Cohesin haploinsufficiency and inv(16) cooperate to reinforce Fli1 gene transcriptional programs during acute myeloid leukemia initiation and maintenance Authors: Alison E Meyer, Katelyn E Heimbruch, Kirthi Pulakanti, Cary Stelloh, Nataly Cruz-Rodriguez, Josiah Murray, Dandan Wang, Lisa Latzko, Sebastian Vosberg, Philipp A Greif, Lucio H Castilla, John A Pulikkan, Subramaniam Malarkannan, Sridhar Rao Published: 2026 ABSTRACT AbstractCore binding factor (CBF) acute myeloid leukemias typically harbor the translocations t(8;21) or inv(16). As cohesin mutations are less commonly observed with inv(16) than with t(8;21), we hypothesized that they may negatively impact inv(16)-driven AML. Using a mouse model of inv(16) with haploinsufficiency of the cohesin subunit Smc3, we paradoxically found that inv(16); Smc3Δ/+ mice have a reduced leukemic latency compared to inv(16); Smc3+/+ mice, disproving our initial hypothesis and instead suggesting a role for cohesin loss in enhancing inv(16)-driven disease. Consistent with the known role of cohesin haploinsufficiency in altering chromatin accessibility, we demonstrated an increase in chromatin accessibility in inv(16); Smc3Δ/+ hematopoietic stem and progenitor cells (HSPCs) prior to leukemia development, with an enrichment for Fli1 DNA binding motifs. Through scRNA-seq on pre-leukemic HSPCs, we observe an increase in Fli1 expression and enhanced Fli1 target expression in ST-HSCs. We further show that Fli1 is essential for the maintenance stage of inv(16); Smc3Δ/+ AML. Our data demonstrate a role for cohesin loss in enhancing the aggressiveness of inv(16)-driven AML and identify Fli1 as a previously unrecognized therapeutic vulnerability in cohesin-mutated AML.Subject terms: Cancer models, Cancer epigenetics Introduction Acute myeloid leukemia (AML) is a bone marrow cancer affecting ~20,000 people annually in the United States, with global rates of incidence and mortality rising [1]. Multiple studies utilizing whole genome or whole exome sequencing on patient samples have identified causal or contributing mutations [2–6] and have revealed a genetically diverse disease [4]. On average, each patient carries 5–15 different mutations [4] and therefore has a unique landscape of somatic mutations and clonal composition. How acquired mutations work together to initiate the transformation to leukemia and/or maintain the disease is a critical knowledge gap in AML. Thus, further study has the potential to identify novel therapeutic pathways. Approximately 15% of adult AML patients have mutations in the core binding factor complex [7–10]. Core binding factor (CBF) is a heterodimeric transcriptional complex comprised of RUNX1, 2, or 3 and CBFβ. This complex is critical for establishing and maintaining hematopoietic stem cells and regulating differentiation through the recruitment of lineage-specific transcription factors [7–10]. Both subunits can be altered through translocations in AML, with RUNX1 being the target in t(8;21) and CBFβ in inv(16) [9, 11]. These chromosomal aberrations result in the expression of the fusion proteins RUNX1::RUNX1T1 (also known as AML1::ETO) in t(8;21) and CBFβ::SMMHC in inv(16). Either mutation is necessary but not sufficient for leukemogenesis, and multiple secondary or tertiary mutations can influence the rate of disease initiation or progression [7, 8, 11, 12]. Interestingly, while both fusion proteins are classically thought to act in a dominant negative manner toward native RUNX1 [10, 11, 13, 14], the co-mutational and transcriptional profiles of t(8;21) and inv(16) AMLs differ [14–19], and they often have different French-American-British classifications [8]. Thus, while the two core binding factor AMLs are usually classified together clinically and are typically considered favorable, they are often morphologically and genetically distinct. Of note, mutations in the cohesin complex are predominantly found in t(8;21) AMLs at a rate of 18–27% [15–19]. By contrast, cohesin complex mutations have not been readily detected in inv(16) AMLs [15–19]. The cohesin complex is a four-member ring-like complex comprised of the subunits SMC1A, SMC3, RAD21, and STAG2. Cohesin’s canonical function is to maintain sister chromatid cohesion; it also has well-documented roles in regulating chromatin looping and gene transcription [20, 21]. Cohesin mutations occur in AML at a rate of 10%, are mutually exclusive, and typically result in haploinsufficiency of the complex [3, 22–24]. Because AMLs harboring cohesin mutations show low levels of aneuploidy, disruption of their canonical function is not thought to contribute to disease [25–28]. Instead, cohesin mutations have been hypothesized to disrupt gene regulation in AML [21]. Due to the reported low co-occurrence of cohesin mutations with