ZFTA–RELA ependymomas make itaconate to epigenetically drive fusion expression

Abstract

ZFTA–RELA⁺ ependymomas are malignant brain tumours defined by fusions formed between the putative chromatin remodeller ZFTA and the NF-κB mediator RELA¹. Here we show that ZFTA–RELA⁺ cells produce itaconate, a key macrophage-associated immunomodulatory metabolite². Itaconate is generated by cis-aconitate decarboxylase 1 (ACOD1; also known as IRG1). However, the production of itaconate by tumour cells and its tumour-intrinsic role are not well established. ACOD1 is upregulated in a ZFTA–RELA-dependent manner. Functionally, itaconate enables a feed-forward system that is crucial for the maintenance of pathogenic ZFTA–RELA levels. Itaconate epigenetically activates ZFTA–RELA transcription by enriching for activating H3K4me3 via inhibition of the H3K4 demethylase KDM5. ZFTA–RELA⁺ tumours enhance glutamine metabolism to supply carbons for itaconate synthesis. Antagonism of ACOD1 or glutamine metabolism reduces pathogenic ZFTA–RELA levels and is potently therapeutic in multiple in vivo models. Mechanistically, ZFTA–RELA epigenetically suppresses PTEN expression to upregulate PI3K–mTOR signalling, a known driver of glutaminolysis. Finally, suppression of ACOD1 or a combination of glutamine antagonism with PI3K–mTOR inhibition abrogates spinal metastasis. Our data demonstrate that ZFTA–RELA⁺ ependymomas subvert a macrophage-like itaconate metabolic pathway to maintain expression of the ZFTA–RELA driver, which implicates itaconate as a candidate oncometabolite. Taken together, our results position itaconate upregulation as a previously unappreciated driver of ZFTA–RELA⁺ ependymomas. Our work has implications for future drug development to reduce pathogenic ZFTA–RELA expression for this brain tumour, and will advance our understanding of oncometabolites as a new class of therapeutic dependencies in cancers.

Main

Ependymomas are aggressive brain tumours that arise throughout the neuraxis^3,4. Despite decades of research, surgical resection and adjuvant therapies remain the mainstays of treatment. Although these treatments are effective in prolonging patient survival, they are not curative^4,5. More than 80% of ependymomas that arise from cerebral hemispheres or the supratentorial (ST) brain region have recurrent fusions between zinc finger translocation associated (ZFTA) and RELA (also known as p65), a key component of the NF-κB pathway¹. Rare ependymomas have ZFTA fusions with non-RELA partners, including MAML1, MAML2, NCOA1 and NCOA2 (ref. ⁶). Collectively, these are referred to as ZFTA fusion ependymomas and occur in both children (frequency of 66–84%) and adults (frequency of 20–58%)^3,7,8,9.

Although neither ZFTA nor RELA are oncogenic, ZFTA–RELA acts as a potent oncogene and is sufficient to transform cells both in vitro and in vivo^1,6,10,11. ZFTA–RELA spontaneously translocates to the nucleus, and RELA drives pathological NF-κB signalling^1,12. Moreover, ZFTA–RELA can act as a master chromatin remodeller^{6,10,11,13,14}. Despite these biological insights, there are no effective treatments, targeted therapies or clinical trials available for ZFTA-driven ependymomas.

Our premise for this study is based on metabolic reprogramming as a cancer hallmark^15,16. Many cancers generate metabolites that contribute to cancer pathogenesis, and these are termed oncometabolites. Such oncometabolites include d-2-hydroxyglutarate (D-2HG) in cancers with mutant isocitrate dehydrogenase 1 and 2 (IDH1/2m), α-ketoglutarate (αKG), succinate and fumarate^17,18. Many of these metabolites directly affect chromatin by altering histone and DNA modifications^19,20,21. It is unknown whether the ZFTA–RELA fusion can reprogram integrated metabolic–epigenetic pathways and whether this information can be leveraged for therapeutic development. To address this knowledge gap, we analysed ependymoma tumour samples, patient-derived cells and animal models. As ZFTA–RELA can both activate NF-κB signalling and remodel chromatin^1,6,10,11,14, we focused our efforts on the interplay among signalling pathways, metabolism and epigenetics. Here we show that ZFTA–RELA induces the production of itaconate, an immunomodulatory metabolite secreted by macrophages^2,22,23. We demonstrate that ZFTA–RELA⁺ cells use glutamine carbons for itaconate production and that this glutamine dependency is a metabolic vulnerability. Moreover, interruption of this pathway leads to therapeutic effects in animal models and reduces protein levels of the pathogenic driver ZFTA–RELA. Our results provide a tractable and unexplored therapeutic avenue for these tumours.

ZFTA–RELA induces itaconate production

To define metabolic dependencies in ZFTA–RELA⁺ ependymomas, we performed metabolomics in isogenic mouse neuronal stem cell (mNSC) models expressing either empty vector (EV) (EV mNSCs) or ZFTA–RELA fusion 1 (the most common form, referred to hereafter as ZFTA–RELA mNSCs)^1,10. Itaconate was the highest upregulated metabolite in ZFTA–RELA mNSCs (Fig. 1a and Supplementary Table 1). Itaconate is an immunomodulatory metabolite that is generated by macrophages in response to pathogen or lipopolysaccharide (LPS) exposure^2,22, but its production by tumour cells is less explored. We confirmed our results by measuring itaconate levels in various models, using LPS-stimulated brain macrophages as positive controls. Itaconate levels were higher in ZFTA–RELA mNSCs than in control mNSCs and comparable with, but lower (around 50%) than, LPS-stimulated brain macrophages (Extended Data Fig. 1a). Itaconate is derived from the TCA cycle metabolite cis-aconitate. Glutamine, which can feed the TCA cycle, was also upregulated in ZFTA–RELA mNSCs compared with EV mNSCs (Fig. 1a,b and Supplementary Table 1). We validated this finding in ZFTA–RELA mNSCs and in patient-derived ependymoma cell lines (Fig. 1c and Extended Data Fig. 1b). Itaconate levels were higher in ZFTA–RELA mNSCs than in cells without a ZFTA fusion (non-ZFTA fusion cells) (Extended Data Fig. 1c). We therefore speculated that itaconate has a crucial role in the pathogenesis of ZFTA–RELA⁺ ependymomas.

Fig. 1: ZFTA–RELA induces itaconate production and ACOD1 inhibition suppress tumour growth.

a,b, Heatmap (a) and volcano plot (b) of differential metabolites in mNSCs transfected with ZFTA–RELA or EV (n = 4, each). FC, fold change. c, Itaconate levels (z-scores) in ZFTA–RELA isogenic mNSCs (n = 4 for all) and in the following patient-derived ependymoma (EPN) cell lines: EP1NS (ZFTA–RELA⁺; n = 5), EPN1425 (ZFTA–RELA⁺; n = 7); ST-1 (ZFTA–MAML3⁺; n = 5); and MAF-811 (non-fusion posterior fossa type A (PFA); n = 3). d, Itaconate is synthesized from cis-aconitate by ACOD1. Citraconate competitively inhibits ACOD1 (ref. ²⁶). e, ZFTA–RELA, endogenous (endog) RELA and ACOD1 protein levels relative to GAPDH in ZFTA–RELA and control mNSCs. f, Left, representative haematoxylin and eosin (left) and ACOD1 IHC (right) images from ZFTA–RELA⁺ and YAP1–MAML2⁺ cells. Right, blinded quantification of the following ependymomas: ST ZFTA–RELA⁺ (n = 18), ST non-ZFTA fusion (n = 5), PFA (n = 7), posterior fossa type B (PFB) (n = 2), spinal (n = 3) and myxopapillary (n = 4). Scale bar, 100 μm. g,h, ZFTA–RELA (g) and ACOD1 (h) protein levels relative to vinculin in EP1NS cells with or without ZFTA–RELA knockdown (using the shRNAs shZFTA-1–shZFTA-3, black arrows show best knockdown). i, Correlation between ZFTA–RELA and ACOD1 protein levels (relative to vinculin or GAPDH). H-59, H-57 and H-41 are also mouse ZFTA–RELA⁺ cells. j, Kaplan–Meier analysis of the following mouse models: Acod1^WT (n = 25, median survival of 24 days); Acod1^−/− (n = 18, median survival undefined) and Nes^creAcod ^fl/+ (n = 7, median survival undefined). NS, not significant. k, Kaplan–Meier analysis of IUE ZFTA–RELA Acod1^WT animals treated with vehicle (n = 7, median survival of 32 days) or dimethyl citraconate (DM-Cit; 25mg kg^–1, intravenously (i.v.); n = 7, median survival undefined). Schematic of the treatment schedule is indicated. l, Fold change in tumour volume of mice with ZFTA–RELA mNSC xenografts and treated with vehicle (control; n = 5), an ACOD1-specific inhibitor (IRG1-IN-1, 1 mg kg⁻¹, intraperitoneally (i.p.); n = 6) or a combination of IRG1-IN-1 (1 mg kg^–1, i.p.) and itaconate (25 mg kg⁻¹, i.v.; n = 8). Schematic of the treatment schedule is indicated. m, ZFTA–RELA mRNA levels in the following cells: EV mNSCs; ZFTA–RELA mNSCs treated with vehicle, DM-Cit (10 mM for 24 h) or shAcod1-1 (n = 2 each); and EPN1425 cells treated with vehicle or DM-Cit (n = 3 each). DMSO, dimethyl sulfoxide. Data are the mean ± s.d. and analysed by unpaired, two-tailed, two-sided, t-tests (b,m), one-way analysis of variance (ANOVA) with Tukey’s multiple comparisons test (c,l) or Dunn’s multiple comparisons test (f), Spearman’s correlation (i) or log-rank tests (j,k), all with 95% confidence intervals (CI). n indicates independent samples or animals.

Source data

We confirmed our results using multiple orthogonal approaches. Itaconate is synthesized from cis-aconitate by the enzyme ACOD1 (Fig. 1d). ACOD1 levels were higher in ZFTA–RELA⁺ cells than in controls (Fig. 1e and Extended Data Fig. 1d,e). To determine the relevance of this finding to human disease, we assessed ACOD1 expression by immunohistochemistry (IHC) in ZFTA–RELA⁺ ependymomas and in ependymomas caused by a non-ZFTA fusion. We also examined posterior fossa group A, posterior fossa group B, spinal and myxopapillary ependymomas (Supplementary Table 2). ACOD1 expression was higher in ZFTA–RELA⁺ (showing a range of expression) than in non-ZFTA fusion ependymomas (Fig. 1f and Extended Data Fig. 1f).

To test whether ACOD1 expression was driven by ZFTA–RELA, we used short hairpin RNAs (shRNAs) that target different regions of the fusion protein and confirmed knockdown by blotting for RELA, which detects both endogenous RELA and the higher-molecular-weight ZFTA–RELA (Fig. 1g). Functionally, ZFTA–RELA knockdown decreased the proliferation of patient-derived EP1NS (ZFTA–RELA⁺) cells (Extended Data Fig. 1g). Moreover, ACOD1 levels were reduced proportionally to the extent of ZFTA–RELA knockdown (Fig. 1h). To assess the relationship between the levels of ZFTA–RELA and ACOD1 expression, we generated mNSCs that express high or low levels of the fusion protein (Extended Data Fig. 1h). We also used a panel of ependymoma cells derived from a patient with a ZFTA–RELA⁺ tumour and from mice with varying amounts of ZFTA–RELA protein expression (Extended Data Fig. 1h). ACOD1 and ZFTA–RELA protein levels showed a positive correlation in these models (Fig. 1i and Extended Data Fig. 1i).

ZFTA–RELA spontaneously translocates to the nucleus in a ZFTA-dependent manner to activate NF-κB signalling¹. Because NF-κB upregulates ACOD1 in macrophages²⁴, we tested whether ZFTA–RELA⁺ cells use a similar mechanism. Deletion of the RELA trans-activating domain (TAD) or the REL homology domain (RHD) in ZFTA–RELA⁺ cells reduces NF-κB activation to slow tumour growth¹⁰. Furthermore, deletion of the zinc finger (ZF) domain in the ZFTA partner prevents nuclear translocation to suppress tumour formation¹⁰ (Extended Data Fig. 1j). A previous study¹⁰ showed that Acod1 mRNA is reduced in cells that express all three of these mutants, which suggests that both ZFTA and NF-κB have a role in ACOD1 upregulation (Extended Data Fig. 1k). We therefore treated ZFTA–RELA mNSCs with the NF-κB inhibitor JSH-23. Treatment led to reduced ACOD1 levels in a dose-dependent manner (Extended Data Fig. 1l). These data suggest that ZFTA–RELA upregulates ACOD1 expression and itaconate production in ependymomas.

