Fusion oncoproteins and cooperating mutations define disease phenotypes in NUP98-rearranged leukemia

Abstract

Leukemias with NUP98 rearrangements exhibit heterogeneous phenotypes such as acute myeloid leukemia, T-cell acute lymphoblastic leukemia (T-ALL), or myelodysplastic syndrome/neoplasms associated with fusion partners, whereas the mechanism responsible for this heterogeneity is poorly understood. Through genome-wide mutational and transcriptional analyses of 177 NUP98-rearranged leukemias, we show that cooperating alterations are associated with differentiation status even among leukemias sharing the same NUP98 fusions, such as NUP98::KDM5A acute megakaryocytic leukemia with RB1 loss or T-ALL with NOTCH1 mutations. CUT&RUN profiling of in vitro cord blood CD34+ cell (cbCD34) models of major NUP98 fusions revealed that NUP98-fusion oncoproteins (FOs) directly regulate differentiation-related genes contributing to the disease phenotypes, represented by NUP98::KDM5A binding to MEIS2 or GFI1B for megakaryocyte (MK) differentiation. In patient samples, NUP98-FO binding patterns are heterogeneous, potentially shaped by somatic mutations and differentiation status. Using cbCD34 models and CRISPR/Cas9 gene editing, we show that RB1 loss cooperates with NUP98::KDM5A by blocking terminal differentiation toward platelets and expanding MK-like cells, whereas WT1 frameshift mutations skew differentiation toward dormant lymphoid-myeloid primed progenitor cells and cycling granulocyte-monocyte progenitor cells, providing evidence for NUP98-rearranged leukemia phenotypes affected by cooperating alterations. NUP98::KDM5A cbCD34 models with RB1 or WT1 alterations have different sensitivities to menin inhibition, suggesting that cellular differentiation provides stage-specific menin dependencies and resistance mechanisms that can be leveraged for future treatment strategies for NUP98-rearranged leukemia.

Graphical abstract

Figure 1.

Heterogeneity of pediatric NUP98r leukemia. (A) Details of NUP98r leukemia samples (n = 185, left) and analytical pipelines (right). (B) Numbers and functional annotations of fusion partners in the study cohort. Colors indicate protein functional groups. (C) Age distribution related to fusion partners and disease types. Colors indicate disease types. (D) UMAP plots of the transcriptional cohort (n = 2321) colored according to leukemia subtypes (left), NUP98 fusion partners (middle), and enrichment of fusion partners in transcriptional clusters (right). The shapes of dots indicate disease types (circles, AML; triangles, ALL), and colors in the heat map indicate enrichment of fusions in each cluster. Asterisks indicate statistically significant adjusted P values from 2-sided Fisher’s exact tests and the Benjamini-Hochberg adjustment (∗P < .05; ∗∗∗P < .001; black: enriched, blue: exclusive). In panels A and C, lines of the box plots represent the 25% quantile, median, and 75% quantile, and the upper whisker represents the higher value of maxima or 1.5× interquartile range (IQR), and the lower whisker represents the lower value of minima or 1.5× IQR. Abbreviations in leukemia subtypes are found in supplemental Table 10. ETP, early T-cell precursor ALL; MDS/MPN, myelodysplastic syndrome/myeloproliferative neoplasms; tMN, therapy-related myeloid neoplasm; UMAP, uniform manifold approximation and projection.

Figure 2.

Mutational background of NUP98r leukemia associated with disease phenotypes. (A) Genetic profiles of NUP98r samples in the cohort. Colors indicate patient annotations (top) and types of gene alterations (bottom). (B) Co-occurrence and mutual exclusivity among recurrent alterations (n ≥ 3). (C) Enrichment of somatic alterations in transcriptional clusters (left) and fusion partners (right). In panels B-C, colors indicate adjusted P values by 2-sided Fisher’s exact tests and the Benjamini-Hochberg adjustment (red: co-occurring, blue: mutually exclusive), and asterisks indicate statistically significant values (adjusted P values, ∗P < .05; ∗∗P < .01; ∗∗∗P < .001). Annotations of genes in mutational heat maps depend on known general functions.

Figure 3.

