ZFTA–RELA Dictates Oncogenic Transcriptional Programs to Drive Aggressive Supratentorial Ependymoma

Ependymoma is an aggressive and chemoresistant pediatric brain tumor, with treatment limited to surgical resection and radiation (1). Although histologically similar, ependymomas are divided into at least nine molecular subtypes associated with distinct genetic and epigenetic alterations (2–6). More than 60% of supratentorial ependymomas are characterized by an oncogenic fusion between zinc finger translocation associated (ZFTA; formerly C11orf95) and v-rel avian reticuloendotheliosis viral oncogene homolog A (RELA; ref. 7). Gene fusions involving YAP1 or ZFTA and other gene partners are less frequent (3). ZFTA–RELA fusion (denoted ZR^fus) protein is typically the sole genetic driver detected in supratentorial ependymoma (7). When expressed in the developing mouse brain, or in transplanted neural stem cells (NSC), ZR^fus is capable of cellular transformation and tumor initiation (7, 8). These findings suggest that the ZR^fus protein and its direct target genes may represent candidates for therapy.

RELA (also known as p65 protein) is a transcription factor (TF) that mediates NFκB pathway activation in diverse processes such as inflammation, cellular metabolism, and chemotaxis (9). In a normal cellular context, the majority of RELA is sequestered in the cytosol by IκBα protein. TNFα or IL1 cytokine induction results in phosphorylation of IκBα, proteasomal degradation, and translocation of RELA to the nucleus. Within the nuclear compartment, RELA undergoes posttranslational modifications to recruit transcriptional coactivators, such as BRD4, to activate inflammatory gene expression programs (10). This occurs via binding of RELA to both enhancers and stretch-superenhancers (SE) and accumulation of active chromatin modifications, demarcated by H3K27 acetylation (H3K27ac; refs. 10, 11). RELA protein interacts as both homodimers and heterodimers with nuclear factor kappa B subunit 1 (NFκB1/P50) to activate target gene expression (9).

Although the normal molecular function of ZFTA protein is currently unknown, its gene fusion with RELA results in constitutive nuclear translocation (7). The fragment of ZFTA that undergoes gene fusion contains at least one to four zinc finger domains, suggesting that this fragment may contribute to novel DNA binding and transcriptional regulation (7). In addition, ZFTA fuses to other genes, often TFs or transcriptional coactivators, such as YAP1 and MAML2 in ependymoma or MKL2 in chondroid lipomas (7, 12). Our study reveals that ZR^fus modulates the tumor epigenome through de novo binding of DNA at specific motifs. Accompanying ZR^fus binding are transcriptional complexes associated with widespread patterns of active chromatin, through recruitment of Brd4, Ep300, and Crebbp (Cbp), for which specific small-molecule inhibitors are currently available. Elucidation of transcriptional coregulators and mapping of ZR^fus downstream genes reveal molecular targets and pathways that may contribute to ependymoma development and may represent leads for future therapy.

To investigate the molecular function of the ZR^fus oncogene in ependymoma, we utilized PiggyBAC (pB) transposon–based in utero electroporation (IUE) technology to model the most common Type 1 ZR^fus fusion variant (denoted ZR^fus1; Fig. 1A–D). As a control, we introduced GFP pB vectors by IUE in age-matched mice that did not generate tumors or affect mouse survival. IUE technology provided several advantages: (i) overexpression of ZR^fus1 in a native context during embryonic brain development, (ii) simultaneous introduction of HA-tagged ZR^fus1 along with DNA constructs: GFP, luciferase, and CRISPR/Cas9 deletion of Trp53, and (iii) study of ZR^fus1 tumor biology all within an immune-competent environment (13). The resulting mouse tumors recapitulated histologic features of ependymoma (Fig. 1B) and showed sustained ZR^fus1 nuclear localization (Fig. 1C). Consequently, the mice succumbed to brain tumors with a median age of about 60 days (Fig. 1D). ZR^fus1 tumors were microdissected by GFP-positive signal and compared against contralateral normal brain (non-GFP positive) tissue by RNA sequencing (RNA-seq). Differential gene expression analysis and comparison against an independent RCAS–TVA ZR^fus1–driven ependymoma mouse model demonstrated a significant correlation of neoplastic transcriptional programs compared with matched normal brain (Fig. 1E; R = 0.79, P < 2.2^e-16; Supplementary Table S1; Supplementary Fig. S1; ref. 8). The IUE ZR^fus1 model positively associated with human ZR^fus1 transcriptional programs as opposed to other ependymoma subtypes, including gene expression of Dlk1, Lmx1b, Ephb2, Igf2, Gli2, and Akt1 (Fig. 1F–H). To further support the consistency between our mouse model and its reflection of the human disease, we directly compared the transcriptome of our IUE model to different ZR^FUS1 mouse models and human ZR^fus1 datasets described in this issue in Kupp and colleagues (14) and Zheng and colleagues (15). We observed a significant overlap between the datasets and derived a ZR^fus1 93 gene signature consistent across all datasets, including our IUE model (Fig. 1H; Supplementary Table S2). These findings demonstrate that the IUE ZR^fus1-driven mouse model of ependymoma recapitulates several key aspects of human ZFTA–RELA ependymoma.

