced the ratio of DMG cells with tumor microtubes as well as the length of the tumor microtubes when co-cultured with neurons (Fig. 4I and 4J; Supplementary Fig. S4H and S4I). Taken together, these results suggest that CHD2 depletion in H3.1K27M DMG cells impaired neuron-to-glioma synaptic connections, calcium transients and tumor microtubes of tumor cells in the in vitro co-culture settings.

Finally, we observed that depletion of EPHB3, a CHD2 regulated gene, in H3.1K27M SU-DIPG4 cells also dramatically reduced the neuron-induced proliferation of these cells (Supplementary Fig. S4J). Interestingly, EPHB3 depletion, unlike CHD2 depletion, also slightly reduced the EdU positive SU-DIPG4 cells when cultured alone (Supplementary Fig. S4J). These results support the idea that CHD2 regulates axon-guidance and synaptic genes in H3.1K27M DMG cells, which in turn regulates the neuron-glioma interactions.

CHD2 is critical for neuron-induced proliferation of H3.1K27M DMG xenografts in vivo

Next, we asked whether CHD2 regulates neuron-glioma interactions in vivo. We stereotactically xenografted GCaMP6s-expressing SU-DIPG36 DMG cells into mouse hippocampal CA1 region. After a period of engraftment, we performed proliferation assays in fixed hippocampal slices and in situ calcium imaging in acute hippocampal slices, respectively. Within the hippocampus, SU-DIPG36 DMG xenografts exhibited a proliferation index (ratio of Ki67 positive cells over HNA positive cells) of approximately 18% (Fig. 5A and 5B). Remarkably, CHD2 depletion significantly reduced xenografted glioma cell proliferation, with a decreased proliferation index of approximately 6% (Fig. 5B). This result is in contrast with in vitro observations that depletion of CHD2 had no apparent effects on EdU positive cells in DMG cells cultured alone (Fig. 4B and Supplementary Fig. 1H), consistent with the idea that CHD2 is required for neuron-induced proliferation of H3.1K27M DMG cells. We next performed in situ calcium imaging in GCaMP6s-expressing SU-DIPG36 DMG xenografts. We consistently observed active spontaneous calcium transients in these patient-derived DMG xenografts (Fig 5C and 5D). CHD2 depletion significantly reduced the amplitude and frequency of the spontaneous calcium transients in SU-DIPG36 DMG xenografts (Fig. 5E and 5F; Supplementary Fig. S5A and S5B). We then performed electrical stimulation of Schaffer collaterals and recorded the stimulus-evoked calcium transients of SU-DIPG36 xenografts at the hippocampal CA1 region (Fig. 5G). Successful axonal stimulation was verified by the consistently observed field excitatory postsynaptic potentials (fEPSP) in CA1 (Supplementary Fig. S5C). The stimulus-evoked calcium transients in DMG were blocked by Tetrodotoxin citrate (TTX) treatment (Fig. 5H and 5I; Supplementary Fig. S5D), indicating that the calcium transients in DMG cells were induced by the stimulus-evoked neuronal activity and that DMG cells and neurons were functionally connected through neuron-to-glioma synapses and/or activity-regulated potassium currents (22). Importantly, CHD2 depletion significantly reduced the amplitude of the stimulation-induced calcium transients (Fig. 5J). Taken together, these results suggest that CHD2 is required for the synaptic and electrical integration of H3.1K27M DMG xenografts into neuronal circuits in vivo.

Figure 5. CHD2 regulates neuron-induced proliferation of H3.1K27M DMG xenografts in vivo.

A-B, Representative confocal image (A) and proliferation index (B) for the effects of CHD2 depletion (sgCHD2) on the proliferation of SU-DIPG36 xenografts in hippocampal CA1. Human nuclei antigen (HNA, green) marks glioma cells and Ki67 (red) marks proliferating cells (n > 2,800 cells, 16 slices, 4 mice, per group). C-D, Representative GCaMP6s fluorescence images (C) and calcium trace (D) from in situ GCaMP6s signal imaging in mice brain slices with SU-DIPG36 xenografts in hippocampal CA1. E, Dot plots for log₁₀ value of calcium peaks amplitude in mice brain slices containing wild-type (sgNeg) or CHD2-depleted (sgCHD2) SU-DIPG36 xenografts (sgNeg, n = 1,190 peaks, 10 slices, 4 mice; sgCHD2, n = 1,254 peaks, 9 slices, 4 mice). F, Frequency of calcium peaks in mice brain slices containing wild-type (sgNeg) or CHD2-depleted (sgCHD2) SU-DIPG36 xenografts. Calcium peaks with the amplitude (ΔF_max/F0) more than 0.1 are used in this analysis (sgNeg, n = 10 slices, 4 mice; sgCHD2, n = 9 slices, 4 mice). G, A diagram showing the calcium imaging in coronal slices with SU-DIPG36 xenografts in hippocampal CA1 and with electrical stimulation of Schaffer collaterals. Created with BioRender.com. H-I, Representative GCaMP6s fluorescence images (H) and calcium traces (I) from in situ GCaMP6s signal imaging in mice brain slices with SU-DIPG36 xenografts in hippocampal CA1 and with electrical stimulation of Schaffer collaterals. 1 μM TTX was used as a control. J, Quantification of calcium peaks amplitude in mice brain slices containing wild-type (sgNeg) or CHD2-depleted (sgCHD2) SU-DIPG36 xenografts in hippocampal CA1 and with electrical stimulation of Schaffer collaterals (sgNeg, n = 63 cells, 10 slices, 4 mice; sgCHD2, n = 91 cells, 10 slices, 4). Scale bars, 50 μm. Error bars represent the mean ± SD. P values were calculated using two-tailed Student’s t-test.

CHD2 colocalizes with FOSL1 to regulate gene expression in H3.1K27M DMG cells

To determine how CHD2 is recruited to specific chromatin regions in H3.1K27M DMG cells, we performed de novo motif analysis using the 9,701 CHD2 CUT&RUN peaks found in SU-DIPG4 but not SU-DIPG17 cells. Surprisingly, the top seven motifs all belonged to members of the AP-1 family of transcription factors with FOSL1 binding motif at the top of the list (Fig. 6A). To unbiasedly evaluate the roles of transcription factors in H3.1K27M DMG cells, we performed CRISPR dropout screens using a pooled library of sgRNAs targeting 1,433 transcription factors in H3.1K27M SU-DIPG4, H3.3K27M SU-DIPG17, GBM43 (wild-type histone H3) and H3.3G34V KNS42 cells. Remarkably, we identified three members of AP-1 transcription factors (FOSL1, JUN and JUNB) as the top dependencies of H3.1K27M SU-DIPG4 cells compared to other glioma cells (Fig. 6B). In contrast, SOX10, a master transcription factor in oligodendrocyte precursor cells(57), is a specific dependency of H3.3K27M SU-DIPG17 cells (Fig. 6B), consistent with our previous study showing the critical role of SOX10 in H3.3K27M DMG cells (58). These results reveal differential dependencies of H3.1K27M and H3.3K27M DMG cells on transcription factors, likely reflecting the differential gene expression profiles between these two DMG subtypes and their different origins within the oligodendroglial lineage.

Figure 6. CHD2 colocalizes with FOSL1 to regulated axon-guidance and synaptic genes.

A, De novo motif analysis of the 9,701 CHD2 CUT&RUN peaks unique to H3.1K27M SU-DIPG4 cells. Motifs are ranked by enrichment score (ES) with the top seven motifs shown. B, Heatmap showing the beta-score of top transcription factor dependencies of H3.1K27M SU-DIPG4 cells compared to three other glioma cultures from CRISPR screens using a sgRNA library targeting 1,433 transcription factors. C, Venn diagram showing the overlap of CHD2 and FOSL1 CUT&RUN peaks in SU-DIPG4 cells. D, Representative genome tracks showing the colocalization of CHD2 and FOSL1 CUT&RUN signals from two independent repeats (rep1 and rep2) at NGEF and NEFL loci. Y-axis indicates reads per million reads (RPM). E, CHD2 interacts with FOSL1 in SU-DIPG4 cells using co-immunoprecipitation assays (N = 3). F, Heatmaps (left) and average (right) of CHD2 CUT&RUN signals at each CHD2 CUT&RUN peak in SU-DIPG4 cells without (sgNeg) and with FOSL1 depletion (sgFOSL1). G, GO analysis of the 296 downregulated genes shared between CHD2- and FOSL1-depleted SU-DIPG4 cells. H, Representative genome tracks showing the CHD2 and FLOS1 CUT&RUN and RNA-seq signals at ERBB4 and NGEF loci in SU-DIPG4 cells with/without CHD2 or FOSL1 depletion. Y-axis indicates reads per million reads (RPM). I, Single-cell RNA-seq data from primary pediatric HGG tumors illustrating significantly higher expression of RGMA and ERBB4 in H3.1K27M DMG (green, n = 11,644 cells, 7 patients) than that in H3.3K27M DMG (blue, n = 76,112 cells, 11 patients) or pediatric HGG with wild type histone H3 (pink, n = 9,581 cells, 4 patients). J, Proliferation index for the effects of FOSL1 depletion on neuron-induced proliferation of H3.1K27M DMG cells (n > 1,600 cells, 4 coverslips, per group). Error bars represent the mean ± SD. P values in I were calculated using Wilcoxon test, with P values in J derived using two-tailed Student’s t-test. n.s., not significant; **** P < 0.0001.