inv(16) in leukemia, we originally hypothesized that cohesin haploinsufficiency would be detrimental to inv(16) leukemogenesis or progression. In contrast to our hypothesis, we found that Smc3 haploinsufficiency significantly decreased AML latency in inv(16) mice. Our data suggest that cohesin mutations increase the accessibility of binding sites for the transcription factor Fli1 in pre-leukemic cells. This results in enhanced Fli1 target gene expression, likely contributing to accelerated inv(16)-driven disease. Our data further identify Fli1 as a potential therapeutic vulnerability in inv(16) and cohesin-mutated AMLs. Materials and methods A detailed materials and methods section can be found in the Supplementary Material Results Smc3 haploinsufficiency decreases the latency of inv(16) AML We initially hypothesized that cohesin haploinsufficiency would delay or prevent inv(16)-driven AML, as cohesin mutations and inv(16) are rarely co-occurrent. To test this hypothesis, we combined the inducible inv(16)+/56M [29] and Smc3fl/+ mouse models [26]. In these models, heterozygous expression of inv(16) and loss of one allele of Smc3 are driven by Mx1-Cre activation following pIpC injection, resulting in an approximate 50% loss of Smc3 expression [26, 30](Supplementary Fig. 1A). We monitored 15 inv(16)+/56M; Smc3+/+ (referred to as inv(16); Swt) and 24 inv(16)+/56M; Smc3Δ/+ mice (referred to as inv(16); Shet) for AML development following pIpC injection (Fig. 1A). Moribund mice were euthanized, and AML was confirmed when possible by staining of bone marrow samples, the presence of pale bones, hepatosplenomegaly, and the disruption of normal liver and spleen structure (Fig. 1B and Supplementary Fig. 1B–D). While AML was 100% penetrant in both genotypes, in contrast to our hypothesis, we observed a statistically significant decrease in latency in inv(16); Shet mice (14 weeks) compared to inv(16); Swt mice (20 weeks, Fig. 1A), suggesting Smc3 haploinsufficiency facilitates accelerated inv(16)-driven disease. Leukemias from both genotypes were morphologically similar (Fig. 1B), the mice displayed similar hepatosplenomegaly and blast infiltration of the liver/spleen (Supplementary Fig. 1B–D), and the leukemias were similarly differentiated as measured by flow cytometry (Supplementary Fig. 1E). Our data collectively disprove our initial hypothesis and instead suggest that Smc3 haploinsufficiency and inv(16) cooperatively decrease disease latency through effects on either the rate of leukemic transformation or disease progression. Smc3 haploinsufficiency decreases the latency of inv(16)-driven AML. A Survival curve showing decreased latency of inv(16); Shet mice (red line) vs. inv(16); Swt mice (black line). **p < 0.01, log-rank test. B Examples of bone marrow from a healthy wildtype (WT) animal (left panel), a leukemic inv(16); Swt (middle panel), and a leukemic inv(16); Shet animal (right panel). Images taken at 600x. Smc3 haploinsufficiency increases chromatin accessibility at ETS family binding sites in pre-leukemic inv(16) HSPCs To examine how cohesin mutations might contribute to inv(16)-mediated leukemic transformation, we performed several studies on pre-leukemic hematopoietic stem and progenitor cells (HSPCs, Lin-). Because cohesin has a known role in chromatin accessibility, we first performed ATAC (Assay for Transposase-Accessible Chromatin) sequencing. HSPCs were isolated from Cre- mice and were infected with retrovirus expressing tamoxifen-inducible Cre. Cre expression was induced 48 h after infection by administration of 4-hydroxytamoxifen. Knock-in of inv(16) and excision of one Smc3 allele were confirmed by genotyping, and ATAC sequencing was performed. We found an increased number of accessible chromatin sites in inv(16); Shet vs. inv(16); Swt HSPCs (Fig. 2A). Specifically, 33 133 accessible regions were unique to inv(16); Shet compared to 564 regions unique to inv(16); Swt (Fig. 2A and Supplementary Table 1), demonstrating that cohesin haploinsufficiency promoted a dramatic increase in the number of accessible sites detected. As cohesin mutations have been linked to alterations in hematopoietic transcription factor site accessibility [26–28, 31], we next performed a de novo motif analysis using HOMER to determine which transcription factor binding sites were enriched in the unique peaks in inv(16); Shet cells. Along with CTCF, which is well known to co-regulate chromatin organization with cohesin, the top 10 most enriched transcription factor sites included the AP-1 superfamily members Fos, Fra1, and Fra2, and multiple ETS family transcription factor members (Etv1, Etv4, Fli1, Elf5, PU.1, and Elf4; Fig. 2B). Because cells may have differentiated during retroviral infection with Cre, we verified our ATAC-seq results with