Using brain tumors from the IUE ZR^fus1 mouse model, we acutely dissociated neoplastic GFP-positive cells to perform two orthogonal methods of chromatin profiling, chromatin immunoprecipitation followed by sequencing (ChIP-seq), and CUT&RUN (Fig. 2A–C; Supplementary Fig. S2A). CUT&RUN permitted native cross-linking–free genomic mapping of proteins and histone modifications and reduction of nonspecific cross-linking artifacts often present in ChIP-seq experiments (16). ChIP-seq and CUT&RUN directed against the HA tag was performed on HA-tagged ZR^fus1 protein, as well as CUT&RUN against Rela, H3K27ac, and IgG (negative control), and assay for transposase-accessible chromatin using sequencing (ATAC-seq; to measure accessible chromatin). Combining two independent experiments, we identified 6,845 shared ZR^fus1-binding sites. This set of peaks was further filtered to exclude nonspecific signal using an IgG control. The resulting 5,608 peaks significantly overlapped with regions also identified by ChIP-seq (Fig. 2B and D; Supplementary Fig. S2). CUT&RUN of HA-tagged ZR^fus1 and Rela revealed mostly subnucleosomal binding of ZR^fus1 with small fragment patterns (<120 bp) as expected of TF proteins (Fig. 2A and C; Supplementary Fig. S2B and S2C; ref. 16). ATAC-seq and H3K27ac CUT&RUN demonstrated that HA-ZR^fus1 binding was localized to “open and accessible” and active chromatin, respectively (Fig. 2D and E) and enriched in both active enhancer and promoter loci (Fig. 2F). ZR^fus1 was bound to promoters and proximal enhancers of known ependymoma oncogenes such as Ephb2, Ccnd1, Akt1, and Notch1 (Fig. 2G and H; refs. 17, 18). Interestingly, expression of the ZR^fus1 protein led to a global elevation of H3K27ac, but also H3K27me3, a mark of poised and repressive chromatin (Supplementary Fig. S2D–S2F). These findings support that ZR^fus1 protein directly engages DNA as a TF oncoprotein and is associated with open and active chromatin surrounding many genes, including known ependymoma oncogenes.

To assess whether ZR^fus1 activates gene expression programs, we performed RNA-seq. The ZR^fus1-bound genes showed higher expression levels as compared with mouse contralateral normal brain (non-GFP⁺) tissue (Fig. 3A). Nine hundred twenty ZR^fus1-bound genes that were at least 2-fold upregulated were identified including common ZR^fus1 signature genes such as Ccnd1, Ephb2, and Gli2 (Fig. 3B; P < 0.05; Supplementary Table S3). We reasoned that localization of ZR^fus1 to enhancers and promoters might indicate aberrant recruitment of transcriptional activation complexes, particularly proteins involved in transcriptional initiation and elongation such as Ep300, Cbp, Brd4, and phosphorylated forms of RNA polymerase II (Serine 2 and 5 phosphorylation of the carboxy terminal domain). Supporting this, in the ZR^fus1 IUE model, CUT&RUN signals demonstrated corecruitment of Brd4, Ep300, Cbp, and Ser2/5 phosphorylated RNA polymerase II at most ZR^fus1-binding sites. (Fig. 3C; Supplementary Fig. S3). These findings are consistent with a manuscript cosubmitted to Cancer Discovery by Kupp and colleagues in this issue (14), in which ZR^fus1 was shown to physically bind both Brd4 and Ep300 to promote gene transcription, and ZFTA was reported to facilitate Ep300-mediated gene activation.

To determine whether the effects on gene transcription were direct, we generated knockout (KO) models of cell lines derived from IUE:ZR^fus1 using CRISPR/Cas9 deletion of Rela and ZR^fus1 with varying efficiencies (Fig. 3D). Using single-guide RNA (sgRNA):Rela targeting constructs, ZR^fus1 KO resulted in decreased expression of 872 (23%) of ZR^fus1 target genes, whereas 173 (4%) were upregulated and 2,780 (73%) were unchanged (Fig. 3E). Loss of ZR^fus1 and Rela did not result in a global change in transcriptional output as measured by RNA ERCC spike-in controls (Supplementary Fig. S3A and S3B). Downregulated genes included several ZR^fus1 specific genes such as Dlk1, Lmx1b, and Akt1, and also members of the NFκB pathway such as Cxcl2, Icam1, and Stat1 (Fig. 3F and G; Supplementary Table S4; Supplementary Fig. S3C). Most notably, the predominant pathway downregulated upon ZR^fus1 and Rela KO included genes involved in nervous system development and neural cell differentiation (Fig. 3H). The top genes downregulated were markers of oligodendrocyte (Olig1/2) and radial glial cell identity (Fabp7), whereas the genes marking neuronal cells (Dlx5/6 and Gchfr) were upregulated (refs. 19, 20; Fig. 3F). Similarly, chemical inhibition of Ep300/Cbp using inhibitors A485 versus A486 (inactive isomeric control) significantly impaired expression of ZR^fus1 target genes and was associated with decreased cellular viability at 72 hours of treatment (Supplementary Fig. S3D and S3E; ref. 21). Our data demonstrate that ZR^fus1 recruits transcriptional activation complexes, including Brd4, Ep300, Cbp, and RNA Pol2 to drive neoplastic gene transcription. Genetic ablation of ZR^fus1 or chemical inhibition of transcriptional coregulators impairs ZR^fus1 target gene expression and cell identity programs.

