NUP98-rearranged pediatric AML is sustained by a conserved leukemia stem cell program with SPNS2/3 as a targetable vulnerability

Abstract Pediatric acute myeloid leukemias (AML) are genetically and molecularly diverse. Pediatric AML carry a range of different driver mutations, often including translocations that produce fusion oncoproteins. Furthermore, they arise from diverse progenitor cell populations and at different stages of fetal and childhood development. This molecular and contextual diversity complicates efforts to identify effective targeted therapies, in part because modeling strategies do not capture the genetic heterogeneity observed among patients. We have developed a multipronged strategy to model complex genetic and contextual features of pediatric AML in mice, with the overarching goal of identifying malignant self-renewal programs that afford targetable vulnerabilities. We first used Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITE-seq) and paired functional assays to identify functionally distinct subpopulations of NUP98-rearranged (NUP98r) AML using a previously described NUP98::HOXD13/Flt3ITD transgenic model. Our approach identified a leukemia stem cell (LSC) subpopulation and gene expression signature that was conserved in human NUP98r AML. The LSC signature was also enriched in human UBTF-tandem duplication (UBTF-TD) and NPM1 mutated (NPM1c) AML, but not KMT2A-rearranged AML, suggesting mutation class-specific self-renewal mechanisms. To compare LSC self-renewal mechanisms across molecular subclasses of AML, we developed a novel strategy to generate inducible, genetically diverse AML models. Specifically, we generated induced pluripotent stem cells (iPSCs) and chimeric mice to express NUP98::HOXA9, NUP98::KDM5A, NUP98::NSD1, RUNX1::RUNX1T1, MNX1, or KMT2A::MLLT1 oncoproteins specifically in blood cells at defined stages of ontogeny. We used CRISPR/Cas9 gene editing to introduce cooperating mutations at endogenous loci, including Flt3 internal tandem duplication (FLT3ITD), Nf1 loss of function (NF1LOF), and NrasG12D. With this approach, we could induce expression of any desired oncoprotein at any desired stage of development along with relevant cooperating mutation, all while bypassing complex mouse breeding. We also used gene editing strategies to generate NPM1c/FLT3ITD and UBTF-TD/ FLT3ITD AML, again with mutations at endogenous loci. The resulting AML models are transplantable, genetically manipulable and in some cases gave rise to immortalized in vitro cell lines. To credential the models and identify fusion/cooperating mutation-specific vulnerabilities, we performed CITE-seq focusing on NUP98::HOXA9/FLT3ITD, NUP98::HOXA9/NF1LOF,RUNX1::RUNX1T1/NF1LOF, KMT2A::MLLT1/NRASG12D, UBTF-TD/ FLT3ITD, and NPM1c/FLT3ITD models. Cross species comparisons to human AML revealed faithful representation of heterogeneity and gene expression within the murine AML. Each genetic subclass exhibited distinct but reproducible differentiation trajectories and transcriptional profiles. NUP98::HOXA9/FLT3ITD, NUP98::HOXA9/NF1LOF, NPM1c/FLT3ITD, UBTF-TD/ FLT3ITD, and KMT2A::MLLT1/NRASG12DAML all displayed intraleukemic heterogeneity and had putative LSC populations that expressed known self-renewal genes. RUNX1::RUNX1T1/NF1LOF AML was homogeneous and lacked a discernible LSC population. Presumptive LSC populations from NUP98::HOXA9/FLT3ITD, NUP98::HOXA9/NF1LOF, NPM1c/FLT3ITD, and UBTF-TD/ FLT3ITD were all enriched for a shared gene expression signature. These highly tractable models thus capture key aspects of human pediatric AML biology.We sought to identify actionable, mutation class-specific vulnerabilities within putative LSC populations, focusing primarily on the NUP98r models. We found that genes encoding sphingosine transporters SPNS2 and SPNS3were highly expressed in NUP98r LSCs relative to non-LSCs, identifying them as potential therapeutic targets. SPNS2/3 were also expressed, to a lesser degree, in NPM1c/FLT3ITD and UBTF-TD/ FLT3ITD AML. Pharmacological SPNS2/3 inhibition severely impaired NUP98rAML growth in vitro, and it significantly extended survival in vivo. In some cases, mice with NUP98r AMLwere cured with SPNS2/3-directed monotherapy. Other genetic subclasses were less sensitive to SPNS2/3 inhibitors, as were normal human and murine hematopoietic progenitors. Our data indicate a therapeutic window for targeting SPNS2/3 in NUP98r AML. They raise the possibility of identifying other mutation class-specific vulnerabilities through the use of well-credentialed, iPSC-derived chimeric AML models.