freshly-isolated HSPCs following pIpC treatment. HOMER analysis again revealed the enrichment of ETS family transcription factors in inv(16);Shet differentially accessible chromatin regions (Supplementary Fig. 2), confirming that Smc3 haploinsufficiency results in increased chromatin accessibility at ETS binding sites. Smc3 haploinsufficiency increases chromatin accessibility at ETS family binding sites in pre-leukemic inv(16) HSPCs. A ATAC replicates were combined for all analyses (n = 3 for both genotypes). Venn diagram showing the number of shared and uniquely identified accessible chromatin peaks in inv(16);Swt (red) and inv(16);Shet (blue) HPSCs by ATAC sequencing. B HOMER analysis of the unique peaks in inv(16);Shet HPSCs shows ETS family transcription factor sites are prevalent in the top ten most enriched motifs by p-value. C ATAC replicates were combined for all analyses (n = 3 for each genotype). Venn diagram showing the number of shared and uniquely identified accessible peaks in Npm1cA/+; Swt (green) and Npm1cA/+; Shet (purple) HSPCs by ATAC sequencing. D HOMER analysis of the unique peaks in Npm1cA/+; Shet HSPCs shows ETS family transcription factor sites are prevalent in the top five most enriched motifs by p-value. E IGV tracks of the Fli1 gene showing enhanced accessibility (taller peaks) at the TSS in inv(16); Shet (in blue) vs. inv(16); Swt (in red) HSPCs. The scaling for all tracks displayed is identical. Additional peaks throughout the gene body and ~10 kb upstream of the TSS are observable in inv(16); Shet but not in inv(16); Swt cells. Regions of interest are boxed in green. mm10 was used for the genomic build. Tagpileup: top tracks for each genotype, Peaks called: bottom tracks for each genotype. The fusion protein resulting from inv(16), CBFβ::SMMHC, has been shown to interact with WT RUNX1 and the critical myeloid ETS transcription factors FLI1, ERG, PU.1, and ELF1 at promoter regions to regulate gene expression [10, 32], making the enrichment of ETS family motifs intriguing. Furthermore, we observed similar ATAC-seq results in a second model of co-mutational AML that we have previously studied, where the Npm1cA/+ mutation was combined with Smc3Δ/+ (Fig. 2C) [30]. HOMER analysis revealed the top five enriched motifs in the Npm1cA+; Shet uniquely accessible chromatin regions were all ETS family transcription factors (Fig. 2D). Given that ETS factor site accessibility was consistent with Smc3 haploinsufficiency across two separate driver oncogene models and in several published cohesin models [26, 28], we further examined the impact of ETS family transcription factors on inv(16) HSPC biology. Smc3 haploinsufficiency increases Fli1 expression and enhances Fli1 activity in pre-leukemic inv(16) HSPCs As accessible chromatin is generally associated with active transcription, we reasoned that elevated expression levels of ETS family transcription factors may contribute to gene dysregulation. Of the ETS factors in our top 10 most enriched sites in inv(16); Shet accessible chromatin (Fig. 2B), Etv1, Etv4, Fli1, Elf4, and PU.1 (encoded by Spi1) are known to be expressed in hematopoietic cells, while Elf5 is largely expressed in epithelial cells and was thus not considered further. The ETS domain, which facilitates DNA binding, is 96% identical between Erg and Fli1 [33], making separation of their effects difficult. For this reason, we included Erg in further analyses. Our ATAC-seq data revealed increased chromatin accessibility near the transcription start site (TSS) of Fli1, Elf4, Etv4, and Spi1, but not in Etv1 or Erg, in inv(16); Shet vs. Swt cells (Fig. 2E and Supplementary Fig. 3A, B, C and data not shown), consistent with increased gene expression. For Fli1, additional uniquely accessible chromatin regions were present in inv(16); Shet cells, with a peak approximately 10 kb upstream of the TSS and several throughout the gene body (Fig. 2E). Given the increased accessibility of genes encoding Ets family members, we performed qPCR to quantify mRNA levels of these genes in inv(16); Swt and Shet HSPCs. Interestingly, only Fli1 and Erg mRNA were significantly upregulated in inv(16); Shet vs. inv(16); Swt HSPCs (Fig. 3A and Supplementary Fig. 4A), however only Fli1 was upregulated at the protein level (Fig. 3B and Supplementary Fig. 4B). We also assayed the expression levels of several known Fli1 target genes and found Ccl2, Ccng1, Mdm2, Rab27b, and Lmo2 were upregulated in inv(16); Shet vs. inv(16); Swt HSPCs (Fig. 3C). Collectively, this work demonstrates that the Fli1 protein is overexpressed and drives the increased expression of known target genes in Smc3 haploinsufficient inv(16) HSPCs. Smc3 haploinsufficiency increases Fli1 expression and enhances Fli1 activity in pre-leukemic inv(16) HSPCs. A qPCR was performed