Inhibition of ACOD1 suppresses tumour growth

To determine the functional role of itaconate and ACOD1 in ZFTA–RELA⁺ cells, we targeted ACOD1 in ZFTA–RELA models. Partial ACOD1 knockdown using independent shRNAs was toxic to ZFTA–RELA⁺ cells (Extended Data Fig. 2a,b). Toxicity was fully prevented through itaconate supplementation (Extended Data Fig. 2c). We next examined in vivo effects of targeting ACOD1 using mouse models in which ZFTA–RELA expression was induced via in utero electroporation (IUE ZFTA–RELA mouse models)^10,11. IUE ZFTA–RELA animals were generated in Acod1 wild-type (Acod1^WT) or whole-body Acod1^−/− mice²⁵. To restrict Acod1 knockout to ependymomas, we developed Nes^creAcod1^fl/+ IUE ZFTA–RELA animals, in which Acod1 knockout is confined to tumour cells. Both Acod1^−/− and Nes^creAcod1^fl/+ knockout mice exhibited a significant (P < 0.0001) increase (around fourfold) in overall survival compared with Acod1^WT animals (Fig. 1j). Moreover, there was no significant difference in survival (P = 0.8311) between Acod1^−/− and Nes^creAcod1^fl/+ IUE ZFTA–RELA animals, which suggests that targeting Acod1 in ependymoma cells is sufficient to suppress tumour growth (Fig. 1j).

To validate our findings, we leveraged a recently characterized itaconate isomer metabolite called citraconate that competitively inhibits ACOD1 (ref. ²⁶). The cell-permeable form dimethyl citraconate potently inhibits ACOD1 in supraphysiological concentrations²⁶. Dimethyl citraconate was toxic to ZFTA–RELA cells in vitro (Extended Data Fig. 2d). Mesaconate is similar in structure to itaconate but does not inhibit ACOD1 (ref. ²⁶) and did not show toxicity (Extended Data Fig. 2e). To test whether dimethyl citraconate treatment can recapitulate the effect observed in animals with genetic deletion of Acod1, we intravenously injected mice with dimethyl citraconate in IUE (Fig. 1j) and flank (Extended Data Fig. 2f) ZFTA–RELA in vivo models. Treatment with dimethyl citraconate increased overall survival and suppressed tumour growth in both models (Fig. 1j,k and Extended Data Fig. 2f). Next, we tested an ACOD1-specific inhibitor (IRG1-IN-1; also known as ERG344) in ZFTA–RELA mNSCs and in mice with MAF-1329 (ZFTA–RELA⁺) patient-derived xenografts (PDXs) in both flanks. We observed a significant decrease (P < 0.0001) in tumour growth in both models after treatment with IRG1-IN-1 (Fig. 1l and Extended Data Fig. 2g). Notably, in vivo supplementation of itaconate (unmodified itaconate, used throughout) restored tumour growth (Fig. 1l). Our results demonstrate that ACOD1 inhibition is toxic in models of ZFTA–RELA⁺ ependymoma.

Inhibition of ACOD1 reduces ZFTA–RELA levels

To understand the molecular mechanisms that mediate tumour suppression after targeting ACOD1, we performed RNA sequencing (RNA-seq) of ZFTA–RELA mNSCs with or without treatment with dimethyl citraconate or a shRNA targeted against Acod1 (shAcod1) (Extended Data Fig. 2h and Supplementary Tables 3 and 4). We determined commonly upregulated (143) and downregulated (261) genes. Upregulated pathways included neuron projection and cholesterol and lipid metabolism (Extended Data Fig. 2i,j and Supplementary Tables 5 and 6). Suppressed pathways included downregulation of NF-κB signatures as the top hit (Extended Data Fig. 2i,k and Supplementary Tables 5 and 7). Notably, ZFTA–RELA mRNA was downregulated after Acod1 knockdown and dimethyl citraconate treatment (Fig. 1m). We confirmed that ZFTA–RELA mRNA levels decreased after dimethyl citraconate treatment in patient-derived EPN1425 (ZFTA–RELA⁺) cells (Fig. 1m). Similarly, ZFTA–RELA protein levels were reduced after ACOD1 suppression in multiple cell lines, including in an inducible system (Fig. 2a,b and Extended Data Fig. 3a,b). We expressed either ACOD1 cDNA or vector controls in HEK293 cells (Extended Data Fig. 3c) and then transfected them with inducible ZFTA–RELA. In these cells, dimethyl citraconate reduced ZFTA–RELA levels, and this effect was attenuated by ACOD1 overexpression (Extended Data Fig. 3d). We validated our findings using the ACOD1 inhibitor IRG1-IN-1, which also reduced ZFTA–RELA levels (Extended Data Fig. 3e,f). Moreover, IRG1-IN-1 reduced ZFTA–RELA protein levels in vivo, and this effect was rescued by itaconate supplementation (Fig. 2c).

Fig. 2: Inhibition of ACOD1 reduces global H3K4me3 and decreases pathogenic ZFTA–RELA levels.

a, ZFTA–RELA and endogenous RELA levels relative to vinculin or GAPDH in EP1NS cells with or without shACOD1 or DM-Cit (5–10 mM for 24 h) treatment, and in EPN1425 cells with or without DM-Cit (10 mM for 24 h) treatment. b, ZFTA–RELA protein levels (normalized to vinculin or GAPDH) in EP1NS cells with or without shACOD1-1–shACOD1-3 treatment, and ZFTA–RELA mNSCs and EPN1425 cells with or without DM-Cit (10 mM for 24 h) treatment (n = 3 each). c, ZFTA–RELA, endogenous RELA and GAPDH levels in tumours (n = 3 each) from animals in Fig. 1l treated with vehicle, IRG1-IN-1 (1 mg kg^–1, i.p.) or a combination of IRG1-IN-1 and itaconate (25 mg kg^–1, i.v.). Tumour images are shown below. d,e, KDM5 enzyme inhibition activity (shown as mean ± s.d.) (d) and schematic (e) of itaconate (n = 9), D-2HG (n = 5) and L-2HG (n = 5) activity. IC₅₀ values are indicated. f, Blots of H3K4me3 and total H3 in ZFTA–RELA mNSCs with or without shAcod1 treatment (left), and EPN1425 and EP1NS cells with or without DM-Cit (10 mM for 48 h) treatment (right). g, Blot of H3K4me3 and total H3 in EV mNSCs with or without itaconate (5–10 mM for 96 h) treatment. h, H3K4me3, total H3, ZFTA–RELA, endogenous RELA and vinculin in ZFTA–RELA mNSCs with control (lane 1), shAcod1 (lane 2) or shAcod1 treatment plus 1 mM itaconate for 48 h (lane 3). i, Heatmaps of H3K4me3 ChIP–seq (left) and ATAC-seq (right) from ZFTA–RELA mNSCs treated with vehicle or DM-Cit (10 mM for 48 h) showing downregulated genes. j, ZFTA–RELA, endogenous RELA and vinculin in ZFTA–RELA mNSCs treated with DMSO, mesaconate (an itaconate analogue that does not inhibit ACOD1 (refs. ^26,61) used as a control, 10 mM) or DM-Cit (10 mM) along with indicated concentrations of the KDM5 inhibitor CPI-455 for 48 h. Data are the mean ± s.d. and analysed by one-way ANOVA with Tukey’s multiple comparisons test (b) or nonlinear regression-fit analysis (d), all with 95% CI. n indicates independent samples.

Source data

Our RNA-seq and quantitative PCR (qPCR) data demonstrated that ZFTA–RELA mRNA levels were decreased after targeting ACOD1 (Fig. 1m), which suggests that transcriptional regulation is involved in this process. Itaconate is structurally similar to the oncometabolite D-2HG produced by IDH1/2m^27,28,29. D-2HG competitively inhibits αKG-dependent DNA and histone demethylases, which results in epigenetic hypermethylation^30,31. Similar to D-2HG, itaconate competitively inhibits the αKG-dependent DNA demethylase TET2 in macrophages^27,28,29. On the basis of this premise, we proposed that itaconate modifies chromatin in ZFTA–RELA⁺ ependymomas and focused on the transcription-activating mark H3K4me3. The H3K4 histone demethylase KDM5 depends on αKG and is inhibited by D-2HG to increase H3K4me3 levels in tumours with mutant IDH1/2m^30,32. Given its structural similarity to 2HG, we proposed that itaconate could also inhibit KDM5. There are two isoforms of 2HG: D-2HG and L-2HG (the more potent physiological isoform)³³. We tested the ability of itaconate to inhibit KDM5 compared with both L-2HG and D-2HG in cell-free assays with unmodified metabolites. Itaconate had a stronger KDM5 inhibitory activity than D-2HG but showed similar half-maximal inhibitory (IC₅₀) values to the more potent L-2HG (Fig. 2d and Extended Data Fig. 3g).

These data led us to propose that itaconate inhibits KDM5 to increase H3K4me3. Conversely, suppression of itaconate reduces H3K4me3 levels (Fig. 2e). Acod1 knockdown in ZFTA–RELA mNSCs and dimethyl citraconate treatment in ZFTA–RELA⁺ cells reduced H3K4me3 levels (Fig. 2f and Extended Data Fig. 3h). Both unmodified citraconate and cell-permeable dimethyl citraconate reduced ZFTA–RELA and H3K4me3 levels, with a greater effect observed with cell-permeable dimethyl citraconate (Fig. 2f and Extended Data Fig. 3i). Treatment with dimethyl citraconate reduced H3K4me3 levels in cell lines without the ZFTA–RELA fusion, which suggested that there are additional regulatory mechanisms involved in this process (Extended Data Fig. 3j). Conversely, supplementation of itaconate to mNSCs that do not express a fusion protein (non-fusion mNSCs) was sufficient to increase H3K4me3 levels (Fig. 2g). Similarly, ACOD1 overexpression upregulated H3K4me3 in HEK293 cells (Extended Data Fig. 3k). The addition of exogenous itaconate to ZFTA–RELA⁺ cells with Acod1 knocked down increased H3K4me3 levels, but not H3K27ac or H3K9me3, and simultaneously rescued ZFTA–RELA expression (Fig. 2h and Extended Data Fig. 3l). Consistent with our hypothesis, exogenous addition of cell-permeable dimethyl αKG to ZFTA–RELA cells reduced both H3K4me3 and ZFTA–RELA levels (Extended Data Fig. 3m,n). Conversely, cell-permeable octyl-L-2HG increased ZFTA–RELA expression in EPN1425 cells (Extended Data Fig. 3n).

To assess genome-wide H3K4me3 alterations, we performed H3K4me3 chromatin immunoprecipitation with sequencing (ChIP–seq) and assessed chromatin accessibility in parallel using assay for transposase-accessible chromatin with sequencing (ATAC-seq). The addition of dimethyl citraconate reduced chromatin accessibility globally in ZFTA–RELA mNSCs (n = 12,117 reduced accessibility sites compared with n = 369 gained accessibility sites) (Fig. 2i, Extended Data Fig. 4a and Supplementary Table 8). Moreover, dimethyl citraconate enhanced chromatin accessibility of genes involved in neurodevelopment and lipid metabolism and suppressed genes involved in growth factor signalling and cell migration (Extended Data Fig. 4b,c and Supplementary Tables 9 and 10). Similarly, H3K4me3 ChIP–seq of ZFTA–RELA⁺ cells revealed that dimethyl citraconate treatment reduced H3K4me3 enrichment globally (n = 9,897 reduced sites compared with n = 230 enriched sites) (Fig. 2i, Extended Data Fig. 4d and Supplementary Table 11). Genes that gained H3K4me3 marks belonged mainly to neurogenesis and neurodevelopmental pathways (Extended Data Fig. 4e and Supplementary Table 12). Genes with reduced H3K4me3 marks corresponded to chromatin organization and remodelling (Extended Data Fig. 4f and Supplementary Table 13). Patient-derived EPN1425 cells treated with dimethyl citraconate showed similar changes in H3K4me3 enrichment (Extended Data Fig. 4g–i and Supplementary Tables 14–16). Overlapping of ATAC-seq and H3K4me3 ChIP–seq data revealed genes that were commonly epigenetically upregulated (n = 17) or downregulated (n = 6,820) after dimethyl citraconate treatment, and included chromatin remodelling and DNA repair pathways (Extended Data Fig. 4j,k and Supplementary Tables 17 and 18).

