Over 95% of ependymomas (EPN) that arise in the cortex are characterized by gene fusions commonly involving the zinc finger translocation associated (ZFTA) protein. These fusion oncoproteins (FOs) join ZFTA with either a transcription factor or co-activator, leading to a chimeric protein capable of novel functions. To understand the core mechanism of ZFTA FOs we leveraged super resolution and lattice light sheet microscopy. We show that ZFTA FOs undergo liquid-like phase separation to form dynamic nuclear condensates on chromatin. Nuclear condensate formation is critical for activation of ZFTA FO oncogenic targets and the initiation of brain tumors in mice. Furthermore, we show that DNA binding is necessary for proper condensate distribution in the nucleus and provide the first NMR data of a key DNA binding domain of ZFTA as a resource for drug discovery efforts. Using machine-learning models trained on protein sequences of fusion oncoproteins in pediatric and adult cancers, we identified critical amino acid residues in ZFTA FOs necessary for neoplastic transformation. Critically, insights gained from AI-driven mutagenesis studies enabled the creation of synthetic ZFTA FOs, never seen in biology, and capable of driving oncogenic gene activation and brain tumor development. These findings reveal the molecular building blocks important for the assembly of ZFTA fusion oncoproteins, informing our general understanding of brain tumor initiation and providing opportunities for therapeutic intervention.

Supplementary Figure S8 shows that KAT6A/7 and Menin inhibition have combinatorial effects in vivo.

Supplementary Methods, Supplementary Results

Mutation of F and FG residues in the FG-rich IDR of NHA9 alters LLPS behavior in vitro. A–C, Confocal fluorescence micrographs of Alexa 488–labeled NHA9–8FA (A), NHA9–21FGAA (B), and NHA9Midi (C) condensates in vitro with increasing protein concentration. Micrographs are presented as maximum intensity projections of 13 Z-stack images acquired over 6 μm with 0.5 μm resolution. D, Confocal micrograph of Alexa 488–labeled NHA9Midi–21FGAA at a concentration of 20 μmol/L. Saturation concentration (Csat) is less than 10 nmol/L (A), between 160 nmol/L and 315 nmol/L (B), and between 40 nmol/L and 80 nmol/L (C).

Supplementary Figure S1 shows that some NUP98 fusions drive cell transformation and leukemogenesis.

Additional leukemia-associated NUP98 FOs form nuclear puncta and transform lin− HSPCs. A, Schematic of NHA9 and three additional NUP98 FOs (NUP98–PRRX1, NUP98–KDM5A, and NUP98–LNP1). Numbers indicate the amino acid residue. Numbers above the schematic reflect NUP98 residues, whereas numbers beneath reflect the fusion partner's residues. B, Representative images of live lin− HSPCs expressing G-NUP98–PRRX1, G-NUP98–KDM5A, and G-NUP98–LNP1. EGFP empty vector and G-NHA9 are included for comparison. C, Average number of colonies per 2,000 cells for lin− HSPCs expressing negative control empty vector and NUP98 FOs. Data shown are mean ± SD from triplicate technical replicates of a representative experiment. D, Representative images of a fixed, nontransduced human CD34+ (hCD34+) cell (top) and a NUP98–KDM5A PDX cell (bottom) stained with an antibody against NUP98. NUP98 is magenta, and DNA is blue. The heat map is a normalized representation of NUP98 fluorescence intensity across the PDX cell. E, Conceptual scheme illustrating how LLPS by NHA9 mediates the formation of aberrant transcriptional condensates in hematopoietic cells. NHA9 (top, left cell image) undergoes LLPS to form many small, chromatin-associated puncta that drive aberrant expression of Hox and other genes and transform hematopoietic cells. LLPS is driven by both homotypic and heterotypic interactions. Mutation of FG motifs (NHA9–21FGAA, bottom left) weakens both homotypic and heterotypic protein–protein interactions, yielding less numerous, larger, and less dense puncta that do not activate Hox gene expression or transform HSPCs. Mutation of residues in the HOXA9 homeodomain (HD; NHA9–ΔDNA, bottom right) weakens heterotypic interactions with DNA, yielding less numerous, larger, and more dense puncta that also do not induce the leukemogenic phenotype in HSPCs.

Supplementary Figure S2 shows that Nup98::Kdm5a;Vav-Cre mice spontaneously develop myeloid disease.

Supplementary Figure S3 shows that NUP98 fusion oncoproteins interact with MYST family histone acetyltransferase complex members.

