Epigenome Programming by H3.3K27M Mutation Creates a Dependence of Pediatric Glioma on SMARCA4

Pediatric gliomas are the leading cause of cancer-related mortality in children worldwide. Among this group of aggressive tumors, diffuse midline gliomas that are H3K27 altered (DMG; ref. 1), including diffuse intrinsic pontine gliomas (DIPG), are particularly devastating, with a median survival of 9 to 12 months from the time of diagnosis (2–4). Although DMGs initiate from the pons and midline of the brain, neuroimaging and autopsy studies indicate that in almost all DMG cases, tumor cells spread to adjacent tissues and other brain regions where they mingle with normal cells (5, 6). The localization at pons and pervasive invasion preclude surgical resection as a treatment option for patients with DMG. Thus, the current clinical management of patients with DMG is limited to radiation, which provides only transient relief from disease progression. Moreover, other targeted and conventional chemotherapies have yet to prove beneficial in this patient population. This highlights an urgent need to understand the molecular mechanisms driving tumorigenesis DMG cells and to identify novel targets.

Approximately 80% of DMG cases contain a heterozygous mutation at either H3-3A or HIST1H3B/C, which encodes histone H3 variant, H3.3 and canonical H3.1, respectively, resulting in the replacement of lysine 27 with methionine (H3K27M; refs. 7–9). In various preclinical animal models, the H3.3K27M mutation cooperates with TP53 and PDGFR mutations to drive the formation of DMG (10–13). Furthermore, expression of an H3K27M transgene together with TP53 mutation in mice also drives tumorigenesis of other tissues including lymphoma and carcinoma (14, 15). Together, these studies validate H3K27M as a driver of oncohistone mutation for DMG as well as other cancers.

At the molecular level, expression of H3K27M results in a global reduction of di- and trimethylation of histone H3 lysine 27 (H3K27me2/me3) on wild-type histone proteins (16–19). H3K27me2/me3 is catalyzed by Polycomb Repressive Complex 2 (PRC2) and plays an important role in silencing gene transcription during development (20). The global reduction of H3K27me2/me3 by H3K27M likely arises from multiple mechanisms including reduced enzymatic activities of the PRC2 complex, sequestered PRC2 complex at poised enhancers, and an inability of H3K27me2/me3 to spread to distal sites (18, 21–23). Despite the global reduction of H3K27me2/me3 levels, the residual occupancies of this methyl mark continue to be essential for silencing of tumor suppressor genes, including p16 and WT1 (16, 17, 21). Supporting this idea, it has been shown that Ezh2, the catalytic subunit of the PRC2 complex, is required for the viability of DMG cells in vitro and in vivo (24, 25).

In addition to H3K27me2/me3, other histone modifications are altered in DMG cells. For instance, H3K27 acetylation is increased in DMG cells (25, 26). Furthermore, H3K36 methylation, a mark that is counterbalanced by H3K27me2/me3, is also altered in DMG cells (27). These changes in histone modifications likely contribute to the global changes in gene expression and superenhancers marked by genes associated with oligodendrocyte precursor cells (17, 28), thereby rendering the DMG cells vulnerable to the inhibition of demethylases, deacetylases, and bromodomains that recognize acetylated histone lysine residues (25, 28–30). However, it remains elusive how DMG cells disseminate and invade surrounding brain tissues.

The mammalian BAF (SMARCA4- or BRM-associated factors) chromatin remodeling complexes belong to the SWI/SNF family of chromatin remodeling complexes. Biochemically, the SWI/SNF chromatin remodeling complexes slide or eject nucleosomes using the energy from ATP hydrolysis to increase chromatin accessibility for transcription factors and coactivators (31, 32). The BAF chromatin remodeling complexes can be classified as three different complexes, cBAF, PBAF, and ncBAF, based on subunit compositions, with either SMARCA4 [Brahma-related gene 1 (BRG1)] or SMARCA2 [Brahma-homolog (BRM)] as the catalytic subunit (33, 34). Each complex also contains unique subunits that specify its functions in a chromatin context– and/or cell type–dependent manner. Indeed, a recent study indicates that histone modifications not only influence the binding of the SWI/SNF complexes but also affect the activity of these complexes in vitro (34–36). Thus, it is not surprising that mammalian SWI/SNF complexes regulate the expression of genes in a context-dependent manner.

The context dependence is in full display in SMARCA4's roles in tumorigenesis. SMARCA4 mutations are frequently observed in diverse cancer types, including lung (37), small cell carcinoma of the ovary (38, 39), and skin (40). Most SMARCA4 mutations are missense mutations that are scattered within the highly conserved ATPase domain, suggesting that the chromatin remodeling activity of SMARCA4 is critical for tumor suppression. Thus, SMARCA4 is a bona fide tumor suppressor gene. In addition, SMARCA4 plays a protumorigenic role in several tissue contexts. For instance, the expression of SMARCA4 in breast cancer correlates with poor survival, and SMARCA4 is important for the proliferation of triple-negative breast cancer cells (41). In melanoma, SMARCA4 works with lineage-determinant microphthalmia-associated transcription factor (MITF) for melanocyte transformation (42). In acute myeloid leukemia, SMARCA4 governs enhancer-mediated gene regulation for tumor maintenance (43). The roles of SMARCA4 in tumor suppression and promotion are also exemplified in pancreatic cancer. In duct cells, SMARCA4 prevents dedifferentiation and inhibits tumor initiation. In contrast, SMARCA4 regulates a mesenchymal-like transcription state and promotes pancreatic ductal adenocarcinoma (44). However, to what extent SMARCA4 functions in DMG tumors is largely unknown.

Through a CRISPR/Cas9 screen targeting genes involved in chromatin regulation, we found that SMARCA4 is selectively essential to support the proliferation of DMG cells. We show that genetic or pharmacologic targeting of SMARCA4 results in reduced proliferation, migration, and invasion of H3.3K27M DMG cells in vitro and reduced tumor growth in vivo. Mechanistically, we found that SMARCA4 colocalizes with SOX10 at gene regulatory elements in H3.3K27M DMG cells to regulate the expression of genes involved in proliferation and ECM deposition. Finally, we provide evidence that an epigenomic rewiring by H3.3K27M can drive SMARCA4 dependency in the DMG context. Together, these studies reveal a previously overlooked mechanism that regulates the fitness of DMG cells and supports a hypothesis that drug targeting of SMARCA4 would have a therapeutic index in patients with DMG.

We hypothesized that the epigenome rewiring in H3K27M DMG cells creates a dependence of DMG cells on other chromatin regulators. To test this hypothesis, we performed a CRISPR/Cas9 screen using a library of 1,303 single-guide RNAs (sgRNA) targeting domains of 180 chromatin-modifying enzymes (45). The screen was performed in four DMG patient-derived cell lines, with three cell lines containing the H3.3K27M mutation and one the H3.1K27M mutation. As controls, we utilized three adult glioblastoma (GBM) cell lines with wild-type histone H3 (Supplementary Fig. S1A). Briefly, we stably expressed Cas9 in each of these cell lines, followed by lentiviral transduction with the sgRNA library. We collected cells 3 days after infection (T0) and after 10 population doublings (T1) and isolated genomic DNA for library preparation and next-generation sequencing. We then compared the abundance of each sgRNA at T1 to T0 in DMG cells and in adult GBM cells. We found that sgRNAs targeting eight genes were depleted at T1 compared with T0 in DMG cells, but not in three adult GBM lines (Supplementary Fig. S1B and Supplementary Table S1 for detailed information about the screen). EZH2, the catalytic subunit of the PRC2 complex, SMARCA4, the catalytic subunit of BAF chromatin remodeling complexes, and ELP3 were the top three among the most striking DMG-specific chromatin regulator dependencies. Previously, it has been shown that EZH2 is required for the proliferation of H3.3K27M DMG cells (17, 21, 24, 25). Further, ELP3 is needed for the fitness of H3.3K27M Drosophila cells (46). However, it was unknown whether SMARCA4 has any role in DMG, which consequently became the focus of our study.

To validate the results of the screen, we used GFP-based competition assays to analyze the impact of SMARCA4 depletion by two sgRNAs (sgSMARCA4-1 and sgSMARCA4-2) on the cell fitness of five H3K27M DMG cells (DIPG17, DIPG4, DIPG36, DIPG6, and SF8628), three GBM cells with wild-type H3 (GBM22, GBM43, and SF9427), and one GBM cell line with H3.3G34V (KNS42). In this assay, sgRNA and GFP are expressed from the same plasmid and the effect of a gene knockout on cell fitness is reflected in GFP⁺ cells becoming outcompeted over time. As shown in Fig. 1A and Supplementary Fig. S1C and SD, expression of an sgRNA for PCNA, an essential gene involved in DNA replication, led to cell fitness defects in all nine lines. In contrast, expression of sgRNA targeting the ROSA26 locus (sgNeg) had no apparent effects on any of the lines. Importantly, DMG cells expressing two sgRNAs targeting SMARCA4 were outcompeted by nontransduced cells, whereas SMARCA4 sgRNAs had no apparent effects on GBM cells. We also noticed that the effects of EZH2 depletion on proliferation appeared to be more pronounced on H3.3K27M-mutant cell lines than H3.1K27M-mutant lines (Fig. 1A). These results support the idea that DMG cells with both H3.1- and H3.3-K27M mutation are vulnerable to SMARCA4 depletion.