These data indicate that an increase in H3K4me3 levels, independent of itaconate treatment, would be sufficient to reverse the effects of dimethyl citraconate. To test this idea, we used CPI-455, a small-molecule KDM5 inhibitor. Treatment of ZFTA–RELA mNSCs with CPI-455 increased H3K4me3 levels (Extended Data Fig. 3o) and reversed the effects of dimethyl citraconate (Fig. 2j). These results demonstrate that ACOD1 inhibition decreases both global H3K4me3 and pathogenic ZFTA–RELA expression.

Citraconate epigenetically reduces ZFTA–RELA

The most common ZFTA–RELA fusion in patients with ependymomas is formed between exons 1 and 2 of ZFTA and the entire open reading frame of RELA¹ (Extended Data Fig. 5a). Regulatory elements essential for the expression of ZFTA–RELA are restricted to the ZFTA partner (shared with wild-type ZFTA and ZFTA–MAML3)^1,34,35 (Extended Data Fig. 5a,b). Moreover, overlap of ATAC-seq and H3K4me3 ChIP–seq data showed that Zfta, but not Rela, was epigenetically repressed by dimethyl citraconate in ZFTA–RELA⁺ cells (Fig. 3a). On the basis of this premise, we proposed that dimethyl citraconate reduces genomic H3K4me3 at ZFTA regulatory elements to reduce ZFTA–RELA transcription.

Fig. 3: DM-Cit reduces genomic H3K4me3 at the ZFTA regulatory region of ZFTA–RELA.

a, H3K4me3 enrichment and chromatin accessibility (ATAC-seq) in ZFTA–RELA mNSCs treated with DMSO or DM-Cit (10 mM for 24 or 48 h). b, Endogenous ZFTA, RELA and vinculin in EV mNSCs treated with itaconate (at the indicated concentrations for 96 h). c, ZFTA–RELA, endogenous ZFTA and vinculin in ZFTA–RELA mNSCs with or without the KDM5 inhibitor CPI-455 (at the indicated concentrations for 48 h). Both high and low exposure of the blots are shown. d,e, Representative H3K4me3 ChIP tracks (left) and quantification (right) of the ZFTA locus (also known as 2700081O15Rik; purple box) in ZFTA–RELA mNSCs (d) and EPN1425 cells (e) treated with DMSO or DM-Cit (10 mM for 48 h) (n = 2 each). f, Fold change (citraconate (10 mM) or DMSO for 48 h) for H3K4me3 ChIP–PCR (left) and mRNA qPCR (right) at ZFTA exon 5 (n = 3), ZFTA exons 1–2 (fusion breakpoint that is part of the ZFTA–RELA fusion, n = 6) and RELA exon 4 (n = 3) in EPN1425 cells. g, Schematic of KDM5A and KDMAB (KDM5) recognition sites (fusion unaltered, purple boxes)^36,37,39 and the mutant (KDM5δ, green boxes) in the regulatory ZFTA portion of the ZFTA–RELA fusion protein. h,i, ZFTA–RELA, endogenous RELA, RFP and GAPDH in EV mNSCs, ZFTA–RELA (unaltered) and ZFTA–RELA mutant (KDM5δ) (h). ZFTA–RELA relative to GAPDH was quantified (n = 5 each) (i). j,k, Cell numbers (n = 8 each) at 72 h, normalized to cells expressing the unaltered fusion protein (j), and average tumour volume (n = 8 samples each) in mice with xenografts of the unaltered fusion or the mutant KDM5δ protein (k). l, ZFTA–RELA, endogenous RELA and GAPDH in cells with the unaltered fusion protein (lane 1) or the mutant KDM5δ protein treated with vehicle (lane 2) or DM-Cit (lanes 3 and 4, at the indicated concentrations for 48 h). Data are the mean ± s.d. and analysed using two-way ANOVA with Šídák’s multiple comparisons test (f), one-way ANOVA with Tukey’s multiple comparisons test (i), unpaired, two-tailed, two-sided t-test (j) or two-way ANOVA with Šídák’s multiple comparisons test (k), all with 95% CI. n indicates independent samples.

Source data

Dimethyl citraconate reduced endogenous ZFTA (and ZFTA–RELA) levels in ZFTA–RELA⁺ cells and in mNSCs expressing ZFTA or RELA (Extended Data Fig. 5c–e). Conversely, itaconate supplementation increased both H3K4me3 levels (Fig. 2g) and endogenous ZFTA (Fig. 3b). We did not observe a similar consistent pattern with endogenous RELA (Fig. 3b). Moreover, treatment of ZFTA–RELA mNSCs with CPI-455 increased both endogenous ZFTA and ZFTA–RELA levels (Fig. 3c). Other ZFTA fusion ependymomas also contained the same ZFTA regulatory region (Extended Data Fig. 5a,b). ZFTA–MAML3 ST-1 ependymoma cells produced itaconate and expressed ACOD1, but at lower levels than in ZFTA–RELA⁺ cells (Extended Data Fig. 1c,e). Similar to ZFTA–RELA⁺ cells, dimethyl citraconate treatment reduced endogenous ZFTA and ZFTA–MAML3, and caused dose-dependent toxicity in ST-1 ependymoma cells (Extended Data Fig. 5f,g).

These data prompted us to focus on genomic changes at the ZFTA regulatory region. H3K4me3 ChIP–seq revealed decreased H3K4me3 enrichment at this region after dimethyl citraconate treatment in both ZFTA–RELA mNSCs and EPN1425 cells (Fig. 3d,e and Extended Data Fig. 5h). This was accompanied by decreased chromatin accessibility at the Zfta regulatory locus (Extended Data Fig. 5i). To investigate this ZFTA regulatory locus in the fusion gene, we conducted H3K4me3 ChIP–PCR in EPN1425 cells, using primers specific to the fusion breakpoint ZFTA regulatory region (exons 1 and 2). We observed a similar decrease in H3K4me3 enrichment after dimethyl citraconate treatment at the fusion ZFTA regulatory elements (Fig. 3f). As controls, primers that spanned exon 5 of ZFTA (excluded from the fusion protein) and exon 4 of RELA did not show a similar significant decrease in H3K4me3 enrichment after dimethyl citraconate treatment (Fig. 3f and Extended Data Fig. 5j). In parallel, dimethyl citraconate treatment decreased mRNA levels of fusion-specific ZFTA exons 1 and 2, whereas minimal or no change was observed in ZFTA exon 5 or RELA exon 4 (Fig. 3f and Extended Data Fig. 5k).

Because our data showed that itaconate can affect KDM5 activity (Fig. 2d), we investigated whether changes in H3K4me3 levels at the ZFTA regulatory region depend on KDM5. We identified two putative KDM5 recognition sites^36,37,38,39 in the ZFTA regulatory region (Fig. 3g). We proposed that mutating these KDM5 recognition sites would reduce H3K4 demethylation and lead to three potential outcomes: (1) increased H3K4me3 enrichment at the ZFTA regulatory region; (2) increased ZFTA–RELA expression; and (3) reduced sensitivity to dimethyl citraconate. mNSCs were transfected with red fluorescent protein (RFP)-tagged unaltered ZFTA–RELA or plasmids with a KDM5 mutant (KDM5δ). In this mutant, the KDM5A recognition site in ZFTA exon 1 was changed from CCGCCC to CCACCA, and the KDM5B recognition site in ZFTA exon 2 was changed from GCACAC to GCAAAC (Fig. 3g). We performed H3K4me3 ChIP–PCR using specific primers for the ZFTA regulatory region. Compared with mNSCs transfected with unaltered ZFTA–RELA, mNSCs transfected with KDM5δ showed increased H3K4me3 enrichment at the ZFTA regulatory region but did not have altered global H3K4me3 levels (Extended Data Fig. 5l,m). This was accompanied by an increase in ZFTA–RELA and downstream ACOD1 protein levels (Fig. 3h,i and Extended Data Fig. 5n). Functionally, there was increased proliferation in vitro and enhanced tumour growth in vivo with mNSCs expressing KDM5δ compared with mNSCs expressing unaltered ZFTA–RELA (Fig. 3j,k and Extended Data Fig. 5o). Moreover, KDM5δ-expressing mNSCs treated with dimethyl citraconate did not show decreased ZFTA–RELA expression (Fig. 3l) and had attenuated toxicity (Extended Data Fig. 5p). Overall, our data suggest that dimethyl citraconate reduces enrichment of H3K4me3 at ZFTA regulatory regions, which in turn leads to a decrease in ZFTA fusion expression.

Glutamine is used to produce itaconate

We mapped metabolic pathways used by ZFTA–RELA⁺ cells to generate itaconate using ¹³C-isotope labelling. Both glucose and glutamine can be used for itaconate production in macrophages^40,41. ZFTA–RELA⁺ cells produced higher glutamine levels than non-fusion cells (Fig. 1a,b and Extended Data Fig. 6a). Furthermore, ZFTA–RELA mNSCs and EPN1425 cells showed enhanced glutamine uptake and metabolism compared with non-fusion controls (Extended Data Fig. 6b,c). Glutamine can give rise to itaconate through oxidative decarboxylation (M+4) (Fig. 4a and Extended Data Fig. 6d) or through reductive carboxylation (M+5) (Fig. 4a and Extended Data Fig. 6e). Isotope tracing showed that glutamine gave rise to itaconate mainly through oxidative decarboxylation (M+4), with the fifth carbon derived from glucose (Fig. 4b and Extended Data Fig. 6f–h). Glutamine can be taken up by cancer cells via the importer SLC1A5 and is metabolized to glutamate by glutaminase (GLS)⁴² (Fig. 4a). GLS levels were higher in ZFTA–RELA mNSCs than in non-fusion mNSCs (Fig. 4c and Extended Data Fig. 7a). We confirmed higher SLC1A5 and GLS protein levels in ZFTA–RELA⁺ cells than in non-fusion cells (Extended Data Fig. 7b–e). ZFTA–RELA knockdown decreased both GLS and SLC1A5 levels (Fig. 4c and Extended Data Fig. 7e). Furthermore, compared with patients with non-fusion ependymomas, ZFTA–RELA⁺ cells showed higher mRNA levels of SLC1A5 and GLS from published bulk³ and single-cell RNA-seq⁴³ datasets (Extended Data Fig. 7f,g), and increased SLC1A5 expression by IHC (Extended Data Fig. 7h,i). Together, our data demonstrate that ZFTA–RELA⁺ ependymomas upregulate glutamine metabolism to generate itaconate.

Fig. 4: ZFTA–RELA epigenetically suppresses PTEN and drives glutamine metabolism to generate itaconate.

a,b, Schematic (a) and quantification (b) of itaconate biosynthesis from glutamine (U-¹³C-GLN) through oxidative phosphorylation (M+4) or reductive carboxylation (M+5) in ZFTA–RELA and EV mNSCs. c, GLS and GAPDH or vinculin in isogenic mNSCs (left) and EP1NS cells with or without shZFTA treatment (arrows show best knockdown, right, relative to vinculin). d,e, Schematic (d) and PTEN, pAKT (S473), total AKT, pS6RP (S235/S236), total S6RP, pGSK3α/β (S21/S9), total GSK3α/β and GAPDH in isogenic mNSCs (e). f, Representative PTEN IHC images (top) and blinded quantification (bottom) from mouse ZFTA–RELA⁺ ependymomas versus normal brain (n = 3 each), and from human ZFTA–RELA⁺ (n = 18) and YAP1–MAML2⁺ (n = 5) ependymomas. Scale bar, 50 μm. g, Kaplan–Meier analysis of progression-free (PFS, top) and overall (OS, bottom) survival in patients with PTEN high and GLS and SLC1A5 low expression (n = 17) and patients with PTEN low and GLS and SLC1A5 high expression (n = 11) ZFTA–RELA⁺ ependymomas, as determined by gene expression and k-means clustering. h, pAKT (S473), total AKT, pGSK3α/β (S21/S9), total GSK3α/β, MYC, GLS and vinculin in ZFTA–RELA mNSCs after treatment with the dual PI3K–mTOR inhibitor PI-103 (at the indicated concentrations for 24 h). i, H3K27me3, H3K9me3, H3K27ac and total H3 in isogenic mNSCs. j, Representative H3K27me3 IHC images (left) and blinded quantification of ZFTA–RELA⁺ (n = 18) and YAP1–MAML2⁺ (n = 5) ependymomas (right). Scale bar, 12.5 μm. k, H3K27me3, total H3, PTEN and vinculin in ZFTA–RELA (right) and EV mNSCs (left) with or without EZH2 inhibitor (GSK-343) treatment (at the indicated concentrations for 48 h). Data are the mean ± s.d. and analysed by unpaired, two-tailed, two-sided t-test (b,j), Mann–Whitney test (f) or log-rank test (g), all with 95% CI. n indicates independent samples.