We hypothesized that ZR^fus1, which harbors a C2H2 zinc finger domain, would promote aberrant DNA binding in addition to its function to drive nuclear localization (7). To mark and remove canonical NFκB-binding sites from tumor-specific sites, we compared ZR^fus1 binding against mouse embryonic fibroblasts treated with TNFα, a method of stimulating Rela nuclear localization and studying inflammatory gene expression (22). The majority of Rela-bound sites in the context of inflammation were absent or weakly bound by ZR^fus1 (Fig. 4A). Tumor-specific DNA binding of ZR^fus1 was enriched for the PLAGL1 or PLAGL2 TF family motifs consisting of a core “GGGCC” DNA sequence (Fig. 4B, enrichment: P < 1e⁻¹⁴⁸⁸, Supplementary Table S5). Also, shared programs between tumor and inflammatory programs were enriched for both Rela and PLAGL2 motifs (Fig. 4B, Rela: 1e⁻³⁸⁷ and PLAGL2: 1e⁻¹¹⁸, Supplementary Table S6). An enrichment of PLAGL TF family motifs was also observed in the ZR^fus1 HA ChIP-seq data, thus further validating our findings using an orthogonal technique (Supplementary Fig. S4A–S4C). Interestingly, in a small number of cases we identified RELA binding sites directly flanked by adjacent PLAGL family motifs, suggesting possible corecruitment and utilization of endogenous Rela proteins (Supplementary Fig. S4D–S4G). Furthermore, Tead family motifs were found to be enriched at a subset of tumor-specific and shared inflammatory loci, and also bound by H3K27ac, Brd4, Ep300, and Pol2 (Supplementary Fig. S4C). This is potentially relevant, as other subtypes of supratentorial ependymoma are driven by YAP1 gene fusions, suggesting potentially shared oncogenic mechanisms (3, 23).

ZR^fus1-binding profiles in a tumor-specific or shared context were comparable between enhancers and promoters, unlike enhancer- and SE-centric profiles observed in canonical TNFα-driven Rela activation (Fig. 4C; ref. 10). As an example, we observed ZR^fus1 binding at the proximal enhancer of Lmx1b, along with Brd4 and Ep300 recruitment and presence of three central “GGGCC” motifs (Fig. 4D). Lmx1b expression was elevated in ZR^fus1-driven mouse tumors and human ependymoma, both associated with the presence of a proximal ZR^fus1-specific SE (Fig. 4E). To confirm our findings in human tumor tissue, we mapped nucleosome-free regions of H3K27ac ChIP data in ZFTA–RELA versus non-ZFTA–RELA primary ependymoma samples (18). We found that PLAGL family TF motifs were the top-ranking DNA motif observed in a ZFTA–RELA-specific context (Fig. 4F; Supplementary Figs. S4H and S5A–S5E). Therefore, our findings demonstrate that ZR^fus1 directly binds a tumor-specific program distinct from the canonical NFκB pathway and converges on activation of PLAGL TF target genes in both mouse and human ependymoma.

We have previously shown that enhancers and SEs regulate subtype-specific gene expression programs in human ependymoma, including ZFTA–RELA tumors (18). To elucidate the mechanisms of oncogene activation in ZR^fus1 ependymoma, we mapped SEs by Ranking of Super Enhancer (ROSE) analysis and identified several genes such as Dlk1, Lmx1b, and Ephb2 which also harbor SEs in human ZFTA–RELA ependymoma (Fig. 5A; Supplementary Table S7; ref. 18). Of note, 73% (511 of 703) of SEs identified harbored at least one overlapping ZR^fus1-binding site, suggesting a significant contribution of ZR^fus1 binding to SE-driven gene expression (Fig. 5B and C). As an example, we observed a 5′ SE predicted to regulate Dlk1, which contained ZR^fus1 binding as well as Brd4 recruitment to sites of open chromatin (Fig. 5D). Dlk1 was overexpressed in IUE:ZR^fus1 tumors compared with normal brain, and had significantly higher expression in human ZFTA–RELA fusion–driven ependymoma in comparison with other ependymoma subtypes (Fig. 5E). These findings are in line with single-cell RNA-seq characterization of human ependymoma that pinpoints DLK1 as a top signature gene in a unique and tumor cell–specific transcriptional program in ZR^fus-driven human tumors (24).

To dissect the TFs that regulate ZR^fus1 SEs, we conducted core regulatory circuitry analysis (CRC; Fig. 5F). This method searches regions of open chromatin (demarcated by ATAC-seq) within H3K27ac-defined SEs (i.e., nucleosome-free regions). Core transcriptional circuits are defined by TFs that are regulated by SEs, which are transcribed and translated, and subsequently bind other SEs (25, 26). CRC analysis identified the top-ranking set of 11 TFs that form a putative regulatory circuit, including PLAGL2, RELA, SOX7, SP2, SOX3, FOXO1, SRF, SMAD1, STAT3, ZFP3, and HIC1 (Fig. 5G, rank = 1, score = 284.091; Supplementary Table S8). To distinguish between ZR^fus1 and PLAGL2 binding at SEs and regulation of CRC transcriptional circuits, we performed CUT&RUN of PLAGL2 (Fig. 5H; Supplementary Fig. S5F). Although ZR^fus1 shared some binding sites with PLAGL2, the occupancy at SEs, containing a PLAGL2 motif or Rela motif, was largely a result of ZR^fus1 binding. These findings support that ZR^fus1 may act as an oncogenic TF by invading a CRC network, binding directly to PLAGL1/2 and RELA motifs, and engaging SEs that regulate downstream ependymoma gene expression programs.