Next, we performed FOSL1 CUT&RUN in H3.1K27M SU-DIPG4 cells and identified 33,969 FOSL1 CUT&RUN peaks from two independent repeats (Supplementary Fig. S6A and S6B). Moreover, most of these peaks are located at promoters (39.4%, 13,374/33,969) and enhancers (36.3%, 12,335/33,969) (Supplementary Fig. S6C). Importantly, 76.8% (10,719/13,960) of CHD2 CUT&RUN peaks overlapped with FOSL1 CUT&RUN peaks in SU-DIPG4 cells (Fig. 6C), such as peaks at the gene loci of EPHB3, NGEF and NEFL (Fig. 6D; Supplementary Fig. S6D). These results imply that FOSL1 helps recruit CHD2 to the gene regulatory elements. Supporting this idea, we found that endogenous CHD2 co-immunoprecipitated with FOSL1 in SU-DIPG4 cells (Fig. 6E), and furthermore, FOSL1 depletion led to a global decrease of CHD2 binding (Fig. 6F; Supplementary Fig. S6E and S6F). Thus, FOSL1 interacts with CHD2 and helps recruit CHD2 to gene regulatory elements in H3.1K27M DMG cells.

Effects of FOSL1 depletion on the expression of synaptic genes and neuron-induced proliferation

To further investigate the roles of FOSL1 in DMG cells, we analyzed the effects of FOSL1 depletion on the fitness of H3.1K27M and H3.3K27M DMG cells. Like CHD2 depletion, FOSL1 depletion resulted in reduced cell viability and colony formation in two H3.1K27M DMG cultures (SU-DIPG4 and SU-DIPG36), while showing limited effects on two H3.3K27M DMG cultures (Supplementary Fig. S6G–S6J). These results show that like CHD2, FOSL1 is also important for the fitness of H3.1K27M DMG cells in vitro.

Gene expression analysis using RNA-seq revealed that FOSL1 depletion in SU-DIPG4 cells resulted in upregulation of 2,310 genes and downregulation of 1,498 genes. Importantly, we observed that a significant number of genes (423 upregulated genes and 296 downregulated genes, respectively) were shared between CHD2 depletion and FOSL1 depletion (Supplementary Fig. S6K). GO analysis and KEGG pathway analysis of the 296 shared downregulated genes indicated the enrichment of genes involved in axonogenesis (e.g., neuron projection regeneration, axon regeneration, axon guidance) (Fig. 6G; Supplementary Fig. S6L; Supplementary Table S3). Like CHD2 depletion, FOSL1 depletion also led to the downregulation of synaptic genes (Fig. 6G and 6H; Supplementary Fig. S6M). Of the 42 downregulated axon-guidance and synaptic genes shared between FOSL1 depletion and CHD2 depletion, expression levels of 14 genes were markedly higher in primary H3.1K27M HGG tumors than either H3.3K27M DMG or other pediatric HGG tumors with wild-type histone H3 based on published single-cell RNA-seq datasets (20) (Fig. 6I; Supplementary Fig. S6N). These results suggest that the FOSL1-CHD2 axis regulates the expression of axon-guidance and synaptic genes, which in turn promotes neuron-glioma interaction. Consistent with this idea, FOSL1 depletion in SU-DIPG4 cells resulted in a marked reduction in neuron-induced proliferation, while having no detectable effects on the proliferation of these cells when cultured alone (Fig. 6J). Therefore, FOSL1 and CHD2 work together to govern the neuron-glioma interactions and tumor progression in H3.1K27M DMG.

Depletion of CHD2 and FOSL1 alters histone modifications

To gain insight into how CHD2 and FOSL1 regulate gene expression in H3.1K27M DMG cells, we analyzed histone modifications in wild-type, CHD2- and FOSL1-depleted DMG cells. CHD2 or FOSL1 depletion resulted in a global increase in H3K27me3 levels in two H3.1K27M DMG cultures, with no increase in the H3.3K27M DMG culture tested (Fig. 7A). The increase of H3K27me3 in SU-DIPG4 cells upon CHD2 depletion was confirmed by analysis of modifications on histones purified by acid-extraction (Fig. 7B). Furthermore, CHD2 or FOSL1 depletion resulted in a dramatic reduction of H3K27ac, while having little impact on three other modifications (H3K4me3, H3K36me2 and H3K36me3) globally (Fig. 7B). Because the expression of three core subunits of PRC2 complex (EZH2, EED and SUZ12) and two H3K27 demethylases (JMJD3 and UTX) was not affected by CHD2 or FOSL1 depletion (Supplementary Fig. S7A), the increase in H3K27me3 in H3.1K27M DMG cells is unlikely due to changes in the expression of the proteins involved in writing and erasing H3K27me3.

Figure 7. CHD2 and FOSL1 regulate histone modifications in H3.1K27M DMG cells.

A, Effects of CHD2 or FOSL1 depletion on H3K27me3 levels based on Western blot analysis of proteins in whole-cell lysates (N = 3). B, Western blot analysis of indicated histone modifications and H3K27M using acid-extracted histones (N = 3). C-E, Effects of CHD2 depletion on H3K27me3 (C), H3K27ac (D) and H3K4me3 (E) on chromatin in SU-DIPG4 cells, with heatmaps representing CUT&RUN signals at each peak for each histone mark and the average of CUT&RUN signals of all peaks of each histone mark shown at right (N = 2). F, Boxplots showing log₂ fold-change (LFC) of H3K27me3 (left), H3K27ac (middle) and H3K4me3 (right) CUT&RUN signals at indicated group of genes upon CHD2 depletion. The DEGs with CHD2 binding at their promoters in wild-type SU-DIPG4 cells were used for analysis with the number of DEGs shown (N = 2). G, GO analysis of the 115 downregulated genes with increased H3K27me3 signal upon CHD2 depletion, with those synaptic program-related terms highlighted. H, A model for the roles of CHD2 and FOSL1 in regulating chromatin dynamics and gene expression, which in turn controls neuron-glioma interactions and tumor progression. Created with BioRender.com.

Next, we analyzed the effects of CHD2 depletion on three key histone modifications (H3K27me3, H3K27ac and H3K4me3) on chromatin in H3.1K27M SU-DIPG4 cells using CUT&RUN. Because of the substantial changes in H3K27me3 and H3K27ac levels, we spiked-in mouse DMG cells for normalization. We identified 5,321 and 6,704 H3K27me3 peaks in wild type and CHD2-depleted SU-DIPG4 cells, respectively (Fig. 7C). Furthermore, H3K27me3 levels increased at 77.5% (5,173/6,672) of H3K27me3 peaks in CHD2-depleted SU-DIPG4 cells (Fig. 7C). Conversely, CHD2 depletion resulted in reduced H3K27ac and H3K4me3 levels at 31.1% (7,604/24,449) and 25.8% (7,875/30,483) of H3K27ac and H3K4me3 peaks, respectively (Fig. 7D and 7E). Importantly, the CHD2 CUT&RUN density at the genomic loci with increased H3K27me3 levels, or decreased H3K27ac and H3K4me3 levels upon CHD2 depletion were significantly higher than those without the corresponding changes in each histone modification (Supplementary Fig. S7B), suggesting that the changes in these three histone marks likely originate from the loss of CHD2 binding upon CHD2 depletion. We also analyzed the effects of FOSL1 depletion on these histone modifications in SU-DIPG4 cells. We found that FOSL1 depletion resulted in increased H3K27me3 at 76.1% (3,252/4,276) of H3K27me3 peaks, and reduced H3K27ac and H3K4me3 at 41.8% (11,462/27,413) and 34.5% (10,418/30,163) of their total peaks, respectively (Supplementary Fig. S7C–S7E). Interestingly, a majority of H3K27me3 peaks with increased signals as well as a large fraction of H3K27ac and H3K4me3 peaks with reduced signals were shared between CHD2 and FOSL1 (Supplementary Fig. S7F). Finally, FOSL1 CUT&RUN density prior to FOSL1 depletion at the promoters of genes with increased H3K27me3 and reduced H3K27ac were higher than those with the corresponding changes upon FOSL1 depletion (Supplementary Fig. S7G), whereas FOSL1 CUT&RUN density at promoters of genes with and without changes of H3K4me3 were similar (Supplementary Fig. S7G). These results indicate that CHD2 and FOSL1 are important to regulate H3K27me3 and H3K27ac levels in H3.1K27M DMG cells.

