FLT3^ITD drives context-specific changes in cell identity and variable interferon dependence during AML initiation

FLT3-internal tandem duplication (FLT3^ITD) is one of the most common mutations found in both pediatric and adult acute myeloid leukemia (AML).^1,2 It typically arises late in AML evolution,³ and it is insufficient to cause AML by itself.⁴ Instead, Flt3^ITD cooperates with a range of different initiating mutations to drive AML.^5-12 When paired with Tet2, Runx1, or Npm1 mutations, Flt3^ITD induces changes in gene expression and epigenome organization that are not observed with any of these individual mutations alone.^9-11,13 The target gene profiles overlap,¹⁰ suggesting that Flt3^ITD may activate a core set of AML effectors irrespective of which mutations it cooperates with. However, the prior studies all described interactions between Flt3^ITD and mutations that occur more commonly in adult AML than in pediatric AML. Approximately, 70% of early childhood AML carry translocations that express fusion oncoproteins, whereas only ∼10% of older adult AML are driven by fusion proteins.¹ FLT3^ITD may therefore convey distinct mutation profile- and age-specific gene dependencies when it pairs with pediatric fusions as opposed to adult initiating mutations.

Cryptic NUP98 translocations (NUP98-t) are AML driver mutations that arise most often in young children and frequently cooccur with FLT3^ITD.^1,14-16 NUP98 has several fusion partners, including NSD1, KDM5A, HOXA9, and HOXD13.¹⁴ NUP98 fusion proteins create phase-separated condensates at target gene loci to physically approximate enhancer and promoter elements.^17,18 They recruit transcriptional coactivators, including MLL1 and NSL complexes, to HOX genes to promote transcript expression.^19-22,HOX gene products then reprogram hematopoietic progenitors and drive aberrant self-renewal programs.^19-21 NUP98 fusions require cooperating mutations to induce AML, with FLT3^ITD being the most common cooperating mutation (>50%).^{1,14,15,22,23} Mouse and human models confirm that NUP98-t is not sufficient to initiate AML by itself.^12,22,24,25 FLT3^ITD interactions with NUP98-t therefore offer a prototypical case for comparing pediatric and adult mutation cooperation.

To understand how pediatric and adult initiating mutations might differentially orient FLT3^ITD-driven transcriptional states, we tested whether Flt3^ITD has similar or distinct functions in the setting of a pediatric-biased NUP98-HOXD13 (NHD13) mutation as compared to an adult-biased Runx1 loss-of-function mutation. Through bulk and single-cell sequencing approaches, we found that Flt3^ITD activated vastly different target genes and different target enhancers when paired with NHD13 as compared to Runx1 mutations. Flt3^ITD/NHD13 elicited a robust type I interferon (IFN-1) response that was not evident in Flt3^ITD/Runx1 mutant progenitors and has not been described for other models of Flt3^ITD cooperation.^6,7,10,11,13 Ifnar1 (IFN-1 receptor) deletion depleted NHD13-dependent pre-AML cells and reduced expression of genes that have been shown to drive AML self-renewal (eg, Cdk6 and Msi2). This impeded AML initiation and extended survival. Our findings show that FLT3^ITD coopts different mechanisms of transformation when it pairs with a pediatric-biased NUP98 fusion as compared to adult-biased mutations. Furthermore, the data uncover a therapeutically actionable IFN-1 sensitivity in NUP98-translocated AML.

All mouse lines are previously published and were obtained from the Jackson Laboratory.^26-29 

Cells were isolated, stained, analyzed, and transplanted as previously described.^10,30-35 Western blotting, cell cycle, and annexin V assays were performed as previously described in the study by Porter et al.¹⁰ Expanded descriptions of these previously published methods are available in the supplemental Methods, available on the Blood website. Extreme Limiting Dilution Analysis was used to calculate repopulating cell frequencies and to compare genotypes.³⁶ 

Bone marrow cells were c-kit selected and stained with tagged Total-Seq B antibodies (BioLegend). We isolated Lineage⁻c-Kit⁺ cells (LK) by using flow cytometry and generated libraries with 10× Genomics Chromium Next GEM Single Cell 3′ v3.1, Chromium Single Cell 3′ Feature Barcode Library, Dual Index NT Set A, and Dual Index TT Set A kits according to the manufacturer’s instructions. Libraries were sequenced, aligned, and filtered as previously described.^30,37 We performed Iterative Clustering and Guide-gene Selection (ICGS, version 2) with the AltAnalyze toolkit.³⁸ Gene Set Variation Analysis was performed using a Gaussian kernel.³⁹ Human RNA-seq data were procured from TARGET,¹ St. Jude Cloud,⁴⁰ or The Cancer Genome Atlas.² 