for Fli1 mRNA on HSPCs isolated from inv(16); Swt and inv(16); Shet animals. The expression of Fli1 was compared to β-actin as a control, and relative expression was plotted. B Western blot showing Fli1 protein expression for inv(16); Swt and inv(16); Shet HSPCs (top panel). Fli1 expression was compared to β-actin as a control and quantified (lower panel). Due to the significant sequence homology between Fli1 and Erg, an antibody specific to Fli1 was used. C qPCR was performed for five known Fli1 transcription factor targets on inv(16); Swt and inv(16); Shet HSPCs. The expression of Ccl1, Ccng1, Mdm2, Rab27b, and Lmo2 mRNA was compared to β-actin as a control and relative expression was plotted. D Fli1 binding sites were mapped to their nearest genes (+/– 50 kb) in the 33 133 unique peaks in inv(16); Shet and the 14 027 peaks shared between inv(16); Swt and inv(16); Shet from Fig. 2A. GO pathway analysis was performed using ShinyGO v0.80 for the inv(16); Shet genes, using the shared genes as background. The pathways were ranked from the largest to the smallest number of genes. Only those pathways with an FDR < 10% were retained. The top five pathways are shown. All assays in this figure were performed for three inv(16); Swt and three inv(16); Shet animals four weeks post-pIpC injection. *p < 0.05, **p < 0.01, ***p < 0.001, Student’s t test. Data in A, B, C represent the mean +/– SEM. Given the enrichment of Fli1 binding sites in uniquely accessible chromatin in inv(16) Shet HSPCs, we hypothesized that such sites may identify pathways of potential importance to AML transformation. To this end, we identified chromatin accessible regions present only in inv(16);Shet and isolated those with a consensus binding motif for Fli1 using the HOMER motif discovery tool. We next mapped these sites to their nearest genes within +/– 50 kb and performed genome ontology (GO) pathway analysis using the ShinyGO gene set enrichment tool (Supplementary Table 2). Interestingly, the top five enriched pathway terms associated with Fli1 binding sites in the unique inv(16); Shet accessible chromatin included “Ras signaling pathway” (Fig. 3D). Several Ras pathway genes including Pla2g6, Kit, Igf1r, and Insr showed increased accessibility near the TSS in inv(16); Shet vs. inv(16); Swt cells (Supplementary Fig. 5A, B and data not shown). Of note, the most common co-occurring mutations in inv(16) AMLs are in NRAS and KRAS, which are less common in t(8;21) [15–19]. However, active RAS activity levels were similar in inv(16); Swt and inv(16); Shet HSPCs (Supplementary Fig. 5C), suggesting that Smc3 haploinsufficiency does not result in the direct activation of RAS itself. Instead, our data suggest that a subset of genes downstream of RAS activation may be similarly regulated by Fli1 in pre-leukemic inv(16); Shet cells. Other interesting GO pathways included “pathways in cancer”, “cytokine-cytokine receptor interaction”, and “human cytomegalovirus infection” (Fig. 3D), which were comprised of overlapping gene sets with known links to the regulation of AML (ex, cell cycle, apoptosis, and immune signaling; Supplementary Table 2). Collectively, our data suggest that Smc3 haploinsufficiency results in increased chromatin accessibility at Fli1 transcription factor binding sites and elevated Fli1 target expression, which increases RAS and other cancer-associated gene expression pathways in inv(16) HSPCs and may contribute to an increased rate of leukemogenesis. Fli1 is overexpressed and drives an enhanced gene expression program in inv(16); Shet HSCs As AML is thought to originate with mutations in hematopoietic stem cells or myeloid progenitors [34–37], we hypothesized that Fli1 and its effector genes would be differentially expressed in the most primitive hematopoietic cells. To address this hypothesis, we sorted Lin-, Sca1 + , c-Kit+ (LSK) cells from two independent inv(16); Swt and inv(16); Shet animals four weeks post-pIpC treatment and analyzed these cells by scRNA sequencing, prior to the animals developing AML. The LSK population contains short and long-term-hematopoietic stem cells (ST/LT-HSCs) and multipotent progenitors (MPPs) [38–40]. Cells clustered into nine main groups, with all groups containing cells from both genotypes (Fig. 4A and Supplementary Fig. 6A). Cells were annotated using the Haemopedia mouse RNAseq database [41]. Only cells that were annotated as MPPs (Lin-, Sca1 + , c-Kit + , CD34 + , Flt3 + ) and ST-HSCs (Lin-, Sca1 + , c-Kit + , CD34 + , Flt3-; hereafter referred to as HSCs) were retained. We were unable to identify LT-HSCs due to their low frequency, consistent with their lack of expansion in inv(16) mice [42]. In total, we analyzed 5 499 inv(16); Swt cells and 2 557 inv(16); Shet cells, which were separated into the MPP and HSC populations