To determine whether the changes in gene expression upon CHD2 or FOSL1 depletion were associated with histone modification changes, we separated genes into three groups (upregulated, unchanged, and downregulated) based on the effects of CHD2 or FOSL1 depletion and analyzed the changes of the three histone modifications within each group in SU-DIPG4 cells. Overall, the changes in each histone modification at the promoters of down-regulated genes correlated well with gene expression changes. Specifically, upon CHD2 depletion, the abundance of H3K27me3, a mark critical for gene silencing, at the promoters of downregulated genes increased compared to that at unchanged genes (Fig. 7F), whereas the levels of H3K4me3 and H3K27ac, two marks associated with active gene transcription, decreased at these chromatin regions (Fig. 7F). Similarly, upon FOSL1 depletion, the levels of H3K27ac and H3K4me3 were also reduced at downregulated genes compared with those at unchanged genes (Supplementary Fig. S7H). At the promoters of upregulated genes, depletion of either CHD2 or FOSL1 led to a notable increase in H3K27ac and H3K4me3 compared with those at unchanged genes (Fig. 7F and Supplementary Fig. S7H). However, we also observed an increase in the levels of H3K27me3 at the promoters of this group of genes (Fig. 7F and Supplementary Fig. S7H). The lack of correlation of H3K27me3 changes at genes with increased expression upon CHD2 or FOSL1 depletion likely reflects the global increase in H3K27me3 upon CHD2 or FOSL1 depletion. Importantly, GO analysis showed that the downregulated genes with increased H3K27me3 were enriched in axonogenesis and regulation of neuron projection (Fig. 7G). Similarly, GO analysis using those downregulated genes with reduced H3K27ac or H3K4me3 also indicated their enrichment in axonogenesis and neuron projection processes (Supplementary Fig. S7I and S7J). Taken together, these results indicate that CHD2 controls the gene expression in H3.1K27M DMG cells, at least in part, by regulating histone modifications including H3K27me3, H3K27ac and H3K4me3 levels at the promoters of these genes.

Discussion

Previous studies have established that synaptic and electrical integration of gliomas into neural circuits drives glioma progression (22–24,27). However, how chromatin regulators govern this process is largely unknown. We discovered the FOSL1-CHD2 axis as a crucial regulator for the expression of axon-guidance and synaptic genes in H3.1K27M, but not H3.3K27M DMG cells, thereby mediating the neuron-glioma interactions for H3.1K27M DMG cells specifically (Fig. 7H). Because H3.1K27M DMG cells used in these studies were cultured alone in the absence of neurons over several passages, the requirement of FOSL1 and CHD2 for their ability to interact with neurons indicates the CHD2-FOSL1 axis regulates the intrinsic gene expression program of this glioma subtype for the neuron-glioma interactions and for tumor progression.

CHD2 regulates the expression of axon-guidance and synaptic genes in H3.1K27M DMG cells, which in turn controls the neuron-glioma interactions.

Neurons are one of the dominant cell types in the microenvironment of glioma cells. Recently, it has been shown that neurons promote the proliferation and progression of multiple subgroups of HGG via both paracrine signaling pathways and neuron-glioma synaptic integration (22–30). Here, we provide multiple lines of evidence to support the idea that CHD2 regulates the expression of axon-guidance and synaptic genes specifically in H3.1K27M DMG cells, which in turn controls neuron-glioma interactions for this HGG subtype. First, depletion of CHD2 impairs the neuron-to-glioma synaptic connections in vitro, decreases activity-dependent calcium transients and compromises the neuron-induced proliferation of H3.1K27M DMG in vitro and in vivo, but not H3.3K27M DMG cells. Second, we found that CHD2 binds and regulates the expression of genes involved in axon guidance. For instance, EPHB3, an EPH receptor with known roles in axon guidance and excitatory synapse formation (51–53) and previously implicated in H3K27M DMG pathophysiology (48), was specifically downregulated in H3.1K27M DMG cells upon CHD2 depletion. Furthermore, depletion of EPHB3 abrogated neuron-induced proliferation of H3.1K27M DMG cells in vitro. Interestingly, inspection of RNA-seq datasets indicated that CHD2 depletion did not affect the expression of genes including NLGN3, BDNF, TRKB, IGF1, GPC3 and TSP-1 in H3.1K27M DMG cells, which are known to be involved in neuron-glioma interactions (22,25–28,59–61). Third, CHD2 depletion in H3.1K27M DMG cells results in a broad dysregulation of synaptic genes, especially those postsynaptic genes involved in synapse assembly and remodeling such as NGEF and ERBB4. Together, these results demonstrate that CHD2 plays a critical role in regulating the expression of axon-guidance and synaptic genes in H3.1K27M DMG cells, and this regulation is essential for the interactions between glioma cells and the neurons in the neural circuits they infiltrate.

CHD2 depletion did not affect the proliferation of H3.1K27M cells as detected by the EdU incorporation assay and yet reduced the fitness of H3.1K27M DMG cells based on the cell viability and colony formation assays in vitro. The apparently conflicting phenotypes displayed in H3.1K27M DMG cells upon CHD2 depletion can be partially explained by the increase of apoptosis in these cells upon CHD2 depletion. We also noticed that in addition to regulating the expression of genes involved in neural development, CHD2 is also important in regulating the expression of genes involved in PI3K-Akt signaling (Supplementary Table S2). Previously, it has been shown that Ephrin-EPH signaling also regulates PI3K-Akt signaling and MAPK signaling (53,62). Therefore, it is possible that CHD2 regulates cell fitness and neural integration of H3.1K27M DMG cells through multiple pathways.

Different subgroups of DMG tumors likely adopt distinct mechanisms to control their gene expression for their interactions with neurons.

Gliomas can be classified into different subgroups/subtypes partly based on gene expression, genetic mutations, DNA methylation and histology (1). It is not clear whether different glioma subtypes utilize similar or distinct mechanisms to control their interaction with neurons. We have shown that FOSL1 interacts with CHD2 and helps recruit CHD2 to chromatin to regulate cell fitness and neuron-glioma interactions in H3.1K27M, but not H3.3K27M DMG cells. This unique dependence of H3.1K27M DMG tumor cells on FOSL1 and CHD2 is likely due to the unique gene expression program and/or cell-of-origin of this glioma subtype (14,20,21,48,63–65). Consistent with this idea, previous studies showed that H3.1K27M DMG tumors arise from oligodendroglial lineage precursors at a stage of maturation distinct from H3.3K27M DMG tumors (14,20). Furthermore, we and others have shown that chromatin remodeler SMARCA4 regulates the expression of genes involved in neural development in H3.3K27M DMG cells (58,66). It would be interesting to determine whether SMARCA4, like CHD2, controls the neuron-glioma interactions in H3.3K27M DMG cells. Therefore, the unique dependence of H3.1K27M DMG tumors on FOSL1 and CHD2 likely reflect the fact that this subgroup of DMG tumor relies on FOSL1 and CHD2 to regulate the unique gene expression pattern that governs cell fitness and glioma-neuron interactions.

CHD2 regulates H3K27me3 in H3.1K27M DMG cells

We observed that CHD2 depletion resulted in a marked increase in H3K27me3 in H3.1K27M, but not H3.3K27M DMG cells. The increase of H3K27me3 correlated with the reduced expression of genes in H3.1K27M DMG cells upon CHD2 depletion. CHD2 depletion did not affect the protein level of H3.1K27M mutation proteins and the H3K27me3 writers and erasers, suggesting that the increase in H3K27me3 is unlikely due to the changes in expression of these proteins that directly impact H3K27me3 levels. We suggest that CHD2 likely regulates the H3K27me3 levels in H3.1K27M DMG cells via multiple mechanisms. First, previous studies discovered that EZH2 was sequestered at poised enhancers in H3K27M DMG cells (8,9,67). In this study, we observed about 30% of CHD2 CUT&RUN peaks located at enhancers. Therefore, it is possible that CHD2 depletion in H3.1K27M DMG cells partially impair the EZH2 sequester at poised enhancers. Second, it has been shown that CHD2 depletion in mouse embryonic stem cells led to increased H3K27me3 levels at bivalent genes through changes in histone H3.3 (including H3.3K27M) occupancy at these genes (36). While most of the increased H3K27me3 peaks in H3.1K27M DMG cells were not localized at bivalent genes, it is possible that CHD2 depletion affects histone H3.3 occupancy, which in turn regulates H3K27me3 levels. Finally, it has been shown that CHD2 depletion results in increased nucleosome occupancy (68), which in general can increase PRC2 enzymatic activity locally (69). Therefore, the increase in H3K27me3 in H3.1K27M DMG cells upon CHD2 depletion is also likely due to an increase in nucleosome occupancy. Future studies are needed to test these and other hypotheses. Nonetheless, our results highlight differential impact of CHD2 and FOSL1 depletion on the chromatin landscape of H3.1K27M and H3.3K27M DMG cells, which in turn contributes to their differential impacts on gene expression, cell viability and neuron-glioma interactions in these two subgroups of HGG tumors.