RNA was isolated with RNeasy plus micro kits (Qiagen). Libraries were generated with Clontech SMARTer kits, sequenced on a HiSeq3000 sequencing system, and aligned to mouse Ensembl release 76 top-level assembly using STAR version 2.0.4b.⁴¹ Linear modeling (limma/voom) was used to compare gene expression across samples.⁴² False discovery rates (FDRs) were calculated using the Benjamini-Hochberg method. Pathway analysis was performed with Reactome,⁴³ Generally Applicable Gene-set Enrichment⁴⁴ or Gene Set Enrichment Analysis.⁴⁵ The IFN-1 target gene set was previously described in the study by Li et al.³⁰ ATAC-seq and ChIPmentation libraries were generated and analyzed as previously described^30,31,46,47 using established pipelines.^48-50 

Survival of Vav1-Cre; Flt3^ITD; NHD13; Ifnar1^+/− and Vav1-Cre; Flt3^ITD; NHD13; Ifnar1^−/− mice was compared with the log-rank test. For retrovirus driven leukemias, 500 000 LK cells were isolated from P0 mice, transduced with MSCV-Flt3^ITD-IRES-mCherry retrovirus, and split among 5 recipients per transduction. Three independent replicates were performed.

We used transgenic and targeted mouse alleles to test whether Flt3^ITD has distinct effects on hematopoiesis when it cooperates with a pediatric-biased NUP98 fusion protein as compared to an adult-biased Runx1 deletion. We crossed Flt3^ITD knock-in mice with Vav1-NUP98-HOX13 (NHD13) transgenic mice to yield Flt3^ITD/NHD13 compound mutants.^27,29 We separately generated Vav1-Cre;Flt3^ITD/+;Runx1^fl/+ (Flt3^ITD/Runx1^DEL) mice.¹⁰ We quantified hematopoietic stem cells (HSCs; CD150⁺CD48⁻Lineage⁻Sca1⁺Kit⁺) and multipotent progenitors (MPPs; CD48⁺Lineage⁻Sca1⁺Kit⁺) in compound and single-mutant littermates at postnatal day (P)14. This age was chosen because NUP98-t is most frequently associated with early-mid childhood AML. Young adult mice were subsequently analyzed as well. Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL mutation combinations had similar effects on HSCs but dichotomous effects on MPPs. HSCs were severely depleted in both Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL mice at P14 (Figure 1A; supplemental Figure 1). In contrast, Flt3^ITD/Runx1^DEL caused a dramatic expansion of the MPP population, whereas Flt3^ITD/NHD13 did not (Figure 1B). The FLT3^ITD inhibitor, gilteritinib, partially suppressed MPP expansion in Flt3^ITD/Runx1^DEL mice after 10 days of treatment (P21-31), whereas it did not significantly reduce MPP numbers in similarly treated Flt3^ITD/NHD13 mice (supplemental Figure 2). Similar changes in HSC and MPP numbers were also observed in 8-week-old adult mice (Figure 1C-D), indicating that the impact of these mutations was not restricted to early postnatal stages of ontogeny.

Because HSCs were almost completely depleted in both Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL mice, we tested whether Flt3^ITD/NHD13 or Flt3^ITD/Runx1^DEL MPPs acquire ectopic repopulating activity that could sustain hematopoiesis. We transplanted 200 sorted MPPs from P14 Flt3^ITD/NHD13 or Flt3^ITD/Runx1^DEL mice, along with 300 000 wild-type CD45.1 competitor bone marrow cells, into lethally irradiated CD45.1 recipient mice. Flt3^ITD/NHD13 MPPs were capable of long-term myeloid and lymphoid reconstitution (Figure 1E-I). In contrast, Flt3^ITD/Runx1^DEL MPPs failed to repopulate a majority of irradiated recipient mice despite their expanded numbers (Figure 1E-I). Thus, Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL MPPs are functionally distinct, with NHD13, but not Runx1^DEL, conveying ectopic repopulating activity.