Next, we analyzed the effects of SMARCA4 depletion on cell proliferation using titer-blue cell viability assay and colony formation assays in three H3.3K27M DMG cell lines (DIPG6, DIPG17, and SF8628) and two GBM cell lines (GBM22 and KNS42). SMARCA4 proteins were efficiently ablated (Fig. 1B), and ablation of SMARCA4 reduced cell viability (Supplementary Fig. S1E) and the ability of three DMG lines to form colonies (Fig. 1C and D). In contrast, SMARCA4 depletion showed minimal effects on the proliferation of KNS42 and GBM22 lines (Fig. 1C and D; Supplementary Fig. S1E). We also found that SMARCA4 depletion in DMG lines (DIPG6 and DIPG17) resulted in reduced incorporation of EdU, a nucleotide analogue incorporated into DNA during S phase (Fig. 1E), and increased apoptosis (Fig. 1F), while having minimal effects on control GBM22 and KNS42 lines. These results indicate that SMARCA4 promotes proliferation and inhibits apoptosis of H3K27M DMG cells.

To rule out potential off-target effects of SMARCA4 sgRNAs, we expressed SMARCA4 tagged with the Flag epitope in both DMG lines (DIPG6 and DIPG17) from a plasmid that was resistant to sgSMARCA4-1 (Fig. 2A; Supplementary Fig. S2A). Exogenously expressed SMARCA4 fully rescued the proliferation defects caused by the depletion of endogenous SMARCA4 in both lines based on cell viability and colony formation assays (Fig. 2B–D; Supplementary Fig. S2B–S2D). Because of the large size of SMARCA4 proteins (∼180 kDa), the endogenous SMARCA4 could not be separated from exogenously expressed SMARCA4 tagged with the Flag epitope on SDS-PAGE gel (Fig. 2A; Supplementary Fig. S2A). We, therefore, transduced DIPG6 and DIPG17 cells expressing SMARCA4-Flag with a second-round lentivirus of sgSMARCA4-1 [sgSMARCA4-1 (+)] and monitored cell growth. Under these conditions, exogenously expressed SMARCA4 also restored the proliferation defects of DIPG6 and DIPG17 caused by the depletion of endogenous SMARCA4. Together, these results provide additional evidence that SMARCA4 is essential for the proliferation of DMG cells in vitro.

To test whether SMARCA4 is important for the proliferation of DMG cells in vivo, we infected DIPG17 cells containing the luciferase reporter gene with sgNeg or sgSMARCA4-1, implanted these cells into the pons of immunocompromised mice, and monitored the growth of DIPG17 xenografts using bioluminescence imaging. Analysis of the tumor growth of the five sgNeg and six sgSMARCA4-1 mice showed that depletion of SMARCA4 dramatically reduced tumor growth and improved mouse survival (Fig. 2E–G). In contrast, the depletion of SMARCA4 in the GBM22 patient-derived xenograft (PDX) line did not affect tumor growth and mouse survival (Supplementary Fig. S2E–S2G). The differential effects of SMARCA4 depletion on the cell viability of H3K27M DMG cells and glioma cells with wild-type H3 were unlikely due to high expression of SMARCA4 in H3K27M DMG lines (Supplementary Fig. S2H). Together, these results show that SMARCA4 is important for the proliferation of DMG cells in vitro and in vivo.

SMARCA4/BRG1 and SMARCA2/BRM, which share about 75% sequence identity, are mutually exclusive as the catalytic subunit of three mammalian SWI/SNF complexes, cBAF, PBAF, and ncBAF (34, 47). Therefore, we addressed whether SMARCA2 is required for cell viability of DMG using the competition-based proliferation assay. Cells with efficient depletion of SMARCA2 by two sgRNAs did not affect the growth of the three DMG cell lines tested (Supplementary Fig. S3A and S3B). We noticed that SMARCA2 was expressed in five H3K27M DMG cell lines and four other glioma lines analyzed (Supplementary Fig. S2H). Thus, SMARCA2 is not required for the proliferation of DMG cells. Next, we tested several unique and shared subunits of the cBAF, PBAF, and ncBAF complexes and asked whether they are required for the viability of DMG cells. We observed that depletion of SS18 (shared subunit of cBAF and ncBAF) and BRD9 (unique subunit of ncBAF) did not affect the proliferation of DMG cells (Supplementary Fig. S3C). Furthermore, depletion of ARID1A (unique subunit of cBAF) had little impact on three DMG cell lines (DIPG6, DIPG17, and SF8628) based on three independent assays (Supplementary Fig. S3D–S3H). In contrast, SMARCE1, a core component of mammalian SWI/SNF chromatin remodeler, was needed for the proliferation of DMG cells (Supplementary Fig. S3D), supporting the idea that SMARCA4 functions within a SWI/SNF complex. Finally, depletion of PBRM1, a unique subunit of the PBAF complex, reduced the fitness of DMG cells compared with nontransduced cells (Supplementary Fig. S3I), as well as the ability of three DMG cell lines (DIPG6, DIPG17, and SF8628) to form colonies, albeit to a lesser extent than SMARCA4 (Fig. 3A and B). Together, these results suggest that SMARCA4 functions as the catalytic subunit of the mammalian SWI/SNF chromatin remodeling complex to promote the viability of DMG cells.

BRM014, a newly synthesized ATPase inhibitor of SMARCA4 and SMARCA2, exhibits an efficient inhibition on both SMARCA4 and SMARCA2 in vitro (48). Therefore, we investigated the effects of BRM014 treatment on DMG cells. We observed a dose-dependent reduction of cell viability of five H3K27M DMG cells (Fig. 3C and D; Supplementary Fig. S4A and S4B) upon treatment with BRM014 based on cell viability assays. In contrast, three H3 wild-type GBM cell lines (GBM22, GBM43, and SF9427) showed minimal effects upon BRM014 treatment. Interestingly, KNS42 cells, which contain the H3G34V mutation, were also sensitive to BRM014 treatment but to a lesser extent compared with H3K27M DMG cells (Supplementary Fig. S4A and S4B). Using these assays, we also observed that BRM014 treatment of five human DMG cells led to reduced ability to form colonies (Fig. 3E), decreased EdU incorporation (Supplementary Fig. S4C), and increased apoptosis (Supplementary Fig. S4D), while having minimal effects on three GBM cell lines (GBM22, GBM43, and SF9427). Furthermore, combined treatment of BRM014 with radiation, the current standard therapy for patients with DMG, significantly reduced the viability of both DMG lines compared with either treatment alone (Supplementary Fig. S4E and S4F). Finally, depletion of SMARCA4 using an sgRNA in the H3.3K27M murine brain stem glioma (BSG) line (49) and treatment of these cells with BRM014 also reduced cell viability (Supplementary Fig. S4G–S4J) in vitro. Taken together, inhibition of SMARCA4 genetically and pharmacologically in multiple human and mouse DMG lines reduces the proliferation of H3K27M DMG cells in vitro.

To determine the anti-DMG tumor activity of BRM014 in vivo, we used a subcutaneous SF8682 xenograft mouse model and treated the mice with BRM014 (30 mg/kg) by intraperitoneal injection 5 days a week for 2 weeks. We found that BRM014 treatments led to significant inhibition of tumor growth and improved animal survival compared with vehicle controls (Fig. 3F and G). We also analyzed the effects of BRM014 treatment on tumor growth using an independent subcutaneous mouse xenograft model and observed that BRM014 treatment reduced tumor growth (Fig. 3H and I). There were no significant adverse effects on body weight during this course of the treatment (Supplementary Fig. S4K). Together, these results support that SMARCA4 ATPase catalytic activity, and by extension its chromatin remodeling activity, is important for the proliferation of DMG cells in vitro and in vivo.

The SMARCA4-containing chromatin remodeling complexes regulate gene transcription in a context-dependent manner (34). Moreover, H3.1K27M and H3.3K27M DMGs show distinct transcription profiles (9, 50). To understand why DMG cells are susceptible to SMARCA4 depletion, we decided to focus on analyzing the effects of SMARCA4 depletion on gene expression, histone modification, and transcription factor binding in H3.3K27M lines (DIPG6 and DIPG17), but not H3.1K27M DMG lines. Based on RNA sequencing (RNA-seq) analysis, we observed that depletion of SMARCA4 resulted in a marked perturbation of the transcriptome in both the DIPG6 and DIPG17 cell lines, with 1,304 and 1,381 downregulated genes, respectively (Fig. 4A). Interestingly, 198 genes were downregulated in both H3.3K27M DMG cell lines (Fig. 4B). Gene ontology analysis (GO) of these 198 downregulated genes indicated that genes associated with cell growth and extracellular matrix (ECM) pathways were enriched (Fig. 4C; Supplementary Table S2). We focused our subsequent analysis on these downregulated genes.