In summary, our results show for the first time that the intrinsic gene expression of glioma cells must be maintained by chromatin regulators for neuron-glioma interactions. Furthermore, different glioma subtypes may utilize distinct mechanisms to maintain the intrinsic expression programs for their interactions with surrounding neurons. As CHD2 is highly expressed in H3.1K27M DMG tumors, it would be interesting to determine whether targeting CHD2 will provide precise therapeutic opportunities for this glioma subtype.

METHODS

Xenograft Studies

All procedures for animal experiments were approved by the Institutional Animal Care and Use Committee at Columbia University Irving Medical Center and Northwestern University, and were following national and institutional guidelines. For mice survival experiments, six-week-old female NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ mice (Jackson Laboratory, strain # 005557) and athymic mice (rnu/rnu genotype, BALB/c background, Envigo, cat no. 069) were used for in vivo xenograft experiments. Orthotopic pontine injection was performed as previously described (70,71). Briefly, SU-DIPG36 cells (72,73) or HSJD-DIPG-007 cells (provided by Dr. Angel Montero Carcaboso at Hospital Sant Joan de Déu, Barcelona, Spain) infected with lentivirus for lentiCRISPR-v2 (Addgene #52961) sgRNA vector (sgNeg, sgCHD2) were harvested on day 5 post infection. 1 μl of 1 × 10⁵ cell suspension was injected using a sterile Hamilton syringe into the pontine tegmentum at 1.5 mm right lateral of Bregma and 4.6 mm posterior to Bregma and 4.5 mm deep. After cell injection, the syringe was left in place for 1 min before it was removed slowly. Intracranial tumor growth was monitored by bioluminescence imaging using SII Lagos bioluminescence/fluorescence imaging machine (Spectral Instruments Imaging Company). For hippocampal xenografts, 2.5 μl of 3 × 10⁵ SU-DIPG36 cells with (sgCHD2) or without CHD2 depletion (sgNeg) were injected using a sterile Hamilton syringe into the CA1 region of the hippocampus of four-week-old female NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ mice (Jackson Laboratory, strain # 005557) at 1.5 mm lateral to midline, 1.8 mm posterior to Bregma and 1.4 mm deep to cranial surface. After cell injection, the syringe was left in place for 2 min before it was removed slowly.

Plasmid Construction and sgRNA Cloning

Full-length coding sequence (CDS) of CHD2-FLAG was obtained by replacing EGFP with 3x Flag tag and then digestion with SalI and NotI restriction enzymes from CHD2-FL-pEGFP-N1 vector (gift from Dr. Haico van Attikum at Leiden University Medical Center). CDS of CHD2-FLAG was then ligated into pENTR4 vector (Invitrogen, cat no. A10465). Sit-directed mutagenesis was performed at PAM site of sgCHD2–1 for sgRNA resistance. Gateway recombination was then carried out by using LR Clonase II Enzyme mix (Invitrogen, cat no. 11791020) to clone sgRNA-resistant full-length CDS of CHD2-FLAG to pLX_TRC313 lentiviral vector (gift from Dr. William C. Hahn at Dana-Farber Cancer Institute). sgRNAs were designed either based on the sequences of the pooled sgRNA library or through Chopchop (https://chopchop.cbu.uib.no/) and CCTop (https://cctop.cos.uni-heidelberg.de:8043/) and were cloned into LRG2.1T lentiviral vector (gift from Dr. Christopher R. Vakoc at Cold Spring Harbor Laboratory) or lentiCRISPR-v2 vector (Addgene #52961) through BsmbI restriction site. The primers and sgRNAs used in this study were shown in Supplementary Table S4.

Cell Lines and Cell Culture

Patient-derived DMG cells (SU-DIPG4, SU-DIPG6, SU-DIPG17, SU-DIPG36) (72,73) were obtained from Dr. Michelle Monje at Stanford University and maintained as neurospheres in knockout DMEM/F12 media (Gibco) supplemented with StemPro Neural Supplement (Gibco), 20 ng/mL human FGF-basic (bFGF) (Gibco), 20 ng/mL human EGF (Gibco), 2 mM L-Glutamine (Gioco), 1% penicillin/streptomycin (Invitrogen) and plasmocin (InvivoGen). Alternatively, the DMG cells were maintained as adherent culture using 24 μg/mL fibronectin-coated dishes in the same medium. Patient-derived glioblastoma cells (GBM22, GBM28, GBM6) were obtained from Dr. Jann N. Sarkaria at Mayo clinic (74). GBM6 cells were maintained as neurospheres using the same medium with that for DMG cells. GBM22, GBM28, KNS42 (Japanese Collection of Research Bioresources, IFO50356), SF8628 (75) and HEK293T (ATCC) cells were maintained as adherent culture in DMEM (Corning) containing 10% FBS (Gemini Bio), 2 mM L-Glutamine and 1% penicillin-streptomycin. Mouse DMG cells with H3.3K27M mutation were obtained from Dr. Oren Becher at Mount Sinai (76) and maintained as neurospheres in NeuroCult^™ Basal Medium (STEMCELL Technologies) containing Neurocult proliferation supplement (STEMCELL Technologies), 20 ng/mL human bFGF (Gibco), 10 ng/mL human EGF (Gibco), 2 μg/mL Heparin (STEMCELL Technologies) and 1% penicillin/streptomycin (Invitrogen). HSJD-DIPG-007 cells were kindly provided by Dr. Angel Montero Carcaboso (Hospital Sant Joan de Déu, Barcelona, Spain) and maintained as neurospheres in 1:1 mixed Neurobasal-A medium (Invitrogen) and D-MEM/F-12 medium (Invitrogen) containing 2% B27 (-A) (Invitrogen), 1% HEPES Buffer Solution (Invitrogen), 1% MEM Sodium Pyruvate Solution (Invitrogen), 1% MEM Non-Essential Amino Acids Solution (Invitrogen), 1% GlutaMAX-I Supplement (Invitrogen), 20 ng/mL human FGF-basic (bFGF) (Gibco), 20 ng/mL human EGF (Gibco), 10 ng/mL human PDGF-AA (Shenandoah), 10 ng/mL human PDGF-BB (Shenandoah), 2 μg/mL Heparin (STEMCELL Technologies), 1% penicillin/streptomycin (Invitrogen) and plasmocin (InvivoGen). All cells were cultured at 37°C with 5% CO₂ and were tested for mycoplasma contamination monthly by PCR. All experiments in this study were performed within 10 passages of the cells described above, starting from the acquisition of cells from the original source.

Lentivirus Production

Lentivirus was produced by transfecting HEK293T cells with plasmids using Polyethylenimine MW 25,000 (PEI, Polysciences, cat no. 23966–1). Briefly, 6 μg of plasmid DNA, 4.5 μg of psPAX2 (Addgene #12260), 1.5 μg of pMD2.G (Addgene #12259) and 18 μl of 2 mg/mL PEI were mixed in Opti-MEM (Gibco), the mixture was then added to each 10-cm-dish culture of HEK293T cells. 12 hours post transfection, the culture was changed with fresh medium. Virus-containing supernatant was collected at 48 hours and 72 hours post transfection, filtered with 0.45-micron strainer and then pooled together. Virus was precipitated by PEG-6000 (Santa Cruz Biotechnology, cat no. sc-302016A) and suspended in appropriate media. For lentivirus infection, target cells were mixed with lentivirus and then centrifuged at 1,200 rpm for 1 hour at room temperature (RT), with culture media changed with fresh ones at 20 hours post infection. Puromycin (1 μg/mL, InvivoGen, cat no. ant-pr) or Hygromycin (40 μg/mL, InvivoGen, cat no. ant-hg) was added at 48 hours post infection if needed.