Prior studies have described overlapping cooperative target gene profiles for Flt3^ITD when it pairs with Tet2, Runx1, or Npm1 mutations.^10,11,13 This suggests that Flt3^ITD might engage a core set of AML effectors, irrespective of which mutations it cooperates with. However, the distinct effects of Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL on MPP function suggest that Flt3^ITD may engage different effectors when it cooperates with pediatric fusion proteins, such as NHD13. To test this, we performed RNA-seq on P14 Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL MPPs, as well as wild-type and single-mutant littermate controls. We initially focused on P14 to maintain consistency with the phenotypic and functional studies described above. Principal component analysis revealed strikingly different gene expression profiles in Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL MPPs relative to one another and to single-mutant controls (Figure 2A). We identified subsets of genes that were significantly activated in either Flt3^ITD/NHD13 or Flt3^ITD/Runx1^DEL MPPs relative to wild-type controls (defined as FDR <0.05, fold change ≥2, Figure 2B; supplemental Table 1). We then defined “cooperative target genes” as those that were significantly increased in compound mutant MPPs relative to each respective single mutant population (eg, increased in Flt3^ITD/NHD13 vs Flt3^ITD and NHD13 alone; FDR <0.05). We observed very little overlap between Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL cooperative target gene sets (Figure 2C). As an alternative approach, we calculated “interaction scores” for each expressed gene as previously described by Liu et al.⁵¹ This scoring system identified genes within Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL MPPs that exhibited greater changes in expression than could be explained through additive effects of each mutation (supplemental Figure 3A). Again, we observed very little overlap between genes that were cooperatively regulated by Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL (supplemental Figure 3B-C; supplemental Table 1).

Next, we tested whether differences between Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL target gene expression were age specific. We previously showed that Flt3^ITD/Runx1^DEL has only modest effect on gene expression in P0 MPPs.¹⁰ In contrast, gene set variation analysis (GSVA)³⁹ showed that Flt3^ITD/NHD13 cooperative target genes were already induced by P0 in Flt3^ITD/NHD13 mice (supplemental Figure 3D). Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL cooperative target genes were maintained with similar, genotype-specific patterns in 4-week-old MPPs (supplemental Figure 3E-F). Thus, Flt3^ITD has distinct cooperative interactions with NHD13 and Runx1^DEL across multiple ages.

We used ChIPmentation and ATAC-seq to test whether Flt3^ITD has cooperating mutation-specific effects on the epigenome. To this end, we identified active enhancers (ie, ATAC-seq peaks with overlapping histone H3K4me1 and histone H3K27ac peaks) in Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL MPPs. Among these, 2928 Flt3^ITD/NHD13 MPPs, and 1518 enhancers in Flt3^ITD/Runx1^DEL MPPs, were hyperactivated relative to wild-type MPPs based on H3K27ac levels that were significantly elevated in the compound mutants with FDR <0.1 (Figure 2D-E; supplemental Table 2). We identified 90 enhancers that were cooperatively activated by Flt3^ITD/NHD13 mutations and 244 enhancers that were cooperatively activated by Flt3^ITD/Runx1^DEL (ie, significantly greater H3K27ac levels in double mutants compared with every single mutant with FDR <0.1) (Figure 2F-G; supplemental Figure 4A-D). Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL cooperative target enhancers were largely nonoverlapping (Figure 2H). Both enhancer classes were enriched for ETS and RUNX1 binding sites. Flt3^ITD/NHD13 target enhancers were also enriched for NKX, GATA, and Zinc-finger family transcription factor binding sites that were not observed in Flt3^ITD/Runx1^DEL target enhancers (supplemental Figure 4E). In contrast, Flt3^ITD/Runx1^DEL target enhancers were enriched for STAT, TCF, AP-1, and IRF binding sites that were not observed in Flt3^ITD/NHD13 target enhancers (supplemental Figure 4F). The latter finding is consistent with data from human AML showing that in adults, FLT3^ITD activates AP-1–dependent enhancers via the MAP-kinase pathway.⁵² Together, these data show that Flt3^ITD activates distinct enhancers, via distinct complements of transcription factors, when paired with a pediatric-biased NUP98 fusion as compared to an adult-biased–Runx1 loss-of-function mutation.