Source data

ZFTA–RELA epigenetically reduces PTEN

We analysed in detail mechanisms by which ZFTA–RELA enhances glutamine metabolism. Glutaminolysis is regulated by master metabolic factors, including PI3K–AKT signalling and MYC that upregulate GLS⁴⁴ (Fig. 4d). Compared with control mNSCs, ZFTA–RELA mNSCs exhibited PI3K–AKT–mTOR activation, as evidenced by increased phosphorylated AKT (pAKT) and phosphorylated S6RP (pS6RP) levels and downstream phosphorylated GSK3α and GSK3β (pGSK3α/β) levels (Fig. 4e and Extended Data Fig. 8a,b). Moreover, pAKT and MYC were increased in patient-derived cells and in a dose-dependent manner when ZFTA–RELA expression was increased in mNSCs (Extended Data Fig. 8c,d). PTEN is a key tumour suppressor that antagonizes PI3K–AKT signalling. PTEN is often deleted in tumours, which leads to activation of PI3K–AKT, MYC and glutamine metabolism, which prompted us to analyse PTEN expression patterns in more detail^44,45,46. Compared with control mNSCs, ZFTA–RELA mNSCs had lower Pten mRNA and protein levels (Fig. 4e and Extended Data Fig. 8e,f). PTEN reduction was not due to genetic deletions or mutations^1,3 (Extended Data Fig. 8g). We verified that PTEN protein levels were lower in ZFTA–RELA⁺ cells than in non-fusion cells (Extended Data Fig. 8h). PTEN immunostaining was lower in ZFTA–RELA⁺ mouse tumours than in normal brain (Fig. 4f and Extended Data Fig. 8i). Notably, IHC showed that ZFTA–RELA⁺ ependymomas had reduced PTEN levels compared with non-ZFTA fusion ependymomas (Fig. 4f and Extended Data Fig. 8i). Higher levels of PI3KA, GSK3B and MYC mRNA levels in ZFTA–RELA⁺ tumour samples than in non-ZFTA fusion tumour samples were also observed in published datasets³ (Extended Data Fig. 8j).

We proposed that low PTEN increases GLS levels via PI3K–AKT signalling. In support of this hypothesis, low PTEN expression was associated with increased GLS and SLC1A5 expression from published ependymoma single-cell RNA-seq data⁴³ (Extended Data Fig. 8k,l). Moreover, patients with ZFTA–RELA⁺ ependymomas with low PTEN but high GLS and SLC1A5 expression showed worse progression-free and overall survival compared with patients with tumours with high PTEN but low GLS and SLC1A5 expression (grouped by k-means clustering) (Fig. 4g and Extended Data Fig. 8l). Treatment of cells with two well-characterized, small-molecule inhibitors, MK-2206 (an AKT inhibitor) and PI-103 (a dual PI3K–mTOR inhibitor), both decreased pAKT levels, which was accompanied by reduced MYC and GLS expression (Fig. 4h and Extended Data Fig. 8m).

As ZFTA–RELA⁺ ependymomas do not exhibit genetic PTEN deletions or mutations^1,3 (Extended Data Fig. 8g), we wanted to analyse the mechanism of PTEN downregulation by ZFTA–RELA. Previous studies have suggested that ZFTA–RELA is a chromatin remodeller^6,10,11,14. We examined PTEN mRNA levels in ZFTA–RELA-expressing cells with either nonfunctional ZFTA or RELA (TAD or RHD mutants) from published studies¹⁰. Only the ZFTA nonfunctional mutant showed high PTEN mRNA expression, which suggested that ZFTA, but not RELA, is required to downregulate PTEN (Extended Data Fig. 9a). In support of this idea, cells expressing ZFTA fusion proteins exhibited reduced PTEN mRNA levels (Extended Data Fig. 9b). These data, along with the known chromatin remodelling function of ZFTA–RELA, led us to speculate that ZFTA–RELA epigenetically represses PTEN expression.

We analysed ZFTA–RELA mNSCs and isogenic mNSCs for repressive H3K27me3 and H3K9me3 marks. ZFTA–RELA mNSCs showed increased H3K27me3 (ref. ¹¹), but not H3K9me3, levels compared with non-fusion mNSCs (Fig. 4i). Similarly, global H3K27me3 levels were higher in ZFTA–RELA⁺ ependymomas than in non-ZFTA fusion ependymomas (Fig. 4j and Extended Data Fig. 9c). ZFTA–RELA knockdown reduced H3K27me3 levels, which was accompanied by an increase in PTEN protein and mRNA expression (Extended Data Fig. 9d,e). Similarly, reducing ZFTA–RELA through ACOD1 inhibition (Fig. 1m) increased Pten expression (Extended Data Fig. 9f). These data suggest that ZFTA–RELA represses PTEN expression in a H3K27me3-dependent manner. The addition of GSK-343, an EZH2 (H3K27 methyltransferase) inhibitor, led to reduced global H3K27me3 levels in both ZFTA–RELA and EV mNSCs (Fig. 4k). ATAC-seq and H3K27me3 ChIP–seq of ZFTA–RELA⁺ cells (Extended Data Fig. 9g) showed that there was reduced chromatin accessibility at the Pten locus, which was reversed by GSK-343 treatment (Extended Data Fig. 9h). Moreover, H3K27me3 enrichment was reduced at the Pten locus after EZH2 inhibition (Extended Data Fig. 9h), which corresponded to increased PTEN expression in ZFTA–RELA mNSCs (Fig. 4k). Likewise, PTEN levels in EP1NS cells also increased with GSK-343 treatment (Extended Data Fig. 9i,j), which was accompanied by a decrease in downstream AKT activation and reduced GLS levels (Extended Data Fig. 9j,k). EV mNSCs treated with GSK-343 also showed a reduction in global H3K27me3, whereas no changes in PTEN protein were observed (Fig. 4k). Our data suggest that ZFTA–RELA epigenetically represses PTEN expression to activate AKT signalling and drive glutamine metabolism.

Targeting ACOD1 and glutamine metabolism

We assessed the effect of glutamine withdrawal from culture media. Glutamine withdrawal led to increased toxicity in all ZFTA–RELA⁺ cells compared with control cells (Fig. 5a). This toxicity was partially reversed by itaconate supplementation (Extended Data Fig. 10a). Glutamine withdrawal also decreased H3K4me3, but not H3K9me3, levels (Extended Data Fig. 10b). We next tested various small-molecule inhibitors of glutamine metabolism (Extended Data Fig. 10c). Treatment with the SLC1A5 inhibitor V9302 (ref. ⁴⁷) or the GLS inhibitor CB-839 (ref. ⁴⁸) resulted in a greater toxicity in ZFTA–RELA mNSCs than in non-fusion mNSCs (Extended Data Fig. 10d,e). The glutamine antagonist 6-diazo-5-oxo-l-norleucine (DON) reduced tumour growth in mice that had ZFTA–RELA mNSCs transplanted into their flank (Extended Data Fig. 10f,g). Overcoming the blood–brain barrier is a major therapeutic obstacle for brain tumours. We therefore focused our efforts on the glutamine antagonist JHU-083, which is highly CNS-penetrant^49,50. ZFTA–RELA⁺ cells showed greater toxicity after JHU-083 treatment than non-fusion cells (Fig. 5b). Consistent with glutamine feeding itaconate biosynthesis, JHU-083 treatment reduced itaconate levels in ZFTA–RELA mNSCs (Extended Data Fig. 11a). Notably, JHU-083 decreased ZFTA–RELA protein levels in vitro (Fig. 5c) and in vivo (Extended Data Fig. 11b). Moreover, itaconate supplementation reversed JHU-083 toxicity and restored ZFTA–RELA protein levels (Extended Data Fig. 11c,d).

Fig. 5: Targeting glutamine metabolism leads to therapeutic effects in ZFTA–RELA ependymomas.

a,b, Cell numbers (normalized to control) in ZFTA–RELA⁺, ZFTA–MAML3⁺ (ST-1) and non-ZFTA fusion cells with or without glutamine (a) or JHU-083 treatment (b, at the indicated concentrations) at 96 h. n = 4 (ZFTA–RELA mNSC, EV mNSC, ST-01 and EPD-210), n = 6 (ST-04, CPITT-1, ST-02 and ST-1) or n = 8 (EP1NS and EPN1425). c, ZFTA–RELA, endogenous RELA and GAPDH in ZFTA–RELA mNSCs (left) and in EP1NS patient-derived cells (right) treated with or without JHU-083 (at the indicated concentrations for 96 h). d,e, Representative bioluminescence images (left) and quantification (right, fold change, medians with interquartile ranges, box plot ends show highest and lowest observations) (d) and Kaplan–Meier analysis (e) of IUE ZFTA–RELA mice treated with vehicle (n = 9) or JHU-083 (n = 10, orally, 20 mg kg^–1, 2 doses per week for 5 weeks). f, Tumour volume (fold change) in mice with MAF-1329 (ZFTA–RELA⁺) PDXs treated with vehicle (n = 5), JHU-083 (n = 5, 20 mg kg^–1 orally, 2 doses per week), DM-Cit (n = 5, 25 mg kg⁻¹, i.v., 3 doses per week) or a combination of JHU-083 and DM-Cit (n = 5) for 4 weeks. g,h, ZFTA–RELA and endogenous RELA and GAPDH (g) and quantification (h) of MAF-1329 PDX tumour tissues treated with vehicle (n = 4), DM-Cit (n = 3) or their combination (n = 3). i,j, Representative bioluminescence images (left) and quantification (right) (i) and haematoxylin and eosin images (j, dotted region) from spontaneous spinal metastatic tumours in mice with ZFTA–RELA mNSC xenografts treated with vehicle (n = 8), DM-Cit (n = 6) or a combination of JHU-083 and PI-103. Scale bar, 124 µm. k, Schematic illustrating the proposed working model of ZFTA–RELA-driven metabolic reprogramming in ST ependymomas. Data are the mean ± s.d. (except d, mean ± minimum to maximum) and analysed by Šídák’s multiple-comparison test (a), two-way ANOVA with Tukey’s multiple comparisons test (b), Mann–Whitney (d), log-rank (e), two-way (f) or one-way ANOVA (h) with Dunnett’s multiple-comparison test, or one-way ANOVA with Holm–Šídák’s multiple-comparisons test (i), all with 95% CI. n indicates independent samples or animals.

Source data

We tested the efficacy of targeting glutamine metabolism with JHU-083 in vivo using multiple animal models. JHU-083 suppressed tumour growth and increased overall survival in mouse models of orthotopic ZFTA–RELA mNSC transplant and in IUE models (Fig. 5d,e and Extended Data Fig. 11f,g). In vivo itaconate supplementation partially rescued JHU-083-mediated tumour suppression (Extended Data Fig. 11e). JHU-083 along with standard-of-care radiation treatment in IUE ZFTA–RELA animals extended survival compared with radiation monotherapy (Extended Data Fig. 11h). We also tested combination therapy with JHU-083 and dimethyl citraconate in mice with a PDX of MAF-1329 tumour cells (Fig. 5f). Intravenous treatment of PDXs with dimethyl citraconate or JHU-083 monotherapy or combination therapy suppressed tumour growth in vivo (Fig. 5f). We confirmed that ZFTA–RELA protein levels were reduced in vivo in animals treated with dimethyl citraconate alone or in combination with JHU-083 (Fig. 5g,h).