ECM is composed of structural proteins including collagen, proteoglycan, lamina, fibronectin, ECM regulators, and integrins (51, 52). We selected several genes involved in ECM composition and regulation [secreted frizzled-related protein 2 (SFRP2), latent transforming growth factor binding protein 4 (LTBP4), HSPG2, COL2A1, COL18A1, and ITGA7] and tested whether SMARCA4 depletion reduced the expression of these genes in DIPG6 and DIPG17 cell lines and two GBM cell lines without H3K27M mutation (GBM22 and KNS42) using qRT-PCR. We found that SMARCA4 depletion resulted in reduced expression of these genes in both DIPG6 and DIPG17 cells, but not in GBM cells (Fig. 4D; Supplementary Fig. S5A). Moreover, exogenous expression of SMARCA4 restored the reduced expression of SFRP2, HSPG2, and LTBP4 in both DIPG6 and DIPG17 cells caused by deletion of endogenous SMARCA4 (Supplementary Fig. S5B). The reduced expression of these three genes was also detected in DIPG6 and DIPG17 cells treated with BRM014 (Supplementary Fig. S5C). Analysis of published RNA-seq datasets (53) indicated that SFRP2 and LTBP4 were highly expressed in H3K27M DMG patient samples compared with high-grade glioma with wild-type H3 (Fig. 4E). SFRP2 is a secreted factor that modulates the WNT signaling pathway and is linked to tumor growth and ECM signaling. LTBP4 interacts with ECM and regulates TGFβ signaling. Therefore, we analyzed the effects of depletion of these two genes on the proliferation of DIPG17 cells and found that depletion of SFRP2 and LTBP4 significantly reduced the viability and colony formation ability of DIPG17 cells (Fig. 4F–I; Supplementary Fig. S5D–S5G). Together, these results indicate that SMARCA4 regulates the expression of genes involved in multiple pathways including cell growth and ECM in H3.3K27M DMG cells, which in turn contributes to the growth of these cells.

In addition to tumor growth, ECM also plays an important role in cell migration and invasion. Therefore, we tested whether SMARCA4 is needed for the migration and invasion of DIPG17 in vitro. Using a 3D migration assay and an invasion assay established by others (54, 55), we found that depletion of SMARCA4 in DIPG17 cells using two sgRNAs led to a dramatic reduction of the ability of DIPG17 cells to migrate on matrigel (Fig. 4J and K) and to invade into surrounding matrigel (Fig. 4L and M). Similarly, BRM014 treatment also significantly compromised the ability of DIPG17 cells to migrate and invade (Supplementary Fig. S5H–S5K). Taken together, these results indicated that SMARCA4 is also important for promoting migration and invasion of DMG cells in vitro, probably through regulating the expression of genes involved in ECM. However, we cannot rule out the possibility that the reduced migration and invasion of DIPG17 cells upon SMARCA4 depletion are due to reduced proliferation.

To identify genes that are regulated by SMARCA4 directly, we first analyzed chromatin binding of SMARCA4 in both DIPG6 and DIPG17 cells using the Cleavage Under Targets and Release Using Nuclease (CUT&RUN; ref. 56), followed by next-generation sequencing (Fig. 5A). More than 63% and 80% of SMARCA4 peaks in DIPG6 and DIPG17 from two independent SMARCA4 CUT&RUN experiments overlapped, respectively (Supplementary Fig. S6A). Moreover, of the 17,580 SMARCA4 peaks identified in DIPG6 cells and the 17,102 peaks identified in DIPG17 cells, more than 50% (8,734 peaks) were found in both DMG lines (Supplementary Fig. S6A). Based on our H3K4me3 CUT&RUN and published H3K27ac chromatin immunoprecipitation sequencing (ChIP-seq) datasets in both lines (28), we classified the 8,734 SMARCA4 peaks identified in both lines as promoters (marked by high levels of H3K4me3, 29%) and enhancers (marked with H3K27ac and low levels of H3K4me3, 47.5%; Fig. 5A). Similar analysis of total SMARCA4 peaks in either DIPG6 (17,580 SMARCA4 peaks) or DIPG17 (17,102 peaks) showed that SMARCA4 is largely localized at these two gene regulatory elements (Supplementary Fig. S6B and S6C). These results indicate that, like in other cell types (42, 57, 58), SMARCA4 likely binds to gene regulatory elements to regulate gene expression in DMG cells.

Next, we integrated the RNA-seq and SMARCA4 CUT&RUN datasets and asked whether SMARCA4 occupied promoters and enhancers of genes downregulated by SMARCA4 depletion in DIPG6 and DIPG17. We found that 325 and 367 downregulated genes in DIPG6 and DIPG17, respectively, contained at least one SMARCA4 CUT&RUN peak at their promoters or enhancers (Supplementary Fig. S6D). GO analysis on these genes indicated that these genes were also enriched with the ECM, cell growth and migration, and DNA replication in both DIPG6 and DIPG17 (Supplementary Fig. S6E and S6F). Finally, SMARCA4 binding at gene regulatory elements of three genes involved in ECM (SFRP2, HSPG2, and COL2A1) was confirmed by CUT&RUN quantitative PCR, and depletion of SMARCA4 resulted in a dramatic reduction of SMARCA4 CUT&RUN signals at these loci (Fig. 5B). Together, these results support the idea that SMARCA4 binds to gene regulatory elements of genes involved in cell proliferation and the ECM pathway and promotes their expression in H3.3K27M DMG cells.

To evaluate the consequence of SMARCA4 depletion on chromatin in DIPG17 cells, we analyzed the effects of SMARCA4 depletion on chromatin accessibility using assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) and on chromatin binding of H3K27ac and H3K4me3 using CUT&RUN. We found that depletion of SMARCA4 reduced the ATAC-seq signals at 34% of ATAC-seq peaks (18,156 out of 53,400 peaks; Supplementary Fig. S6G), and decreased the H3K27ac and H3K4me3 CUT&RUN signals at 86% (22,468/26,141 peaks) and 83% (16,881/20,434 peaks) of H3K27ac and H3K4me3 peaks, respectively (Supplementary Fig. S6H and S6I). Moreover, SMARCA4 levels based on CUT&RUN signals prior to SMARCA4 depletion at regions with reduced ATAC-seq signals, reduced H3K27ac, or reduced H3K4me3 were higher than at those without corresponding changes upon SMARCA4 depletion (Supplementary Fig. S6J–S6L), suggesting that the reduction of ATAC-seq signals, H3K27ac, and H3K4me3 upon SMARCA4 depletion is linked to SMARCA4. Next, we asked whether gene expression changes correlate with changes in ATAC-seq, H3K27ac, and H3K4me3 upon SMARCA4 depletion. We observed that ATAC signals and H3K27ac, but not H3K4me3 CUT&RUN signals, around genes with reduced expression were decreased compared with genes without changes in gene expression (Supplementary Fig. S6M–S6O), indicating that reduced gene expression upon SMARCA4 depletion is associated with reduced chromatin accessibility and H3K27ac, but not H3K4me3. Interestingly, genes with increased expression were associated with changes in H3K27ac and H3K4me3 signals, but not changes in ATAC-seq signals (Supplementary Fig. S6M–S6O). Finally, we calculated H3K27ac and H3K4me3 CUT&RUN density around 18,156 reduced ATAC peaks and observed that the levels of H3K27ac, but not H3K4me3, were reduced, indicating that the reduction of ATAC signals upon SMARCA4 depletion was likely accompanied by reduced H3K27ac (Supplementary Fig. S6P and S6Q). Together, these results indicate that SMARCA4 regulates chromatin accessibility, H3K27ac, and H3K4me3 chromatin binding in DIPG17 cells, which in turn affects gene expression.

SMARCA4 is recruited to gene regulatory elements likely through both transcription factors and local chromatin modifications. To gain insights into how SMARCA4 is recruited to gene regulatory elements in DIPG6 and DIPG17 cell lines, we analyzed DNA sequence motifs in 8,734 SMARCA4 peaks that overlapped in two DMG cell lines and found that these SMARCA4 peaks are enriched with a set of transcription factors including two SRY-box transcription factors (SOX10 and SOX9), basic leucine zipper ATF-like (BATF), nuclear factor 1 A-type (NFIA), and nuclear factor of activated T cells 5 (NFAT5), with SOX10 at the top of the list (Fig. 5C).

SOX10 is expressed in neural crest cells during early development (59). Genes regulated by SOX10 are activated in mouse brain cells expressing H3.3K27M-mutant proteins (14). To test whether SOX10 helps recruit SMARCA4 to gene regulatory elements, we first asked whether SOX10 interacts with SMARCA4. We found that SMARCA4 coimmunoprecipitated with SOX10 in DIPG17 cells (Fig. 5D). Next, we tested whether SOX10 colocalizes with SMARCA4 in DIPG6 and DIPG17 cells. Using CUT&RUN, we identified 14,766 and 17,197 SOX10 peaks in DIPG6 and DIPG17 cells, respectively, from two independent SOX10 CUT&RUN experiments in each cell line. Moreover, more than 50% of SOX10 peaks (9,106) were shared between DIPG6 and DIPG17 cells, and 31.8% and 44% of these shared peaks are localized at the promoter and enhancers, respectively (Fig. 5E). Importantly, we observed that in both DIPG6 and DIPG17 cell lines, SOX10 and SMARCA4 colocalized based on analysis of SOX10 CUT&RUN density at 17,580 SMARCA4 peaks in DIPG6 cells and at 17,102 SMARCA4 peaks in DIPG17 cells (Fig. 5F–I). Of note, colocalization of SOX10 with SMARCA4 at three selected SMARCA4 target genes was shown (Fig. 5J). These results indicate that SOX10 interacts and colocalizes with SMARCA4 in DIPG6 and DIPG17 cells.