We next sought to identify molecular pathways that might be uniquely activated in Flt3^ITD/NHD13 MPPs, given the limited overlap between Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL cooperative target genes. We performed Reactome pathway analysis on cooperative Flt3^ITD/NHD13 target genes and found striking enrichment for IFN-1 target genes (Figure 3A). Indeed, almost half of all Flt3^ITD/NHD13 cooperative target genes were IFN-1 targets, based on a previously compiled gene set (Figure 3B).³⁰ IFN-1 target expression was much higher in Flt3^ITD/NHD13 MPPs than in any other genotype (Figure 3C). Gene set enrichment analysis confirmed that IFN-1 targets were activated by both Flt3^ITD and NHD13 alone, but they were further activated when the mutations cooperated (Figure 3D). Interestingly, Flt3^ITD/Runx1^DEL cooperation suppressed the IFN-1 signature relative to Flt3^ITD alone (Figure 3D). Serum from P14 Flt3^ITD/NHD13 contained high levels of IFN-α (Figure 3E) but not IFN-β (data not shown). IFN-1 target gene activation was evident even by P0 in Flt3^ITD/NHD13 MPPs (Figure 3F). Thus, Flt3^ITD cooperates with NHD13 but not Runx1^DEL to induce IFNα expression and IFN-1 target genes.

Our functional studies and bulk RNA-seq data raised the question of whether Flt3^ITD differentially reprograms the differentiation trajectories of hematopoietic progenitors in the setting of NHD13 and Runx1^DEL mutations. Since bulk RNA-seq and MPP surface markers might not adequately capture the transcriptional heterogeneity of mutant progenitors, we performed CITE-seq to better delineate how Flt3^ITD/NHD13 and Flt3^ITD/Runx1^DEL mutations cooperate to reprogram cell identity. We profiled control (×2), Flt3^ITD (×2), Runx1^DEL, Flt3^ITD/Runx1^DEL, NHD13, and Flt3^ITD/NHD13 LK cells at P14 (supplemental Table 3). We used Iterative Clustering and Guide Gene Selection, version 2 (ICGS2) to cluster the cells in an unsupervised manner and then overlaid surface marker phenotypes for various progenitor populations based on tagged feature barcodes (Figure 4A; supplemental Figure 5A).^38,53 We observed strong concordance between the replicate control and Flt3^ITD samples (Figure 4B). Several clusters within the control LK cells aligned with well-defined progenitor populations, based on both surface marker phenotypes and gene expression (Figure 4A-B; supplemental Figure 5B; supplemental Table 4). These included cluster 18 (HSCs/MPPs), cluster 25 (MPP), and clusters 2 and 11 (granulocyte-monocyte progenitors [GMPs]) (supplemental Figure 5B). Control and Runx1^DEL samples had very similar cluster distributions (Figure 4B), indicating that heterozygous Runx1^DEL has a very limited effect on hematopoiesis, consistent with the bulk RNA-seq data.

Next, we compared the cluster distributions of Flt3^ITD, Flt3^ITD/NHD13, and Flt3^ITD/Runx1^DEL progenitors to determine whether each genotype confers unique cell identities, and the degree to which mutant progenitors resemble normal progenitors. By itself, Flt3^ITD caused the expansion of a GMP-like cluster (#11) that was also evident in wild-type mice, and it led to the emergence of MPP-like clusters (#7 and #10) that were not present in wild-type mice (Figure 4A-C). Concomitant Flt3^ITD/Runx1^DEL mutations markedly expanded clusters 10 and 11 at the expense of other normal populations without causing any additional clusters to emerge (Figure 4B-C; supplemental Figure 5C-D). Thus, Runx1^DEL largely amplifies Flt3^ITD-dependent changes in hematopoiesis rather than conveying new emergent transcriptional states.

The Flt3^ITD/NHD13 interaction had markedly different effects on cell identity compared with Flt3^ITD/Runx1^DEL. NHD13 caused a cluster (#13) to emerge that was not present in either wild-type or Flt3^ITD mice (Figure 4A-C). This cluster contained cells with MPP surface marker phenotypes despite clear transcriptional differences between it and normal MPPs (supplemental Figure 5C-D). Flt3^ITD/NHD13 cooperation expanded the emergent NHD13-specific population, but it did not cause any additional populations to form, nor did it cause the Flt3^ITD-specific cluster 10 to emerge (Figure 4B-C). We used marker genes affiliated with clusters 10 and 11 (Flt3^ITD/Runx1^DEL) and cluster 13 (NHD13 and Flt3^ITD/NHD13) to test whether the cooperating mutations activate different transcriptional programs and enhancer elements. Reactome pathway analysis showed enrichment of known FLT3^ITD/STAT5 target genes, as well as p21 and p53 response genes, in the Flt3^ITD- and Flt3^ITD/Runx1^DEL-specific cluster 10 (Figure 4A; supplemental Table 4). In contrast, it showed enrichment of IFN-1 target genes, such as Irf7, in the NHD13- and Flt3^ITD/NHD13-specific cluster 13, in addition to canonical NUP98-fusion protein targets, such as Pbx3 (Figure 4A,D; supplemental Table 4). Enhancers that mapped within 100 kb of cluster 10 and 11 marker genes showed elevated H3K27ac in Flt3^ITD/Runx^DEL mice relative to all other genotypes, whereas enhancers that mapped near cluster 13 genes had elevated H3K27ac in Flt3^ITD/NHD13 mice (Figure 4E; supplemental Figure 6). These data again show that Flt3^ITD has profoundly different effects on gene expression, enhancer activation, and cell identity when it cooperates with NHD13 as compared to Runx1^DEL mutations. Furthermore, they implicate cells within cluster 13 as putative cells of origin for Flt3^ITD/NHD13-developed AML.