Ependymomas are thought to arise from a heterogeneous population of radial glial cells during embryonic brain development (17, 27). To investigate the developmental origins in our IUE:ZR^fus1 model, we compared tumor transcriptomes against a single-cell RNA-seq atlas of the mouse developing cortex (Supplementary Fig. S5G–S5I; ref. 28). As compared with normal brain, we found that cells with the closest and most specific enrichment were embryonic day E15 dorsal and ventral radial glial cells (Supplementary Fig. S5G). Astrocytes of early postnatal stages showed varied degrees of enrichment. PLAGL1 and PLAGL2 gene expression was also found to be present and modestly elevated during this embryonic stage (Supplementary Fig. S5G). We observed more restricted patterns in the human developing brain cortex by querying the CoDEx database, and found PLAGL1 expression to be enriched in outer radial glial cells (Supplementary Fig. S5J; ref. 29). These findings support a possible hypothesis that ZR^fus1 may bind, maintain, and/or activate PLAGL target gene expression in developing radial glial cells that subsequently give rise to ependymoma.

The PLAGL1 and PLAGL2 proteins have diverse roles in both development (30) and tumorigenesis and have been shown across several cell types and tissues to regulate and express an imprinted gene network (including Axl, Dlk1, Igf2), Wnt pathway (Wnt and Fzd family genes), and extracellular matrix target genes (Integrin family; refs. 30–33). Within ZR^fus1 tumors, we detected direct binding enrichment of genes associated with NFκB pathway activation such as TNFα signaling, Epstein–Barr virus infection, toxoplasmosis, and osteoclast differentiation (Supplementary Fig. S6A–S6E). In the tumor-specific program, we identified genes elevated and associated with Wnt signaling pathway activity, Cushing syndrome, hepatocellular carcinoma, focal adhesion, and members of the MAPK signaling pathway (Fig. 6A). Many of the identified target genes identified contained previously unreported therapeutic leads in ZFTA–RELA ependymoma, such as Cdk4/6, Gsk3B, Akt1, Mapkapk2/3, Fgfr4, Ngfr, Brd7, and Ep300. Further integration of ZR^fus1 tumor-specific programs with the Washington University Drug Gene Interaction Database (34) revealed downstream candidate genes that are possibly clinically actionable (Akt1, Cdk4/6, Fgfr1/4), mapped in the “druggable” genome space (Axl, Csf1r, Dlk1, Ephb2/4, Notch1), protein kinases (Cdk1/4/6, Igf2r, Ccnd1/3), proteases (Adamts enzymes, Stat3, Smad3), methyltransferases (Prdm1/16), and cell surface proteins (Acvrl1, Cd40/44/63, Pdgfrb; Fig. 6B). Our findings demonstrate that genes regulated by ZR^fus1 represent a novel set of candidates relevant to ependymoma development, which may represent possible therapeutic leads for future characterization.

Supratentorial ependymomas are often initiated by a single gene fusion (7, 8, 23, 35). ZFTA–RELA is the most common alteration; however, YAP1 and other gene fusions have been reported (see Zheng and colleagues, ref. 15). Together, ependymoma gene fusions have been shown to promote nuclear localization and are thought to lead to promiscuous TF activity, aberrant chromatin regulation, and upregulation of oncogenes. This activity may be tied to developmental programs that are essential for ependymoma cell growth (18). Our study demonstrates, in a natively forming brain tumor mouse model, that ZR^fus1 directly engages accessible DNA and active chromatin. ZR^fus1 binding is associated with expression of known ependymoma oncogenes, such as Ephb2, Notch1, and Ccnd1, and expression patterns are highly consistent with different mouse and human datasets of ZR^fus-driven tumors (see Kupp and colleagues and Zheng and colleagues, refs. 14, 15). Our collective analysis across these different studies led to synthesis of a core set of 93 signature genes elevated in ZR^fus-driven mouse and human tumors.