We also found that SOX10 was a top candidate critical for the proliferation of DIPG17 cells in an independent CRISPR/Cas9 screen using a library of sgRNAs targeting transcription factors (Supplementary Fig. S7A). Therefore, we depleted SOX10 from DIPG6 and DIPG17 cells and analyzed the impact of SOX10 depletion on the expression of three ECM genes (SFRP2, HSPG2, and COL2A1). We observed that SOX10 depletion, like SMARCA4 depletion, reduced the expression of these three genes (Fig. 6A and B). Moreover, SOX10 bound to the promoters/enhancers of these genes, and this binding was reduced dramatically upon SOX10 depletion (Fig. 6C). Taken together, these studies indicate that SOX10, like SMARCA4, binds to gene regulatory elements of genes involved in ECM and promotes the expression of these genes directly.

Next, we analyzed the impact of SOX10 depletion on proliferation, migration, and invasion. We found that SOX10 depletion dramatically inhibited the fitness of DIPG6 and DIPG17 cells based on GFP-based competition assays (Supplementary Fig. S7B), cell viability assays (Fig. 6D), and colony formation assays (Fig. 6E; Supplementary Fig. S7C). Moreover, the depletion of SOX10 in DIPG17 cells also reduced the ability of these cells to migrate (Fig. 6F and G) and to invade (Fig. 6H and I). Finally, we determined how SOX10 depletion affects chromatin binding of SMARCA4 in DIPG17 cells. We found that SOX10 depletion dramatically reduced the association of SMARCA4 with chromatin globally (Fig. 6J–L; Supplementary Fig. S7D). These results support a model whereby SOX10 helps recruit SMARCA4 to gene regulatory elements, which in turn controls the expression of genes involved in cell growth and ECM pathways.

As DIPG6 and DIPG17 cells uniquely harbor H3.3K27M mutation compared with other glioma cell lines, we hypothesized that H3.3K27M mutation renders DMG cells dependent on SMARCA4. To test this hypothesis, we followed a published study and depleted the H3-3A gene from DIPG17 using an sgRNA targeting the H3.3K27M-mutant allele (26). We then analyzed the effects of SMARCA4 depletion on the proliferation of these H3.3K27M-depleted DIPG17 cells. We observed that the deletion of H3.3K27M slightly inhibited the proliferation of DIPG17 cells. However, the inhibitory effects of SMARCA4 depletion on cell proliferation were not observed in H3.3K27M-deleted DIPG17 cells compared with DIPG17 with H3.3K27M (Fig. 7A–D). The same results were obtained in the study of the effects of SMARCA4 depletion on H3.3K27M-deleted DIPG6 cells (Supplementary Fig. S8A–S8D). Moreover, in contrast to the reduced expression of three genes (SFRP2, COL2A1, and HSPG2) tested upon depletion of SMARCA4 in DIPG17 cells, SMARCA4 depletion had no additional effects on the expression of these three genes in H3.3K27M-deleted DIPG17 cells (Fig. 7E). We also analyzed the effects of SOX10 on the viability of H3.3K27M-deleted DIPG17 cells. Depletion of SOX10 dramatically reduced the viability of both DIPG17 and H3.3K27M-depleted DIPG17 cells (Supplementary Fig. S8E and S8F). Furthermore, SOX10 deletion also reduced the ability of H3.3K27M-deleted DIPG17 cells to form colonies (Supplementary Fig. S8G). These results indicate that the dependence of H3.3K27M DMG cells on SMARCA4 is most likely due to the presence of H3.3K27M mutation. In contrast, their dependence on SOX10 was not completely dependent on H3.3K27M mutation, suggesting that SOX10 also regulates gene expression in DIPG17 cells in an SMARCA4-independent manner.

To gain further insights into the relationship between H3.3K27M and SMARCA4 in H3.3K27M DMG cells, we first analyzed the distribution of H3.3K27M in DIPG6 and DIPG17 cells using CUT&RUN and asked whether SMARCA4 colocalizes with H3.3K27M-mutant proteins. We found that SMARCA4 peaks are enriched with H3.3K27M CUT&RUN signals in both DIPG17 and DIPG6 cells (Supplementary Fig. S8H and S8I). Next, we asked whether H3.3K27M is required for the expression and chromatin binding of SMARCA4 in DMG cells. Depletion of H3.3K27M-mutant proteins had no apparent effects on the overall levels of SMARCA4 in DIPG17 cells (Supplementary Fig. S8J). However, depletion of H3.3K27M-mutant proteins in DIPG17 cells resulted in reduced SMARCA4 occupancy at 14,302 peaks while having minor effects on 6,665 SMARCA4 peaks with a cutoff fold change (FC; sgH3.3K27M/sgNeg) <0.83 and FDR <0.05 (Fig. 7F and G). These results indicate that H3.3K27M-mutant proteins are needed for the proper chromatin distribution of SMARCA4 in H3.3K27M DMG cells.

To understand how depletion of H3.3K27M in DIPG17 cells affects SMARCA4 occupancy, we analyzed whether reduced SMARCA4 occupancy is associated with changes in H3K27ac and H3K27me3, two histone marks altered the most upon depleting H3.3K27M (Supplementary Figs. S7A and S8K) and SOX10 chromatin binding. The increased H3K27me3 peaks showed little overlap with the reduced SMARCA4 peaks (Supplementary Fig. S8L). Therefore, we asked whether changes in H3K27ac and SOX10 upon deletion of H3.3K27M correlated with changes in SMARCA4 occupancy. Depletion of H3.3K27M in DIPG17 resulted in reduced H3K27ac density at 17,070 peaks while having no apparent effects on H3K27ac levels at 7,135 peaks with a cutoff of FC (sgH3.3K27M/sgNeg) <0.83 and FDR <0.05 (Supplementary Fig. S8M). Moreover, we found that reduced SMARCA4 peaks showed a modest but significant correlation with reduced H3K27ac peaks genome-wide (Supplementary Fig. S8N). Furthermore, the average density of SMARCA4 at H3K27ac peaks with reduced density was reduced slightly more than at H3K27ac peaks without changes in density upon depletion of H3.3K27M-mutant proteins (Supplementary Fig. S8O). These results suggest that reduced levels of H3K27ac in DIPG17 cells after depletion of H3.3K27M play some, but not a major, role in the reduction of SMARCA4 occupancy in DIPG17 cells.

H3.3K27M depletion did not affect SOX10 protein levels (Supplementary Fig. S8J). However, H3.3K27M depletion led to reduced SOX10 occupancy at 6,732 SOX10 peaks while having minor effects on 10,748 SOX10 peaks with the same cutoff of FC (sgH3.3K27M/sgNeg) <0.83 and FDR <0.05 (Supplementary Fig. S8P) compared with control DIPG17 cells. Moreover, reduced SMARCA4 peaks also showed a modest correlation with the reduced SOX10 peaks (Supplementary Fig. S8Q). Furthermore, the average SMARCA4 CUT&RUN density at the 6,732 reduced SOX10 peaks was dramatically decreased, whereas SMARCA4 density was not altered much at the 10,748 unaltered SOX10 peaks upon depletion of H3.3K27M (Fig. 7H). These results are consistent with the idea that reduced SOX10 binding in H3.3K27M-depleted DIPG17 cells plays an important role in the reduction of SMARCA4.

To further analyze the relationships among reduced H3K27ac, SOX10, and SMARCA4 upon depletion of H3.3K27M mutation proteins, we first identified 12,867 regions with reduced H3K27ac alone, 2,450 regions with reduced SOX10 alone, and 4,282 chromatin regions with a reduction of both H3K27ac and SOX10 upon H3K27M depletion. We then calculated SMARCA4 density at each of these regions and found that SMARCA4 was reduced the most at regions with the reduction of both H3K27ac and SOX10 (Fig. 7I), which became more apparent based on the analysis of the average FCs (sgH3.3K27M/sgNeg) of SMARCA4 CUT&RUN density at these three regions (Fig. 7J). Finally, most of the 4,282 chromatin regions with the reduction of both SOX10 and H3K27ac upon depletion of H3.3K27M are enhancers (Fig. 7K), suggesting that epigenome rewiring by H3.3K27M generates novel enhancers marked by SOX10 and H3K27ac, which depend on SMARCA4 the most. Together, these results support a model whereby epigenome reprogramming by H3.3K27M-mutant proteins leads to an increase in H3K27ac and SOX10 chromatin binding, which in turn, alone and in combination, recruits SMARCA4 to gene regulatory elements in H3.3K27M DMG cells to regulate the expression of genes involved in cell growth and ECM pathways (Fig. 7L).

