ZFTA Translocations Constitute Ependymoma Chromatin Remodeling and Transcription Factors

Brain tumors are the most common and lethal childhood cancers (1). Although these tumors frequently resist conventional therapies, they often contain genetic alterations that might serve as targets for novel treatments (2). But these genetic alterations are complex and poorly understood, impeding efforts to target them pharmacologically. Drugging brain tumor drivers effectively will require a better understanding of how they transform neural lineages.

Ependymomas are tumors of the brain and spinal cord that are incurable in up to 40% of patients (3). Although histologically similar, ependymomas from the different regions of the central nervous system are distinct entities, with different lineage origins, transcriptomes, genetic alterations, and clinical outcomes (4–9). More than 70% of supratentorial ependymomas (ST-EP) contain translocations between ZFTA and RELA (ZFTA–RELA) or, more rarely, between ZFTA and YAP1 (ZFTA–YAP1) or MAML2 (ZFTA–MAML2; refs. 6, 8): we term this subgroup of ependymomas ST-EP-ZFTA^FUS. Recurrent translocations between ZFTA and MRTFB have also been described in chondroid lipomas (10, 11), and ZFTA–RELA or ZFTA–YAP1 transgenes drive ST-EP-ZFTA^FUS tumors in mice (6, 12).

Although ZFTA fusion proteins are transforming, how this gene cooperates with its various translocation partners to drive tumorigenesis remains to be determined. Here, we identify a tripartite transformation mechanism of ZFTA-containing fusions that includes active nuclear trafficking; zinc finger (ZF)–dependent chromatin binding and remodeling; and promiscuous activation of gene expression.

ZFTA contains four evenly distributed ZF domains; between one and four of these are incorporated into fusion proteins with RELA or YAP1 (Fig. 1A; ref. 6). To understand how these fusion proteins transform cells, we studied ZFTA–RELA^FUS1, ZFTA–RELA^FUS2, and ZFTA–YAP1^FUS, which are observed in 56%, 33%, and 10% of ST-EP-ZFTA^FUS ependymomas, respectively (6, 8). Because protein–protein and protein–DNA interactions can be species and cell context–specific, we studied fusion proteins in both human (HEK293) and mouse [mouse neural stem cells (mNSC)] cells: the latter are proven cells of origin of ependymoma in mice (5, 6, 12).

Immunoblotting of subfractions of HEK293 cells transduced with hemagglutinin (3xHA)-tagged proteins identified 74% ± 3.3 SE (n = 5) of RELA and 53% ± 2.7 SE (n = 5) of YAP1 in the cytoplasm, while 85% ± 3.4 SE (n = 5) of ZFTA was located in the nucleus (Fig. 1B and C). Fusion of RELA or YAP1 to ZFTA reversed this cellular distribution: 90% ± 1.7 SE of ZFTA–RELA^FUS1 (n = 5), 86% 1.7 ± SE of ZFTA–RELA^FUS2 (n = 5), and 70% ± 1.8 SE of ZFTA–YAP1^FUS (n = 5) protein was located in the nucleus (Fig. 1B and C). Similar distributions of fusion and wild-type partner proteins were observed by immunoblotting in mNSCs, and by immunofluorescence of fixed and living HEK293 cells (Fig. 1D; Supplementary Fig. S1A–S1C).

Deletion of the entire ZFTA-ZF–or just its core C2-H2 or bi-partite nuclear localization sequence—blocked ZFTA–RELA^FUS1 nuclear trafficking in HEK293 and mNSCs (Fig. 2A–E; Supplementary Fig. S1B). Only one of the four ZFTA-ZFs within ZFTA–YAP^FUS was required for nuclear trafficking (Fig. 2D and E; Supplementary Fig. S1C). Thus, a single ZFTA-ZF, and particularly its C2H2 and NLS motifs, is both necessary and sufficient for spontaneous nuclear trafficking of fusion proteins.

Deletion of the REL homology domain (RHD) reduced nuclear fusion protein content by approximately 30%, suggesting this region might mediate nuclear retention of fusion proteins, possibly through DNA or protein binding (Fig. 2A–E; Supplementary Fig. S1B). Conversely, the RELA transactivation domain (TAD) and its nuclear export sequence (NES) were dispensable for nuclear trafficking.