Previous reports have suggested aberrant NFκB activity as a central driver and mechanism of transformation in ZFTA–RELA ependymoma (7). This hypothesis has prompted clinical trial concepts to inhibit NFκB signaling using known small molecules against the canonical NFκB pathway. While our data report a shared inflammatory program directly activated by ZR^fus1, a major component of ZR^fus1 binding is seen in a tumor-specific context and not observed in canonical TNFα-driven gene expression. This suggests that the ZFTA fragment of the ZR^fus1 fusion protein confers oncogenic properties and is supported by several points: (i) ZFTA is fused with other genes in ependymoma (15), and (ii) hyperactive Rela is unable to drive tumorigenesis (8, 12). On the other hand, the ZFTA fragment alone is unable to initiate tumors, therefore tying its function to its fusion partner, often a TF and/or coactivator (7). Indeed, Kupp and colleagues (14) have demonstrated a high similarity between ZFTA–RELA and ZFTA–Ep300 transcriptional programs, consistent with our data that reveals corecruitment of transcriptional machinery including Brd4, Ep300, Cbp, and Pol2. These findings suggest that therapeutic strategies that disrupt Rela transcriptional activity may be more effective than drugs that target cytosolic regulation of Rela (i.e., IKK inhibitors). An alternate approach could be the disruption of transcriptional cofactors, such as BRD4 and EP300/CBP, that are recruited and associated with transcription of ZR^fus1-bound genes. Indeed, ZFTA–RELA-driven PDX models show sensitivity to BET bromodomain inhibitors, such as JQ1, and IUE:ZR^fus1 tumor cells to Ep300/Cbp inhibitors (18). It is not yet clear whether ZFTA–RELA tumors, in vivo, are more sensitive to perturbation of oncogene transcription (i.e., transcriptional addiction) to an extent where a therapeutic window can be achieved that spares effects on normal cells.

Our data point to tumor-specific DNA binding of ZR^fus1 and convergence on the core binding sequence of PLAGL family TFs. This group of TFs has been shown to be expressed in a variety of cancer types with diverse tumor-suppressive and oncogenic roles and regulate an imprinted gene network along with Wnt pathway activation (30, 36–41). PLAGL2, specifically, has been reported to be amplified and overexpressed in glioblastoma and positively contributes to glioma genesis through inhibition of cell differentiation (33). While our data do not exclude the functional role of PLAGL1/2, they do suggest that ZR^fus1 may be utilizing shared DNA-binding sites for target gene activation. ZR^fus1-directed transcriptional programs converge on upregulation of PLAGL-associated Mapk, focal adhesion, imprinted, and Wnt pathway genes. These PLAGL gene sets are enriched for putative molecular targets that are potentially clinically actionable and hold promise for future lead development, such as druggable genome candidates, kinases, and cell surface proteins.

We hypothesize that PLAGL1/2 programs may be active during embryonic brain development. Supporting this, we show that ZR^fus1 tumor expression matches PLAGL1/2-positive radial glial cells (RGC) of the developing cortex, consistent with data that RGCs contain cells of origin of ependymoma (27, 42). In line with other pediatric brain tumors, we hypothesize that our data support a model of stalled development where stem- or progenitor-cell gene expression programs are aberrantly sustained (28). Related to this, the role and expression patterns of ZFTA in normal development and whether its cis-regulatory elements facilitate cell- and temporal-specific expression of ZR^fus1 expression during tumorigenesis are unclear. Indeed, several examples in other pediatric brain tumors exemplify how structural alterations not only activate oncogenes but also “hijack” developmental enhancers to promote gene expression (43–45).

Our findings shed light on the transcriptional programs deregulated by ZFTA–RELA and reveal potential downstream molecular targets that may hold important cellular vulnerabilities. As a single genetic driver, “drugging” the ZFTA–RELA fusion specifically, as opposed to coregulators and downstream targets, would likely be the most effective. Solving the structure of ZFTA–RELA fusion with emerging protein crystallography approaches and identification of binding surfaces of the ZR^fus protein could facilitate the design of proteolysis targeting chimeras for inhibition and/or protein degradation. However, a crucial question of whether ZR^fus is still required for tumor maintenance in vivo following transformation and establishment of epigenetic marks remains unanswered. Lessons learned from studying ZFTA–RELA are likely to apply to many other fusion-driven pediatric and adult tumors in terms of developmental modeling, epigenetic characterization, and preclinical therapeutic testing.

Western blot analysis was performed using standard techniques. Antibodies used included anti-HA tag antibody (Abcam, catalog no. ab9110, RRID:AB_307019), Lamin B1 (Abcam, catalog no. ab16048), β-Tubulin (Cell Signaling Technology, 86298), β-actin (Millipore, catalog no. A1978), BRD4 (Bethyl Laboratories, catalog no. A301–985100A), Crebbp (Cbp; Cell Signaling Technology, catalog no. 7389s), and EP300 (Abcam, catalog no. ab10485). IHC was performed on deparaffinized formalin-fixed, paraffin-embedded sections with antigen retrieval in sodium citrate (10 mmol/L pH 6) at 95°C for 15 minutes. Sections were blocked with 5% normal horse serum in PBS-T for 1 hour and primary antibodies added for an overnight incubation at 4°C. IHC for the antibodies: H3K27me3 (CST9733 1:250–1:500 dilution), H3K27ac (Abcam 1:250–1:500 dilution), and HA (Sigma – 11867431001 at 1:1000 dilution).