DMG is a fatal disease without an effective therapy in sight. Therefore, there is an urgent need to identify novel treatment targets. Here we show that SMARCA4, the catalytic subunit of BAF chromatin remodeling complexes, is essential for the proliferation, migration, and invasion of DMG cells in vitro and for tumor growth in vivo, providing strong preclinical evidence for targeting SMARCA4 as a viable treatment strategy for DMG tumors with H3K27M mutation.

SMARCA4 functions mutually exclusively with SMARCA2 as the catalytic subunit of BAF chromatin remodeling complexes. In several tumor types, SMARCA4 and SMARCA2 function redundantly to regulate expression of genes involved in proliferation, apoptosis, and migration (34, 60). We presented several lines of evidence indicating that SMARCA4, but not SMARCA2, is important for the proliferation of DMG cells. First, through a CRISPR/Cas9 screen, we found that SMARCA4, but not SMARCA2, is one of the top candidate genes that when depleted results in reduced proliferation of four DMG cell lines. Second, using in vitro competition assays, we observed that SMARCA4 depletion in three DMG cell lines renders these cells less competitive in growth compared with those without SMARCA4 depletion, whereas SMARCA2 depletion has no apparent effects. Third, using four independent assays (cell viability, colony formation, BrdUrd incorporation, and apoptosis), we found that depletion of SMARCA4 or inhibition by the small molecular inhibitor BRM014 in one mouse and multiple human DMG cell lines leads to reduced proliferation and increased apoptosis. Fourth, depletion of SMARCA4 in DIPG17 and inhibition of SMARCA4/SMARCA2 in SF8628 and one mouse H3.3K27M DMG line using the small-molecule inhibitor BRM014 reduce tumor growth in vivo. Finally, we observed that SMARCA4 deletion or inhibition by BRM014 decreases the ability of DMG cells to migrate and invade in vitro. These results indicate that SMARCA4 is essential for the proliferation, migration, and invasion in DMG cells.

SMARCA4 regulates gene expression in a context-dependent manner (34). We found that SMARCA4 depletion in two H3.3K27M DMG lines (DIPG6 and DIPG17) results in reduced expression of genes involved in cell growth and ECM. Furthermore, SMARCA4 binds to the gene regulatory elements of genes involved in these processes. In other tumor types, it is well documented that ECM remodeling plays an important role in tumor cells to migrate and to invade (61). Therefore, we suggest that SMARCA4 regulates the expression of genes involved in cell growth and ECM remodeling, which promotes proliferation, migration, and invasion of DMG tumors. It has been shown previously that several ephrin and EPH receptors are associated with super-enhancers found in DMG cells. Moreover, inhibition of EPH signaling in DMG cells dramatically reduced cell invasion (28). EPH signaling plays an important role in axon guidance during normal development (62). Therefore, it is possible that multiple pathways, including ECM remodeling and EPH signaling, contribute to the migration and invasion of DMG cells.

SMARCA4 is recruited to gene regulatory elements in part through its interaction with transcription factors (34). For instance, in melanoma, SMARCA4 interacts with MITF, the master regulator of melanocyte lineage. Moreover, MITF works with SOX10 to recruit SMARCA4 to MITF-associated gene regulatory elements (42, 63). In hematopoietic stem cells and other settings, SMARCA4 interacts with GATA1 and EKLF transcription factors and regulates the differentiation of stem cells to erythrocytes (57). Here, we show that SOX10 helps recruit SMARCA4 to gene regulatory elements in H3.3K27M DIPG cells (DIPG6 and DIPG17) to regulate expression of genes involved in cell growth and ECM. First, we found that SOX10 binding sites are enriched at SMARCA4 peaks commonly identified in both DIPG6 and DIPG17 cells. Consistent with this observation, SMARCA4 interacts with SOX10 and colocalizes with SOX10 at the gene regulatory elements in both DIPG6 and DIPG17 cells. Second, depletion of SOX10 results in a dramatic reduction in SMARCA4 occupancy and decreased expression of genes that are also regulated by SMARCA4. Third, like SMARCA4, SOX10 is essential for the viability, invasion, and migration of H3.3K27M DMG cells. Together, these studies support a model whereby SOX10 recruits SMARCA4, likely through direct protein–protein interaction, to promoters and enhancers of genes involved in cell growth and ECM pathways, which in turn promotes proliferation, migration, and invasion of H3.3K27M DMG cells. In the future, it would be interesting to determine whether SOX10 and other transcription factors whose consensus DNA sequences are found in SMARCA4 CUT&RUN peaks are involved in regulating SMARCA4 chromatin binding in H3.1K27M DMG cells.

We have shown that SMARCA4 is important for the proliferation of both H3.1K27M and H3.3K27M DMG cells, but not adult GBMs with wild-type histone H3. In this study, we provided multiple lines of evidence indicating that the dependence of H3.3K27M DMGs on SMARCA4 is due to the presence of H3.3K27M mutation. We found that ablation of H3.3K27M in either DIPG6 or DIPG17 eliminates the dependence of the DMG cells on SMARCA4 for proliferation. Furthermore, the depletion of H3.3K27M dramatically reduces SMARCA4 chromatin binding in DIPG17 cells. Together, these results indicate that the dependence of H3.3K27M DMG cells on SMARCA4 is due, at least in part, to the H3.3K27M mutation. In contrast, the effects of SOX10 depletion on the proliferation of DIPG17 cells were not completely dependent on the presence of H3.3K27M mutation, suggesting that SOX10 likely also regulates gene expression in a SMARCA4-independent manner. In the future, it would be interesting to determine whether the dependence of H3.1K27M DMG tumors on SMARCA4 is due to H3.1K27M mutation.

Why is the chromatin binding of SMARCA4 dependent on H3.3K27M mutation? We suggest that epigenome rewiring by H3.3K27M-mutant proteins helps recruit SMARCA4 to gene regulatory elements unique in DMG cells. Supporting this idea, we observed that the reduction of SMARCA4 occupancy correlates with decreased H3K27ac and SOX10 induced by depletion of H3.3K27M. Importantly, the SMARCA4 occupancy is reduced the most at loci with reduction of both H3K27ac and SOX10, and these loci are largely enhancers. These results suggest that multiple changes including SOX10 binding and an increase in H3K27ac induced by H3.3K27M contribute to the recruitment of SMARCA4 to gene regulatory elements in particular enhancers in H3.3K27M DMG cells. Consistent with this idea, it has been shown that the expression of H3.3K27M-mutant proteins in E14 midbrain and hindbrain of mouse activates a transcriptome enriched for PRC2 and SOX10 targets (14). Furthermore, SOX10 is one of the genes enriched at superenhancers identified in DMG cells (28). During normal development, SOX10 promotes the generation of oligodendrocyte precursor cells (OPC), the cell origin of DMG tumors (64). We suggest that rewring of the epigenome of OPC by H3.3K27M-mutant proteins creates a dependence of these cells on SOX10 and SMARCA4 for the regulation of expression of genes involved in proliferation and ECM, which in turn promotes tumorigenesis and invasiveness of H3.3K27M DMG tumors (Fig. 7L).

The patient-derived cell lines (DIPG6, DIPG17, DIPG4, and DIPG36) were kind gifts from Dr. Michelle Monje (Stanford University School of Medicine). DIPG6, DIPG17, DIPG4, and DIPG36 were all cultured in neuro stem medium (KnockOut/F12 medium supplemented with 10% neuro stem supplement, 20 μg/mL EGF, 20 μg/mL human basic FGF, 1% penicillin/streptomycin, and 1% L-glutamine). GBM22, GBM43, and GBM28 were gifts from Dr. Jann Sarkaria (Mayo Clinic) and cultured in 10% FBS DMEM supplemented with L-glutamine and penicillin/streptomycin. SF8628, KNS42, and SF9427 cell lines were cultured in DMEM supplemented with 10% FBS, penicillin/streptomycin, and L-glutamine as previously described (21). 293T cells were purchased from ATCC. Murine BSG spheres expressing H3.3K27M were gifts from Dr. Oren Becher (Mount Sinai). Cells were cultured in mouse Neurocult media supplemented with 10% NeuroCult Proliferation Supplement (Mouse & Rat), 10 ng/mL EGF, 20 ng/mL human basic FGF, and 2 μg/mL Heparin. All the cells were cultured in 37°C, 5% CO₂, and 95% humidity and were tested monthly for Mycoplasma.

Ten-week-old, female NSG mice (NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ, The Jackson Laboratory) were used in SMARCA4 depletion in the DIPG17-PDX study. Six-week-old, female athymic nude mice (Athymic Nude-Foxn1^nu, homozygous) purchased from Envigo (IN, USA) were used in SMARCA4 depletion in GBM22 xenograft, BRM014 treatment on SF8628 and murine BSG study. The experiments involving the SF8628-PDX model were performed at Northwestern University. All other studies were performed at the Irving Cancer Research Center at Columbia University Irving Medical Center. All animals were maintained in pathogen-free facilities. All animal experiments were performed under Columbia University or Northwestern University Institutional Animal Care and Use Committee–approved protocols and complied with the ethical regulations and humane endpoint criteria according to the NIH Guide for the Care and Use of Laboratory Animals.