Varieties of cellular hierarchies in NUP98r leukemia. (A) Strategies for scRNAseq and deconvolution of bulk RNAseq data. (B) UMAP plots of patient samples colored by sample source (left) and transcriptional clusters (right). (C) Enrichment of cells in each cluster indicated by colors and sizes (left), and marker gene expression indicated by colors (averaged expression) and size (ratio of expressing cells: count >0 (right). (D) UMAP plot of reference bone marrow and thymocyte scRNAseq data, colored according to cell labels from the original reference data. (E) Distribution of patient sample scRNAseq on the reference data inferred by the MapQuery function in the Seurat package. Cells in normal hematopoietic cell clusters were excluded. Cells are colored according to the cell density on the UMAP plot. Cooperating mutations found in bulk samples are also shown. (F) Enrichment of cells with each cell label inferred by Seurat, indicated by colors and sizes. (G) Cellular component of bulk RNA samples (n = 185) inferred by CIBERSORT using a signature matrix derived from reference scRNAseq data. Bars are colored by cell populations in each sample. cDC, classic dendritic cell; CLP, common lymphoid progenitor; DP, CD4-CD8 double-positive T cell; GMP, granulocyte-monocyte progenitor; HSPC, hematopoietic stem and progenitor cell; NK, natural killer T cell; pDC, plasmacytoid dendritic cell; UMAP, uniform manifold approximation and projection.

Figure 4.

cbCD34 models recapitulate the phenotypes of NUP98r leukemia. (A) Experimental schema using cbCD34 models. (B) Colony-forming unit assays of cbCD34 models with empty control vectors or NUP98::NSD1, NUP98::KDM5A, or NUP98::HOXA9-expressing vectors. (C) Cell growth assays of cbCD34 models in liquid culture. (D) Flow cytometric analysis of cbCD34 models in liquid culture (top, CD34⁺; middle, CD11b⁺; bottom, CD41a⁺; population ratio, % in mCherry⁺ live cells). (E) Principal component analysis of RNAseq data from liquid culture. Colors indicate NUP98 fusions, and shapes indicate days after transduction. (F) Heat map showing expression of representative genes related to stemness or differentiation of hematopoietic cells. The colors of cells indicate expression levels normalized among samples, and genes are annotated on the left. (G) Comparison of differentially expressed genes in each cbCD34 model compared with empty vector controls at day 42. Venn diagram showing overlaps of highly expressed genes in each model (middle), and gene ontology term analyses of shared or specific differentially expressed genes are shown (left, right). Gray lines show −log10 FDR = 0.05. Data were obtained from 3 biological replicates (different lots of cord blood). In panels B-D, statistical tests were performed by a generalized linear mixed effect model with Poisson (B) and Gaussian (C-D) distributions, followed by comparison with empty vector control and the Benjamini-Hochberg adjustment, asterisks indicating adjusted P values, ∗P < .05. Error bars indicate mean ± standard error of the mean. FDR, false-discovery rate; PDX, patient-derived xenograft.

Figure 4.

cbCD34 models recapitulate the phenotypes of NUP98r leukemia. (A) Experimental schema using cbCD34 models. (B) Colony-forming unit assays of cbCD34 models with empty control vectors or NUP98::NSD1, NUP98::KDM5A, or NUP98::HOXA9-expressing vectors. (C) Cell growth assays of cbCD34 models in liquid culture. (D) Flow cytometric analysis of cbCD34 models in liquid culture (top, CD34⁺; middle, CD11b⁺; bottom, CD41a⁺; population ratio, % in mCherry⁺ live cells). (E) Principal component analysis of RNAseq data from liquid culture. Colors indicate NUP98 fusions, and shapes indicate days after transduction. (F) Heat map showing expression of representative genes related to stemness or differentiation of hematopoietic cells. The colors of cells indicate expression levels normalized among samples, and genes are annotated on the left. (G) Comparison of differentially expressed genes in each cbCD34 model compared with empty vector controls at day 42. Venn diagram showing overlaps of highly expressed genes in each model (middle), and gene ontology term analyses of shared or specific differentially expressed genes are shown (left, right). Gray lines show −log10 FDR = 0.05. Data were obtained from 3 biological replicates (different lots of cord blood). In panels B-D, statistical tests were performed by a generalized linear mixed effect model with Poisson (B) and Gaussian (C-D) distributions, followed by comparison with empty vector control and the Benjamini-Hochberg adjustment, asterisks indicating adjusted P values, ∗P < .05. Error bars indicate mean ± standard error of the mean. FDR, false-discovery rate; PDX, patient-derived xenograft.

Figure 5.