To our surprise, Flt3^ITD/NHD13 mice did not develop rapid, fully-penetrant AML, in contrast to a prior study.¹² This complicated our efforts to test whether the emergent CITE-seq population (cluster 13) contained bona fide pre-AML cells. The Flt3^ITD allele used in this study (Flt3^tm1Dgg)²⁷ differs from the allele used previously to generate AML (Flt3^tm1.1Dosm)¹² in that it retains a floxed PGK-neo cassette in intron 15 that could dampen Flt3^ITD expression (Figure 5A). We therefore deleted the PGK-neo cassette with Vav1-Cre. Mice with the resulting “Flt3^ITDΔ” allele had elevated STAT5 and extracellular signal-regulated kinase (ERK) phosphorylation in MPPs, relative to wild-type and Flt3^ITD mice (Figure 5B). Furthermore, Flt3^ITDΔ/NHD13 mice now developed rapid, fully penetrant AML within 4 to 6 weeks after birth (Figure 5C). AML cells ectopically expressed B220 (CD45R), consistent with the Flt3^tm1.1Dosm model (Figure 5D).¹² 

To test whether the CITE-seq-defined, NHD13-specific population (Figure 4B, cluster 13) was enriched for pre-AML cells, we sought surface markers that would enable isolation by flow cytometry. CD317 (Bst2) expression was significantly elevated in Flt3^ITD/NHD13 cluster 13 cells, relative to NHD13 cells and all other clusters (Figure 5E). CD317 could therefore potentially enrich for pre-AML cells. We isolated CD317⁺LK or unfractionated bone marrow cells from P14 Flt3^ITD/NHD13 mice, treated the cells briefly with Tat-Cre to delete the PGK-neo, and then performed limit dilution assays (Figure 5F-G).³⁶ The CD317⁺LK fraction was highly enriched for engrafting cells (Figure 5H), and 12 of 14 engrafted recipients died of AML. These data confirm that the transcriptionally unique, IFN-1 responsive cell population in NHD13 and Flt3^ITD/NHD13 mice can initiate AML. They also raise the question of whether further cooperative changes in gene expression might be observed if CITE-seq assays were performed with Flt3^ITDΔ (ie, after removal of the PGK-neo cassette) rather than Flt3^ITD. Of note, the Flt3^ITD/Runx1^DEL mice described above carry Vav1-Cre and could be more appropriately described as Flt3^ITDΔ/Runx1^DEL. A summary of these alleles is provided in supplemental Table 5.