Because RELA and YAP1 are transcription factors (13, 14) and ZFTA-fusion proteins drive their translocation to the nucleus, we then reasoned that these fusion proteins might promote aberrant transcription. As a first step to test this, we looked for similarities among the transcriptomes of human ST-EP-ZFTA^FUS ependymomas and mouse models of these tumors. Twenty-six percent (93/359) of genes that were upregulated specifically in two independent cohorts of human ST-EP-ZFTA^FUS relative to other ependymomas (hereafter “ZFTA^FUS-sig”; Supplementary Table S1; refs. 6, 8), were also upregulated in at least two out of four mouse models of these tumors (overlap, P = 1.55e⁻⁰⁸; Supplementary Fig. S2A–S2D; refs. 6, 12, 15, 16). These included the validated ependymoma oncogene EPHB2 (5) and NES, IGF2, and JAG1 that regulate NSCs and ependymoma progenitor–like cells (4, 7, 17). ZFTA^FUS-sig also included GLI2, which, together with data presented in two accompanying articles in this issue, implicates this Hedgehog signaling effector as a critical regulator of ependymoma tumorigenesis (15, 16).

To test more directly whether ZFTA fusions drive aberrant gene expression, we performed total RNA sequencing of mNSCs and HEK293 cells transduced with wild-type partner genes, ZFTA fusions, or empty vector. In keeping with its superior transforming potency (6), ZFTA–RELA^FUS1 drove the greatest transcriptomic change relative to ZFTA–RELA^FUS2, ZFTA–YAP1^FUS, or wild-type partner proteins (Fig. 3A and B). Seventy-five percent (70/93, overlap P = 2.26e⁻³⁹) and 45% (42/93, overlap P = 2.11e⁻⁴¹) of ZFTA^FUS-sig genes were upregulated by ZFTA–RELA^FUS1 and ZFTA–RELA^FUS2 in HEK293, respectively (Fig. 3C). Furthermore, genes upregulated by ZFTA–RELA^FUS1 or ZFTA–RELA^FUS2 in HEK293 cells were enriched for the ZFTA^FUS-sig and “human ST-EP-ZFTA^FUS ependymoma specific” gene sets more than any other (18,225 gene sets tested; Supplementary Fig. S3A–S3D). Less significant enrichment of genes involved in development and neurogenesis was also observed.

Although ZFTA^FUS-sig genes were also upregulated by ZFTA–RELA^FUS1 or ZFTA–RELA^FUS2 in mNSCs, this occurred to a much lesser extent than in HEK293 cells, and neither fusion upregulated the “human ST-EP-ZFTA^FUS ependymoma–specific” gene set in mNSCs (Fig. 3D; Supplementary Fig. S3A–S3D). Thus, ZFTA fusions drive ZFTA^FUS-sig gene expression. This is modeled more faithfully in HEK293 cells than in mNSCs, presumably because of species-specific protein–protein or protein–DNA interactions. Indeed, in an accompanying article, we show that ZFTA-fusion proteins display species-specific selectivity for certain transcription factor binding sites (15).

In keeping with its less frequent selection in human ST-EP-ZFTA^FUS and reduced transforming potency in mice (6), ZFTA–YAP1^FUS upregulated fewer ZFTA^FUS-sig genes in HEK293 than did ZFTA–RELA^FUS1 (55/93, overlap P = 1.08e⁻²⁶; Fig. 3E). However, comparison of all genes upregulated by ZFTA–RELA^FUS1, ZFTA–RELA^FUS2, or ZFTA–YAP1^FUS in HEK293 identified a common set of 273 genes that was highly enriched for ZFTA^FUS-sig genes (adjusted enrichment P = 2.26e⁻⁴⁵; Supplementary Fig. S3E and S3F). Thus, ZFTA-fusion proteins drive a core set of genes, highly enriched for the ZFTA^FUS-sig transcriptome and regulators of development and neurogenesis, that is likely important for ependymoma tumorigenesis.

Deletion of either the ZFTA-ZF, which is required for fusion protein nuclear trafficking (Fig. 2; Supplementary Fig. S1), or the RELA–TAD, which recruits transcriptional coregulators, blocked ZFTA^FUS-sig gene expression in ZFTA–RELA^FUS1-transduced HEK293 and mNSCs (Fig. 3C and D). Conversely, the RELA–RHD that mediates DNA binding and dimerization of NFκB proteins was dispensable for ZFTA^FUS-sig expression. Interestingly, deletion of all but one ZF from ZFTA–YAP1 enhanced ZFTA^FUS-sig expression (Fig. 3E), potentially explaining why ZFTA–RELA^FUS1 that also contains a single ZFTA-ZF is the most frequently observed fusion in human ST-EP-ZFTA^FUS (6).