Differential gene regulation by NUP98-FOs. (A) Integrative Genome Viewer (IGV) tracks of the HOXA-B clusters from CUT&RUN using HA, H3K4me3, and H3K27ac antibodies in HA-tagged NUP98r cbCD34 models (top: empty vector control, gray; HA-NUP98::KDM5A, red; HA-NUP98::NSD1, blue; HA-NUP98::HOXA9, black), and heat map showing expression levels of HOXA-B genes (bottom). N-terminal NUP98 antibody was applied for the empty vector control. (B) IGV tracks of differentiation-related gene loci (top: RUNX1, GFI1B, and the HOXC cluster) and Venn diagram showing overlap of protein-coding genes with annotated peaks (bottom: false-discovery rate <0.00001). (C) CUT&RUN strategy from primary patient samples or NUP98::KDM5A cell lines (CHRF-288-11 and ST1653). (D) Counts of peaks from the N-terminus NUP98 antibody in primary samples (top) and overlaps of target genes among non-NUP98::KDM5A and NUP98::KDM5A (bottom). NUP98::KDM5A AMKL-specific 21 target genes are highlighted. Colors indicate peak annotations. (E) IGV tracks of the HOXA-B cluster from CUT&RUN using N-terminal NUP98, H3K4me3, and H3K27ac antibodies in primary leukemia samples and NUP98::KDM5A cell lines. (F) Principal component analysis of genome-wide PBS (probability of being signals) scores of H3K27ac (left) and H3K27me3 (right) from primary samples. Colors indicate expression clusters, and shapes indicate fusion partners. (G) Differential signal analysis using H3K27ac PBS scores between NUP98::KDM5A and other (NUP98::NSD1 and NUP98::RAP1GDS1) samples (left) and NUP98::KDM5A AMKL and non-AMKL (right) calculated by limma, followed by the Benjamini-Hochberg adjustment. Only regions with significant enrichment (adjusted P < .05) are shown. (H) IGV tracks of the MECOM and MEIS2 gene loci from CUT&RUN using N-terminal NUP98, H3K4me3, and H3K27ac antibodies in primary leukemia samples and NUP98::KDM5A cell lines. IgG, immunoglobulin G.

Figure 6.

Functional characterization of recurrent somatic alterations in NUP98r leukemia. (A) Experimental schema of induction of cooperating alterations (RB1, WT1) in cbCD34/Cas9 models. (B) Cell growth assays of cbCD34/Cas9 NUP98::KDM5A with gRNAs targeting the AAVS, RB1, or WT1 loci (left), and cytospin of cells on day 35 (right; scale bars, 50 μm). (C) Rates of indel (insertions and deletions) at days 4 and 39 in each condition. Bars represent fractions of indel rates in all target sequence reads, and dots represent the out-of-frame indel ratio among total indels. (D) Flow gating (left), CD34⁺ CD41a⁺ positivity (middle), and CD34⁺ CD41a^– (right) among mCherry⁺ GFP⁺ mAmetrine⁺ live cells. (E) Principal component analysis of RNAseq of gRNA-transduced NUP98::KDM5A cbCD34 models at day 35. (F) Differentially expressed genes analysis between AAVS controls and RB1-gRNA conditions (left), and gene ontology term analysis of differentially expressed genes (right). Colors indicate differentially expressed genes and gene ontology terms (red: high in RB1-gRNA conditions, blue: low in RB1-gRNA conditions). (G) Differentially expressed genes analysis between AAVS controls and WT1-gRNA conditions as shown in panel F. (H) UMAP plots of scRNAseq data from gRNA-transduced NUP98::KDM5A cbCD34 models at day 35, showing marker gene expression (left), annotated clusters (middle), and cell distributions among conditions (right). Colors in plots indicate relative expression levels, clusters, and cell density, respectively. (I) Enrichment of cells with each cluster indicated by colors and sizes. (J) Pseudotime along myeloid (HSC → GMP → monocytes) and platelet (HSC → MEP → MK → platelet) trajectories. Colors represent pseudotime scores of each single cell inferred by Slingshot. (K) RB1 (top) and CDKN2A (bottom) expression along the pseudotime axis in each condition, with red curves showing average expressions. (L) Differentially expressed genes analysis between the platelet-like and MK-like clusters in the AAVS-control condition (left) and gene ontology term analysis (right) of genes high in the platelet-like cluster (red) and the MK-like cluster (blue). (M) Schematics illustrating platelet differentiation in normal hematopoiesis and NUP98::KDM5A models (created in BioRender. Umeda, M. (2025) https://BioRender.com/kreq8kl). Assay data were obtained in technical triplicates from an established NUP98::KDM5A/Cas9 line and independent experiments. One data point in panel C was not obtained due to technical errors. RNAseq data were obtained from 6 independent experiments. In panels B-D, statistical tests were performed by linear mixed effect model (B) or 2-sided Student t test by comparing day 4 and day 39 (C) or gRNA conditions and AAVS controls (D), and limma (F-G), followed by the Benjamini-Hochberg adjustment when applicable. Differentially expressed genes in scRNAseq (L) were identified using the FindMarker function in the Seurat package with default settings, which calculates adjusted P values with limma implementation of the Wilcoxon rank-sum test followed by Bonferroni correction. Asterisks indicating P values or adjusted P values <.05. Error bars indicate mean ± standard error of the mean. HSC, hematopoietic stem cell; UMAP, uniform manifold approximation and projection.

Figure 6.