We also evaluated the dual PI3K–mTOR inhibitor PI-103 in vitro and in vivo. PI-103 showed greater toxicity in ZFTA–RELA mNSCs than EV mNSCs (Extended Data Fig. 12a). Moreover, PI-103 as monotherapy or in combination with DON suppressed tumour growth in mice in which ZFTA–RELA mNSCs were xenografted into their flanks (Extended Data Fig. 12b–e). Similarly, PI-103 as a single agent and in combination with JHU-083 increased overall survival in animals with orthotopic ZFTA–RELA mNSC xenografts (Extended Data Fig. 12f–h). ZFTA–RELA PDX tumours treated with PI-103 or JHU-083 monotherapy showed growth suppression, with the greatest effect observed with combined PI-103 and JHU-083 treatment (Extended Data Fig. 12j,k). We did not observe a tumour-suppressive effect with JHU-083, PI-103 or dimethyl citraconate treatment in non-ZFTA–RELA or in posterior-fossa group A (MAF-811) ependymoma PDXs (Extended Data Fig. 12i).

Ependymomas can metastasize to the spinal cord and pose a major therapeutic challenge^51,52. Mice with orthotopic ZFTA–RELA mNSC xenografts developed spontaneous spinal cord metastases and eventually hindlimb paralysis (Fig. 5i,j, and Supplementary Video 1). Metastatic tumours also retained low PTEN expression (Fig. 4f and Extended Data Fig. 12m). Although JHU-083 or PI-103 monotherapy did not alter metastases (Extended Data Fig. 12l), dimethyl citraconate alone, or a combination of JHU-083 and PI-103, abrogated spinal metastasis (Fig. 5i,j and Supplementary video 1). Overall, these data demonstrate that targeting glutamine metabolism and the downstream itaconate metabolic pathway at multiple nodes singly or in combination (Fig. 5k) lead to therapeutic effects in ZFTA–RELA ependymomas, with reduction in levels of the pathogenic fusion protein both in vitro and in vivo.

Discussion

We investigated and exploited metabolic dependencies to identify itaconate as a metabolite produced by ZFTA–RELA⁺ ependymoma cells. We demonstrated that ACOD1 upregulation by ZFTA–RELA leads to the production of itaconate. Moreover, ZFTA–RELA reduces PTEN to activate PI3K–AKT signalling and stimulate glutamine metabolism. Glutamine serves as a carbon source for generating itaconate. Overall, our results revealed a critical actionable pathway with multiple druggable metabolic vulnerabilities (Fig. 5k). Our findings also have translational relevance, as inhibition of PI3K signalling and glutamine metabolism led to therapeutic effects in multiple ZFTA fusion models.

Itaconate is a macrophage metabolite known for its immunomodulatory properties^2,22. Studies have revealed that itaconate has various roles in macrophages, including anti-inflammatory and antiviral responses, modulating redox balance through NRF2 activation and modifying cysteine residues on proteins^{23,27,53,54,55}. Itaconate can also affect central carbon metabolism by inhibiting succinate dehydrogenase and modifying glycolytic enzymes^{27,28,56,57,58}. However, the expression of ACOD1 and the production of itaconate by tumour cells, and its potential role in cancers, have not been extensively explored. Our data demonstrated that ZFTA–RELA⁺ ependymomas use this pathway to promote the production of itaconate. We also observed that antagonism of itaconate and/or inhibition of glutamine metabolism resulted in reduced levels of pathogenic ZFTA–RELA. Mechanistic investigations indicated that this effect was epigenetic, as itaconate increased H3K4me3 levels and led to an increase in endogenous ZFTA and ZFTA–RELA levels. Our results are in line with recent studies that have also identified itaconate as an epigenetic modifier that inhibits the DNA demethylase TET2 in macrophages^27,29. Our results suggest that ZFTA–RELA⁺ ependymomas rely on itaconate-mediated epigenetic feedback to sustain the expression of ZFTA–RELA. This pro-oncogenic effect of itaconate to drive integrated metabolic and epigenetic pathways suggests that it could function as an oncometabolite like 2HG, succinate and fumarate in other cancers^{17,18,19,59,60}. Because of its immunoregulatory role, itaconate could also potentially modulate the immune microenvironment in ependymomas.

Collectively, our results suggest that ZFTA–RELA⁺ ependymomas rely on itaconate for sustained expression of ZFTA–RELA. Our data uncovered new leads for the therapeutic potential of targeting glutamine metabolism and itaconate biosynthesis in ZFTA fusion ependymomas. These findings hold promise for the development of effective treatment strategies against this aggressive cancer.

Methods

Cell lines and culture methods

All cell lines were cultured in a humidified incubator under normoxia at 37 °C with 5% CO₂. Cell lines were validated by STR analysis and were routinely screened and tested negative for mycoplasma.

Mouse cell lines

Immortalized mNSCs were provided by R.J.G.⁶². RCAS-TVA control (NS-1, NS-2 and NS-3) cell lines or mouse tumour cell lines with the ZFTA–RELA fusion (H-57, H-41 and H-59)¹² were provided by E.C.H. Mouse cells were cultured in neurobasal medium (Gibco, 21103049) with 0.2% heparin solution (StemCell Technologies, 07980), 20 ng μl^–1 EGF (Shenandoah, 100-26) and 20 ng μl⁻¹ FGF-basic 154aa (Shenandoah, 100-146).

Patient-derived cell lines

EP1NS (ZFTA–RELA⁺) cells were obtained from T. Milde⁶³. EPD-210 (PFA) cells were obtained from the Brain Tumour Resource Laboratory, Fred Hutchinson Cancer Research Center. Both cell lines were grown in neurobasal medium without vitamin A (Thermo Fisher, 12587-010) with 0.2% heparin solution, 20 ng μl^–1 EGF and 20 ng μl^–1 FGF-basic 154aa (Shenandoah, 100-146). ST-1 (ZFTA–MAML3⁺), ST-2 (ZFTA–RELA⁺) and ST-4 (ZFTA–RELA⁺, CDKN2A^−/−) cells were provided by K.A.M. The cell lines were cultured in neurobasal medium without vitamin A supplemented with Glutamax (Thermo Fisher Scientific, 35050-061), 200 µg ml^–1 human EGF and 4 µg ml^–1 human FGF and 0.2% heparin in T-75 flasks coated with poly-l-ornithine solution (Sigma-Aldrich, P4957) and laminin from Engelbreth–Holm–Swarm murine sarcoma basement membrane (Sigma-Aldrich, L2020). The CPITT-1 (ZFTA–RELA⁺) cell line was provided by S.A. and was cultured in neurobasal medium supplemented with 0.2% heparin solution, 20 ng μl^–1 EGF and 20 ng μl^–1 FGF-basic 154aa. EPN1425 cells were provided by S. Mack and were cultured in DMEM medium (Gibco, 11965092) supplemented with 10% fetal bovine serum (VWR, 89510-186) and 200 mM l-glutamine (Thermo Fisher Scientific, A2916801). MAF-1329 (ZFTA–RELA⁺) and MAF-811 (PFA) cell lines were provided by A.G. and N.K.F. and were cultured in Opti-MEM media (Gibco, 31985070) supplemented with 10% fetal bovine serum and 200 mM l-glutamine. All cell lines were cultured in media supplemented with penicillin–streptomycin (10,000 U ml–¹) (Thermo Fisher Scientific, 115140122) and plasmocin prophylactic (InvivoGen, antmpp).

Lentiviral-transduction-mediated gene silencing using shRNA

Isogenic mNSCs were generated by transfecting cells with ZFTA WT, RELA WT, ZFTA–RELA or EV backbone⁴ using lentiviral particles (SBI LentiStarter 3.0 kit, V060A). The following lentiviral transfection protocol was used to express ZFTA WT, RELA WT, EV or ZFTA–RELA (which contains 200 bp upstream of the initiating start codon) plasmids into immortalized mNSCs. Similarly, the same protocol was used to knockdown genes with shRNAs. First, 2 × 10⁶ HEK293T cells were plated on 100 mm dishes 36–48 h before transfection. A change of medium was performed the next day to a volume of 5 ml antibiotic-free medium. All lentiviruses were prepared using a Lentistarter 3.0 kit (System Biosciences, LV060A-1). In brief, 2 μg transfer DNA (shACOD1), 20 μg pPackH1 mix and 24 μl PureFection reagent were mixed in 500 μl serum-free DMEM medium and incubated at room temperature for 30 min. The mixture was added dropwise to HEK293T cells and gently swirled to distribute. The transfected cells were then incubated at 37 °C and 5% CO₂ to produce virus for 48 h. During this period (around 48 h before intended infection), 1 × 10⁶ ZFTA–RELA mNSCs or EP1NS cells were plated in T-75 flasks. Lentiviral particles were then collected and filtered using a 0.45 µm PVDF filter. Next, lentiviral particles were evenly distributed onto target cells. Lentiviral medium was removed after 24 h and replaced with suitable ZFTA–RELA mNSC or EP1NS cell culture medium. After 48 h, the transfected cells were treated with the appropriate antibiotic for selection. EP1NS cells were treated with 2 μg ml^–1 puromycin, whereas ZFTA–RELA mNSCs were selected using 15 μg ml^–1 blasticidin. Lentiviral plasmids used for shRNA-mediated knockdown are as follows: shAcod1 (mouse, access ID: NM_008392.1) (Gentarget) and shACOD1 (human) (Horizon Discovery). The following sequences were targeted to knockdown Acod1 in ZFTA–RELA mNSCs: shAcod1-1 (GAGAGCTTTGCTGGTATGATT) and shAcod1-2 (GAGGCATTGGCTATTGCTGTT).

The following sequences were targeted for knocking down ZFTA–RELA in EP1NS (ZFTA–RELA⁺) cells. The following ZFTA–RELA shRNAs were custom-designed and obtained from Gentarget: shZFTA Fus-1, GCTTGCCCGCCCAAGGGCCCA; shZFTA Fus-2, AGGGCCCAGAACTGTTCCCCC; and shZFTA Fus-3, CAGAACTGTTCCCCCTCATCT.

Human ACOD1 lentiviral cDNA (NM_001258406) and scrambled vector controls were purchased from Origene (SKU RC232825L4).

Snapshot metabolomics

To obtain a comprehensive overview of metabolites, a total of 2 million cells were cultivated in T-75 flasks and incubated for 24 h. Before metabolite collection, a complete change of medium was performed 2 h before the samples were collected. Our metabolomic analysis involved liquid chromatography with tandem mass spectrometry (LC–MS/MS) per a previously described method⁶⁴. We used an Agilent 1290 UHPLC and 6490 Triple Quadrupole (QqQ) mass spectrometer (LC–MS) for label-free targeted metabolomic analysis. Agilent MassHunter Optimizer and Workstation software LC–MS data acquisition for 6400 series QqQ B.08.00 were used for standard optimization and data acquisition.

For each multiple reaction monitoring (MRM) transition, a retention time with a 1-min left delta and right delta was assigned. Additional parameters included a mass extraction window of 0.05 Da on both sides of the extracted m/z, Agile2 integrator algorithm, peak filter of 100 counts, noise algorithm RMS, noise SD multiplier of 5 min, S/N of 3, accuracy max of 20% max %Dev and quadratic/cubic Savitzky–Golay smoothing algorithm with a smoothing function width of 14 and a Gaussian width of 5.

In reversed-phase liquid chromatography (RPLC), we used a Waters Acquity UPLC BEH TSS C18 column (2.1 × 100 mm, 1.7 μm) in the positive ionization mode. Mobile phase A consisted of 0.5 mM NH4F and 0.1% formic acid in water, whereas mobile phase B contained 0.1% formic acid in acetonitrile. The gradient program involved initially holding mobile phase B at 1% for 1.5 min, followed by an increase to 80% over 15 min, further increasing to 99% over 17 min and holding for 2 min before returning to the initial condition and holding for 10 min.