All animal procedures in this study were performed with Institutional Animal Care and Use Committee approval. IUE was performed as described previously, and plasmids provided as a gift from Dr. Joseph LoTurco PhD (13). After anesthesia with 5% isoflurane, pregnant mice, at E16.5, were subjected to abdominal incision to expose the uterus. DNA plasmid cocktail (pBCAG-HA-ZR^FUS1, pbCAG-eGFP, pX330-sgTp53, GLAST-PBase, pBCAG-Luc) was injected into the lateral ventricles with a glass pipette. Electric pulses were then delivered to the embryos by gently clasping their heads with forceps-shaped electrodes. Six 33 V pulses of 55 ms were applied at 100-ms intervals. The uterus was then repositioned into the abdominal cavity, and the abdominal wall and the skin were then sutured. Following birth, pups were traced on a weekly basis by bioluminescence imaging to monitor brain tumor formation. Mouse tumors were collected on the basis of isolation of GFP⁺ tumors and dissociation into single cells using a Brain Tumor Dissociation Kit (MACS, #130–095–942) following manufacturer's instructions.

Cells were grown in neural basal medium (Invitrogen) supplemented with sodium pyruvate, glutamine, B27, N2, bFGF (10 ng/mL), and rhEGF (20 ng/mL). Cells were grown on treated cell culture dishes coated with Matrigel (Corning). All lines tested negative for Mycoplasma contamination, as assessed using a PCR based approach. Tumor-derived cell lines were confirmed to be authentic and unique by short tandem repeat fingerprinting.

Mouse tumor cells were grown to full confluency and washed with ice-cold PBS and then collected with a cell scraper in ice-cold E1 buffer (50 mmol/L HEPES-KOH, 140 mmol/L NaCl, 1 mmol/L EDTA, 10% glycerol, 0.5% NP-40, 0.25% Triton X-100, 1 mmol/L DTT supplemented with 1× protease inhibitor cocktail) for each volume of cell pellet. Resuspended cells were spun down at 1,100 × g at 4°C for 2 minutes and supernatant (cytoplasm fraction) was collected in a fresh tube. Cell pellet was resuspended in the same volume of the E1 buffer and spun down at 1,100 × g for 2 minutes. Supernatant was discarded and pellet was resuspended in the same volume of the E1 buffer and incubated on ice for 10 minutes before spinning down at 1,100 × g for 2 minutes at 4°C. Supernatant was discarded and pellet was gently resuspended in 2 volumes of ice-cold E2 buffer (10 mmol/L Tris-HCl, 200 mmol/L NaCl, 1 mmol/L EDTA, 0.5 mmol/L EGTA, supplemented with 1× protease inhibitor cocktail) and centrifuged at 1,100 × g for 2 minutes at 4°C. The supernatant (nuclear fraction) was collected in a fresh tube.

sgRNA sequences targeting Rela and nontargeting control (D98) were designed using CRISPR tool and the constructs were generated by plasmid cloning following GECKO protocol (RRID:SCR_015935). sgRNA sequences are provided in Supplementary Table S9. To produce lentiviral particles, HEK293T cells were seeded 24 hours prior to transfection and were transfected with 5 μg pMD2.G, 5 μg psPAX2, and 5 μg Lenti-CRISPR-v2 using calcium phosphate transfection method (RRID:Addgene_12260). Medium was refreshed 24 hours post-transfection. The media containing the viral particles were collected at 48- and 72-hour time-points and filtered with 0.45-μm filters. Mouse tumor cells were seeded into 10-cm plates 24 hours prior to infection and were infected with freshly collected lentiviruses with 8 μg/mL polybrene for 24 hours. Infected cells were selected with 2 μg/mL puromycin for 3 days. Knockout efficiencies were tested with Western blot analysis.

RNA was extracted using TRIzol or Qiagen miRNA Easy Kit according to the manufacturer's instructions, and cDNA was prepared from approximately 1 μg total RNA using iScript gDNA Clear cDNA Synthesis Kit (Bio-Rad, #1725034). Quantitative PCR was performed using iTaq Universal SYBR Green Supermix (Bio-Rad, #172–5120) using the manufacturer's recommended protocol. Primers are listed in Supplementary Table S7. In the case of RNA-seq, ERCC spike-ins (Thermo Fisher Scientific) were added according to the manufacturer's recommendations and protocols before proceeding with library preparation. RNA-seq library preps were performed using Kapa RNA Hyperprep Kit with RiboErase (HMR) according to the manufacturer's recommendations.

CUT&RUN assay was performed as described in Skene and colleagues (2018; ref. 16). Briefly, 0.5–1 million cells were captured with BioMagPlus Concanavalin A beads and incubated with primary antibody for 10–20 minutes at room temperature. After washing away the EDTA in the buffer and unbound antibody with dig-wash buffer (20 mmol/L HEPES pH 7.5, 150 mmol/L NaCl, 0.5 mmol/L Spermidine, 1× Complete Protease Inhibitor EDTA-Free, and 0.05% Digitonin), protein A-MNase was added and incubated for 10–20 minutes. The cells were washed again and placed in an ice-water prechilled metal block at least 5 minutes. CaCl₂ was added to the final concentration of 2 mmol/L to activate protein A-MNase for 30 minutes on the ice-water chilled metal block. The reaction was stopped by the addition of equal volume of 2X STOP buffer (340 mmol/L NaCl, 20 mmol/L EDTA, 4 mmol/L EGTA, 0.02% Digitonin, 5 μL/mL RNase A, 50 μg/mL glycogen, and 2 pg/mL heterologous spike-in DNA). The protein–DNA complex was released and DNA was extracted with Gel and PCR Clean-up kit (Macherey-Nagel NucleoSpin, catalog no. 740609.250) or Phenol–chloroform–isoamyl alcohol precipitation (for small fragment DNA), followed by Qubit fluorometer and Agilent 4200 Tapestation quality and size distribution control.