as shown in Table 1 and Fig. 4B. Cells were next analyzed for Fli1 expression based on cluster (Supplementary Fig. 6B). We found a significant upregulation of Fli1 in inv(16); Shet cells vs. inv(16); Swt cells in clusters 0, 1, 3, and 6 (Fig. 4C and Supplementary Table 3). Fli1 expression in clusters 2, 4, 5, 7 and 8 was not statistically different between genotypes (Supplementary Fig. 6C). Interestingly, when examined by cell type, although a trend toward increased Fli1 expression was observed in MPPs, this expression increase reached statistical significance only in HSCs (Fig. 4D). Collectively our data indicate that, while the overall cell types and numbers between inv(16)Shet and inv(16); Swt were similar, inv(16); Shet HSCs showed elevated Fli1 mRNA. Fli1 is overexpressed and drives an enhanced gene expression program in inv(16); Shet HSCs. A Seurat was used to group LSK-sorted cells from two inv(16); Swt and two inv(16); Shet into nine distinct clusters (0-8). Cells from inv(16); Shet mice are shown in blue, and those from inv(16); Swt mice are shown in salmon. In total, 2557 inv(16); Shet and 5 499 inv(16); Swt cells were used for analysis. B The SingleR tool was used to annotate the clusters using the Haemopedia mouse RNA sequencing reference transcriptomics dataset. C DESeq2 was used to test differential expression of Fli1 in each cluster using a significance cutoff of adj. p-val <0.05. The expression levels of Fli1 were plotted, and significantly increased expression was observed for inv(16); Shet cells (blue) vs. inv(16); Swt cells (red) in clusters 0, 1, 3, and 6. D DESeq2 was used to test differential expression of Fli1 mRNA by cell type in clusters 0, 2, 3, and 6. Significantly increased Fli1 expression was observed in inv(16); Shet (blue) vs. inv(16); Swt (red) HSCs. ***p < 0.001, **p < 0.01, Wilcoxon test. Table 1Number of cells of each type analyzed by scRNAseq.HetWTTotalMPP107525393614STHSC148229604442Total255754998056The number of cells identified as MPPs or HSCs is noted for each genotype. Number of cells of each type analyzed by scRNAseq. The number of cells identified as MPPs or HSCs is noted for each genotype. To determine if the increased expression of Fli1 was associated with increased expression of its target genes, or regulon, we applied the single-cell regulatory network inference and clustering (pySCENIC) program [43] to our scRNA sequencing data to identify transcription factor-driven regulons that were different between the two genotypes [43]. The activity of hundreds of regulons was scored for every single cell (regulon activity score). Consistent with the upregulation of several Fli1 target genes (Fig. 3C), Fli1 was identified as a significantly enriched regulon (normalized enrichment score ≥ 3) in all cells from clusters 0, 1, 3, and 6 of our scRNA seq data, with 203 high-confidence Fli1-activated genes identified (Supplementary Table 4). We hypothesized that there would be a quantitative difference in Fli1 regulon activity between the two genotypes, given the genotype-specific activation of select Fli1 targets observed in Fig. 3C. To address this hypothesis, we calculated the Z-score deviation from the average regulon activity score for the significant regulons that were also present as enriched motifs in our ATAC analysis from Fig. 2B. As shown in Fig. 5A, Fli1 showed a prominent Z-score increase in inv(16); Shet vs. inv(16); Swt cells (0.27 vs –0.22; p = 2.2 x 10–16), demonstrating that this regulon has increased activity in inv(16); Shet as compared to inv(16); Swt cells. The Fli1 regulatory network is a major contributor to gene expression in inv(16) HSCs. A PySCENIC was used to apply the mouse motif annotation database to our scRNA gene expression data. The area under the curve for each cell, representing the regulon activity score, was calculated with AUCell. Regulons that met a normalized enrichment score threshold ≥3, an AUC threshold of >0.05, and a rank threshold of 5000 were retained. The Z-score deviation from the AUC mean is shown in the heatmap for those motifs also identified by ATAC analysis in Fig. 2B. inv(16);Swt is shown in pink and inv(16);Shet in light blue. Average Z-score for inv(16);Swt = –0.22; Average Z-score for inv(16);Shet = 0.27. p = 2.22 × 10–16, Wilcoxon. B Box and whisker plot comparing log-normalized Fli1 expression between inv(16); Swt (red) and inv(16); Shet (blue) in clusters 0, 1, 3, and 6. The central solid horizontal line represents the median, the box indicates the interquartile range (IQR), and the whiskers represent 1.5*IQR. Individual data points (dots) represent single-cell expression values. Statistical significance was determined by a two-tailed Wilcoxon rank-sum test (p = 6.9 × 10–6). n = 5 650 total cells. The highest 25% of Fli1-expressing