We repeated CITE-seq assays with P14 control, Flt3^ITDΔ, NHD13, and Flt3^ITDΔ/NHD13 LK cells so that we could test whether enhanced FLT3^ITD signaling elicits additional cooperative changes in gene expression, and for further comparison with Flt3^ITDΔ/Runx1^DEL. We included B220 and CD317 as surface barcode markers and excluded B220 from the flow cytometry lineage cocktail. Both markers were more highly expressed in Flt3^ITDΔ/NHD13 LK cells than in other genotypes by both flow cytometry and CITE-seq (supplemental Figure 7). The B220⁺ population may mark nascent AML cells (supplemental Figure 7A). We integrated single-cell transcriptomes from the original analysis for comparison purposes. Flt3^ITDΔ/NHD13 dramatically reprogrammed hematopoiesis, yielding 2 prominent clusters, numbers 15 and 33, that expressed MPP3/4 surface markers but did not cluster near normal hematopoietic progenitors (Figure 5I; supplemental Figure 8A,E; supplemental Table 4). They accounted for a significantly greater percentage of LK cells in Flt3^ITDΔ/NHD13 mice than was observed in single mutants or Flt3^ITD/NHD13 (Figure 5J), thus demonstrating greater cooperation between the Flt3^ITDΔ and NHD13 alleles than was observed with the weaker Flt3^ITD allele. Indeed, only 23% of the cells from the initial CITE-seq cluster 13 (Figure 4) mapped to cluster 33, and <0.2% mapped to cluster 15 (supplemental Figure 8B). Runx1 deletion amplified Flt3^ITDΔ-driven expansion of MPP-like cluster 26 (supplemental Figure 8D-E). There was no overlap between clusters that emerged in Flt3^ITDΔ/NHD13 and Flt3^ITDΔ/Runx1^DEL mice. Reactome pathway analysis redemonstrated enrichment for IFN-1 targets among cluster 33 marker genes and RAS pathway genes among cluster 15 markers (Figure 5K). Single-cell GSVA confirmed that IFN-1 target genes are significantly elevated in Flt3^ITDΔ/NHD13 MPPs (based on CITE-seq tags) relative to Flt3^ITDΔ/Runx1^DEL and single-mutant MPPs, consistent with the prior analyses (Figure 5L).

We next tested whether genes that mark Flt3^ITDΔ/NHD13 clusters are selectively expressed in NUP98-t mouse and human AML. We performed bulk RNA-seq on Flt3^ITDΔ/NHD13 and previously collected Flt3^ITDΔ/Runx1^DEL/DEL AML specimens⁵⁴ (deletion of both Runx1 alleles was required for frank transformation). Principal component analysis revealed dramatic differences in gene expression between these genetically distinct leukemias (supplemental Figure 9A). GSVA showed that cluster 15 and 33 marker genes were enriched in Flt3^ITDΔ/NHD13 AML, but not in Flt3^ITDΔ/Runx1^DEL/DEL AML, as expected (Figure 5M). Importantly, cluster 33 (but not cluster 15) marker genes were highly expressed NUP98-translocated AML from 2 large pediatric AML RNA-seq data sets (TARGET and St. Jude),^1,40 relative to KMT2A-rearranged and NPM1-mutated AML (Figure 5N; supplemental Figure 9B). The latter 2 mutations hyperactivate HOX genes, much like NUP98-t, and NPM1 frequently cooperates with FLT3^ITD making these good benchmarks for comparison. Cluster 33 (but not cluster 15) marker genes were also highly expressed in pediatric but not adult FLT3^ITD-positive AML (The Cancer Genome Atlas) (Figure 5O; supplemental Figure 9C). Together, these observations show that FLT3^ITD and NUP98-t cooperate to elicit changes in gene expression and cell identity that are maintained in fully transformed AML and conserved in humans. Moreover, the changes are specific to pediatric FLT3^ITD-positive AML. Flt3^ITDΔ/NHD13 mutations hyperactivate IFN-1 target genes, raising the question of whether IFN-1 promotes NUP98-t–AML initiation.

We tested whether IFN-1 sustains Flt3^ITD/NHD13 pre-AML cells by measuring HSC, MPP, pregranulocyte-monocyte progenitor (pGM), and GMP numbers (supplemental Figure 10A-B) in Flt3^ITD/NHD13 mice, on either Ifnar1^+/− or Ifnar1^−/− backgrounds, at P0 or P14. Ifnar1 encodes the IFN-1 receptor. Ifnar1^+/− served as a control, rather than Ifnar1^+/+, to simplify breeding strategies. We confirmed that Ifnar1 heterozygosity had no perceivable effect on hematopoiesis in Flt3^ITD/NHD13 mice (supplemental Figure 10C-G). Ifnar1 deletion caused a near-complete loss of MPP2 and MPP3/4 subpopulations (CD150⁺CD48⁺Lineage⁻Sca1⁺Kit⁺ and CD150⁻CD48⁺Lineage⁻Sca1⁺Kit⁺, respectively) at P14, particularly in mice with NHD13 or Flt3^ITD/+; NHD13 genotypes (Figure 6A-B). Ifnar1 deletion also caused a concomitant expansion of phenotypic pGMs (Lineage^–Sca1^–Kit⁺CD150^–CD105^–CD16/32^–) and GMPs (Lineage⁻Sca1⁻Kit⁺CD150⁻CD105⁻CD16/32⁺) in Flt3^ITD/NHD13 mice (Figure 6C-D), without altering pGM or GMP proliferation or death rates (supplemental Figure 11A-D). Similar changes were observed at P0 (supplemental Figure 11E-H). Ifnar1 deletion reduced the colony-forming activity of Flt3^ITD/NHD13 MPP3/4s (Figure 6E), but competitive transplantation assays and limiting dilution transplants unexpectedly showed no difference in the repopulating activity of Flt3^ITD/+; NHD13; Ifnar1^+/− and Flt3^ITD/+; NHD13; Ifnar1^−/− bone marrow cells despite profound MPP depletion in the absence of Ifnar1 (Figure 6F-G). Instead, phenotypic Flt3^ITD/NHD13 pGMs repopulated recipient mice and offset the loss of MPPs (Figure 6H-J). These observations suggest that functional properties of hematopoietic progenitors, such as repopulating activity, may be decoupled from conventional surface marker phenotypes in Flt3^ITD/NHD13 mice, especially after Ifnar1 deletion (Figure 6K).