Pooled CRISPR Screens

Cas9-expressing cells were established by infecting DMG or GBM cells with lentivirus from LentiV_Cas9_puro vector (gift from Dr. Christopher R. Vakoc at Cold Spring Harbor Laboratory). Cas9 expression was confirmed by Western Blot using Rabbit monoclonal anti-Cas9 (S. pyogenes) antibody (Cell Signaling Technology, cat no. 65832) and cutting efficiency was confirmed by using sgRNA targeting PCNA and CDK1. Epigenetic and transcription-factor domain-focused sgRNA libraries (gift from Dr. Christopher R. Vakoc at Cold Spring Harbor Laboratory) were amplified using ElectroMAX Stbl4 Competent Cells (Invitrogen, cat no. 11635018) with a coverage more than 1,000 and were used to produce lentivirus as described above. Cas9-expressing cells were infected with the lentivirus of pooled sgRNA libraries at a MOI ≈ 0.3 and 1,000 coverage. Cells were collected at the initial time point (day 3 post infection, T0) and end point (10 doubling times after the initial time point, T1). Genomic DNA was extracted using QIAGEN DNeasy Blood and Tissue Kit (QIAGEN, cat no. 69504). Sequencing library was prepared as previously described (47). Briefly, sgRNA-containing genomic DNA fragments (about 200 bp) were amplified by PCR, followed by end repair using T4 DNA polymerase (NEB), DNA Polymerase I (Klenow) fragment (NEB) and T4 Polynucleotide Kinase (NEB), added with 3’ A-overhang and then ligated with customized barcodes (47) using T4 DNA Ligase (NEB). Barcoded libraries were pooled and analyzed by paired-end sequencing using Nextseq (Illumina) with xGen^™ ssDNA & Low-Input DNA Library Prep Kit (IDT, cat no. 10009817).

Data Analysis for Pooled CRISPR screens

MAGeCK (version 0.5.7) (77) program was used for the analysis. Briefly, the sequencing datasets were de-barcoded, merged, and then the 43 bp single-end reads contained 20-bp sgRNA sequence was aligned to the epigenetic and transcription factor domain-focused sgRNA libraries (47) using ‘count’ function from MAGeCK without allowing for any mismatches. The essential gene was evaluated from the rankings of gRNAs (by their p-values and log₂ fold-change) between initial time point (T0) and end point (T1) using the α-RRA (α-Robust Rank Aggregation) algorithm. With the raw counts, RRA fits a negative binomial model (NB) to test whether the loss-of-function screen is significant from initial sgRNA counts (T0) to the final sgRNA counts (T1) via false discovery rates and guide-level log₂ fold-change. We use the beta-score, generated from MAGeCK-VISPR algorithm (78), as a measurement of essential score in Fig. 1B and Fig. 6B to compare genes in different cell lines. FDR and beta-score of each gene in each sgRNA library were estimated to detect the candidate hits using MAGeCK-MLE algorithm by normalizing against negative control sgRNAs, which first fits a negative binomial generalized linear model with log-link to guide level counts, and then fits a coefficient at the guide level to evaluate the guide RNA abundance changes between T0 and T1. Beta-score is similar with the term guide-level log₂ fold-change, where a positive and negative beta-score indicating that the corresponding gene is positively and negatively selected, respectively. For the comparison of different cell lines (Fig. 6B), a further loess normalization is applied to the beta-scores to ensure that beta-scores have similar distributions across different cell lines.

GFP-based Competition Assay

Cas9-expressing cells were infected with lentivirus expressing GFP and sgRNA from the same vector LRG2.1T (gift from Dr. Christopher R. Vakoc at Cold Spring Harbor Laboratory) with a MOI ≈ 0.5. The GFP positive cells were measured every 3–4 days from day 3 to day 17 by using Attune NxT flow cytometer (Invitrogen). The percentage of GFP positive cells at each time point compared to day 3 was calculated.

Western Blot

For knockout experiments, Cas9-expressing DMG or GBM cells were harvested on day 7 post infection with sgRNA lentivirus. Cell pellets were suspended in 1X SDS sample buffer containing 2-mercaptoethanol and boiled for 5 min. Proteins in whole cell lysates were separated by SDS-PAGE, transferred to nitrocellulose membrane, and followed by immunoblotting. Antibodies used in Western Blot: Rabbit polyclonal anti-CHD2 antibody (Cell Signaling Technology, cat no. 4170S), Rabbit polyclonal anti-CHD2 antibody (Millipore, cat no. HPA060744), Mouse monoclonal anti-alpha-tubulin (Developmental Studies Hybridoma Bank, cat no. 12G10), Mouse monoclonal anti-FLAG antibody (Sigma-Aldrich, cat no. F1804), Mouse monoclonal anti-FOSL1 antibody (Santa Cruz Biotechnology, cat no. sc-376148X), Mouse monoclonal anti-FOSL1 antibody (Santa Cruz Biotechnology, cat no. sc-28310), Rabbit anti-Histone H3K4me3 antibody (Abcam, cat no. ab8580), Rabbit monoclonal anti- tri-Methyl-Histone H3 (Lys27) antibody (Cell Signaling Technology, cat no. 9733), Rabbit monoclonal anti-acetyl-Histone H3 (Lys27) antibody (Cell Signaling Technology, cat no. 8173), Rabbit polyclonal anti-Histone H3 antibody (Abcam, cat no. ab1791), Rabbit monoclonal anti-Histone H3K27M antibody (RevMAb Biosciences, cat no. 31–1175-00-L), Rabbit monoclonal anti-Di-Methyl-Histone H3 (Lys36) antibody (Cell Signaling Technology, cat no. 2901), anti-Histone H3K36me3 antibody (Active Motif, cat no. 61021), Rabbit anti-EphB3 antibody (Abcam, cat no. ab133742), Rabbit monoclonal anti-EZH2 antibody (Cell Signaling Technology, cat no. 5246), Rabbit monoclonal anti-SUZ12 antibody (Cell Signaling Technology, cat no. 3737), Rabbit polyclonal anti-EED antibody (Millipore, cat no. 09–774), Rabbit polyclonal anti-JMJD3 antibody (Millipore, cat no. 07–1534), Rabbit polyclonal anti-JARID2 antibody (Novus Biologicals, cat no. NB100–2214), Rabbit anti-UTX antibody (Bethyl Laboratories, cat no. A302–374A), Rabbit monoclonal anti-Cas9 (S. pyogenes) antibody (Cell Signaling Technology, cat no. 65832), Goat polyclonal anti-mouse IgG (H+L) HRP-linked (Jackson ImmunoResearch, cat no. 115–035-003) and Goat polyclonal anti-rabbit IgG (H+L) HRP-linked (Jackson ImmunoResearch, cat no. 111–035-144).

MTT Cell Viability Assay

Cas9-expressing DMG or GBM cells were seeded into 96-well plates on day 7 post infection with sgRNA lentivirus, at a density of 1 × 10³ cells per well with 3–6 replicate wells for each condition. Cell viability was measured every other day from day 8 post infection (day 1 of MTT) by using CellTiter-Blue assay kit (Promega, cat no. G8081). The experiment was then repeated at least three times independently.

Colony Formation Assay

Cas9-expressing DMG or GBM cells were seeded into 1:60 diluted Matrigel (Corning, cat no. 354234)-coated 6-well plates on day 7 post infection with sgRNA lentivirus, at a density of 800 cells per well with 3 wells for each condition. After culturing for 2 weeks, the culture was stained with 0.1% crystal violet (Sigma-Aldrich). Quantitation of the colony numbers was done by using Image J. This colony formation assay for each datapoint presented was then repeated at least three times independently.

Annexin V Apoptosis Assay

Cas9-expressing DMG or GBM cells were harvested on day 10 post infection with sgRNA lentivirus, stained with Pacific Blue-conjugated Annexin V (Invitrogen, cat no. A35122) according to the manufacturer’s instructions and then analyzed by Attune NxT flow cytometer (Invitrogen).

RNA-seq and RT-qPCR

DMG cells were harvested in RLT plus buffer (QIGEN) on day 7 post infection with sgRNA lentivirus. Total RNA was extracted using RNeasy Plus Mini kit (QIGEN, cat no. 74136) according to the manufacturer’s instructions. For RNA-seq, libraries were prepared and then sequenced at Columbia genomic center. For RT-qPCR, cDNA was synthesized by using SuperScript III Reverse Transcriptase (Invitrogen, cat no. 18080093) in a reaction containing 0.5 μg total RNA and 50 ng random hexamers according to manufacturer’s instructions. RT-qPCR was performed in 12 μl reactions containing 0.15 μM primers and iTaq Universal SYBRGreen Supermix (Bio-Rad, cat no. 1725124). The primers for ACTB were use as normalization control. All primers used in this study were shown in Supplementary Table S4.

