Leukemic fusion: partners and mutations shape NUP98r leukemia Free

In this issue of Blood, Umeda et al¹ show that the diverse phenotypes of NUP98-rearranged (NUP98r) leukemias are shaped by specific fusion partners, which display different patterns of genome localization and are associated with distinct cellular states and therapeutic sensitivities, highlighting the importance of context-dependent molecular mechanisms in leukemogenesis and treatment response.

NUP98r leukemias have been puzzling part of pediatric and young adult hematologic malignancies since they were first described.^2,3 Although relatively uncommon, comprising up to 7% of pediatric acute myeloid leukemia (AML) cases,⁴ these cases display striking clinical heterogeneity, manifesting as AML, T-cell acute lymphoblastic leukemia (T-ALL), or myelodysplastic syndromes,⁵ and account for up to half of the primary induction failures in pediatric AML.⁶ The diversity of fusion partners (over 30 have been described), combined with varied cooperating mutations, raises an essential question: what determines phenotype and therapeutic vulnerability in NUP98r leukemias?

By integrating genome-wide mutational profiling, single-cell transcriptomics, CUT&RUN chromatin profiling, and CRISPR/Cas9 mutagenesis in cord blood CD34⁺ (cbCD34) models, Umeda et al provide a detailed map of how fusion partners and cooperating mutations interact with cellular differentiation status to shape the disease landscape of NUP98r leukemia. The authors began with a large cohort of 185 NUP98r samples from 177 patients, primarily children and young adults. Within this cohort, NUP98::NSD1 and NUP98::KDM5A emerged as the most frequent c-terminal fusions. The former was closely associated with a myelomonocytic AML phenotype and the latter was enriched in acute megakaryocytic leukemia (AMKL) in infants. Other fusion partners, including RAP1GDS1, were more common in T-ALL and were enriched in NOTCH1 mutations.

A critical insight of the study is that even leukemias with the same fusion, such as NUP98::KDM5A, can exhibit strikingly different phenotypes depending on the secondary mutations. For instance, RB1 loss skewed differentiation toward megakaryocytic progenitors and blocked terminal platelet production, mimicking features of AMKL. In contrast, WT1 mutations shifted cells toward lymphoid-myeloid primed progenitors and granulocyte-monocyte progenitors, favoring a more undifferentiated or lymphoid-biased phenotype. These relationships were modeled using CRISPR/Cas9–edited cbCD34⁺ cells and tracked via single-cell RNA sequencing (RNA-seq) and pseudotime analyses. The authors’ application of CUT&RUN to both the cbCD34 models and patient samples further show how NUP98 fusion oncoproteins bind chromatin in a context-dependent manner. All fusions demonstrated binding to canonical loci, such as HOXA/B clusters, consistent with previous studies.^7,8 However, NUP98::KDM5A displayed unique binding to megakaryocyte-regulating genes, like GFI1B and MEIS2, particularly in AMKL samples, suggesting that fusion-specific targeting contributes to the consequent lineage phenotype, especially when combined with deletions in RB1. These data suggest that fusion oncoprotein binding is not only partner-specific but may also be modulated by differentiation state and comutation profile.⁸ 

Beyond these biologic insights, the study offers translational implications. Treatment of cbCD34 models with the menin inhibitor revumenib showed that sensitivity is contingent on differentiation status and mutational background. Although NUP98::KDM5A models with wild-type RB1 or WT1 showed partial differentiation and growth inhibition, cells with RB1 loss were markedly less responsive. Differentiation status was also associated with differential sensitivity, raising the possibility that immunophenotyping leukemias could be used as a predictive biomarker for menin treatment. The study showed that leukemic identity is driven by the complex interplay of fusion partner, chromatin occupancy, mutational comutations, and hematopoietic context. It also challenges a one-size-fits-all approach to treatment; menin dependency, for example, cannot be assumed solely based on the presence of a NUP98 fusion. Rather, future therapeutic trials should consider incorporating transcriptional or epigenetic biomarkers to identify patients most likely to benefit from targeted therapies.

As with all studies, this work raises new questions. To what extent are the in vitro cbCD34 models reflective of in vivo hematopoietic microenvironments, particularly in fetal vs adult niches? Can the stage-specific menin resistance be reversed by combining differentiation therapy with menin inhibitors? More broadly, should leukemia subclassification under the World Health Organization or International Consensus Classification be based on the presence of a single fusion partner (ie, NUP98r) or should classifications be more granular and take into account the comutations or differentiation status? The work also highlighted the technical complexity required to render modern clinical leukemia diagnostics, especially given that most NUP98 rearrangements are cryptic by conventional cytogenetics, necessitating the use of RNA-seq or whole-genome sequencing–based evaluation, which is not yet routine at all centers.^9,10 

In summary, Umeda et al provide a compelling framework for understanding the complex clinical heterogeneity of NUP98r leukemias. Their findings reinforce the view that it is not just the fusion but also the cellular context that drives leukemia and predicts the therapeutic response.

Conflict-of-interest disclosure: E.D. and D.S. report receiving royalties from Caris Life Sciences.