For hydrophilic interaction liquid chromatography (HILIC), we used a Waters Acquity UPLC BEH amide column (2.1 × 100 mm, 1.7 μm) in the negative ionization mode. Mobile phase A consisted of 20 mM ammonium acetate in water at pH 9.6, whereas mobile phase B comprised acetonitrile. The gradient program involved initially holding mobile phase B at 85% for 1 min, followed by a decrease to 65% over 12 min, further decreasing to 40% over 15 min and holding for 5 min before returning to the initial condition and holding for 10 min.

Both columns were maintained at a temperature of 40 °C, and each sample (3 μl) was injected into the LC–MS system with a flow rate of 0.2 ml min^–1. Calibration was performed using an Agilent ESI-low concentration tuning mix. Optimization was carried out on a 6490 QqQ in either the RPLC-positive or HILIC-negative mode for each of the 245 standard compounds (215 compounds for RPLC-positive and 217 compounds for HILIC-negative).

Immunoblotting

Cells were lysed in 1× RIPA buffer containing 1× protease (100× stock, Sigma Aldrich, P8340) and 1× phosphatase inhibitors (100× stock, Sigma Aldrich, P5726). Protein concentrations were quantified using colorimetric bicinchoninic acid assay (BCA) (Pierce BCA protein assay, 23227). Equal amounts of total protein from cell lysate or histone extracts were loaded on 4–15% Mini-Protean TGX precast gels (Bio-Rad, 3450027). Proteins were transferred to PVDF membranes using a Bio-Rad Trans-Blot Turbo transfer system (Bio-Rad, 1704150). Membranes were blocked with 5% bovine serum albumin (BSA) dissolved in TBST (TBS buffer containing 0.1% Tween-20) and incubated in primary antibody diluted in 5% BSA at 4 °C overnight. The following antibodies were used in the immunoblotting experiments: RELA (Cell Signaling Technology, 8242, 1:1,000); GAPDH (Cell Signaling Technology, 2118, 1:10,000); vinculin (Sigma Aldrich, V9264, 1:40,000); ACOD1-human (Abcam, ab222411, 1:1,000 and Novus Biologicals, NBP3-06244, 1:1,000); ACOD1-mouse (Cell Signaling Technology, 17805, 1:1,000); ZFTA (C11orf95) (VWR, 89379-010, AP11349B, 1:1,000); RFP (Abcam, Ab124754, 1:1,000); MAML3 (Invitrogen, PA5-13678, 1:1,000); SLC1A5 (Cell Signaling Technology, 5345, 1:1,000); GLS-human (Cell Signaling Technology, 49363, 1:1,000); GLS-mouse (Invitrogen, PA5-35365, 1:1,000); MYC (Abcam, 32072, 1:1,000); PTEN (Cell Signaling Technology, 9559, 1:1,000); pAKT (S473) (Cell Signaling Technology, 9271, 1:1,000); AKT (Cell Signaling Technology, 4056, 1:1,000); pS6RP (S235/236) (Cell Signaling Technology, 4858, 1:1,000); S6RP (Cell Signaling Technology, 2217, 1:1,000); pGSK3α/β (Cell Signaling Technology, 9331, 1:1,000); GSK3α/β (Cell Signaling Technology, 5676, 1:1,000); H3K4me3 (Cell Signaling Technology, 9751, 1:1,000); H3K9me3 (Cell Signaling Technology, 13969, 1:1,000); H3K27Ac (Cell Signaling Technology, 8173, 1:1,000); H3K27me3 (EMD Millipore, 07-449, 1:1,000); and total H3 (Cell Signaling Technology, 3638, 1:5,000). Next, membranes were washed with TBST three times. Membranes were incubated with species-matched secondary antibodies (1:5,000), goat-anti-mouse (Bio-Rad, 1706516) or goat anti-rabbit (Bio-Rad, 1706515) conjugated to horseradish peroxidase (HRP) for 2 h at room temperature and washed with TBST 3 times (about 5 min each) and TBS one time (around 10 min). Immunoreactivity was detected using Pierce ECL Western blotting substrate (Thermo Fisher Scientific, 32106). Tumour tissues from mice were processed for immunoblotting analysis after lysing red blood cells with RBC lysis buffer (Miltenyi Biotech, 130-094-183). Uncropped and unprocessed scans of all the immunoblots in the study are provided in Supplementary Fig. 1.

IHC analysis

IHC was conducted on ZFTA–RELA⁺ and control ependymoma patient-derived samples and on mouse tumours using established procedures. The patient tumours were classified by DNA methylation and fusions were determined by RNA-seq. For IHC of metastatic spinal cord, the vertebral column was dissected and decalcified before haematoxylin and eosin staining. Immunohistochemical staining was performed using either a Discovery XT processor from Ventana Medical Systems or a Leica Bond automated staining processor from Leica Biosystems. Each tissue section was blocked with a mixture of 10% normal goat serum and 2% BSA in PBS for 30 min. Subsequently, a rabbit monoclonal anti-ACOD1 antibody (1:200, Abcam, ab238580) or anti-PTEN (1:200, Abcam, ab170941), H3K27me3 (EMD Millipore, 07-449, 1:150), Ki-67 (Invitrogen, MA5-14520, 1:400) and SLC1A5 (Sigma Aldrich, HPA035240, 0.1 µg ml^–1) was applied to each section for 5 h. The tissue sections were then treated with biotinylated goat anti-rabbit IgG (PK6101, Vector Labs) at a dilution of 1:200 for 60 min. Chromogens were detected using a DAB detection kit along with streptavidin–HRP and blocker D to minimize background signal (Ventana Medical Systems) following the manufacturer’s instructions. Afterwards, each section was mounted, dried and visualized using an Aperio Vista scanning system (AperioScanscope Scanner). The accompanying Aperio ImageScope software program was used to view each slide at ×40 magnification. An experimenter, who was unaware of the study design, captured JPEG images from three randomly selected areas of each control-stained and ZFTA–RELA-stained section. To quantify the images, an automated analysis program previously published by our laboratory⁶⁵ was used. This MatLab-based program uses techniques such as k-means clustering, colour segmentation based on RGB colour differentiation and Otsu’s threshold-based background–foreground separation. It generates a quantitative score by multiplying the extracted pixels with the average intensity for each JPEG image.

Cumate-inducible ZFTA–RELA fusion system

The SPARQ 2 Cumate Switch system (System Biosciences, QM822B-1) was used to design the inducible ZFTA–RELA fusion model. The EV and ZFTA–RELA fusion plasmids were transduced in both mNSCs and in HEK293T cells via lentiviral transfection as described above. Cumate (10,000×) purchased from System Biosciences (PBQM100A-1) was used at 1× concentration for 48 h to induce expression of either EV or the ZFTA–RELA fusion.

KDM5 activity measurement

Assessment of KDM5 enzyme activity was performed using a JARID1A Homogenous Assay kit (BPS Bioscience, 50510-2) following the manufacturer’s instructions in three independent experiments. For enzymatic inhibition, KDM5 enzyme (25ng µl^–1) was incubated in quadruplicate with different concentrations (100, 250, 500, 1,000, 2,000, 5,000 or 10,000 µM) of itaconic acid (Sigma-Aldrich, I29204), D-2G (Cayman, 16366) or L-2HG (Cayman, 16367). For enzymatic activation, KDM5 enzyme at different concentrations (1.5, 3.0, 6.5, 12.5 or 25.0 ng µl^–1) was incubated in quadruplicate with 2,000 µM αKG (Sigma-Aldrich, 349631), 2,000 µM itaconic acid (Sigma-Aldrich, I29204), 2,000 µM D-2HG (Cayman, 16366) or 1,000 µM L-2HG (Cayman, 16367). Compounds were diluted in either PBS or DMSO, which were also used as negative controls. Assays were read in an Envision plate reader (PerkinElmer) using the AlphaLisa 615/Alphascreen 444 method.

ChIP–seq

ZFTA–RELA mNSCs and EPN1425 cells were seeded at a density of 2 × 10⁶ cells in a 100 mm dish and treated with 10 mM dimethyl citraconate (TCI, C0364) or DMSO for 24–48 h. Following this, the cells were detached and dissociated using Accutase (Corning, 25-058-CI). ChIP experiments were performed per the manufacturer’s protocol using an Ideal ChIP–seq kit for transcription factors (Diagenode). For histone ChIPs, 1 million cells were used per ChIP reaction with 2 µg antibody. In brief, cells were crosslinked for 10 min in a 1% formaldehyde solution, followed by quenching with 1/10 the volume of 1.25 M glycine for 5 min at room temperature. Following this, the cells were lysed and sonicated (Bioruptor Pico, Diagenode) to a desired chromatin size of about 200 bp using the Easy mode cycle. Sheared chromatin was then incubated with the following antibodies overnight at 4 °C: H3K4me3 (CST, 9751S) and H3K27me3 (Millipore, 07-449). ChIP DNA was de-crosslinked and purified the next day using a Diagenode iPure kit V3 following the manufacturer's protocol. Purified DNA was prepared for sequencing using the manufacturer’s instructions (Illumina). About 1–10 ng ChIP DNA was converted to blunt-ended fragments using T4 DNA polymerase, Escherichia coli DNA polymerase I large fragment (Klenow polymerase) and T4 polynucleotide kinase (New England Biolabs (NEB)). Klenow fragment (3′ to 5′ exo minus; NEB) was used to add a single adenine base to fragment ends, followed by ligation of Illumina adaptors (Quick ligase, NEB). PCR enriched the adaptor-ligated DNA fragments using Illumina Barcode primers and Phusion DNA polymerase (NEB). PCR products were size-selected using 3% NuSieve agarose gels (Lonza), followed by gel extraction using QIAEX II reagents (Qiagen). Quantified libraries were quality-checked using a Bioanalyzer 2100 (Agilent) and sequenced on an Illumina HiSeq 2500 Sequencer (125-nucleotide read length).

Reads were processed as previously described⁶⁶. To summarize, reads were first processed using Trimmomatic (v.0.39) (settings TruSeq3-PE-2.fa:2:30:10, minlen 50) followed by alignment with bwa (bwa mem, options -5SP -T0, v.0.7.17-r1198-dirty) to the mm10 (GRCm38) genome reference or the hg38 (GRCh38) reference^67,68. After alignment, the reads were filtered using MarkDuplicates from Picard and then by a quality score of >20 using SAMtools⁶⁹. MACS2 was used to call peaks, filtered using bedtools and converted to bigwigs with UCSC wigtoBigwig^70,71. Cistrome overlap analysis was performed in R (v.3.6.0) using ChipSeekAnno (v.3.0.0) and ChipSeeker (v.1.29.1)^72,73. Enrichment heatmaps were generated using DeepTools⁷⁴.

ATAC-seq

ATAC-seq was performed as previously described⁷⁵. In brief, 50,000 mNSC and EPN1425 cells were washed in cold PBS and resuspended in lysis buffer (10 mM Tris HCL, pH 7.4, 10 mM NaCl, 3 mM MgCl₂, digitonin, NP-40, Tween-20 and protease and phosphatase inhibitor). This single-cell suspension was incubated on ice for 5 min with gentle mixing by pipetting every 2 min, followed by quenching in resuspension buffer. The lysate was centrifuged at 1,300g for 5 min at 4 °C. Nuclei were resuspended in a 50 µl reaction containing 25 µl of 2× TD buffer and 1 µl Tn5 enzyme for 30 min at 37 °C using a Nextera DNA Library Preparation kit. Samples were immediately purified using a Qiagen minElute column and PCR-amplified with NEBNext High-Fidelity 2× PCR master mix (NEB, M0541L). Optimal PCR cycles were determined by qPCR, and the amplified libraries were further purified using a Qiagen minElute column and SPRI beads (Beckman Coulter). Libraries were quantified and quality checked using a Bioanalyzer 2100 (Agilent).

The ATAC-seq libraries were sequenced on an Illumina HiSeq 2500 platform, using a 2 × 50-nucleotide paired-end read length with a sequence depth of 40–45 million. Sequencing of ATAC-seq libraries generated fastq files, which were initially processed using Trimmomatic (v.0.39) for trimming⁶⁷. These files were then aligned to the mm10 (GRCm38) mouse genome reference or the hg38 (GRCh38) reference using bwa mem (v.0.7.17-r1198-dirty)⁶⁷, and the alignments were converted to binary format with SAMtools (v.1.9)⁶⁸. Next, we eliminated mitochondrial and duplicated reads using SAMtools and Picard MarkDuplicates (v.2.26.0-1-gbaf4d27-SNAPSHOT)⁶⁹. Peaks in the ATAC-seq data were identified using MACS2 (v.2.1.1.20160309)⁷⁰. Finally, the conversion of data to bigwig format was accomplished using the UCSC tool wigtoBigwig⁷¹. The enrichment heatmaps were generated using DeepTools⁷⁴.