RNA-seq Data Analysis

Trim Galore (version 0.6.7) (https://github.com/FelixKrueger/TrimGalore) was used to trim the adaptors and remove low-quality reads based on the raw sequencing reads (2 × 100 bp) received from the platform of CUGC. The genome sequence and coding gene annotation of hg19 was from GENCODE (79). STAR (version 2.7.6a)5 (80) was used to align clean reads against the hg19 human reference genome assembly. Using SAMtools (version 1.11) (81), we converted sequence alignment map files to BAM files and extracted unique mapped reads with ‘-q 40’ parameter. The gene-level count matrix was generated by featureCounts (version 2.0.1) (82) relying on the gene transfer format (GTF) file of exon annotation. Gene expression level was quantified by transcripts per kilobase million (TPM), which was calculated from counts of all samples using TPM calculator (83). General linear models (GLMs) of edgeR (version 3.34.0) (84) based on negative binomial distributions were used to perform differential gene expression analysis between wild-type and each mutant cell culture. The P value of each gene was adjusted using the Benjamini–Hochberg (BH) procedure. Genes with a BH-adjusted P value less than 0.05 and a fold change greater than 1.5 (upregulated) or less than 0.67 (downregulated) were considered as differentially expressed genes (DEGs). Enrichment analysis of Gene Ontology (biological process) and KEGG pathway was determined via the enrichGO and enrichKEGG function from the clusterProfiler package (version 4.3.0) (85), respectively, with adjusting P values obtained with the BH procedure. To generate a volcano plot for DEGs, log₂ (fold-change) was plotted against -log₁₀ (adjusted P value) using EnhancedVolcano package (https://github.com/kevinblighe/Enhanced-Volcano).

Immunoprecipitation

8 million DMG cells were harvested for each reaction, washed once with cold PBS, and then suspended in ice-cold lysis buffer (50 mM HEPES-KOH pH=7.4, 100 mM NaCl, 1% NP40, 10% Glycerol, 1 mM EDTA, 1 mM PMSF and one Roche complete protein inhibitor tablet per 50 ml). Cells were lysed on ice by using Dounce homogenization with 30 strokes and then incubated on ice for 30 minutes. After centrifugation at 19,000 x g at 4°C for 10 minutes, supernatants were separated from pellet fractions. Supernatants were transferred into new 1.5 ml tubes and spun one more time at 19,000 x g at 4°C for 20 minutes. Supernatants were then collected and incubated for 1 hour with 3 μg mouse IgG antibody (Diagenode, cat no. C15400001) while rotating at 4°C. Samples were then incubated for 40 minutes with Protein G Sepharose beads (Millipore, cat no. GE17–0618-01) while rotating at 4°C. Pellet was separated from supernatant and discarded. The resulting supernatants were mixed with 3 μg anti-FOSL1 antibody (Santa Cruz Biotechnology, cat no. sc-28310) or mouse IgG antibody (Diagenode, cat no. C15400001) and incubated overnight while rotating at 4°C. Protein G Sepharose beads were added and incubated for 1 hour while rotating at 4°C. The pellet was then collected and washed 6 time with wash buffer (50 mM HEPES-KOH pH=7.4, 100 mM NaCl, 0.01% NP40, 10% Glycerol, 1 mM EDTA, 1 mM PMSF and one Roche complete protein inhibitor tablet per 50 mL). Samples were eluted in 1x SDS sample buffer, boiled for 5 min and subjected to SDS-PAGE and immunoblotting with the indicated antibodies.

Histone Purification by Acid Extraction

2 million DMG cells were harvested, suspended in 1 mL ice-cold Hypotonic buffer (10 mM Tris-HCl pH=8.0, 1 mM KCl and 1.5 mM MgCl₂) and then spun at 2,000 rpm for 2 minutes at 4°C. Pellet was incubated in 1 mL ice-cold Hypotonic buffer for 30 minutes while rotating at 4°C. Pellets was harvested by centrifugation at 10,000 rpm for 10 minutes at 4°C and then incubated in 300 μL 0.2 N HCl for 30 minutes while rotating at 4°C. Supernatants were collected by centrifugation at 16,000 g for 10 minutes at 4°C. 100% Trichloroacetic Acid (Fisher Chemical, cat no. A322–100) was added drop by drop while flicking the supernatants to a final concentration of 33%. Samples were then rotated overnight at 4°C. Histone pellets were obtained by centrifugation at 16,000 g for 10 minutes at 4°C, washed twice with ice cold acetone, dried at RT for 20 minutes and then dissolved in 200 μL H2O for 1 hour while rotating at RT.

Neuron–DMG Co-culture Experiments

Primary mouse cortical neurons were isolated from the cortices of E15.5 embryos, plated at a density of 240,000 cells per 12 mm glass coverslip in 24-well plates coated with poly-L-lysine (DIV 0), and maintained in neuron medium (1 mL per well) consisting of neurobasal medium (Invitrogen) supplemented with 1x B27 plus (Gibco), 1x L-Glutamax and 2.5% FBS. On DIV 5, 0.4 mL medium was removed from each well in the morning. In the afternoon, DMG cells were digested with Accutase (BD Biosciences), filtered with 40 μm strainer (Fisher Scientific), washed once with serum-free neuron medium and then suspended in serum-free neuron medium. 60,000 DMG cells in 0.6 mL serum-free neuron medium were seeded into each well of the neuron culture. For monocultures of DMG cells, 60,000 DMG cells were seeded onto each 12 mm glass coverslip in 24-well plates coated with poly-L-lysine in the same medium with the neuron-DMG co-culture. NBQX (Tocris, cat no. 0373) was added at a final concentration of 10 μM as the control. DMG cells cultured alone or co-cultured with neurons for 72 hours were pulsed with 20 μM (final concentration) of EdU (Invitrogen, cat no. A10044) for 1 hour. The cell cultures were washed twice with PBS and then fixed with 4% paraformaldehyde for 20 min at RT before immunofluorescence.

EdU Click-iT Reaction and Immunofluorescence

After DMG cells cultured alone or co-cultured with neurons were fixed, the cells were washed 3 times with PBS, incubated in PBS containing 0.2% Triton X-100 and 2.5% goat serum at 37°C for 1 hour. The cells were then washed twice with PBS, 5 min each. The Click-iT reaction cocktail was prepared by adding 100 mM CuSO4, 5 mM BDP TMR azide (Lumiprobe, cat no. 32430) and freshly prepared 0.5 M L-Ascorbic acid sequentially into PBS for a final concentration of 1 mM, 2 μM and 50 mM respectively. The Click-iT reaction was performed by incubating the cells in each well of 12 well plate with 400 μL reaction cocktails at RT for 30 min. After the cells were washed twice with PBS, mixtures containing anti-H3K27M primary antibody (RevMAb Biosciences, cat no. 31–1175-00-L), 0.2% Triton X-100 and 2.5% goat serum in PBS buffer were added and incubated at 4°C overnight. The cells were then washed 3 times with PBS (5 min each), incubated at 37°C for 1 hour in goat anti-rabbit Alex-Flu 488 secondary antibody (Invitrogen, cat no. A-11034) diluted in PBS containing 2.5% goat serum. The coverslips were then mounted with Fluoro-gel II with Dapi (Electron Microscopy Sciences, cat no. 17985–50) for imaging using Eclipse Ni-U Fluorescent Microscope (Nikon). Proliferation index was obtained by calculating the percentage of EdU labelled DMG cells (identified by EdU+/H3K27M+) over total number of DMG cells (identified by H3K27M+).

Synaptic Puncta Staining and Visualization

Fixed coverslips were incubated in blocking solution (2.5% normal goat serum, 0.2% Triton X-100 in PBS) at 37°C for 1 h. Primary antibodies including guinea pig anti-synapsin1/2 (1:400; Synaptic Systems, cat no. 106004), and rabbit anti-RFP (1:500; Rockland Immunochemicals, cat no. 600–401-379) were diluted in blocking solution and incubated overnight at 4 °C. The cells were then washed 3 times with PBS (5 min each), incubated at RT for 1 hour in PBS containing 2.5% goat serum, goat anti-rabbit Alex-Flu 594 (1:500, Invitrogen, cat no. A-11012) and Donkey anti-Guinea pig Alexa Flu 647 (1:300, Jackson ImmunoResearch, cat no. 706–605-148) secondary antibodies, and then washed three times with PBS (5 min each). The coverslips were then mounted with Fluoro-gel II with Dapi (Electron Microscopy Sciences, cat no. 17985–50) for imaging using a 20× objective with 4× zoom on Yellow A1 confocal microscope (Nikon) at Confocal and Specialized Microscopy core in Columbia University Medical Center.

Confocal Puncta Quantification

The colocalization of synaptic puncta in confocal images was analyzed using the reported method and code (22) with modifications. Scanning confocal images were collected using a 20x/0.75 objective lens with 4x zoom set to achieve a pixel size of 0.155 μm, and the pinhole set to 1.0 Airy unit (corresponding to an optical section of 2 μm). Z-stacks were collected with an interval of 1 μm over the depth of the cell. However, most processes appeared within a depth range of 1–3 μm, reducing the probability of spurious colocalization. Confocal images of PSD95-RFP and Synapsin were analyzed in Fiji (86) using the “Find Close Peaks” script (22) (https://github.com/bioimage-analysis/find_close_peaks) with slight modifications. Images were pre-processed using a rolling-ball background subtraction. Pre- and post-synaptic puncta were identified and localized using a Difference of Gaussians (DoG) detector from the ImgLib2 algorithm library (87). PSD95-RFP puncta within 1.5 μm of Synapsin puncta were scored as colocalized, and a glioma cell was scored as having colocalization if at least one synaptic colocalization was observed.