To generate the lentivirus for depletion using sgRNA or protein expression, 6 μg plasmid used was cotransfected in 293T cells with 4.5 μg psPAX2 and 1 μg pMD2.G by PEI reagents (Polysciences, 23966). Supernatants containing virus particles were collected at 48 hours and 72 hours following transfection. After filtration through a 0.45-μm filter, virus particles were concentrated by centrifugation at 3,000 rpm for 30 minutes after incubation with polyethylene glycol solution at 4°C overnight. The precipitated virus particles were dissolved in KnockOut DMEM/F12.

To generate Cas9-expressing cell lines, DMG cells (DIPG6, DIPG17, DIPG4, DIPG36, and SF8628) and GBM cells (GBM22, GBM43, GBM28, KNS42, and SF9427) were digested and seeded at a density of 2 × 10⁵ cells in one well of a 6-well plate in the presence of LentiV-Cas9-puro virus. Infection was performed with a spinning of 1,200 rpm for 60 minutes at room temperature. Following centrifugation, cells were transferred to incubators for regular culture. The medium was refreshed 24 hours after infection and 0.5 to 1 μg/mL puromycin was added 48 hours after infection. After selection for 3 days, a second round of spin infection was performed and 0.75 to 1.5 μg/mL of puromycin was added for further selection. The expression of Cas9 in each line was analyzed by Western blot, and the activity of Cas9 in each line was measured by GFP depletion (competition) assays using an sgRNA targeting the essential gene PCNA, which is coexpressed with GFP from the same plasmid. Normally, at least 20-fold reduction of GFP-expressing cells as determined by flow cytometry (Attune NxT Acoustic focusing cytometry, Thermo Fisher Scientific) after 5 doublings was used as the standard of robust Cas9 activity suitable for screening.

To generate SMARCA4- or SOX10-depleted cells, the virus titer of each sgRNA for SMARCA4 or SOX10 was measured. The spin infection described above was performed with a high multiplicity of infection (MOI; ∼2.1). To generate DIPG6 and DIPG17 cells exogenously expressing SMARCA4, Cas9-expressing DMG cells (DIPG6 and DIPG17) were transduced with virus for sgSMARCA4-1 to remove endogenous SMARCA4. These cells were then infected with virus for the expression of SMARCA4-FLAG. Forty-eight hours after infection, cells were selected with hygromycin for a duration of 12 to 14 days.

To generate DIPG17 cells expressing firefly luciferase (FLUC), DIPG17 cells were infected with different amounts lentivirus expressing FLUC. Forty-eight hours after infection, cells were screened for transfection efficiency by treatment of luciferin (D-luciferin potassium salt) in vitro and examined by the IVIS spectrum imaging system. Cells with over 80% infection efficiency were used for implantation in mice.

Each of the Cas9-expressing DMG cells (DIPG17, DIPG6, DIPG4, and SF8628) and GBM cells (GBM22, GBM43, and GBM28) were infected using an sgRNA library consisting of 1,303 sgRNAs targeting domains of 180 chromatin-modifying enzymes (45) at a low MOI (∼0.3) to ensure that each cell received about one sgRNA. Three days after infection, a fraction of cells were collected as reference (T0) for the sgRNA abundance. The remaining cells were then passaged and grown for an additional 10 doublings (T1) for each line. Genomic DNA at T0 and T1 samples of each line was isolated for library preparation by following published protocols. Genes whose depletion resulted in reduced fitness were identified using MAGeCK (65).

The constitutive Cas9 expression construct (LentiV-Cas9-puro) and LRG2.1 backbone (U6-sgRNA-EFS-GFP) were provided by C.R. Vakoc. To produce plasmid expressing sgRNAs, two DNA oligos were designed using http://chopchop.cbu.uib.no and were synthesized at IDT and cloned into Bsmb1-digested LRG2.1T. A majority of the sgRNAs have a high score to minimize the off-target effect. shLTBP4 was purchased from Sigma (TRCN0000055829 and TRCN0000055830). All sgRNAs and short hairpin RNAs used in this study are provided in Supplementary Table S3.

To express SMARCA4 exogenously, the SMARCA4-pLX313 plasmid was subcloned from pCMV5 SMARCA4 (Jun Qin, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, China) into a lentivirus-based pLX-TRC313 vector using the Gateway LR Clonase II enzyme mix (Thermo Fisher Scientific, 11791-020). Site-directed mutagenesis was performed to alter the CCCGAGCTC to CCGGAACTG (synonymous mutation), so that the plasmid was resistant to sgSMARCA4-1. The FLUC plasmid was a gift from Dr. Hashizume (Northwestern University).

The GFP depletion assay was performed using sgRNAs in the LRG2.1 vector. Cas9-expressing cells were infected with relevant sgRNA at MOI ∼0.3 with spinning infection as described above in a 6-well plate. After medium changes, cells at different times after infection were collected for flow cytometry analysis (Attune NxT Acoustic focusing cytometry, Thermo Fisher Scientific). GFP-positive cells at each time point were analyzed by FCS Express software (De Novo software). The percentage of GFP-positive cells at each time point over those after 3 days of infection was calculated.

To analyze cell proliferation, cells were plated in 96-well plates at least triplicate at a density of 1 × 10³ cells per well in 100 μL culture medium. Cell proliferation was measured by the CellTiter-Blue assay kit (Promega, cat. #G8081) according to the manufacturer's protocol. Briefly, 20 μL of CellTiter-Blue reagent was added into each well. After 2 to 4 hours of incubation at 37°C, fluorescence was measured by GloMax explorer (Promega) at 560/590 nm. Cell viability was calculated by subtracting the average fluorescence value of the related culture medium background and normalized against those at starting time (day 0). To test the effect of BRM014 on cell proliferation, cells were seeded at a density of 1 × 10³ cells per well in 50 μL culture medium, and another 50 μL medium containing either DMSO or 2-fold serial dilution of different concentration of BRM014 was added. Fluorescence intensity was measured each day from day 0 to day 6 after the addition of BRM014.

To analyze apoptosis, cells with SMARCA4 depletion (or treated with 1 μmol/L BRM014 for 5 days) were seeded at 2 × 10⁵/well (6-well plate) for 24 hours and were collected for double staining with Pacific Blue annexin V and propidium iodide (PI) according to the manufacturer's protocol, followed by flow cytometry analysis. Cells stained by both PI and annexin V were gated and defined as the late apoptosis population.

To analyze the effects of SMARCA4 depletion or SMARCA4 inhibition by BRM014 on DNA synthesis, cells with SMARCA4 depletion or treated with 0.1 μmol/L and 1 μmol/L BRM014 and DMSO controls for 5 days were seeded at 2 × 10⁵/well (6-well plate) for 24 hours. Cells were then pulsed with 20 μmol/L EdU (diluted in the warm culture medium) for 4 hours, and collected and digested with accutase solution. Seventy percent ice-cold ethanol was used to fix the cells overnight. After incubation with extraction buffer (0.1% Triton X-100) for 5 minutes at 4°C, click reactions were performed in the dark by adding the reaction cocktail [100 mmol/L Tris-HCL (pH = 8.5), 100 mmol/L CuSO₄, 1 mmol/L BDP TMR azide, and 0.5 M ascorbic acid] for 30 minutes at room temperature. BrdUrd-positive cells were analyzed by flow cytometry.

Cells were disassociated by accutase solution and filtered with 40-μm strainers (Fisher Scientific, cat. #08-771-1) for single-cell suspension. At least triplicate of each sample was seeded in 6-well plates coated with a low concentration matrigel (at 37°C for 1 hour) at a density of 1,000 to 4,000 cells/per well depending on the cell line used (1,000 cells/well for DIPG17, DIPG6, GBM22, GBM43, and KNS42; 2,000 cells/well for SF9427; and 4,000 cells/well for SF8628). After incubation for 2 to 3 weeks, colonies were stained with 0.1% crystal violet in 10% ethanol for 20 to 30 minutes and quantified. Relative colony formation was calculated by normalizing to corresponding controls (e.g., for SMARCA4/SOX10 depletion, relative colony formation was normalized by comparing with sgNeg of each cell line, and for BRM014 treatment, relative colony formation was normalized with samples treated with DMSO).

3D migration assays were performed as previously described (28, 55) with modifications. Briefly, DIPG17 cells were digested by accutase solution at 37°C and filtered by a 40-μm strainer for a single-cell suspension. Cells (1 × 10³) in the 200-μL medium were seeded into an Ultra-Low Attachment (ULA) round-bottom, 96-well plate (COSTAR, 7007) and cultured for 4 days to facilitate spheroid formation. The regular flat 96-well plate (SARSTEDT, 83.3924) was coated with matrigel (125 μg/mL) in the stem cell medium for 1 hour at 37°C. Once coating was completed, 200 μL medium/well was added in the coated plate (for inhibitor testing, the medium contains DMSO or 1.5× final concentration of BRM014). After removing 100 μL medium, spheres in the remaining 100 μL medium with a diameter around 250 μm to 300 μm were transferred into the matrigel-coated, 96-well plate. At different times (0, 24, and 48 hours) after transfer to the matrigel-coated plate, images were taken under a microscope (Life Technology, EVOS FS) with standard exposure and gamma setting. The area of migration was quantified by ImageJ, and relative migration distance was normalized by 0-hour time point.