To reconcile this discrepancy, we performed additional CITE-seq experiments to clarify population-specific, IFN-1-dependent transcriptional changes using an approach that would not be confounded by changes in surface marker expression. We profiled control, Flt3^ITD, NHD13, and Flt3^ITD/NHD13 progenitors, all on Ifnar1^+/− or Ifnar1^−/− backgrounds, clustered the cells with ICGS2 and overlaid established surface marker phenotypes for MPP and myeloid progenitor populations. As in Figure 4, we observed an emergent NHD13-dependent population (cluster #19 in this case) that expanded in the presence of Flt3^ITD (Figure 7A-B; supplemental Figure 12). The surface marker phenotypes of NHD13 and Flt3^ITD/NHD13 cells changed with Ifnar1 deletion, shifting from a predominantly MPP2 phenotype to pGM/GMP phenotypes (Figure 7C-D). This likely explains why Flt3^ITD/+; NHD13; Ifnar1^−/− bone marrow cells retained repopulating activity despite near-complete loss of MPPs; the pGM/GMP-like cells were transcriptionally similar to Flt3^ITD/+; NHD13; Ifnar1^+/− MPPs, irrespective of their different surface marker phenotypes. Ifnar1 deletion caused cluster 19 to contract in both NHD13 and Flt3^ITD/NHD13 mice (Figure 7B). It reduced the expression of the direct NUP98-NSD1 target Cdk6;⁵⁵ as well as self-renewal genes such as Gata2 and Msi2, and inflammatory genes such as Trim30a and Irf7 (Figure 7E; supplemental Table 6). Thus, IFN-1 modulates the size of an emergent pre-AML population in NHD13 and Flt3^ITD/NHD13 mice, and it enhances the expression of potential effectors of AML initiation.

We next tested whether IFN-1 signaling promotes Flt3^ITD/NHD13-driven AML. We generated Flt3^ITDΔ; NHD13; Ifnar1^+/− and Flt3^ITDΔ; NHD13; Ifnar1^−/− mice and compared their survival. Flt3^ITDΔ; NHD13; Ifnar1^−/− mice survived significantly longer than Flt3^ITDΔ; NHD13; Ifnar1^+/− mice (Figure 7F). We next took an orthogonal strategy that allowed us to activate Flt3^ITD after NHD13, rather than in parallel, to better mimic the sequential acquisition of mutations that occurs in humans. We isolated P0 LK cells from NHD13; Ifnar1^+/− and NHD13; Ifnar1^−/− mice and transduced them with MSCV-FLT3^ITD-IRES-mCherry retrovirus (Figure 7G). We transplanted the transduced cells into irradiated recipient mice and monitored survival. We observed similar engraftment efficiencies for each genotype (Figure 7H), but recipients of NHD13; Ifnar1^−/− cells survived significantly longer than recipients of NHD13; Ifnar1^+/− cells (Figure 7I). The mice died of acute megakaryoblastic leukemia (AMKL), irrespective of the Ifnar1 genotype, based on CD41 expression within the mCherry-positive population (supplemental Figure 13A-B). STAT5 and ERK phosphorylation in the AMKL cells was comparable to levels in Flt3^ITDΔ; NHD13 AML (supplemental Figure 13C). RNA-seq followed by GAGE pathway analysis showed that Ifnar1-deficient leukemias had the reduced expression of genes associated with cell cycle progression and RNA processing/splicing (supplemental Figure 13D; supplemental Table 7). IFN-1, therefore, promotes AML/AMKL initiation.