RNA-seq

RNA was isolated from 1 × 10⁶ cells seeded in 100 mm dishes using Trizol (Invitrogen 15596-026) and treated with DNase (Sigma, 9003-98-9). RNA-seq libraries were prepared according to the Illumina TruSeq protocol and were sequenced on a HiSeq 2000. RNA-seq data generated were aligned to the mouse reference genome using bowtie and analysed using the RSEM software package with default parameters. DEseq2 was used to identify differentially upregulated and downregulated genes⁷⁶. Differentially enriched pathways were identified using Enrichr (https://maayanlab.cloud/Enrichr/)⁷⁷.

Generation of the ZFTA–RELA KDM5δ mutant

The following sequences with RFP tags were transfected into mNSCs to derive ZFTA–RELA and ZFTA–RELA KDM5δ cells (shown in Fig. 3 and Extended Data Fig. 5). Putative KDM5 recognition sites in the ZFTA portion of the fusion were identified based on previous studies^36,37,39 and mutated as indicated below.

The KDM5A site in ZFTA exon 1, normally CCGCCC, was changed to CCACCA. The KDM5B site in ZFTA exon 2, normally GCACAC, was changed to GCAAAC. In detail, the parts of the sequence involved are highlighted in bold:

ZFTA–RELA unaltered: GATCCCATGGAGCCCGGCGGGGACCACCGTAAGAGCCGGAGCAGCGGCGGCAGGGGCGGCCCCGGGCCAGCAGTGGCCTCGGCACGGGGCCGACGGCTGCCGCCCGCCGGATCGAGCGGCAGCGCGGAGCCAGAGGAAGACGAAGGCGGGCAAGATCTTCAGCTGGAAGGGGGTGCCTTGGGGTCCTGGGGGAGTGCCCCCCTGCCCTCCTCCAGGGCCAGGGGACCAGCATCTTCAGGCAGGAAATATTCAGACCACTGTGAGGCCCGGGCCTCGAGGCCTGGAAAGAGCCGCATCCCTGGCCGTGACCACCGGCGCTACTACCACGACCACTGGCGGCTGGAGTACCTGATGGACTTCAACCCTGCCCGGCACGGCATGGTGTGCATGGTGTGCGGCAGCTCCCTGGCCACCCTCAAGCTCAGCACCATCAAGCGCCACATCCGCCAAAAGCACCCCTACTCCTTGCATTGGAGTCCCCGGGAGAAGGAAGTCATCAGCAACAGCTGGGATGCACACATGGGGCTGGGGGCCTGCGGAGAGGCCGAGGGCCTGGGGGTCCAGGGGGCTGAGGAGGAGGAGGAGGAGGAAGAAGAGGAGGAGGAGGAGGGGGCCGGTGTCCCAGCTTGCCCGCCCAAGGGCCCAG.

ZFTA–RELA KDM5δ: GATCCCATGGAGCCCGGCGGGGACCACCGTAAGAGCCGGAGCAGCGGCGGCAGGGGCGGCCCCGGGCCAGCAGTGGCCTCGGCACGGGGCCGACGGCTGCCACCAGCCGGATCGAGCGGCAGCGCGGAGCCAGAGGAAGACGAAGGCGGGCAAGATCTTCAGCTGGAAGGGGGTGCCTTGGGGTCCTGGGGGAGTGCCCCCCTGCCCTCCTCCAGGGCCAGGGGACCAGCATCTTCAGGCAGGAAATATTCAGACCACTGTGAGGCCCGGGCCTCGAGGCCTGGAAAGAGCCGCATCCCTGGCCGTGACCACCGGCGCTACTACCACGACCACTGGCGGCTGGAGTACCTGATGGACTTCAACCCTGCCCGGCACGGCATGGTGTGCATGGTGTGCGGCAGCTCCCTGGCCACCCTCAAGCTCAGCACCATCAAGCGCCACATCCGCCAAAAGCACCCCTACTCCTTGCATTGGAGTCCCCGGGAGAAGGAAGTCATCAGCAACAGCTGGGATGCAAACATGGGGCTGGGGGCCTGCGGAGAGGCCGAGGGCCTGGGGGTCCAGGGGGCTGAGGAGGAGGAGGAGGAGGAAGAAGAGGAGGAGGAGGAGGGGGCCGGTGTCCCAGCTTGCCCGCCCAAGGGCCCAG.

ChIP–PCR

ZFTA–RELA mNSCs and EPN1425 cells were seeded at a density of 5 × 10⁶ cells in a 100 mm dish and treated with 10 mM dimethyl citraconate (TCI, C0364) or DMSO for 48 h. Chromatin fragmentation was achieved by sonication (QSonica, 800R) and by micrococcal nuclease (Cell Signaling Technologies, 10011) digestion to a final size of 150–300 bp. ChIP was performed using a Simple Chip Enzymatic Chromatin IP kit (Cell Signaling Technologies, 9002) and a Simple Chip Plus Sonication Chromatin IP kit (Cell Signaling Technologies, 56383) as per the manufacturer’s protocols.

Two independent set of primers were designed and used for the ChIP PCR and RT–qPCR experiments shown in Fig. 3h and Extended Data Fig. 5e,f,j. They were custom designed and obtained from Integrated DNA Technologies (IDT). The following ZFTA-based primers were used: ZFTA exon 5 set 1 FWD: 5′-AGAGGAGGACGAAGAGGAC-3′; ZFTA exon 5 set 1 REV: 5′-CCGTCGTAGTCCATCAGGTA-3′; ZFTA exon 5 set 2 FWD: 5′-TGAGGAGCGCCAGACTAT-3′; ZFTA exon 5 set 2 REV: 5′-CACACAGCGCCAGACTT-3′; ZFTA exons 1–2 set 1 FWD: 5′-GAGCCAGAGGAAGACGAAGG-3′; and ZFTA exons 1–2 set 1 REV: 5′-GAAGATGCTGGTCCCCTG-3′. The following RELA-based primers were used: RELA exon 4 set 1 FWD: 5′-ACAGGACCAGGGACAGT-3′; RELA exon 4 set 1 REV: 5′-CTCAGCCTCATAGAAGCCATC-3′; RELA exon 4 set 2 FWD: 5′-CGCATCTCCCTGGTCAC-3′; and RELA exon 4 set 2 REV: 5′-CAT CCC GGC AGT CCT TT-3′. The following mouse primers were used: H3f3a FWD: 5′-ACAAAAGCCGCTCGCAAGAGTG-3′; and H3f3a REV: 5′-TTCTCGCACCAGACGCTGAAAG-3′.

RT–qPCR

RNA was isolated from 1 × 10⁶ cells seeded in 100 mm dishes using Trizol (Invitrogen 15596-026) and treated with DNase (Sigma, 9003-98-9) per the manufacturer’s instructions. qPCR was performed with n = 5 samples of 20 ng mRNA each using a TaqMan RNA-to-C_T kit (Applied Biosystems, 4392938). TaqMan probes (Thermo Fisher) targeting mouse Pten (Mm00477208), human PTEN (Hs02621230), human GAPDH (Hs99999905, Thermo Fisher Scientific, 4326317E) and mouse Gapdh (Mm99999915_g1, Thermo Fisher Scientific, 4352339E) were used. The following primers from IDT were used to detect ZFTA–RELA mRNA in patient-derived cell lines (shown in Fig. 1m): fusion set 1 FWD: 5′-GAGGAGGAGGAGGAAGAAGAG-3′; fusion set 1 REV: 5′-GCTGCTCAATGATCTCCACATA-3′; fusion set 2 FWD: 5′-GAAGAAGAGGAGGAGGAGGAG-3′; and fusion set 2 REV: 5′-CTTGGGCTGCTCAATGATCT-3′.

RT–qPCR was conducted using an ABI7900HT Sequence Detection system, and relative expression was calculated using the ${2}^{-\Delta {\Delta }_{\mathrm{CT}}}$ method, with Gapdh (mouse) or TBP (human) as a housekeeping control for normalization.

Glutamine uptake and intracellular glutamine and glutamate measurement

Glutamine uptake and intracellular glutamine levels were estimated using a Glutamine/Glutamate-glo assay kit (Promega, J8022). In brief, 25,000 cells per ml medium per well were seeded in triplicate in a 24-well plate for each time point as indicated. Both cell culture media and cells were collected at 0, 12, 24, 36, 48, 60 and 72 h. Following extraction, intracellular and extracellular glutamine and glutamate were measured by bioluminescence detection for all the time points. The concentrations of glutamine and glutamate were calculated using the standards provided by the manufacturer. The values obtained were used to estimate percentage glutamine uptake and intracellular glutamate to glutamine ratios.

Isotope tracing of radiolabelled glucose and glutamine

For ¹³C tracing, 2 million cells were plated and cultured using medium supplemented with either ¹³C-U-labelled glucose (25 mM) or glutamine (2 mM, both from Cambridge Isotope Laboratories, CLM-99-1396-1 and CLM-1822-H, respectively) for 16 h. Cells were collected by centrifugation and the medium was aspirated. The cell pellet was resuspended in 300 μl ice-cold methanol followed by thorough pipetting. Next, 300 μl water containing 1 μg norvaline as internal standard and 600 μl chloroform was added to the microfuge tube. The extracts were vortexed thoroughly for 30 min at 4 °C, followed by centrifugation. The polar layer was dried in a SpeedVac for 3–4 h without heat and stored at −80 °C. For derivatization, 30 μl methoxyamine hydrochloride (Fisher Scientific, PI45950) was added to the dried samples and incubated at 45 °C for 30 min with constant mixing. Next 30 μl MBTSTFA and 1% TBDMCS (Sigma Aldrich, M-108-1243 5×1ML) was added, and samples were incubated at 45 °C for 1 h. Derivatized samples were transferred to GC vials with glass inserts. Samples were analysed using an Agilent 7890 GC equipped with a 30-m HP-5MSUI capillary 1246 column connected to an Agilent 5977B MS in scan mode. In total, a 1–2 μl sample was injected at 270 °C with helium as the carrier gas at 1 ml min^–1 flow. The temperature gradient was maintained at 100 °C for 1 min, increased at 3.5 °C min^–1 to 255 °C, increased to 320 °C at 15 °C min^–1 and held for 3 min (method total time of 52.6 min). MS detector was operated in scan mode (70–600 m/z). Relative metabolite abundances were calculated by normalizing the integrated peak area of ions to the internal standard signal (norvaline) and protein content. For ¹³C-labelled samples, isotopic correction of raw GC–MS peaks for all reported metabolites was performed using the IsoCorrectoR package (v.1.5.1) available as part of the Bioconductor library (BioC 3.8) and implemented in R (CRAN 3.6.1).

Cell proliferation and viability assays

Proliferation and viability of cells after drug treatment was assessed by trypan blue exclusion assays. Cells were seeded at a density of 2.5 × 10⁴ cells per well in a 24-well plate and were left overnight. They were then treated with vehicle or drug (JHU-083, V-9302, CB-839 or PI-103) at varying concentrations and time points for the indicated time periods, following which they were counted using a Countess III cell counter (Thermo Fisher Scientific). Cell number (y axis) for each cell line was calculated as a percentage of living cells normalized to the untreated or vehicle-treated controls (average number of cells alive in the untreated or vehicle-treated wells).

Chemical compounds

The drugs and chemical compounds used in the study are listed as follows: dimethyl citraconate (Fisher Scientific, C0364-5G), mesaconic acid (Millipore-Sigma, 131040-10G), itaconic acid (Sigma Aldrich, I29204), dimethyl 2-ketoglutarate (Cayman Chemical, 28394), octyl-l-2-hydroxyglutarate (Cayman Chemical, 16367), LPS (Cayman Chemical, 19661), IRG1-IN-1 (MedChem Express, HY-148335), GSK-343 (Cayman Chemical, 14094), CPI-455 (Cayman Chemical, 22127), DON (Cayman Chemical, 17580), JHU-083 (MedChem Express, HY-122218), V-9302 (Selleckchem, S8818), CB-839 (MedChem Express, HY-12248), PI-103 (Selleckchem, S1038), MK-2206 (Cayman Chemical, 11593) and DMSO (Sigma Aldrich, D2650).