NUP98 fusion oncoproteins (FO) are a hallmark of childhood acute myeloid leukemia. NUP98 FOs drive leukemogenesis through phase-separated condensate formation and maintenance of an active chromatin landscape at stem cell-associated genes in cooperation with epigenetic regulators. In this study, we show that MYST family histone acetyltransferase (HAT) complex proteins, including KAT6A/MOZ, KAT7/HBO1, and the common KAT6A/7 complex subunit BRPF1, associate with NUP98 FOs on chromatin and within condensates. MYST HATs are molecular dependencies in NUP98-rearranged (NUP98-r) leukemia, and genetic inactivation or pharmacologic inhibition of KAT6A and KAT7 impairs NUP98-r cell fitness. KAT6A/7 inhibition decreased global H3K23ac levels, displaced NUP98::HOXA9 from chromatin at the Meis1 locus, and led to myeloid cell differentiation. Additionally, KAT6A/7 inhibition decreased leukemic burden in multiple NUP98-r leukemia xenograft mouse models, synergized with menin inhibitor treatment, and was efficacious in menin inhibitor-resistant cells. In summary, we show that MYST family HATs are therapeutically actionable dependencies in NUP98-r acute myeloid leukemia. SIGNIFICANCE:KAT6A and KAT7 associate with NUP98 FOs to drive leukemogenesis. Inhibition of their HAT activity is an effective therapeutic strategy in NUP98-r leukemias, including those resistant to menin inhibition. Moreover, combined KAT6A/7 and menin inhibition is synergistic, supporting clinical translation to improve outcomes for NUP98 FO-driven leukemias.

Supplementary Figure S5 shows that NUP98-rearranged cells respond to KAT6A/7 inhibition in vitro and in vivo.

Supplementary Table S1. Exonic mutations/indels in Nup98::Kdm5a;Vav-Cre tumors developing after bone marrow transplant Supplementary Table S2. Exonic mutations/indels in spontaneous Nup98::Kdm5a;Vav-Cre tumors Supplementary Table S3. Enrichment/depletion of gRNAs for in vivo epigentic CRISPR/Cas9 screen Supplementary Table S4. Enrichment/depletion of genes for in vivo epigenetic CRISPR/Cas9 screen Supplementary Table S5. Gene ontology analysis for in vivo epigenetic CRISPR/Cas9 screen Supplementary Table S6. Upregulated gene sets in NUP98::HOXA9 mouse leukemia cells after 72-hour treatment with 2 uM SNDX-5613 + 500 nM PF9363 versus vehicle treatment Supplementary Table S7. Downregulated gene sets in NUP98::HOXA9 mouse leukemia cells after 72-hour treatment with 2 uM SNDX-5613 + 500 nM PF9363 versus vehicle treatment Supplementary Table S8. Statistical analysis of Meis1 ChIPseq peaks Supplementary Table S9. Upregulated gene sets in NUP98::KDM5A AMKL PDX (CPCT0021) after treatment with PF9363 versus vehicle Supplementary Table S10. Downregulated gene sets in NUP98::KDM5A AMKL PDX (CPCT0021) after treatment with PF9363 versus vehicle Supplementary Table S11. Average expression of KAT6A and Menin targets in scRNAseq data Supplementary Table S12. Motif analysis of regions with differentially upregulated accessibility after combination treatment versus vehicle Supplementary Table S13. Motif analysis of regions with differentially downregulated accessibility after combination treatment versus vehicle Supplementary Table S14. Primers used in this study Supplementary Table S15. gRNA sequences used in this study Supplementary Table S16. Antibodies used in this study Supplementary Table S17. Fusion oncoprotein amino acid sequences used in RIME

Expression of high FG motif valence NHA9 constructs in lin− HSPCs leads to hematopoietic cell transformation and aberrant expression of Hox family and other genes. A, Average number of colonies per 2,000 cells for lin− HSPCs expressing negative control empty vector and mEGFP-tagged NHA9 and mutant constructs. The values of colony numbers shown are mean ± SD from triplicate technical replicates of a representative experiment. B–D, RNA-seq was performed for lin− HSPCs expressing empty vector, G-NHA9, or mutants after 1 week of growth in methylcellulose containing myeloid and erythroid growth factors (n = 5 for each condition). B, Heat map for differentially expressed genes of interest. C, PCA of the 500 most variable genes. D, Gene set enrichment analysis for cells expressing G-NHA9, G-NHA9–8FA, or G-NHA9Midi—each versus empty vector. Pathways of interest are shown, with a complete list of significantly upregulated or downregulated gene sets in Supplementary Table S3. The most significantly dysregulated genes from each pathway are marked in B.

Supplementary Figure S7 shows that KAT6A/7 and Menin inhibition alter gene expression and remodel chromatin in NUP98-rearranged cells.