The 3D invasion assay was performed as previously described (28, 54) with modifications. Briefly, after the formation of a spheroid in ULA round-bottom, 96-well plates as described above, 50 μL matrigel (Corning Science, 354234) was added to the neurosphere of DIPG17 cells with a diameter around 300 to 500 μm and then incubated at 37°C for 1 hour. After the medium matrigel was solidified, 50 μL neuro stem cell medium was added (for inhibitor testing, the medium contains DMSO or 3× final concentration of BRM014). Images were taken every 24 hours and the invasion area was quantified using ImageJ, and relative invasion distance was normalized by 0-hour time point.

To perform Western blotting, cells were counted and resuspended in Laemmli buffer at a density of 1 × 10⁵ cells/10 μL. Gel electrophoresis was performed using 4% to 15% SDS-PAGE gels, and proteins were transferred to the PVDF membrane (Bio-Rad). After blocking with 5% milk for 1 hour in room temperature, membranes were incubated with antibodies against protein of interest in TBS with 5% milk at 4°C overnight. The following antibody dilutions were used for Western blot analysis: SMARCA4 (1:1,000, Santa Cruz Biotechnology, cat. #sc-17796), PBRM1 (1:2,000, Bethyl Laboratories, cat. #A301-591A), SOX10 (1:2,000, Abcam, cat. #ab155279), H3K27M (1:1,000, ReVMab Bioscience, cat. #31-1175-00), H3K27me3 (1:1,000, Cell Signaling Technology, cat. #9733), ARID1A (1:1,000, Cell Sig­naling Technology, cat. #12354), SMARCA2 (1:1,000, Cell Signaling Technology, cat. #11966), LTBP4 (1:100, Santa Cruz Biotechnology, cat. #sc-393666), SFRP2 (1:500, Abcam, cat. #ab137560), FLAG (1:500, Sigma-Aldrich, cat. #F1804), and α-Tubulin (1:5,000, Sigma-Aldrich, cat. #T9026).

To perform immunoprecipitation assays, 1 × 10⁷ cells were lysed with the buffer (50 mmol/L HEPES–KOH, pH 7.4, 200 mmol/L NaCl, 0.5% NP40, 10% glycerol, 1 mmol/L EDTA and proteinase inhibitors) and homogenized 25 times by dounce homogenizer. After clarification by centrifugation, the lysates were incubated with the SMARCA4 antibody at 4°C overnight. Sepharose protein G beads were then added for incubation for more than 1 hour. The beads were washed using washing buffer (50 mmol/L HEPES–KOH, pH 7.4, 100 mmol/L NaCl, 0.01% NP40, 10% glycerol, 1 mmol/L EDTA, and proteinase inhibitors) 5 times, each for 5 minutes. Samples were dissolved in Laemmli buffer and analyzed by Western blotting.

Patient-derived DIPG17 cells expressing FLUC were infected with high titer sgRNA targeting the ROSA26 locus (sgNeg) wild-type or sgSMARCA4-1 targeting SMARCA4. On day 4 after infection, DIPG17 cells in neurospheres were digested into single-cell suspensions and injected stereotactically into the right pons of the 10-week-old, female NSG mice. To start the intracranial injection, mice were injected intraperitoneally with buprenorphine 0.05 mg/kg and anesthetized with 87.5 mg/kg ketamine and 12.5 mg/kg xylazine in 0.9% saline. The top of the head was swabbed with betadine and alcohol, and the skin was opened by a 1.5-cm incision with a scalpel. A small burr hole was created at 1.5 mm to the right of the midline and 1 mm posterior to the lambdoid suture. Mice were injected with 1 μL cell suspension (100,000 cells/μL) into the pontine tegmentum at 3.0-mm deep from the inner base of the skull using a sterile Hamilton syringe and quintessential stereotaxic injector. After delivery, the syringe was left in place for another 3 minutes before it was removed. The wound was closed with 5′o vicryl sutures. After surgery, mice were allowed to recover from anesthesia on a warming pad. Mice were checked daily for the signs of stress and euthanized when they developed neurologic symptoms, including loss of 20% body weight, loss of balance, and ataxia. Twenty-four hours later, 5 mg/kg carprofen was administered every 24 hours and continued for 4 days.

For the GBM22-PDX model, almost all procedures were performed the same way with DIPG17 except that 6-week-old, athymic female mice were used, and the injection was performed at the supratentorial brain, which was located at 3 mm right side from bregma and 3 mm depth from the skull.

For survival studies, two criteria were used: a reduction of body weight by 20% or severe neurologic motor deficits consistent with brain stem dysfunction.

Tumor growth was monitored using the IVIS spectrum optical in vivo imaging system (PerkinElmer) starting at week 11 after injection of DIPG17 cells with every 3- or 4-week frequency and starting at week 3 after injection of GBM22 cells twice weekly. Briefly, mice were injected with 150 mg/kg luciferin-D potassium salt (Golden Biotechnology) dissolved in PBS and anesthetized with 2% isoflurane in medical air. Bioluminescence imaging was acquired serially 12 minutes after luciferin-D potassium was injected, and peak signal intensity at selected regions of interest was quantified by Living Imaging Software (Xenogen Corp.) and expressed as photon flux (p/sec/cm²/sr). Relative tumor size/volume in each mouse was calculated by normalizing luminescence readings at specific time points against the corresponding one at week 11 for DIPG17 and at week 3 for GBM22. The growth curve was drawn by GraphPad software. Representative images shown were adjusted with minimal and maximal thresholds.

For the subcutaneous xenograft model, SF8628 cells were implanted into the flank of athymic mice as previously described (29). Briefly, 4 × 10⁶ cells in 0.4 mL of cell culture media with matrigel (BD Biosciences) at a 1:1 ratio were injected in the right flank of mice under anesthetization by isoflurane. Mice were randomly assigned to vehicle (DMSO, n = 7) and BRM014 treatment (30 mg/kg for 5 days a week for 2 weeks, n = 8) groups when the size of tumor reached 100 mm³. The tumor sizes were measured twice a week, and the mice were euthanized when the tumor size reached 1,000 mm³. All protocols described here were approved by Northwestern University Institutional Animal Care and Use Committee.

For the subcutaneous brain stem glioma mouse model, 1 × 10⁶ murine BSG cell spheres were digested and resuspended in 0.1 mL culture medium with matrigel and were injected in the right flank of mice. Treatments with BRM014 were initiated when tumor xenografts reached around 90 mm³. Eighteen mice were randomly assigned to vehicle [10% DMSO + 10% Kolliphor HS + 80% (10% HP-β-CD sterile water), n = 9] and BRM014 (n = 9) treatment groups. BRM014 was administered by intraperitoneal injection 5 days per week at 20 mg/kg for 5 weeks. BRM014 was dissolved before use in 10% DMSO + 10% Kolliphor HS + 80% (10% HP-β-CD sterile water). Tumor length/width and mouse weight were measured twice weekly. Mice were sacrificed after the treatment, and tumors were extracted and weighed.

Total RNA was extracted using an RNeasy Mini Kit (Qiagen, 74104). cDNAs were synthesized by the SuperScript III First-Strand kit (Invitrogen, cat. #18080-051) using 500 ng total RNA. Real-time PCRs were performed using 0.1 μmol/L primers and SYBR Green PCR Master Mix (Bio-Rad, cat. #1725124). β-Actin was used as an internal control for the normalization of gene expression.

To analyze the effect of SMARCA4 depletion on gene expression using RNA-seq, DMG cells (DIPG17 and DIPG6) were infected with sgSMARCA4-1, sgSMARCA4-2, and sgNeg. Cells were collected at day 8 after infection. Total RNAs were purified. After quality control analysis, RNA-seq library preparation and deep sequencing were performed by the Columbia University Genome Center. Paired-end sequence reads from the sequencing platform of the Columbia University Genome Center were downloaded. Sequence reads were aligned to the hg19 genome using STAR (version 2.7.6a; ref. 66) after trimming low-quality reads using Trim Galore (version 0.6.7). The coding gene annotation of hg19 was downloaded from GENCODE (67). Unique mapped reads were extracted using SAMtools (version 1.11; ref. 68). The reads within exon regions were annotated using featureCounts (version 2.0.1; ref. 69). Transcripts per kilobase million (TPM), calculated from counts of all samples, were used to determine the expression levels using TPMCalculator (70). Differentially expressed genes (DEG) were identified with edgeR (version 3.34.0; ref. 71) using generalized linear models (GLM) and quasi-likelihood F-tests. First, the dispersion was estimated from the samples in duplicate, with a GLM with a factor for every condition. A GLM was fitted with a factor for each cell line (wild-type vs. mutant) using these dispersion estimates. P values were adjusted after multiple testing by FDR correction. DEGs were defined according to the cutoff criteria: FDR <0.05 and FC >1.5. GO term enrichment analysis was performed using the “enrichGO” function from the clusterProfiler package (version 3.18.0; ref. 72).