CUT&RUN

CUT&RUN experiments were carried out as previously described (50,88), with slight modifications. Briefly, DMG cells were crosslinked at RT on an orbital shaker using 1% Formaldehyde (Sigma-Aldrich) for 2 min for CHD2 CUT&RUN, 0.5% Formaldehyde for 2 min for FOSL1, H3K4me3, H3K27me3 or H3K27ac CUT&RUN. Formaldehyde was quenched by 125 mM glycine for 5 min at RT. Cells were then collected by spinning 600 g at RT for 3 min and then rinsed 3 times with 1 mL wash buffer (20 mM HEPES pH=7.5, 150 mM NaCl, 0.5 mM spermidine, 1% Triton X-100, 0.05% SDS and one Roche complete protein inhibitor tablet per 50 mL). In general, 5 × 10e5 cells were used for each CHD2 or FOSL1 CUT&RUN, 2.5 × 10e5 cells for each H3K4me3, H3K27me3 or H3K27ac CUT&RUN experiments. Mouse DMG cells (76) were added as spike-in controls. Cells were suspended in 1 mL wash buffer and then bound to BioMag Plus Concanavalin A beads (Polysciences, cat no. 86057–10) by gentle rotating at RT for 20 min. Cell-beads slurry was incubated overnight at 4°C with primary antibodies diluted in wash buffer containing 0.02% digitonin (Dig wash buffer). Secondary antibody-pAG-Mnase complex was assembled at a molecular ratio of 1:2 in 50% glycerol at 4°C on an orbital shaker for 1 hour. Cell-beads slurry was washed twice with Dig wash buffer, incubated at 4°C with secondary antibody-pAG-Mnase complex on an orbital shaker for 1 hour. After washed 3 times with Dig wash buffer, the slurry was placed on an ice-cold block and incubated with Dig wash buffer containing 2 mM CaCl₂ to initiate the CUT&RUN experiments. After incubation on the ice-cold block for 30–60 min (time varies for different antibodies), one volume of 2x stop buffer (340 mM NaCl, 20 mM EDTA, 4 mM EGTA, 0.02% Digitonin, 0.05 mg/mL glycogen, 0.05 mg/mL RNase A, 1% Triton X-100 and 0.05% SDS) was added to stop the reaction. DNA fragments were released by incubation at 37°C for 30 min with shaking at 400 rpm. Supernatant was collected, supplemented with 0.1% SDS, 0.25 mg/mL proteinase K, 10 mM 1M Tris-HCl pH=8.0 and 10 mM DTT and then incubated overnight at 65°C with shaking at 700 rpm to reverse crosslink. CUT&RUN DNA was extracted with Phenol-chloroform-isoamyl alcohol (Invitrogen, cat no. 15593049) and ethanol precipitation and then dissolved in 12 μL 1 mM Tris-HCl pH=8 0.1 mM EDTA. CUT&RUN-qPCR was performed as described above. Sequencing libraries were prepared as previously described (50,89). Antibodies used in CUT&RUN: Rabbit polyclonal anti-CHD2 antibody (Proteintech, cat no. 12311–1-AP), Mouse monoclonal anti-FOSL1 antibody (Santa Cruz Biotechnology, cat no. sc-376148X), Rabbit anti-Histone H3K4me3 antibody (Abcam, cat no. ab8580), Rabbit monoclonal anti- tri-Methyl-Histone H3 (Lys27) antibody (Cell Signaling Technology, cat no. 9733), Rabbit monoclonal anti-acetyl-Histone H3 (Lys27) antibody (Cell Signaling Technology, cat no. 8173), Donkey anti-Rabbit IgG (H+L) antibody (Sigma-Aldrich, cat no. SAB3700932) and Rabbit anti-Mouse IgG H&L antibody (Abcam, cat no. ab46540).

CUT&RUN Data Analysis

CUT&RUN libraries were sequenced using the paired-end methods using Illumina NextSeq 500 platforms. Raw reads were filtered via trimming Illumina adapters and removing low-quality reads using Trim Galore (version 0.6.7) with default parameters. Trimmed reads were aligned to human reference genome (hg19) using Bowtie2 (version 2.2.4) (90) with following flags: with --no-mixed --no-discordant --no-dovetail --no-contain --local --maxins 1000. Sequence alignment map files were transformed into BAM files and extracted unique mapped reads using SAMtools (version 1.11). Duplicates reads were removed using MarkDuplicates function of Picard (version 2.23.8) (https://github.com/broadinstitute/picard) with REMOVE_DUPLICATES = true. Alignment files in the BAM format were converted to read coverage files in Bigwig format using deepTools bamCoverage (version 3.2.1) (91) with RPM normalization after removing blacklisted regions. The SEACR software (version 1.3) (89) was used for peak identification with relaxed mode. Overlapped peaks between replicates were retained and then merged to obtain an unambiguous peak set for each CUT&RUN target using BEDTools (version 2.29.2) (92). Peak annotation was performed using the R package ChIPSeeker (version 1.26.0) (93) with minor modifications for promoter annotation. Promoter regions were redefined by H3K4me3 CUT&RUN peaks in corresponding cells, instead of the default distance with TSS. The adjacent gene of each peak was defined with the cut off less than 50 kb distance between its genomic location and its nearest transcriptional start site (TSS). To combine CUT&RUN signals with expression levels of genes, CUT&RUN density at the promoters (TSS ± 3 kb) was calculated. Heatmaps and line plots of occupancy were generated using the deepTools function plotHeatmap and plotProfile (version 3.2.1), respectively. Motif analysis was performed using Haystack motif enrichment tool (94) to scan for putative TF binding sites within the identified specific peaks in SU-DIPG4 cells. The most recent version of the non-redundant JASPAR database (version 2022) (95) and HOCOMOCO database (version 11) (96) with hg19 sequences were used for the motif analysis.

The reliable reads of each sample with spike-in DNA were re-mapped against the spike-in mouse genome (mm10) downloaded from GENCODE database using bowtie2. Reads that mapped to both hg19 and mm10 genome were removed. The scale factor for spike-in mouse was calculated as SF = 1,000,000/mapped reads to mm10.The scale factors were used for scaling BAM files to Bigwig files with bamCoverage function from deepTools setting 10 bp window size. For differential binding analysis, reads in the consensus peak set were counted for each replicate and analyzed with edgeR package (version 3.34.0) using total library sizes (no spike-in DNA) or scale factors (spike-in DNA) as the size factors. The threshold of fold change was established as FC > 1.2 and adjusted P-value < 0.05. For differential CHD2 CUT&RUN signals between wild-type (sgNeg) and FOSL1-depleted (sgFOSL1) SU-DIPG4 cells, the decreased peaks were identified via differentially binding analysis with the significance threshold as (sgFOSL1/sgNeg < 0.83 and FDR < 0.05). For histone modifications, the increased H3K27me3 and decreased H3K4me3 and H3K27ac CUT&RUN signals were detected via differential binding analysis with the significance threshold (fold change > 1.2 and FDR < 0.05) and (fold change < 0.83 and FDR < 0.05), respectively. Bar plots were generated with the R package ggplot2. For representative genome track snapshots, reference sequence track with annotated genes was obtained from hg19 genome and Y-axis indicated reads per million reads (RPM) using SparK (https://github.com/-harbourlab/SparK).

Single-cell RNA sequencing (scRNA-seq) analysis

The scRNA-seq datasets including counts matrix, barcode and feature files were downloaded from recently published study (20). scRNA-seq data processing and quality control were performed as described previously (20). We assessed 22 primary pediatric HGG tumors, including 11,644 H3.1K27M DMG cells of 7 participant patients, 76,112 H3.3K27M DMG cells of 11 participant patients and 9,581 wild type HGG cells of 4 participant patients. We employed a global-scaling normalization method “LogNormalize” that normalizes the gene expression measurements for each cell by the total expression using NormalizeData function from Seurat (version 4.0) (97), scaled counts to 10,000 UMIs per cell, and log2-transformed the counts using ScaleData function. The Uniform Manifold Approximation and Projection (UMAP) dimensional reduction technique was run on the scaled data to seek a low-dimensional embedding of cells using RunUMAP function. VlnPlot function was used to show expression probability distributions across groups with p-value between different group estimated by Wilcoxon tests.

Gene Set Enrichment Analysis (GSEA)

To evaluate the significance of transcriptional changes at the gene group level, GSEA was performed according to the original report (98) using the GSEA function from the clusterProfiler package. The gene lists of our RNA-seq data were ranked by the log2 (fold-change) value of their expression between wild type and depletion conditions, then used as the input for GSEA.