Mutation of multiple F and FG residues in the FG-rich IDR of NHA9 alters puncta formation in cells. A, Schematic of NHA9 and mutant constructs used in this study. FG motif valence is shown on the left. B, Representative image of live HEK293T cells expressing G-NHA9–8FA (top, green) and G-NHA9–21FGAA (bottom, green). DNA is stained with Hoechst dye (blue). C–F, Plots of puncta # (/103 μm3; C), Vp (μm3; D), Kp (${K_p}{\rm{\ }} = {\rm{\ }}\frac{{[ {{\rm{DP}}} ]}}{{[ {{\rm{LP}}} ]}}$ (E), and ΔGTr (kcal/mol; F) versus [G-NHA9 construct] for G-NHA9 (green), G-NHA9–8FA (purple), and G-NHA9–21FGAA (blue) from data represented in B. Data are plotted on a semi-log (y-axis: log10) scale in C–E. Refer to Supplementary Table S2 for mean values ± standard error. The pairwise P value between G-NHA9 versus G-NHA9–8FA and G-NHA9 versus G-NHA9–21FGAA is shown in each plot (C–F; n = 935, 683, and 865 in C including the cells with zero punctum and n = 378, 273, and 159 in D–F excluding the cells with zero punctum, respectively, for G-NHA9, G-NHA9–8FA, and G-NHA9–21FGAA).

The NHA9 constructs form puncta in lin− HSPCs. A–G, Representative images of live lin− HSPCs expressing EGFP empty vector as a control (A), G-NHA9 (B), G-NHA9–ΔDNA (C), G-NHA9–8FA (D), G-NHA9–21FGAA (E), G-NHA9Midi (F), and G-NHA9Midi–21FGAA (G).

DNA binding by the HOXA9 homeodomain of NHA9 influences puncta morphology and behavior. A, Representative confocal microscopy image of live HEK293T cells expressing G-NHA9–ΔDNA (green). DNA is stained with Hoechst dye (blue). An overlay of the G-NHA9–expressing cell from Fig. 1B is included for comparison (right). B and C, Plots of puncta # (/103 μm3; B) and Vp (μm3; C) versus [G-NHA9 construct] for G-NHA9 (green) and G-NHA9–ΔDNA (red) from data represented in A. Data are plotted on a semi-log (y-axis: log10) scale. D, Still images of multiple time points taken from a time-lapse confocal fluorescence microscopy video (Supplementary Video S2) of a fusion event in an HEK293T cell expressing G-NHA9–ΔDNA. E, Confocal micrographs of FRAP of a G-NHA9–ΔDNA punctum in HEK293T cells at different time points after photobleaching (left). Fluorescence recovery curves are shown for bleached (red, right) and unbleached puncta (black, right). The recovery curve for G-NHA9 is also provided for comparison (green). Individual puncta were manually tracked at different times, and the G-NHA9–ΔDNA fluorescence intensity versus recovery time was plotted as the mean ± SD (n = 20). The pairwise P value for the recovery curves between G-NHA9 and G-NHA9–ΔDNA is 2.2 × 10−16 using the t test. F–H, Plots of the concentration of the NHA9 construct in the nuclear light phase ([LP], μmol/L; F) and within puncta (termed the dense phase; [DP], μmol/L; G), and the Kp (${K_p}{\rm{\ }} = {\rm{\ }}\frac{{[ {{\rm{DP}}} ]}}{{[ {{\rm{LP}}} ]}}$ (H) versus [G-NHA9 construct] for G-NHA9 (green) and G-NHA9–ΔDNA (red). Data are plotted on a semi-log (y-axis: log10) scale. I, 1D-density distribution of PCC per cell for G-NHA9 and G-NHA9–ΔDNA to analyze the linear relationship of the signal between mEGFP and Hoechst. Refer to Supplementary Table S2 for mean values ± standard error. The pairwise P values between G-NHA9 and G-NHA9–ΔDNA are shown in each plot (B, C, F–I; see Methods; n = 935 and 780 in B, F, and I including the cells with zero punctum, and n = 378 and 254 in C, G, and H excluding the cells with zero punctum, respectively, for G-NHA9 and G-NHA9–ΔDNA).

Supplementary Figure S4 shows that MYST family histone acetyltransferase complex members are molecular dependencies in NUP98-rearranged cells.

Membraneless cellular assemblies termed biomolecular condensates — into which diverse biopolymers partition — mediate myriad biological processes. A study now reveals that physicochemical features, not specific stereochemistry, influence whether small molecules are enriched within or excluded from a diverse panel of condensates.