To understand how ZFTA-fusion proteins drive aberrant transcription, we used chromatin immunoprecipitation sequencing (ChIP-seq) to map ZFTA, RELA, and ZFTA–RELA^FUS1 binding sites across the genome. Because the ZFTA^FUS-sig was driven most faithfully in HEK293 cells, we then used these cells for ChIP-seq experiments. Both ZFTA (n = 29,477 sites) and ZFTA–RELA^FUS1 (n = 32,447 sites) displayed extensive genome-wide binding that was distinct from that of RELA (n = 1,156; Fig. 4A–C). Remarkably, 75% (n = 24,497/32,447) of ZFTA-binding sites overlapped with those of ZFTA–RELA^FUS1, suggesting ZFTA governs fusion protein chromatin binding (overlap, P < 1.0e⁻²⁵⁰; Fig. 4B and C). Overlapping binding sites included 91% (85/93) of ZFTA^FUS-sig genes, including EPHB2 and GLI2, as well as L1CAM, which is used as a diagnostic marker of human ST-EP-ZFTA^FUS ependymoma (6); however, only ZFTA–RELA^FUS1 binding was associated with increased gene expression (Fig. 4D; Supplementary Fig. S4A). In stark contrast, RELA bound only 11 ZFTA^FUS-sig genes, with no detectable gene induction. Together, these data support the notion that ZFTA tethers ZFTA–RELA^FUS1 to the genome, allowing the RELA–TAD to promote aberrant gene transcription.

Following chromatin binding, transcription factors recruit chromatin-modifying complexes that create a permissive or suppressive transcriptional environment (18). Review of published ChIP-seq profiles (19) revealed that 59% (38/93 overlap, P = 4.61e⁻⁴⁹) of ZFTA^FUS-sig genes in human ST-EP-ZFTA^FUS ependymomas are marked as “active” with H3K27ac, including EPHB2, GLI2, CCND1, and L1CAM (Fig. 4E; Supplementary Fig. S4A and S4B). Furthermore, H3K27ac marking of genes in ZFTA–RELA^FUS1-transduced HEK293 cells and mNSCs were significantly enriched for ZFTA^FUS-sig genes (Fig. 4D), and our mouse model of ST-EP-ZFTA^FUS, as well as a patient-derived xenograft of ST-EP-ZFTA^FUS, expressed much greater levels of H3K27ac than did a non–fusion-driven ST-EP ependymoma (Supplementary Fig. S4C). Together, these data are compatible with the notion that fusion proteins bind and remodel chromatin to an active state, promoting the transcription of ZFTA^FUS-sig genes to drive ependymoma tumorigenesis. Indeed, an accompanying study also identified ZFTA–RELA^FUS1 binding to, and activation of, ZFTA^FUS-sig genes (15).

To assess further the extent to which ZFTA–RELA^FUS1 binding modifies chromatin, we performed ChIP-seq of H3K4me1 (poised/active enhancers), H3K36me3 (gene elongation), H3K27me3 (downregulated genes), and H2AK119ub (repressed developmental genes) in HEK293 cells and mNSCs (Supplementary Fig. S4D and S4E). Robust ChIP-seq profiles of these marks could not be obtained in HEK293 cells, presumably because they did not tolerate prolonged expression of ZFTA–RELA^FUS1. However, the histone landscape of mNSCs, which did tolerate longer exposure to fusion proteins, was extensively remodeled by ZFTA–RELA^FUS1, resulting in overwhelming gains of H3K4me1 (2,614/3,016 significant peaks) and H3K36me3 (639/1,071) and loss of H3K27me3 (n = 407/459). Thus, ZFTA–RELA^FUS1 appears to function as an aberrant transcription factor, binding and remodeling chromatin to drive expression of a transforming transcriptome.

To characterize how ZFTA–RELA^FUS1 modifies chromatin to drive gene transcription, we quantified its protein–protein interactions on chromatin using a cross-linking immunoprecipitation and mass spectrometry technique termed qPLEX-RIME (20). We again focused our studies in HEK293 cells in which ZFTA–RELA^FUS1 bound and activated ZFTA^FUS-sig genes most faithfully.