KAPA Hyper Prep Kit (catalog no. KK8504, KAPA Biosystems) and KAPA Dual-indexed Adapter Kit (catalog no. KK8722 KAPA Biosystems) were used to construct the CUT&RUN DNA library for sequencing on an Illumina platform following the manufacturer's instructions. After adaptor ligation, 2× volume of AMPure XP beads was used to recover the small fragments. After 12–14 cycles of PCR amplification, the product was cleaned up with AMPure XP beads. 3% gel purification was performed if the PCR dimers were too much. Paired-end Illumina sequencing on the barcoded libraries was performed as per the manufacturer's instructions. Briefly, the mixed libraries were denatured according to the standard protocol from Illumina. 1.3 mL of 1.8 pmol/L diluted library pool was loaded to a NextSeq 500/550 Mid Output Kit v2 (150 cycles), and sequenced in the NextSeq 500 platform. Paired-end sequencing was then performed (2 × 75 bp, 8-bp index).

Paired-end reads were adapter and quality trimmed using Trimgalore (v0.6.5, default parameters, http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/; RRID:SCR_011847) and aligned to mouse genome mm10 using Bowtie2 (v2.3.5.1, parameters: –local -D 20 -R 3 -N 0 -L 20 -i S,1,0.50 –no-unal –no-mixed –no-discordant –phred33 -I 10 -X 700; RRID:SCR_016368). Duplicated reads were then marked and removed using picard MarkDuplicates (v2.21.1; http://broadinstitute.github.io/picard/), and Samtools (v1.9; RRID:SCR_002105), respectively, with default parameters. For Rela and HA samples, only fragments of size less than 120 bp were retained. Deeptools (v3.4.3; RRID:SCR_016366) was used to convert all the resulting BAM files to Bigwig format for visualization. MACS2 (v2.2.7.1) was used to call peaks, on the resulting BAM files, with aP value threshold of 1e-3. A set of 6,845 peaks were inferred in the HA CUT&RUN by overlapping the called peaks from the two independent mice using bedtools2. This set was further filtered to remove any overlaps with nonspecific IgG peak signals (from both mice), resulting in 5,608 peaks. The peaks were then annotated to nearest genomic features using annotatePeaks.pl from Homer (v4.11.1; ref. 46).

RNA-seq and ATAC-seq analysis were performed using Genialis Expression software (https://www.genialis.com) deployed locally on BCM computational infrastructure. Briefly, the RNA-seq pipeline run on the Genialis platform comprised the following steps. The raw reads were filtered to remove adapters and poor-quality reads using BBDuk (v37.9; https://sourceforge.net/projects/bbmap/). The resulting reads were mapped to the reference genomes (ENSEMBL 92) using STAR (v2.7.0; RRID:SCR_015899). FeatureCounts (v1.6.3; RRID:SCR_012919) was used for gene expression level quantification followed by DEseq2 (RRID:SCR_000154) for differential gene expression analysis (47). Low-expressed genes with expression count summed over all samples below 10 were filtered out from the input matrix to DESeq2. The paired-end reads from ATAC-seq were trimmed using BBDuk (v37.9) and mapped to the reference genome mm10 using Bowtie2 (v2.3.4.1). MACS2 (v2.1.1.20160309) was then used to call peaks on the aligned reads using P value cutoff of 0.01 (parameters –shift -75 –extsize 150 –nomodel –call-summits –nolambda –keep-dup all -P = 0.01)

To find motifs enriched around the binding sites, Homer findMotifsGenome.pl and MEME suite (v5.1.0; ref. 48) were used. Furthermore, we utilized the Rela ChIP-seq data in the Gene Expression Omnibus (GEO) accession GSM3895240 (48) to differentiate the tumor-specific binding profile of ZR^fus1 from the canonical inflammatory context. For this, the overlaps between the peaks in GSM3895240 and the 5,608 HA CUT&RUN peaks were inferred using bedtools2. Homer motif analysis was performed on each of the peaks sets; unique to HA:ZR^fus1, shared with and unique to the GSM3895240 Rela peaks. Enrichment of specific pathways in each of these groups was then computed using the R package Clusterprofiler (RRID:SCR_016884). To identify DNA motifs in human H3K27ac datasets, first peaks were identified in H3K27ac peaks then valleys (nucleosome-free regions) mapped that contain TF-binding sites. Human ZFTA–RELA-specific H3K27ac valley regions were compared against non-RELA ependymoma (as background) using HOMER (RRID:SCR_010881).

SEs were identified on the basis of H3K27ac and IgG control CUT&RUN data using ROSE (49) with a stitching distance of 12.5 kb and exclusion of peaks within 2.5 kb of a promoter. Regions of H3K27ac overlapping ATAC peaks, along with the identified super enhancers, were provided as an input to determine CRC (https://github.com/linlabcode/CRC; ref. 25).