NUP98 fusion oncoproteins(FOs) are a hallmark of high-risk leukemia and are present in approximately 5% of pediatric and 3% of adult acute myeloid leukemia (AML). NUP98 fusion oncoproteins involve the N-terminal, intrinsically disordered region of NUP98 and the C-terminal portion of one of over 30 fusion partners. Approximately one-third of fusion partners have DNA-binding homeodomains, and many others have domains involved in gene regulation. Outcomes in children with NUP98-rearranged (NUP98-r) AML are poor, with relapse rates of nearly 70%. We and others recently demonstrated the importance of NUP98 FO-mediated liquid-liquid phase separation in NUP98-r leukemic transformation (Cancer Discov 2022;12:1152, Nature 2021;595:591), but the composition of NUP98 FO-associated condensates and role of FO-interacting proteins in leukemogenesis are incompletely understood. To identify FO interacting proteins, we performed rapid immunoprecipitation mass spectrometry of endogenous proteins (RIME) for HA-tagged wildtype NUP98 and eight NUP98 FOs (NUP98::HOXA9, PMX1, KDM5A, NSD1, JADE2, LNP1, RAP1GDS1, SETBP1) in transfected HEK293T cells. We identified that many known NUP98 FO interactors (e.g. XPO1, RAE1, KMT2A, Menin) were shared across all tested NUP98 FOs, and we observed novel FO interactors including MOZ and other histone acetyltransferase (HAT) complex members. Imaging of HEK293T cells expressing GFP-tagged NUP98::LNP1 demonstrated colocalization of FO with MOZ and associated histone acetylation marks within FO condensates. To investigate the functional importance of chromatin remodeling complexes including HATs for NUP98::KDM5A-driven leukemogenesis, we performed an in vivo CRISPR/Cas9 screen with a guide RNA library targeting 337 epigenetic genes. Our results suggested that BRPF1, a chromatin writer of histone H3 acetylation marks that associates with MOZ and other HAT complexes,is a molecular dependency in NUP98::KDM5A AML. Competitive co-culture of Nup98::Kdm5a;Vav-Cre;Cas9 cells expressing control or HAT complex-targeting gRNAs further validated that inactivation of Brpf1 and a subset of other MOZ/MORF HAT complex members (Moz, Hbo1, Brd1 or Meaf6) decreased cell fitness. We next tested efficacy of pharmacologic inhibition of MOZ in NUP98-r leukemia models. The commercially available MOZ/HBO1 inhibitor PF9363 resulted in reduced viability and myeloid differentiation in mouse and human models of NUP98-r leukemia with multiple fusion partners. In hematopoietic stem and progenitor cells from Nup98::Kdm5a;Vav-Cre mice, PF9363 treatment resulted in global loss of H3K14ac and H3K23ac as determined by mass spectrometry, and reduction in H3K9ac at NUP98 FO target genes by CUT&RUN. In vivo, PF9363 (3 mg/kg daily) reduced leukemic burden and/or prolonged survival in 1 NUP98::HOXA13 and 2 NUP98::NSD1 PDX models. As disease progressed upon treatment cessation, we tested PF9363 in combination with Menin inhibitor SNDX-5613, since Menin inhibitors have shown activity in NUP98-r leukemia (Blood 2022;139:894) and are currently being evaluated in clinical trials. In mouse NUP98 FO leukemia cells, combined in vitro treatment with PF9363 and SNDX-5613 led to decreased cell viability and heightened myeloid differentiation as compared to either agent alone. Single-agent treatment with PF9363 or SNDX-5613 resulted in loss of the NUP98 FO and BRPF1 from chromatin at select FO target genes, including Meis1. Combination treatment decreased both expression of select Hox transcription factors and NUP98 FO chromatin occupancy at HoxA/B cluster genes, which was not observed with either PF9363 or SNDX-5613 monotherapy. Finally, we tested the combination of PF9363 and SNDX-5613 in vivo using NUP98-r PDX models that are sensitive or resistant to Menin inhibition. In a Menin inhibitor sensitive NUP98::KDM5A PDX, combined treatment with PF9363 and SNDX-5613 resulted in increased expression of myeloid differentiation markers compared to SNDX-5613 alone. In a NUP98::NSD1 PDX that is resistant to Menin inhibition (Blood 2022;139:894), PF9363 monotherapy extended median latency by 21%. In combination with SNDX-5613, PF9363 increased median latency by 50% compared to vehicle and by 18% compared to PF9363 alone. In summary, our studies show that MOZ is a therapeutic vulnerability in NUP98 FO-driven AML and that MOZ inhibitor treatment may improve responses to Menin inhibition.