CUT&RUN was performed as previously described (56) with minor modifications. DMG cells (5 × 10⁵; 10% mouse embryonic stem cells or murine BSG cells were added if a spike in was needed) were harvested using accutase and fixed with freshly prepared 0.5% PFA for 2 minutes. After stopping the cross-linking using 500 μL 2.5 M glycine, cells were washed 3 times with digitonin wash buffer (20 mmol/L HEPES, pH 7.4, 150 mmol/L NaCl, 0.5 mmol/L spermidine, protein inhibitor cocktail, 0.02% digitonin, 0.05% SDS, and 1% Triton X-100) and bound to Concanavalin A–coated beads (Polysciences, cat. #86057-10) prewashed with binding buffer (20 mmol/L HEPES, pH 7.4, 10 mmol/L KCl, 1 mmol/L CaCl₂, 1 mmol/L MnCl₂). Cells on beads were then incubated with antibodies on a nutator overnight at 4°C [1:100 SMARCA4 antibody (Cell Signaling Technology cat. #37966, 1:250 for SOX10 antibody (Abcam, ab155279), 1:500 for H3K27ac antibody (Cell Signaling Technology, cat. #8173), 1:1,000 for H3K27M antibody (ReVMab Bioscience, cat. #31-1175-00), 1:1,000 for H3K4me3 antibody (Abcam, cat. #ab8580), and 1:100 for H3K27me3 (Cell Signaling Technology, cat. #9733)]. After overnight incubation, cells were washed 3 times with 0.02% digitonin wash buffer, resuspended in 100 μL digitonin buffer containing Protein A-Micrococcal Nuclease fusion protein (pA-MNase) and second antibody complex assemblies (pA-MNase and second antibodies were preassembled at a 2:1 ratio in 50% glycerol at 4°C for 1 hour) and nutated at 4°C for 1 hour. After pA-MNase–second antibody binding, cells were washed 3 times with digitonin wash buffer for 5 minutes. Finally, cells were resuspended in 100 μL ice-cold digitonin buffer in a heating block precooled in the ice water bath. Digestion of the chromatin was initiated by the addition of 2 mmol/L CaCl₂ in the tube and lasted for 60 minutes at 0°C. The digestion was stopped by addition of 100 μL 2× STOP buffer (340 mmol/L NaCl, 20 mmol/L EDTA, 4 mmol/L EGTA, 0.02% digitonin, 100 μg/mL RNase A, and glycogen 50 μg/mL). Digested DNA was released by incubation at 37°C for 30 minutes, and the supernatant was collected. Then, 5 μL proteinase K and 200 μL 2× elution buffer [20 mmol/L Tris-HCl (pH = 8.0), 300 mmol/L NaCl, 20 mmol/L EDTA, 2% SDS, 10 mmol/L DTT] was added to each sample for reverse cross-linking at 65°C overnight. DNA was extracted by phenol–chloroform and dissolved in 12 μL low EDTA TE buffer. Libraries were prepared by the ACCEL-NGS 1S plus DNA library kit, and samples and were sequenced using an Illumina NextSeq 500 platform.

The ATAC-seq assays were performed in duplicates according to previously described protocols with minor adaptations (73). Briefly, 50,000 DIPG17 cells were washed with cold PBS and resuspended in lysis buffer (10 mmol/L Tris-HCl, pH 7.5, 10 mmol/L NaCl, 3 mmol/L MgCl₂, 0.1% NP40, 0.1% Tween-20, and 0.01% digitonin) and incubated on ice to extract the nuclei. The nuclei were centrifuged at 500 × g for 10 minutes, and the nuclei pellets were resuspended with 50 μL transposition reaction mix (25 μL 2XTD buffer, 16.5 μL 1 × PBS, 1% Tween-20, 0.01% digitonin, and 2.5 μL Tn5 transposase) and incubated for 30 minutes at 37°C. DNA was purified by using a MinElute PCR purification kit (Qiagen) and eluted in 10 μL elution buffer. The eluted DNA was submitted to PCR with NEBnext High-Fidelity master mix and was amplified with PCR (5 minutes at 72°C, 1 minute at 98°C, 12 cycles of 10 seconds at 98°C and 10 seconds at 63°C), and purified using Agencourt AMPure XP beads. DNA concentration was measured using the Qubit dsDNA HS assay kit and Bioanalyzer instrument and pooled. Libraries were sequenced using NextSeq 500.

CUT&RUN and ATAC libraries were sequenced via the paired-end method using Illumina sequence platforms (NextSeq 500) at the Columbia University Irving Medical Center, supported by the Herbert Irving Comprehensive Cancer Center. Raw reads were trimmed to remove sequencing adapters, and low-quality reads were removed using Trim Galore (version 0.6.7) with default parameters, and aligned to human (hg19) reference genome using Bowtie 2 (version 2.2.4; ref. 74) with –no-mixed –no-discordant –no-dovetail –no-contain –local parameters. Reads were indexed and sorted, and then multimapped reads were filtered using SAMtools (version 1.11) and read duplicates were removed using Picard (version 2.23.8; http://broadinstitute.github.io/picard/). Genome coverage was calculated using deepTools bamCoverage (version 3.2.1; ref. 75) and normalized to library size (reads per million) after removing blacklisted regions. Peaks were called using SEACR (version 1.3; ref. 76) in relaxed mode by selecting the top 0.1% of regions by the AUC. Overlapped peaks were extracted using BEDTools (version 2.29.2; ref. 77). SMARCA4 CUT&RUN peaks were annotated to their nearest genes using the R package ChIPSeeker (version 1.26.0; ref. 78) to assess changes in gene expression associated with SMARCA4. All heat maps were drawn using deepTools (version 3.2.1). The HOMER (version 4.11; ref. 79) findMotifsGenome.pl function was used to identify motifs enriched at selected regions. We used nonredundant human DNA-binding motifs from the JASPAR (version 2020; ref. 80) and HOCOMOCO (version 11; ref. 81) databases. The top five motifs of transcription factors were identified. To analyze datasets with spike-in mouse cells, the same procedures were followed to align fastq files to a mouse (mm10) reference genome downloaded from GENCODE. A normalization factor was calculated as follows: To account for differences between sequence depth for different samples, the mm10-scaled hg19 read counts were scaled to 1 million reads. Briefly, the normalization scale factor was equal with 1,000,000/[unique mapped reads from mouse genome], as recommended in the Henikoff pipeline. Next, read counts were calculated from the hg19 aligned bam files and then multiplied by a scaling factor. BedGraphs of these files were generated as intermediary files to facilitate the generation of BigWig coverage maps using the bedGraphToBigWig tool from UCSC (version 4; ref. 82).

To integrate CUT&RUN and RNA-seq datasets, we first annotated each SMARCA4 CUT&RUN peak to its closest transcriptional start site with a cutoff of 50 Kb as the maximum distance. Genes with at least a SMARCA4 CUT&RUN peak were then matched with the list of DEGs identified from RNA-seq datasets.

Statistical analysis was performed using R software. Statistical parameters, statistical tests used, error bar definitions, and sample sizes are indicated in each figure and corresponding figure legend. In cases in which outliers were removed for plotting purposes, the removed datapoints were still used for statistical analysis.

RNA-seq, SMARCA4, SOX10, H3K4me3, H3K27ac, H3K27me3, and H3.3K27M CUT&RUN and ATAC-seq datasets in DIPG6 and DIPG17 cells are available through the Gene Expression Omnibus (GEO; accession GSE185280). H3K27ac ChIP-seq datasets in DIPG6 and DIPG17 cell lines were downloaded from published datasets (GSE94259). Please refer to supplementary data.

C.R. Vakoc reports grants and personal fees from Treeline Biosciences and Syro Pharmaceuticals, personal fees from Flare Therapeutics, Roivant Sciences, C4 Therapeutics, and KSQ Therapeutics, and grants from Boehringer Ingelheim outside the submitted work. No disclosures were reported by the other authors.

Y. Mo: Conceptualization, validation, investigation, writing–original draft. S. Duan: Formal analysis, writing–original draft. X. Zhang: Conceptualization, data curation, writing–review and editing. X. Hua: Data curation, formal analysis. H. Zhou: Data curation. H.-J. Wei: Data curation. J. Watanabe: Data curation, writing–review and editing. N. McQuillan: Data curation. Z. Su: Methodology. W. Gu: Methodology. C.-C. Wu: Supervision. C.R. Vakoc: Resources, writing–review and editing. R. Hashizume: Supervision, writing–review and editing. K. Chang: Resources. Z. Zhang: Conceptualization, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank Dr. Oren Becher for the H3.3K27M mouse DMG line and Dr. Jun Qin for the SMARCA4 plasmid. This work is supported by NCI R01 CA204297 (to Z. Zhang), by the Gary and Yael Fegel Family Foundation, the Matheson Foundation (UR010590), St. Baldrick's Foundation, and the Star and Storm Foundation (to C.-C. Wu). The genome sequencing facility at Columbia University is supported by the Herbert Irving Cancer Center Support Grant (P30CA013696).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.