Histone extraction

Cells were plated at densities below confluence. Following the specified treatment, cells were rinsed with PBS. The cell pellet resulting from centrifugation was resuspended in 1 ml hypotonic lysis buffer (containing 10 mM Tris HCl pH 8.0, 1 mM KCl, 1.5 mM MgCl₂ and protease and phosphatase inhibitors) and incubated for 30 min on a rotator at 4 °C. The pellet was then collected by centrifugation at 10,000g, 4 °C for 10 min. It was subsequently resuspended in 400 μl of 0.4 N H₂SO₄ and left to incubate overnight on a rotator at 4 °C. After centrifugation, the resulting supernatant was transferred to a new tube, and 132 μl trichloroacetic acid was gradually added. The mixture was then incubated on ice for 30 min. The histone pellet was obtained by centrifugation at 16,000g, 4 °C for 10 min and washed with cold acetone. After another round of centrifugation at 16,000g, 4 °C for 5 min, acetone was removed. The histone pellet was further washed with acetone and subsequently allowed to air dry with the caps open at room temperature for 20 min to eliminate any remaining acetone. Finally, the dried histone pellet was resuspended in an appropriate volume of ddH₂O, supplemented with protease inhibitors and kept on ice.

Animal handling and housing conditions

Animal experiments were performed after approval from the University of Michigan Committee on the Use and Care of Animals (PRO00010599) and were conducted as per National Institutes of Health (NIH) guidelines for animal welfare. All animals used in these studies were housed in pathogen-free conditions as per IACUC guidelines with continuous access to both food and water in addition to veterinary care. NOD-SCID-IL2R γ-chain deficient (NSG) mice (NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ, 005557) aged 4–6 weeks old were used for all experiments involving subcutaneous or orthotopic injections of ZFTA–RELA mNSCs and for the subcutaneous MAF-1329 and MAF-811 PDX models. IUE models were CD-1 (Charles River, Crl:CD1(ICR), 022) or C57BL/6 animals (The Jackson Laboratory, 003771). For all drug trials, the animals were randomized and sorted into treatment groups in an unbiased manner, ensuring that all cohorts contained equal proportions of male and female animals.

Nes ^cre Acod1 ^fl/+ models

Acod1^fl/fl animals were provided by M. Diamond²⁵. B6.Cg.Tg.(Nes Cre)1KlnJ animals were crossed with Acod1^fl/fl animals. The pups were weaned after 21 days and genotyped to establish Nes^creAcod1^fl/+ mice.

Subcutaneous ZFTA–RELA xenograft models

In this study, two subcutaneous xenograft models (mouse and human patient-derived) ZFTA–RELA fusion tumours were established in NSG mice. Xenograft models were generated by injecting 1 × 10⁶ ZFTA–RELA mNSCs on either flank of the animal and tumour volumes were measured once palpable tumours (around 200 mm³) appeared (about 1 week after injection). MAF-1329 (ZFTA–RELA⁺) and MAF-811 PDXs were provided by A.G. and N.K.F. PDX cells were injected in NSG mice and were allowed to grow. Once the tumours reached the end-point volumes of 1,000–1,500 mm³, they were excised and homogenized into single-cell suspensions using a gentleMACS Octo dissociator (Miltenyi Biotech, 130-096-427), following which they were counted and serially passaged in NSG animals to generate a suitably sized cohort for the trial. For subcutaneous tumours, growth was measured using Vernier calipers, and tumour volumes were calculated using the formula L × W × W/2, where L is the longer dimension and W is the shorter dimension of the tumour. At the end point, the animals were humanely euthanized and perfused before collecting the tumour tissues. Tumour volumes were not allowed to exceed these limits (as per IACUC guidelines) in any of the experiments.

Orthotopic ZFTA–RELA tumour models

Orthotopic models of ZFTA–RELA tumours were established in NSG animals by injecting 2 × 10⁵ ZFTA–RELA bioluminescent mNSCs in the cortex, 3 mm posterior and 2 mm lateral right from the bregma using a stereotaxic apparatus. Before implantation, mice were anaesthetized by intraperitoneal injection of ketamine (90 mg kg^–1) and dexmedetomidine (0.6 mg kg^–1). Carprofen (5.5 mg kg^–1) was used as an analgesic to alleviate pain after surgery. A total volume of 4 μl of cells resuspended in sterile PBS was injected using a Hamilton syringe through a burr hole drilled at the location described. The surgery site was treated with iodine and sealed with a wound clip to avoid infection. The animals were then revived using atipamezole (1.25 mg kg^–1).

IUE ZFTA–RELA tumour models

IUE ZFTA–RELA immunocompetent ependymoma animal models were established using previous protocols to generate piggy-bac (PB) ZFTA–RELA plasmids containing the luciferase gene along with the plasmid transposase (PBase)^6,10,11. Pregnant CD1 or B6 mice (Charles River) mice at embryonic day 15 were used. IUE was performed by injecting a concentrated mixture of DNA (1 μg μl^–1 ZFTA–RELA plasmid) along with 0.05% Fast Green (Sigma) into the lateral ventricle of the embryos using a pulled glass capillary pipette. Electroporation was performed using 5 square pulses (45 V, 50 ms pulses with 950 ms intervals) and applied using a 3 mm tweezer electrode, with the positive electrode directed towards the cortex (BTX/Harvard Bioscience). After the procedure, the embryos were carefully placed back into the abdominal cavity, the incision was sutured, and the female mouse was monitored until she fully recovered.

Pharmacological and radiation treatment

In both the implantation and IUE models, tumour engraftment and establishment was verified by bioluminescence measurements. The flux density was calculated, and treatment regimens were initiated only after tumours were established (predefined as >10⁵ photons s^–1). Two independent baseline measurements were recorded for every animal in the trial to ensure tumour growth and to avoid possible technical artefacts. Subsequent bioluminescent readings were normalized to the baseline measurements to calculate the fold change in the signal to assess tumour progression. The glutamine antagonist JHU-083 was dissolved in 10% DMSO and 90% corn oil to be dosed orally at 20 mg kg^–1 twice a week. The dual PI3K–mTOR inhibitor was dissolved in DMSO and was i.p. injected every other day during weekdays with scheduled breaks in treatment on the weekend. Dimethyl citraconate was prepared in sterile saline solution was i.v. injected on alternate days for the entire duration of the treatment paradigm. For experiments involving radiotherapy, whole-brain irradiation was performed on mice twice per week (single fraction of 2 Gy per day) for two consecutive weeks.

Bioluminescence measurement and analysis

Successful tumour formation was verified by checking for bioluminescence after injecting the pups with luciferin (15 mg ml^–1, GoldBio, LUCK 115144-35-9) using an IVIS spectrum (Perkin Elmer) instrument. Once anaesthetized, a sequence of bioluminescence images was recorded to capture the peak flux intensity value for each animal. This value was then used to normalize to baseline for fold change calculation. In the implantation model in which ZFTA–RELA tumours metastasized to the spinal cord, the fold change in tumour bioluminescence signal was calculated by defining a region of interest around the spinal cord distal and distinct to the head. The end points of the animal trial were defined based on IACUC guidelines and animal welfare was prioritized. Swelling of the head, lack of mobility, loss of weight or appetite, among other parameters, were all considered as key criteria for euthanasia of tumour-bearing animals. Tumour volumes were not allowed to exceed these limits (as per IACUC guidelines) in any of the experiments.

Statistics and reproducibility

The Department of Bioinformatics provided consultation for the statistical analyses conducted. Each figure and its corresponding legend indicate the sample size (n), the specific statistical test performed and the corresponding P values. No data were excluded in analyses of samples or animals. Owing to the rarity of ZFTA–RELA ependymomas, the sample sizes were determined based on the available tumour samples. The figure legends provide the exact n value for each experiment. The data are presented as the mean ± s.d., and the box and whisker plots show the maximum, mean and minimum values. Graphs were generated and statistical analyses were performed using Prism software (v.9.5.1, GraphPad). The data analysis methods included unpaired, two-sided, two-tailed Student’s t-test, one-way or two-way ANOVA followed by post hoc Bonferroni multiple comparison analysis along with Dunnett’s correction, as indicated in each figure. Data in Fig. 1h was analysed using nonparametric Spearman’s correlation with 95% confidence intervals and correlation coefficient (R²) and P values are indicated.

Overall survival data, including data from animal models, were visualized using Kaplan–Meier curves and the association of various factors with overall survival was assessed using the log-rank test. Data were considered significant if P values were below 0.05 (95% CI).

Data shown in the following figures are representative images of independent experiments or samples as indicated. Fig. 1e (n = 3 independent experiments), Fig. 2a (n = 5 independent experiments), Fig. 2g (n = 2 independent experiments), Fig. 2h (n = 2 independent experiments), Fig. 2j (n = 2 independent experiments), Fig. 3b (n = 2 independent experiments), Fig. 3c (n = 2 independent experiments), Fig. 3h (n = 5 independent experiments), Fig. 3l (n = 2 independent experiments), Fig. 5c (n = 2 independent experiments), Fig. 5j (n = 3 independent samples were used for immunohistochemical staining), Extended Data Fig. 1e (n = 2 independent experiments), Extended Data Fig. 1f (ZFTA–RELA (n = 18 independent samples), non-ZFTA fusion (n = 5 independent samples), posterior fossa type A (n = 7 independent samples), posterior fossa type B (n = 2 independent samples), spinal (n = 3 independent samples) and myxopapillary (n = 4 independent samples) of ependymoma tumours from patients). Extended Data Fig. 1h (n = 3 independent experiments), Extended Data Fig. 1i (n = 3 independent experiments), Extended Data Fig. 1l (n = 2 independent experiments), Extended Data Fig. 3a (n = 5 independent experiments), Extended Data Fig. 3b (n = 2 independent experiments), Extended Data Fig. 3c (n = 2 independent experiments), Extended Data Fig. 3d (n = 2 independent experiments), Extended Data Fig. 3e (n = 2 independent experiments), Extended Data Fig. 3f (n = 2 independent experiments), Extended Data Fig. 3h (n = 3 independent experiments), Extended Data Fig. 3i (n = 2 independent experiments), Extended Data Fig. 3j (n = 2 independent experiments), Extended Data Fig. 3k (n = 2 independent experiments), Extended Data Fig. 3l (n = 2 independent experiments), Extended Data Fig. 3m (n = 2 independent experiments), Extended Data Fig. 3n (n = 3 independent experiments), Extended Data Fig. 3o (n = 2 independent experiments), Extended Data Fig. 5c (n = 3 independent experiments), Extended Data Fig. 5d (n = 3 independent experiments), Extended Data Fig. 5e (n = 2 independent experiments), Extended Data Fig. 5f (n = 2 independent experiments), Extended Data Fig. 5n (n = 2 independent experiments), Extended Data Fig. 7b (n = 2 independent experiments), Extended Data Fig. 7c (n = 2 independent experiments), Extended Data Fig. 7d (n = 2 independent experiments), Extended Data Fig. 8c (n = 2 independent experiments), Extended Data Fig. 8d (n = 2 independent experiments), Extended Data Fig. 8h (n = 2 independent experiments), Extended Data Fig. 8i (n = 3 independent samples each for mouse ZFTA–RELA fusion tumours and cortex tissues, n = 18 independent samples for ZFTA–RELA⁺ tumours from patients and n = 5 independent samples for non-ZFTA (YAP1) fusion tumours from patients), Extended Data Fig. 8m (n = 2 independent experiments), Extended Data Fig. 9c (n = 17 independent samples for ZFTA–RELA⁺ tumours from patients and n = 5 independent samples for non-ZFTA (YAP1) fusion tumours from patients), Extended Data Fig. 9i (n = 2 independent experiments), Extended Data Fig. 9j (n = 2 independent experiments), Extended Data Fig. 9k (n = 2 independent experiments), Extended Data Fig. 10b (n = 2 independent experiments), Extended Data Fig. 11d (n = 3 independent experiments), Extended Data Fig. 12m (n = 3 independent samples).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.