cells from inv(16); Swt and cells in inv(16); Shet with a similarly high expression level (above the dashed line) were taken for the analysis shown in (C). C. Fli1 mRNA expression vs. regulon activity score from pySCENIC analysis was plotted for each genotype. Cells from inv(16); Shet mice are shown in blue, and those from inv(16); Swt mice are shown in red. 661 inv(16);Swt and 625 inv(16);Shet cells were used for the analysis. Linear regression models indicate a significant positive correlation in both genotypes; the regulatory slope is significantly steeper in inv(16); Shet cells (slope=11.71, R2 = 0.309) compared to inv(16); Swt cells (slope=6.13, R2 = 0.091). Shaded areas represent 95% confidence intervals. Significance was tested with Anova (p = 2.89 × 10–7). We next identified the Fli1 level in the top quartile of inv(16); Swt cells from clusters 0, 1, 3, and 6 and then isolated inv(16); Shet cells with a similar high expression of Fli1 (above the dashed line, Fig. 5B). This allowed us to isolate cells with comparable, high-level expression of Fli1 under the two genetic conditions. We then plotted Fli1 expression vs. regulon activity score for each cell to determine the relationship between the two (Fig. 5C). Applying a linear interaction model revealed a significant interaction between Fli1 regulon activity and genotype (Fig. 5C, p = 2.89x10–7). Post-hoc trend analysis showed that while Fli1 regulon activity was positively correlated with Fli1 expression in both of the genotypes, the association was significantly more pronounced in the inv(16); Shet group compared to the inv(16); Swt group (inv(16); Shet slope = 1.9x that of inv(16); Swt). Collectively, our data indicate enhanced Fli1 target gene expression in inv16; Shet vs. inv16; Swt cells, even in cells where the expression level of Fli1 itself is comparable between the two genotypes, suggesting that cohesin loss enhances the ability of Fli1 to drive gene expression on a per-cell basis. Our data support the hypothesis that re-enforced, deregulated Fli1 expression programs may contribute to the reduced latency observed in inv(16); Shet genotype by driving changes in gene expression in primitive hematopoietic progenitors. Knockdown of Fli1 disrupts disease maintenance in inv(16);Shet leukemia While our data demonstrate a potential role for Fli1 and its gene expression programs in accelerating inv(16);Shet leukemic development, we were curious if Fli1 also had a role during disease maintenance. Interestingly, the upregulation of Fli1 observed in HSPCs (Fig. 3B) was not maintained after leukemia development at either the mRNA or the protein level (Fig. 6A, B), suggesting that elevated Fli1 expression in response to Smc3 haploinsufficiency occurs early, in the pre-leukemic state, and then wanes during leukemic progression. However, consistent expression of the Fli1 transcriptional program could still be required during the active disease stage. To determine if Fli1 is necessary for inv(16) leukemic maintenance, we lentivirally infected inv(16); Shet leukemic cells with an empty vector and two different shRNAs targeting Fli1 (shFli1). We achieved an approximate 70–80% knockdown at the mRNA level (data not shown) and 60–80% knockdown at the protein level (Supplementary Fig. 7A). We next transplanted 200,000 pLKO (empty vector/EV), shFli1 #1, or shFli1 #2 (all CD45.2) alongside 200,000 CD45.1 support cells into lethally irradiated CD45.1 recipients. Three mice for each condition were transplanted. All three mice receiving empty vector succumbed to leukemia 37 days post-transplant, while 5/6 mice receiving shFli1 #1 or shFli1 #2 did not present any evidence of leukemia 100 days post-transplant (Fig. 6C and Supplementary Fig. 7B). One shFli1 #1 recipient did have evidence of leukemia upon sacrifice 100 days post-transplant, including pale bones, an enlarged spleen, and excess blasts in the bone marrow (data not shown). As assessed by genotyping, the marrow of this animal was positive for inv(16) and Smc3Δ/+, indicating that the leukemia developed from the transplanted cells; however, a significant portion of the marrow was CD45.1+ ( ~ 60%, data not shown), indicating the leukemia was not as aggressive as in pLKO recipients. No CD45.2 cells were present in any of the five remaining shFli1 recipients (Supplementary Fig. 7C). We conclude that Fli1 expression is critical for inv(16); Shet leukemic maintenance. Fli1 knockdown disrupts disease maintenance in inv(16); Shet leukemia. A qPCR was performed on cDNA from three inv(16); Swt and three inv(16); Shet leukemias for Fli1 mRNA using β-actin as a control. The relative expression for each genotype was plotted. B Western blot showing the expression level of Fli1 protein vs. β-actin in two inv(16); Swt and two inv(16); Shet leukemias (left panel). The expression of Fli1 vs. β-actin was quantified and plotted (right panel). C Survival curve showing mice receiving pLKO (empty vector, EV, black), shFli1#1 (blue), and shFli1 #2 (red, overlaps with blue). n = 3 for each condition, *p < 0.01, log-rank test. Data in A and B represent the mean +/– SEM. NS: non-significant, Student’s t test. Smc3 haploinsufficiency results in a more aggressive leukemic phenotype compared to inv(16) alone We next sought to determine if overexpression of Fli1 would be sufficient to accelerate inv(16); Swt leukemogenesis. However, secondary transplants using lenti- or retrovirally infected HSPCs isolated from pIpC-treated inv(16) mice were not feasible due to poor engraftment, likely due to their reported poor contribution to the bone marrow in the presence of competitor cells, in this case, WT CD45.1 support cells [29]. We were therefore unable to definitively test if Fli1 accelerates inv(16)-driven leukemogenesis due to this technical limitation. Interestingly, similar to HSPCs, we also were unable to test the effect of Fli1 loss via shRNA-mediated depletion on inv(16);Swt disease maintenance due to poor engraftment of virally transduced inv(16);Swt leukemias. These data collectively suggest that Smc3 heterozygosity is more detrimental to proper cell function in comparison to inv(16) alone. Discussion In contrast to our initial hypothesis, we demonstrate reduced latency in inv(16); Shet vs. inv(16); Swt mice. While this manuscript was in revision, Conneely et al. published similar results, although statistical significance was not reached, likely due to a smaller number of mice used [44]. The reduced latency could be due to an accelerated rate of leukemogenesis or to a more aggressive disease maintenance stage. We were unfortunately unable to definitively show that Fli1 accelerates inv(16)-driven leukemogenesis due to poor engraftment of inv(16);Swt HSPCs. However, the enhanced expression of Fli1 in primitive inv(16); Shet HSCs (Fig. 4D) combined with increased accessibility of Fli1 binding sites (Fig. 2B) and increased oncogenic, Fli1-mediated gene expression (Figs. 3C, D and 5A–C) in HSPCs suggest that Fli1 may contribute to a more rapid rate of disease initiation in the presence of inv(16), although additional experiments are required to rigorously distinguish the role of Fli1 in inv(16) leukemogenesis. Additionally, our data show a requirement for Fli1 in inv(16);Shet leukemias (Fig. 6C), indicating that Smc3 haploinsufficiency likely reduces the latency of inv(16) AML due to effects on both the rate of leukemogenesis and the maintenance stage of disease. While we were unable to test the effect of Fli1 expression loss on inv(16); Swt disease maintenance due to poor long-term engraftment of virally transduced inv(16); Swt leukemias, our success in the inv(16); Shet genotype underscores the more aggressive nature of the Shet leukemias. Given that Fli1 activation was associated with the expression of Ras pathway genes in inv(16); Shet cells (Figs. 2A, B; 3D), we hypothesize that this may contribute to an enhanced rate of leukemogenesis and/or increased robustness of the resulting leukemia. RAS mutations are known to drive a more aggressive, more proliferative disease [45]. Interestingly, it has been proposed that mutated HSCs may be under chronic stress due to enhanced rates of proliferation or chronic inflammation, potentially resulting in epigenetic modifications that drive the acquisition of RAS mutations [45]. As the cohesin complex is known to regulate epigenetic changes and Smc3 haploinsufficiency drives substantial changes in chromatin accessibility (Fig. 2A), it would be interesting to compare the epigenetic landscape of inv(16); Shet and inv(16); RAS-mutant cells, particularly with regard to sites of Fli1-mediated regulation. Examining changes in the epigenome may thus provide insight into why RAS mutations are more common in inv(16) vs. t(8;21) AML. Of note, we and others [44] have not observed a big difference between leukemic inv(16); Swt and inv(16); Shet gene transcription via bulk RNAseq (data not shown). However, it is possible that scRNA sequencing may reveal changes in clonal heterogeneity and hierarchy between the two genotypes, and it is an important future experiment. While it is also possible that differences exist between mice and humans with regard to inv(16) AML leukemogenesis or maintenance, our results are consistent with previously published effects of Fli1 in a variety of AML and MDS models [46–50]. How might Smc3Δ/+ and inv(16) cooperate to enhance Fli1 activity? We have previously shown that cohesin recruits the histone methylase complex, PRC2, to repress Hoxa7/9 [25]. Under cohesin haploinsufficiency, the repressive mark H3K27me3 is lost at the Hoxa7/9 