Functional characterization of recurrent somatic alterations in NUP98r leukemia. (A) Experimental schema of induction of cooperating alterations (RB1, WT1) in cbCD34/Cas9 models. (B) Cell growth assays of cbCD34/Cas9 NUP98::KDM5A with gRNAs targeting the AAVS, RB1, or WT1 loci (left), and cytospin of cells on day 35 (right; scale bars, 50 μm). (C) Rates of indel (insertions and deletions) at days 4 and 39 in each condition. Bars represent fractions of indel rates in all target sequence reads, and dots represent the out-of-frame indel ratio among total indels. (D) Flow gating (left), CD34⁺ CD41a⁺ positivity (middle), and CD34⁺ CD41a^– (right) among mCherry⁺ GFP⁺ mAmetrine⁺ live cells. (E) Principal component analysis of RNAseq of gRNA-transduced NUP98::KDM5A cbCD34 models at day 35. (F) Differentially expressed genes analysis between AAVS controls and RB1-gRNA conditions (left), and gene ontology term analysis of differentially expressed genes (right). Colors indicate differentially expressed genes and gene ontology terms (red: high in RB1-gRNA conditions, blue: low in RB1-gRNA conditions). (G) Differentially expressed genes analysis between AAVS controls and WT1-gRNA conditions as shown in panel F. (H) UMAP plots of scRNAseq data from gRNA-transduced NUP98::KDM5A cbCD34 models at day 35, showing marker gene expression (left), annotated clusters (middle), and cell distributions among conditions (right). Colors in plots indicate relative expression levels, clusters, and cell density, respectively. (I) Enrichment of cells with each cluster indicated by colors and sizes. (J) Pseudotime along myeloid (HSC → GMP → monocytes) and platelet (HSC → MEP → MK → platelet) trajectories. Colors represent pseudotime scores of each single cell inferred by Slingshot. (K) RB1 (top) and CDKN2A (bottom) expression along the pseudotime axis in each condition, with red curves showing average expressions. (L) Differentially expressed genes analysis between the platelet-like and MK-like clusters in the AAVS-control condition (left) and gene ontology term analysis (right) of genes high in the platelet-like cluster (red) and the MK-like cluster (blue). (M) Schematics illustrating platelet differentiation in normal hematopoiesis and NUP98::KDM5A models (created in BioRender. Umeda, M. (2025) https://BioRender.com/kreq8kl). Assay data were obtained in technical triplicates from an established NUP98::KDM5A/Cas9 line and independent experiments. One data point in panel C was not obtained due to technical errors. RNAseq data were obtained from 6 independent experiments. In panels B-D, statistical tests were performed by linear mixed effect model (B) or 2-sided Student t test by comparing day 4 and day 39 (C) or gRNA conditions and AAVS controls (D), and limma (F-G), followed by the Benjamini-Hochberg adjustment when applicable. Differentially expressed genes in scRNAseq (L) were identified using the FindMarker function in the Seurat package with default settings, which calculates adjusted P values with limma implementation of the Wilcoxon rank-sum test followed by Bonferroni correction. Asterisks indicating P values or adjusted P values <.05. Error bars indicate mean ± standard error of the mean. HSC, hematopoietic stem cell; UMAP, uniform manifold approximation and projection.

Figure 7.

Cooperating alterations and differentiation status affect sensitivity to menin inhibition. (A) Experimental schema showing revumenib treatment. (B) Relative cell growth of cbCD34 NUP98::KDM5A models with gRNA treated with revumenib (0.1-1 μM) compared with dimethyl sulfoxide controls. (C) Flow gating (top), CD34⁺CD41a⁺ population (left), CD34^–CD41a⁺ population (middle), and CD34^–CD11B⁺ populations (right) among mCherry⁺ GFP⁺ mAmetrine⁺ live population at day 15. (D) GSEA analyses between dimethyl sulfoxide and revumenib-treated conditions, colors showing NES (top-left), comparison of expression changes between AAVS and RB1-gRNA1 conditions (right), and gene ontology term analysis of changes enriched (difference of fold changes >1) in AAVS conditions (bottom-left). (E) Relative cell growth of unedited cbCD34 NUP98::KDM5A (control, black), CHRF-288-11 (red), and ST1653 patient-derived xenograft (blue) treated with revumenib (0.1-1 μM) compared with dimethyl sulfoxide controls. (F) Schematics illustrating a cellular hierarchy of NUP98::KDM5A models created in BioRender. Umeda, M. (2025) https://BioRender.com/kreq8kl. Data were obtained from 3 technical replicates using gRNA-transduced cells in Figure 6. One data point at day 15 was excluded for technical errors. Statistical tests were performed by a generalized linear mixed effect model with Gaussian distribution, followed by the Benjamini-Hochberg adjustment (B,E) or Student t test by comparing gRNA conditions with AAVS controls (C). Asterisks indicate P values or adjusted P values; ∗P < .05. Error bars indicate mean ± standard error of the mean. DMSO, dimethyl sulfoxide; NES, normalized enrichment score.