Proliferation Assay in Coronal Slice

Mice were anaesthetized with isoflurane and then transcardically perfused with 20 mL PBS. The whole brain was removed from the skull, washed briefly in ice-cold PBS and then immersed into 4% PFA (Sigma-Aldrich) at 4°C for overnight. The fixed brain was washed twice with ice-cold PBS, transferred into 15% sucrose at 4°C until the tissue sunk (6–12 hours), and then transferred into 30% sucrose at 4°C for overnight. The tissue was then embedded, cryoprotected and sectioned in the coronal plane at 40 μm at Molecular Pathology Core in Columbia University Medical Center. The floating 40 μm coronal slices were then preserved in cryoprotectant buffer (30% sucrose, 30% ethylene glycol and 1% Polyvinyl-pyrrolidone (PVP-40, Sigma-Aldrich) in 0.1 M PB) at −20°C. For immunohistochemistry, 40 μm coronal slices were washed twice with TBS buffer, incubated in block solution (3% normal goat serum, 0.3% Triton X-100 in TBS) at RT for 1 hour. Rabbit anti-Ki67 (Abcam, cat no. ab15580, 1:1000) and mouse anti-human nuclei antigen (Millipore, cat no. MAB1281, 1:400) were diluted in antibody solution (1% normal goat serum, 0.3% Triton X-100 in TBS) and used to incubate coronal slices at 4°C overnight. The slices were then washed 3 times with TBS and incubated in antibody solution with goat anti-rabbit Alexa 594 (Invitrogen, cat no. A-11012, 1:800) and goat anti-mouse Dylight 649 (Biolegend, cat no. 405312, 1:300) at RT for 1 hour. The slices were then washed 3 times with TBS, mounted with Fluoro-gel II with Dapi (Electron Microscopy Sciences) and then imaged using a 20× objective with 2× zoom on Yellow A1 confocal microscope (Nikon) at Confocal and Specialized Microscopy core in Columbia University Medical Center.

Acute Slice Preparation for Calcium Imaging Experiments

Mice were euthanized by cervical dislocation, and brains were immediately removed from skull and incubated in ice-cold sucrose-based slicing solution containing (in mM): 210 Sucrose, 2.5 KCl, 26 NaHCO3, 1.25 NaH2PO4, 0.5 CaCl2, 7 MgCl2, 10 D-glucose. Coronal slices (400 μm) containing the hippocampal region were prepared from mice (7–8 weeks after xenografting) using a Leica VT1000S Vibratome. The slices were recovered at 34°C in oxygenated (95% O₂, 5% CO₂, 3 mL/min) artificial CSF (ACSF) containing (in mM): 125 NaCl, 2.5 KCl, 26 NaHCO3, 1.25 NaH2PO4, 1.5 CaCl2, 1.5 MgCl2, 10 D-glucose for 18 min, and then maintained in oxygenated ACSF at RT for at least 42 min before calcium imaging.

Calcium Imaging Experiments

For calcium imaging in co-culture, GCaMP6s-expressing DMG cells were seeded into neuron culture as shown above in the neuron-DMG co-culture experiments. After co-culture for 72 hours, the co-culture was transferred into the 37°C chamber with 5% CO₂ in Green Spinning-Disk Confocal at Confocal and Specialized Microscopy Core in Columbia University Medical Center. GCaMP6s fluorescence was excited with a 488 nm laser. Time-lapse frames were collected every 0.5 second (2 Hz) in 4 minutes in each recording for spontaneous calcium activity. At least 7 recordings in 7 randomly selected regions in each repeat of each condition were collected.

For calcium imaging in acute coronal slices, the slices were transferred into a submerged recording chamber in a LEICA DM LFS microscope and perfused with oxygenated (95% O₂, 5% CO₂, 3 mL/min) ACSF (similar with the above ACSF but containing 2 mM CaCl2 and 4 mM KCl) at 32°C. GCaMP6s fluorescence was excited with an I3 filter (BP 450–490 nm). GCaMP6s-expressing DMG tumor cells in CA1 hippocampal region were visualized in the IR-DIC mode basing on the fluorescence. Time-lapse frames were collected in the CA1 region every 0.35 second (~2.85 Hz) in 4 minutes in each recording for spontaneous calcium activity. For electrical stimulation, a tungsten concentric bipolar electrode (World Precision Instruments) was placed on Schaffer collaterals and 10 Burst stimulation (TBS) (300 μA, 0.2 ms, 4 pulses at 100 Hz) with 200 ms intervals was delivered 3 times with 30 seconds intervals in each recording using an Iso-Flex Stimulus isolator (A.M.P.I., Israel). Recordings were made in the CA1 region every 0.35 second (~2.85 Hz) in 4 minutes. For the recordings with Tetrodotoxin citrate (TTX, Tocris, cat no. 1069), TTX was diluted into ACSF perfusion media at a final concentration of 1 μM, oxygenated and delivered through the perfusion system. TTX containing ACSF media was perfused for at least 10 minutes prior to starting recordings.

Data Analysis for Calcium Imaging

For spontaneous calcium imaging in co-culture, region of interest (ROI) for individual cells was identified by segmentation using deep learning-based method Cellpose 2.0 (99). Briefly, maximum intensity projection (MIP) of each recording sequence was obtained using Fiji. Five randomly selected MIP images from all recordings in co-culture and DMG alone were selected to perform 5 rounds of training starting from the cyto model in Cellpose 2.0 using green channel for segmentation. Then the trained model was used to segment all the recordings using their MIP images. The ROIs for floating cells and died cells (constant strong fluorescence) were removed manually. Three ROIs without any cells were manually added for each recording to define the background signal. For calcium imaging in acute coronal slices, ROIs were manually labelled in Fiji. Three ROIs without obvious fluorescence were added for each recording to define the background signal.

The mean intensity of each ROIs at every time point in each recording was extracted using Fiji and then subtracted with the average mean intensity of 3 background ROIs without any cells to get F values. We used different methods to define baseline fluorescence F0 for different calcium imaging experiments. For spontaneous calcium imaging in co-culture and coronal slices, F0 was defined as the 8 percentile of the 50-frame sliding window. Then ΔF/F0 trace of each cell was computed by dividing F-F0 with F0. Calcium traces was then smoothed using smooth.spline in the package stats in R 4.3.1. Local peaks of calcium traces in each cell were obtained using findpeaks in the package pracma in R 4.3.1, with the peak amplitudes greater than 2.5 folds of S.D. of the whole ΔF/F0 trace in each cell as the threshold. Full width at half maximum (FWHM) for each peak was then calculated using a customized code in R 4.3.1. For calcium imaging in coronal slices with electrical stimulation, F0 was defined as the average F values of the 20 frames before first stimulation. The ΔF/F0 traces were then calculated without smoothing. Local peaks were identified manually basing on the start points and intervals of electrical stimulation.

Tumor Microtubes Staining, Imaging and Quantification

GFP-expressing DMG cells with (sgCHD2) or without (sgNeg) CHD2 depletion were co-cultured with neurons for 3 days and then fixed as shown above. The fixed coverslips were washed twice with PBS, incubated in blocking buffer (2.5% normal goat serum, 0.2% Triton X-100 in PBS) at 37°C for 1 hour, and then incubated in blocking buffer with rabbit anti-GFP antibody (Abcam, cat no. ab290, 1:1500) at 37°C overnight. The coverslips were washed 3 times with PBS, incubated at RT for 1 hour in PBS with 2.5% goat serum and goat anti-rabbit Alexa 488 (Invitrogen, cat no. A-11034, 1:500) and then washed 3 times with PBS. The coverslips were then mounted with Fluoro-gel II with Dapi (Electron Microscopy Sciences) for imaging using Eclipse Ni-U Fluorescent Microscope (Nikon). Tumor microtubes were manually measured using NIS Elements software (Nikon) and divided into 4 groups (< 50 μm, 50 – 100 μm, 100 – 150 μm, > 150 μm) for comparison.

Statistical Analysis and Data Visualization

All experiments involved in deep sequencing were independently repeated at least twice. Data were presented as the mean ± SD. The statistical significance was calculated by two-sided unpaired Student’s t-test unless otherwise specified in the R program (version 3.6.3) and GraphPad Prism 10. For the box plots, the central line denotes the median, the box limits the upper and lower quartiles, and the whiskers for 1.5× the interquartile range. Sample sizes and replications were reported in figure legends.

Significance:

Neurons drive the proliferation and invasion of glioma cells. Here we show that chromatin remodeler CHD2 controls the epigenome and expression of axon-guidance and synaptic genes, thereby promoting neuron-induced proliferation of H3.1K27M DMG and the pathogenesis of this deadly disease.

Data Availability Statement

The RNA-seq and CUT&RUN data in this study have been deposited in the Gene Expression Omnibus (GEO) under accession number GEO: GSE211074. All other data are available in the main text or the supplementary materials.