Principal component analysis positioned the ZFTA–RELA^FUS1 protein interactome between those of ZFTA and RELA, suggesting these proteins share binding partners (Fig. 5A). In keeping with this notion and their widespread chromatin binding, proteins interacting with ZFTA and ZFTA–RELA^FUS1 were highly enriched for nuclear proteins and/or those involved in chromatin binding, remodeling, and transcription (Fig. 5B; Supplementary Fig. S5A). Conversely, RELA and ZFTA–RELA^FUS1, but not ZFTA, bound components of the NFκB interactome, including NFκB2 (Fig. 5B; Supplementary Fig. S5A). This observation was confirmed by independent coimmunoprecipitation studies without cross-linking (Fig. 5C). Conversely, SMARCA4, SMARCC1, SMARCD1, SMARCD2, and SMARCE1, which are key components of the SWI/SNF (SWItch/sucrose nonfermentable) complex that regulates nucleosome positioning (21), and the histone acetylator CREBBP were unique to the ZFTA–RELA^FUS1 interactome.

To further define the domains of ZFTA–RELA^FUS1 that recruit protein complexes, we repeated our qPLEX-RIME and immunoprecipitation studies in HEK293 cells expressing either ZFTA–RELA^FUS1ΔZF, ZFTA–RELA^FUS1ΔRHD, ZFTA–RELA^FUS1ΔTAD, or the full-length fusion (Fig. 6A and B). Loss of the ZFTA-ZF markedly impaired fusion protein binding of three major chromatin-remodeling complexes (enrichment relative to the CORUM database; ref. 22; Fig. 6A; Supplementary Fig. S5B): SWI/SNF (SMARCA4, SMARCB1, SMARCC2, SMARCD2, SMARCD3, ACTL6A, ARID1A; FDR = 5.89e⁻¹⁶); SAGA (Spt-Ada-Gcn5 acetyltransferase) that acetylates histones (TADA3, TAF5L, TAF6L, TADA2B, TTRAP, SUPT20H, USP22, SUPT3H, SGF29, TADA1, ATXN7L2; FDR = 2.60e⁻²³; ref. 23); and NuA4/Tip60-HAT (NuA4/Tip60 histone acetyltransferase) that acetylates histone H4 and H2A (YEATS4, EP400, EPC1, TRRAP, VPS72; FDR = 1.57e⁻¹⁵; ref. 24). Deletion of this domain also blocked binding of other key chromatin-remodeling proteins, including BRD4 and ATRX. These data were validated by independent coimmunoprecipitation studies of selected proteins in the reverse direction (Fig. 6C). Notably, Cut-n-Run-Sequencing studies in an accompanying article also identify corecruitment of Brd4, Ep300, Crebbp, and RNA polymerase 2 to most ZFTA–RELA^FUS1 binding sites (15).

While failure of ZFTA–RELA^FUS1ΔZF to access the nucleus very likely impeded its binding of chromatin-remodeling complexes (Fig. 2A–C), neither the RELA–RHD (ZFTA–RELA^FUS1ΔRHD) nor TAD (ZFTA–RELA^FUS1ΔTAD) were required to recruit SWI/SNF, SAGA, or NuA4/Tip60 HAT complexes (Fig. 6A; Supplementary Fig. S5B). Thus, we propose that the ZFTA-ZF coordinates both chromatin binding and recruitment of chromatin-remodeling complexes to fusion proteins.

As expected, NFκB complex binding (NFKB1, NFKBIA, NFKBIB, NFKBIE, REL; FDR = 3.98e⁻¹³) was confined largely to the RELA–RHD (Fig. 6A; Supplementary Fig. S5B). Deletion of the RELA–RHD also ablated binding of the minichromosome maintenance (MCM) 2/4/6/7 complex that is required for both DNA replication initiation and elongation (25). Interaction between the NFκB and MCM2/4/6/7 complexes has not been reported, suggesting this is a novel property of the ZFTA–RELA^FUS1 protein. Deletion of the RELA–RHD also disrupted fusion protein recruitment of a several other proteins involved in the NFκB interactome, including components of the 20S core proteasome complex responsible for proteolytic processing of NFKB1 (Figs. 5B and 6A; ref. 26).