The bulk RNA-seq and CITE-seq studies all predict that IFN-1 dependence should be specific to Flt3^ITD/NHD13 AML initiation and not observed in Flt3^ITD/Runx1^DEL AML. The Ifnar1 and Runx1 alleles are closely linked on mouse chromosome 16, making it impractical to test this hypothesis through conventional breeding strategies. Instead, we isolated HSCs/MPPs from Flt3^ITDΔ; Ifnar1^+/− and Flt3^ITDΔ; Ifnar1^−/− mice at P14 and inactivated Runx1 by electroporating cells with a Runx1-specific–gRNA/Cas9 complex (supplemental Figure 14A). The cells were subsequently transplanted into lethally irradiated recipient mice. Editing efficiencies of >70% were confirmed by next-generation sequencing. Recipients began dying of AML at ∼2 months after the transplant (supplemental Figure 14B-D). Ifnar1 deletion did not affect survival. Thus, Flt3^ITD/NHD13 cooperation conveys sensitivity to IFN-1 inactivation that is not observed with Flt3^ITD/Runx1^DEL cooperation.

Our findings add a layer of complexity to established paradigms for oncogenic cooperation during AML initiation. Prior work has shown that Flt3^ITD elicits cooperative changes in gene expression and enhancer activation when paired with Npm1, Tet2, or Runx1 mutations,^10,11,13 and the cooperative target gene profiles for these mutations overlap.¹⁰ Activating Nras mutations (eg, Nras^G12D) demonstrate similar patterns of cooperation with either Tet2 or Ezh2 mutations.^51,56 However, whereas prior studies examined interactions involving mutations that occur most commonly in adult AML, our current study suggests that interactions involving pediatric AML fusion proteins may be fundamentally different. Cooperative Flt3^ITD/NHD13 target genes barely overlap with cooperative Flt3^ITD/Runx1^DEL target genes (Figure 2). Furthermore, Flt3^ITD/NHD13 cooperation creates an emergent progenitor population that does not resemble normal hematopoietic progenitors or Flt3^ITD-mutant MPPs despite the expression of MPP surface markers (Figures 4 and 5). This emergent population can give rise to AML (Figure 5H), and cluster-specific marker genes are conserved in human pediatric NUP98-t AML (Figure 5M-N). The data show that Flt3^ITD coopts different mechanisms of transformation when it pairs with a pediatric-biased NUP98-t than with other, adult-biased AML mutations. These differences may ultimately create age- and cooperating mutation-specific vulnerabilities.

Our data reveal a context-specific role for IFN-1 signaling in AML initiation. We have shown that IFN-1 modulates the size of the emergent NHD13-dependent population of preleukemic progenitors, promotes the expression of self-renewal genes within that population, and potentiates leukemogenesis. Additional studies will be needed to test whether IFN-1 dependence extends to other cooperative interactions, such as those involving Nras^G12D or other NUP98 fusions. IFN-1 signaling is clinically actionable via IFNAR1/2 blocking antibodies or inhibitors of downstream signal transduction proteins (eg, the TYK2 inhibitor deucravacitinib).^57,58 Further studies will help identify patients who are most likely to benefit from IFN-1–directed therapies.

This work was supported by grants to J.A.M. from the National Institutes of Health (NIH), National Heart, Lung, and Blood Institute (R01 HL152180 and R01 HL136504) and the NIH, National Cancer Institute (NCI) (U01 CA267031), Alex’s Lemonade Stand Foundation (“A” Award), Hyundai Hope On Wheels, the Alvin J. Siteman Cancer Center Investment Program (supported by the Foundation for Barnes-Jewish Hospital and NIH, NCI Cancer Center Support Grant P30 CA091842), the Leukemia and Lymphoma Society, and the Children’s Discovery Institute of Washington University and St. Louis Children’s Hospital. J.A.M. is a Leukemia and Lymphoma Society Scholar.

Contribution: J.A.M. designed and oversaw all experiments and secured the funding; J.A.M. and Y.L. conducted experiments, interpreted data and wrote the manuscript; W.Y. performed all bioinformatic analyses; R.M.P., E.B.C., E.D., P.R.-L., and J.M.-C. performed experiments and interpreted data; and all authors reviewed and edited the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Jeffrey A. Magee, Washington University School of Medicine, 660 S. Euclid Ave, Box 8208, St. Louis, MO 63110; e-mail: mageej@wustl.edu.