To project ZR^fus1 tumor and normal brain transcriptomes to the scRNA-seq atlas of the developing mouse forebrain (28), we evaluated enrichment of the gene signature of each atlas cell type in each sample, using single-sample gene set enrichment analysis (ssGSEA; ref. 50). The most enriched cell type signatures in ZR^fus1 tumors were selected for visualization by computing the median rank of each cell type signature across ZR^fus1 tumor samples based on ssGSEA scores, and taking the top 15.

All raw and processed sequencing data from our study may be found at GEO (accession no.: GSE161679).

K. Kong reports grants from F32 Postdoctoral Fellowship during the conduct of the study. S.G. Injac reports non-financial support from Takeda Pharmaceuticals outside the submitted work. F. Sahm reports other support from Illumina outside the submitted work. S.M. Pfister reports grants from IMI-2 project together with various companies (Eli Lilly, Roche, Bayer, Pfizer, Pharmamar, AstraZeneca, Johnson and Johnson) outside the submitted work; in addition, S.M. Pfister has a European Patent (16 710 700.2) for DNA-Methylation Based Method for Classifying Tumor Species issued. C.Y. Lin reports other support from Kronos Bio., Inc. outside the submitted work. R.J. Gilbertson reports a patent for PCT/EP2020/063096 issued to Cancer Research Technology Limited & MedImmune Limited. No disclosures were reported by the other authors.

A. Arabzade: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. Y. Zhao: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S. Varadharajan: Conceptualization, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. H. Chen: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Jessa: Formal analysis, validation, investigation, methodology. B. Rivas: Formal analysis, investigation, methodology. A.J. Stuckert: Formal analysis, investigation. M. Solis: Formal analysis, investigation. A. Kardian: Formal analysis, investigation. D. Tlais: Formal analysis, investigation. B.J. Golbourn: Formal analysis, investigation. A.J. Stanton: Formal analysis, investigation. Y. Chan: Formal analysis, investigation. C. Olson: Conceptualization, formal analysis, investigation, writing–original draft, writing–review and editing. K.L. Karlin: Conceptualization, formal analysis, investigation, writing–original draft, writing–review and editing. K. Kong: Conceptualization, formal analysis, investigation. R. Kupp: Conceptualization, resources, formal analysis, validation, writing–review and editing. B. Hu: Conceptualization, resources, data curation, formal analysis, writing–original draft, writing–review and editing. S.G. Injac: Formal analysis, writing–review and editing. M. Ngo: Formal analysis, writing–review and editing. P.R. Wang: Resources, formal analysis, writing–original draft, project administration, writing–review and editing. L.A. De Leon: Conceptualization, formal analysis, writing–original draft, project administration, writing–review and editing. F. Sahm: Conceptualization, formal analysis, writing–original draft, writing–review and editing. D. Kawauchi: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S.M. Pfister: Conceptualization, formal analysis, investigation, writing–original draft, project administration, writing–review and editing. C.Y. Lin: Conceptualization, resources, data curation, formal analysis. H. Hodges: Formal analysis, writing–review and editing. I. Singh: Formal analysis, writing–review and editing. T.F. Westbrook: Resources, formal analysis, writing–original draft, project administration, writing–review and editing. M.M. Chintagumpala: Formal analysis, funding acquisition, writing–review and editing. S.M. Blaney: Formal analysis, writing–review and editing. D.W. Parsons: Conceptualization, formal analysis, writing–original draft, writing–review and editing. K.W. Pajtler: Formal analysis, writing–review and editing. S. Agnihotri: Conceptualization, formal analysis, writing–original draft, project administration, writing–review and editing. R.J. Gilbertson: Formal analysis, writing–review and editing. J. Yi: Conceptualization, formal analysis, writing–original draft, writing–review and editing. N. Jabado: Conceptualization, formal analysis, writing–review and editing. C.L. Kleinman: Conceptualization, resources, formal analysis, funding acquisition, validation, writing–review and editing. K.C. Bertrand: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. B. Deneen: Conceptualization, formal analysis, investigation, writing–original draft, project administration, writing–review and editing. S.C. Mack: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, writing–original draft, project administration, writing–review and editing.

S.C. Mack is supported by a V Foundation Scholar in Cancer Research Award, Cancer Prevention Research Institute of Texas (CPRIT) scholar award (RR170023), Alex's Lemonade Stand Foundation (ALSF) A award, THINC Seed Grant, and NIH-National Institute of Neurological Disease and Stroke (R01NS116361). We acknowledge the joint participation of the Adrienne Helis Malvin Medical Research Foundation, and Baylor College of Medicine. K.C. Bertrand is supported by grant funding from Hyundai Hope on Wheels, Rally Foundation, and St. Baldrick's Foundation Early Career Development Award. J. Yi is supported by a THINC Seed Grant, K12 award (5K12CA090433-17), and an Alex's Lemonade Stand Foundation Center of Excellence Developmental Therapeutic Scholar Award. C.L. Kleinmann is supported by Canadian Institutes of Health Research (CIHR) #PJT-156086 and NSERC RGPIN-2016-04911. B. Deneen is supported by National Cancer Institute-Cancer Therapeutic Discovery (U01-CA217842) and Cancer Prevention Research Institute of Texas (RP150334). H. Hodges is supported by grants from CPRIT (RR170036), NCI (R00CA187565), and NIGMS (R35GM137996).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.