In keeping with the property of TADs to provide a protein scaffold for transcription coregulators (27), deletion of the RELA–TAD ablated ZFTA–RELA^FUS1 interaction with the Mediator complex (MED11, MED16, MED22, MED27, MED30, MED4, MED14, MED18, MED20, MED23, MED24, MED31, JMJD6; FDR = 1.24e⁻²⁶; Fig. 6A; Supplementary Fig. S5B). This transcription coregulator communicates between active enhancers and promoters by interacting with proteins that bind to either of these two classes of regulatory DNA elements (28). Thus, the RELA–TAD, which is critical for driving ZFTA–RELA^FUS1-induced aberrant gene expression (Fig. 3C and D), likely mediates the recruitment of key transcriptional activators. Proteins within the Mediator complex were enriched to a lesser extent within the ZFTA-ZF and RELA–RHD interactomes, suggesting these domains might also coordinate ZFTA–RELA^FUS1-Mediator complex binding (Fig. 6A; Supplementary Fig. S5B).

To compare the interactomes of different ZFTA-fusion proteins, we also cataloged protein-binding partners of ZFTA–YAP1 using qPLEX-RIME (Supplementary Fig. S6). The interactome of ZFTA–YAP1 was remarkably similar to that of ZFTA–RELA^FUS1 (overlap, P = 3.61e⁻⁴⁰; Supplementary Fig. S6A and S6B) and included chromatin-binding proteins (FDR = 8.2e⁻⁹), SWI/SNF (FDR = 8.2e⁻⁹), MCM (FDR = 2.3e⁻³), and SAGA complexes (FDR = 0.0004), as well as members of the Mediator and NuA4/Tip60 HAT complexes. As expected, ZFTA–YAP1 did not recruit the NFκB complex, but did bind known interactors of YAP1, including CTNNB1, TEAD1, and TEAD3. The interactomes of ZFTA–YAP1 and the ZFTA–YAP1ΔZF2–3 deletion mutant were almost identical, supporting further the notion that a single ZF is both necessary and sufficient for fusion-driven transformation (overlap, <1.0e⁻³⁰⁰; Supplementary Fig. S6A). Thus, ZFTA-fusion proteins recruit similar interactomes regardless of binding partners, explaining why different fusions drive similar transcriptomes.

To test more directly whether the fusion partners of ZFTA recruit transcriptional coregulators, we created an artificial fusion protein between ZFTA and the histone acetyltransferase EP300 that interacts with ZFTA–RELA^FUS1 (Fig. 6D) and was reported recently to colocate with the Mediator complex at superenhancers in NSCs (29). ZFTA–EP300 upregulated 4,171 genes in HEK293 cells including 67% (62/93) of ZFTA^FUS-sig genes (overlap P = 7.21e-²⁵; Fig. 6E). Thirty-seven percent of genes upregulated by ZFTA–RELA^FUS1 in HEK293 cells were also upregulated by ZFTA–EP300 (overlap P = 6.45e⁻⁷⁵; Fig. 6E), including 56% (52/93) of ZFTA^FUS-sig genes. Expression levels of ZFTA^FUS-sig genes correlated closely in ZFTA–EP300- and ZFTA–RELA^FUS1–transduced HEK293 cells (R = 0.72; P < 0.0001; Fig. 6F). Despite these similarities, ZFTA–EP300 failed to upregulate GLI2 that is shown in an accompanying article to be crucial for ZFTA fusion–driven transformation (Fig. 6E; ref. 15). This article further shows that ZFTA–EP300 does not drive brain tumorigenesis in mice, supporting the notion that aberrant hedgehog signaling is an important mediator of fusion-driven ependymoma tumorigenesis.

Together, these data suggest a model in which the ZFTA-ZF promotes genome-wide chromatin binding of ZFTA-fusion proteins and the recruitment of key chromatin-remodeling proteins including the SWI/SNF, SAGA, and NuA4/Tip60 complexes (Fig. 7A). Aberrant transcription is then driven by fusion partners including RELA and YAP1 that involves transcriptional coactivators and the Mediator complex. Although the RELA–RHD recruits the NFκB and the MCM complex, this is not required for fusion-driven gene expression.

We showed previously that mNSCs transduced with either ZFTA–RELA^FUS1, ZFTA–RELA^FUS2, or ZFTA–YAP1^FUS, but not wild-type ZFTA, RELA, or YAP1, form brain tumors in mice (6, 12). Therefore, to test directly which motifs within ZFTA–RELA^FUS1 are required for transformation, we transduced mNSCs with either full-length ZFTA–RELA^FUS1, ZFTA–RELA^FUS1ΔZF, ZFTA–RELA^FUS1ΔRHD, or ZFTA–RELA^FUS1ΔTAD and injected suspensions of 1 × 10⁶ of these cells separately into the brains of 4-week-old immunocompromised mice exactly as described previously (6).

As expected, all mice harboring ZFTA–RELA^FUS1–transduced mNSCs succumbed to brain tumors (n = 14, median survival, 118 days). In stark contrast, neither cells transduced with ZFTA–RELA^FUS1ΔZF (n = 10) or ZFTA–RELA^FUS1ΔTAD (n = 10) formed tumors (surveillance, 225 days postimplant; Fig. 7B). This requirement of ZFTA-ZF to drive ependymomas in mice was also confirmed in an accompanying study (16). Thus, the chromatin binding and remodeling functions of ZFTA-ZF, and the transcription coactivator properties of the RELA–TAD, are indispensable for fusion-driven tumorigenesis.

In stark contrast, all mice injected with ZFTA–RELA^FUS1ΔRHD–transduced mNSCs developed brain tumors, in keeping with its redundancy in fusion chromatin-remodeling complex recruitment and aberrant gene expression. These tumors were histologically indistinguishable from those driven by full-length ZFTA–RELA^FUS1, but took significantly longer to form (P < 0.005; Fig. 7B and C). Thus, although not required for tumorigenesis, recruitment of the NFκB and MCM complexes, and/or nuclear localization of ZFTA–RELA^FUS1 by the RHD, might contribute to tumor development (Figs. 2C and 7A).

We identify three functions conferred by ZFTA on oncogenic fusion proteins: nuclear translocation, chromatin binding, and the recruitment of chromatin modifiers. ZFTA thereby tethers fusion proteins across the genome, modifying chromatin to an active state and enabling its partner transcriptional coactivators (RELA, YAP1, MAML2, or MRTFB) to promote promiscuous expression of a transforming transcriptome.

ZF domains promote nuclear translocation in part by binding the nuclear import protein KPNB1 via their C2H2 and/or NLS motifs (30). A single ZFTA-ZF (its C2H2 or NLS motifs in particular) was necessary and sufficient to traffic fusion proteins to the nucleus. Neither the RELA–RHD nor RELA–TAD were required for nuclear translocation, although deletion of the RELA–RHD reduced nuclear trafficking by approximately 30%. Thus, ZFTA–RELA^FUS may be maintained in the nucleus by RELA–RHD-mediated protein–protein or protein–DNA interactions; these may include the NFκB and/or MCM complexes that bound the RELA–RHD. The MCM complex forms a DNA helicase important in DNA replication, potentially explaining why RELA–RHD deletion prolonged fusion-driven tumor latency.

Our finding that both ZFTA and ZFTA–RELA^FUS1 display widespread, overlapping chromatin binding provides strong evidence that the ZFTA-ZF also mediates DNA binding of fusion proteins. This was associated with the recruitment of the SWI/SNF, SAGA, and NuA4/TIP60 complexes and profound remodeling of the chromatin landscape, converting thousands of genes—including ZFTA^FUS-sig genes—to a transcriptionally active state. Thus, we propose that ZF domains that are known to recruit chromatin modifiers confer this property on fusion proteins (31–33). But important caveats warrant further investigation. Although nuclear trafficking and DNA binding of ZFTA and ZFTA–RELA^FUS1 were similar, only the latter was detected in association with chromatin-remodeling proteins. And SWI/SNF, SAGA, and NuA4/TIP60 complex binding to ZFTA–RELA^FUS1 was revealed only by comparison with the ZFTA–RELA^FUS1ΔZF mutant that does not access the nucleus. Nevertheless, both ZFTA–RELA^FUS1ΔRHD and ZFTA–RELA^FUS1ΔTAD that do translocate to the nucleus as well as ZFTA–YAP1 readily recruited SWI/SNF, SAGA, and NuA4/Tip60 HAT complexes. Furthermore, our artificial ZFTA–EP300 fusion and the array of disparate partner genes in ZFTA-containing fusions drive expression of ZFTA^FUS-sig genes. Thus, we propose that the ZFTA-ZF mediates SWI/SNF, SAGA, and NuA4/TIP60 complex recruitment as well as nuclear translocation and chromatin binding.

The observation that HEK293 cells recapitulate the ZFTA^FUS-sig more faithfully than mNSCs that drive ependymoma in mice was surprising and suggests important species-specific differences in ZFTA-fusion activity. Furthermore, because ZFTA fusions possess the capacity to remodel diverse genomes, then the almost exclusive restriction of these fusions to human supratentorial ependymoma suggests NSCs are uniquely susceptible to the 11q chromothripsis that drives ZFTA translocations, rather than to fusion-driven aberrant gene transcription.

The notion that ZFTA confers critical chromatin binding and remodeling functions on fusion proteins is supported further by the observation that it is the common partner in an array of oncogenic fusions. Fusion partners with ZFTA include RELA, YAP1, MAML2, and MRTFB, which are all transcriptional coregulators. The conjoining of their transcription-promoting activity with the DNA binding and chromatin remodeling of ZFTA likely accounts for the potent induction of gene transcription by fusion proteins. The “selection” of RELA, YAP1, and MAML2 as partners of ZFTA likely represents topological “marriages of convenience”: in human ST-SP-ZFTA^FUS, extensive chromothripsis of 11q brings ZFTA (11q13.1) into close proximity with RELA (11q13.1), YAP1 (11q22.1), or MAML2 (11q21; ref. 6). ZFTA partner genes also share an ability to recruit the Mediator complex. Because the Mediator complex organizes the transcriptional network that determines NSC identity (29) and genes upregulated by ZFTA fusion included regulators of neurogenesis, then perturbation of neural cell fate is likely fundamental to fusion-driven tumor initiation.

Finally, together with complementary data in two accompanying studies, we unmask novel approaches for treating fusion-driven ependymoma (15, 16). Zheng and colleagues (16) identify GLI2 as an important target of fusion-driven transformation that we show is bound and upregulated by ZFTA–RELA^FUS1. Notably, our artificial fusion ZFTA–EP300 that drove expression of most ZFTA^FUS-sig genes did not upregulate GLI2 and lacked transforming capacity in vivo. Thus, aberrant Hedgehog signaling might prove a useful therapeutic target. Similar to Arabzade and colleagues (15), we also show that ZFTA–RELA^FUS1 localizes to enhancers and promoters and recruits transcriptional activation complexes. Thus, thalidomide analogues that cause the rapid ubiquitination and proteasomal degradation of ZF proteins (34) are a second potential therapeutic approach worthy of further study.

A full description of all methods and reagents can be found in Supplementary Methods and Supplementary Table S2.

Embryonic day (E) 14.5 Cdkn2a^−/− mNSCs were harvested from mice (under project license approved by the University of Cambridge Animal Welfare and Ethical Review Body) and expanded as neurospheres as described previously (5) in Neurobasal media supplemented with B27, N2, l-Glutamine, and penicillin–streptomycin, in the presence of 20 ng/mL EGF and FGF2 (all reagents are described in Supplementary Table S2). Human embryonic kidney (HEK) 293 cells were obtained from the ATCC and maintained in standard DMEM with 10% FCS. ELISA tests were performed monthly to ensure Mycoplasma-free cell status, and cell identity was authenticated prior to each set of in vitro and in vivo experiments using Single Tandem Repeat genotyping. mNSCs and HEK293 cells were plated on mouse laminin or poly-d-Lysine–coated plates, respectively. Nuclear and cytoplasmic extractions of cells were prepared using the NE-PER Kit (Thermo Fisher Scientific). Western blotting and fluorescence microscopy of cultured cells were conducted exactly as described previously (4, 5) using antibodies described in Supplementary Table S2.

Deletion mutants of ZFTA–RELA^FUS1 were generated using the In-Fusion system (Clontech) by deleting the first ZF (ΔZF, W109-W155), C2-H2 ZF (ΔC2-H2, M124-H149), REL homology domain (ΔRHD, P229-Y516), RELA transactivation domain (ΔTAD, P625-A754), RELA nuclear localization signal (ΔRELA-NLS, L499-P531), or RELA nuclear export signal (ΔRELA-NES, L646-L660). ZFTA–YAP1 mutants deleted ZFs 3–4 (W344-R649) or ZFs 2–4 (K214-R649). All mutants were subcloned into retrovirus carrying the indicated protein tag.

Retroviruses and lentiviruses were generated in 293T cells (ATCC) by Lipofectamine 2000 transfection with viral packaging plasmids encoding pmd2.G, psPAX2 (Addgene), and the transfer vector expressing the construct of interest. Virus-containing supernatant was collected at 48 and 72 hours posttransfection, filtered and concentrated with PEG-it virus precipitation solution (SBI) overnight at 4°C.

Total RNA libraries were generated using the TruSeq mRNA library prep kit (Illumina), 50 base-pair single-end reads acquired using the HiSeq 4000, and aligned to the human (hg38) or mouse (mm10) genome using TopHat (35). Read counts were normalized and differential gene expression determined using DESeq2 (36).

Cells transfected with Avi-tagged proteins (Genecopoeia) were cross-linked after 48 hours in 1% formaldehyde and nuclei sonicated. For histone ChIP-seq, cells were cross-linked and sonicated in ChIP Lysis Buffer before primary antibody-bead immunoprecipitation. Sheared chromatin was incubated with magnetic streptavidin beads (Thermo Fisher Scientific) to generate ChIP-seq and input libraries using the ThruPlex Sample Prep Kit (Illumina). ChIP-seq sequences were aligned to hg19 or mm10 and peaks analyzed using MACS and DiffBind (37). Profileplyr was used to count overlapping reads, annotate genomic features, and generate plots. Reads were counted in 25 bp bins covering 1,000 bp upstream and downstream of each 500 bp differentially bound site (total width of each region: 2,500 bp).

qPLEX-RIME assays were performed as described previously (20). Briefly, cells expressing the appropriate biotinylated protein were cross-linked, nuclei sonicated, and incubated with HA-coated magnetic beads (Thermo Fisher Scientific). Beads were subjected to protein digestion and peptides dried (via speedvac), reconstituted in 0.1 mol/L triethylammonium bicarbonate, and labeled using TMT-10plex reagents (Thermo Fisher Scientific). The peptide mixture was fractionated with reverse-phase cartridges at high pH (Pierce) and reconstituted peptides analyzed using a Dionex Ultimate 3000 UHPLC system with nano-ESI Fusion Lumos (Thermo Fisher Scientific). CID tandem mass spectra were processed using SequestHT on Proteome Discoverer 2.1 and analyzed using qPLEXanalyzer, R-bioconductor package (20).

All experiments involving animals were carried out under a project license approved by the University of Cambridge Animal Welfare and Ethical Review Body. Orthotopic allografts of ST-EP in mice were generated in accordance with UK Home Office regulations under project license exactly as described previously (5, 6). Formalin-fixed paraffin embedded sections of tumors in mice were analyzed using hematoxylin and eosin, RELA, and H3K27acetyl mark immunostaining as described previously (6).

S.M. Pfister reports grants from IMI-2 project together with various companies outside the submitted work; in addition, S.M. Pfister has a patent for European Patent 16 710 700.2 DNA-Methylation Based Method for Classifying Tumor Species issued. 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.

R. Kupp: Conceptualization, formal analysis, investigation, writing–review and editing. L. Ruff: Investigation, writing–review and editing. S. Terranova: Investigation, writing–review and editing. E. Nathan: Investigation, writing–review and editing. S. Ballereau: Data curation, formal analysis, investigation, writing–original draft. R. Stark: Formal analysis, visualization, writing–review and editing. C. Sekhar Reddy Chilamakuri: Formal analysis, visualization, writing–review and editing. N. Hoffmann: Formal analysis, investigation, writing–review and editing. K. Wickham-Rahrmann: Investigation, writing–review and editing. M. Widdess: Investigation. A. Arabzade: Formal analysis, writing–review and editing. Y. Zhao: Formal analysis, writing–review and editing. S. Varadharajan: Formal analysis, writing–review and editing. T. Zheng: Formal analysis, writing–review and editing. M.K. Murugesan: Investigation. S.M. Pfister: Formal analysis, writing–review and editing. D. Kawauchi: Formal analysis, writing–review and editing. K.W. Pajtler: Formal analysis, writing–review and editing. B. Deneen: Formal analysis, writing–review and editing. S.C. Mack: Formal analysis, writing–review and editing.K.E. Masih: Formal analysis, writing–review and editing.B.E. Gryder: Formal analysis, writing–review and editing. J. Khan: Formal analysis, writing–review and editing. R.J. Gilbertson: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, writing–original draft, project administration, writing–review and editing.

This work was supported by grants to R. Gilbertson (Major Centre Core and Children's Brain Tumour Centre from Cancer Research UK; The Brain Tumour Charity; and P01CA96832 and R01CA129541 from the US National Cancer Research); S. Mack (RR170023, Cancer Prevention and Research Institute of Texas), and K. Pajtler (Collaborative Ependymoma Research Network). We thank the Cambridge Institute genomics, proteomics, biorepository unit (BRU), and histology core facilities for technical assistance. This work was also supported by Cancer Research UK, NCI (R01CA129541 and U54CA243125), SOHN foundation, and Mathile foundation.

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.