promoters, resulting in increased expression, driving self-renewal [25]. Additional studies have shown that loss of the cohesin subunit STAG2 results in a decrease in the H3K27me3 mark at multiple sites [51]. As the FLI1 promoter is a target of PRC2 in breast cancer [52], we hypothesize that Smc3Δ/+ may result in a loss of PRC2-mediated methylation at the Fli1 promoter, driving enhanced expression. Additionally, the CBFB-SMMHC fusion has been shown to interact with the transcriptional coactivator p300 and other hematopoietic transcription factors, including TAL1, ELF1, ERG, and PU.1 [32], which may further drive FLI1 and FLI1-mediated gene expression changes. While the comprehensive effects of Fli1 regulon activation in AML are unknown, Fli1 regulates multiple factors involved in myeloid differentiation, self-renewal, and proliferation [46–48, 53]. Collectively, the literature and our data suggest that cohesin haploinsufficiency results in increased Fli1 expression and Fli1 site accessibility, allowing the CBFβ::SMMHC fusion and Fli1 to cooperatively deregulate gene expression to drive a reduction in disease latency. The negative impact of FLI1 upregulation is evidenced by previous studies linking deregulated FLI1 expression with reduced overall survival and duration of remission in AML [48, 49]. Therefore, we speculate that patients with enhanced FLI1 activity, and not simply elevated FLI1 expression, may have an overall worse prognosis. Interestingly, about half of all patients with an NPM1cA mutation also harbor a cohesin mutation [4, 21, 22], and patients with mutated NPM1 or FLT3-ITD have been documented to have higher Fli1 protein expression and lower overall survival [49]. Similar to inv(16); Shet cells, we observed an enrichment of Fli1-binding motifs in the uniquely accessible chromatin regions in Npm1cA/+; Smc3Δ/+ HSPCs (Fig. 2D), indicating that Fli1 may be a specific target for cohesin-mutated AMLs. Interestingly, others have identified Fli1 as a putative target in erythroleukemia, where it has been shown to regulate proliferation, inflammation, apoptosis, differentiation, and angiogenesis [53–55]. Several molecules with varying levels of specificity have been shown to inhibit Fli1, with the EWS-FLI1 inhibitor trabectedin (ET-743) receiving FDA approval for the treatment of ovarian cancer and soft-tissue sarcomas, and TK216 completing phase I/II trials for Ewing sarcoma [56, 57]. However, significant difficulties persist in identifying a Fli1-specific inhibitor that is clinically efficacious and lacks significant off-target effects. Further, it is unknown if EWS-FLI1 inhibitors will also target wild-type Fli1, which appears to be critical in inv(16) leukemia based upon the data presented here. Additional studies with co-mutational models may aid in achieving mechanistic insights for newer classes of Fli1 inhibitors and will be necessary to explore how broadly related Fli1 expression programs are to cohesin loss. Given the decreased latency we observed upon combining Smc3 haploinsufficiency and inv(16), it is surprising that cohesin mutations are not commonly found alongside inv(16) mutations in AML. It is possible that cohesin mutations co-occur in concert with inv(16) at a higher rate than has been reported. Targeted amplicon sequencing with higher coverage revealed additional mutations in 96% of patients in one CBF AML study [16]. In this study, a total of six cohesin mutations were found by targeted sequencing in inv(16) AML samples, which were not observed by whole exome sequencing [16]. We combined the six cohesin-mutated samples into one group to analyze the effect of cohesin mutation on overall survival or recurrence-free survival vs. cohesin-WT inv(16) AML, but found no difference (Supplementary Fig. 8A, B), likely due to the low number of cohesin-mutated samples. Thus, targeted sequencing of more patient samples may uncover additional cohesin mutations in inv(16) and add more power to this analysis. Our mechanistic data do, however, provide an alternative hypothesis as to why inv(16) and cohesin mutations are rarely observed together. We have shown that the addition of Smc3 haploinsufficiency results in a dramatic increase in chromatin accessibility, with an enrichment of Fli1 binding sites linked to RAS signaling genes (Fig. 2A, B; 3D). The most common co-occurring mutations with inv(16) are activating mutations in N-RAS or K-RAS [15–19], suggesting that inv(16); Shet may mimic some effects of RAS mutations, although not at the level of RAS itself (Supplementary Fig. 5C). In support of this, most prostate cancers do not have RAS mutations, but ~50% express aberrant fusions between TMPSS2 and ETS family members [58, 59]. While fusion with FLI1 is rare [58], fusion-mediated overexpression of ETS family members can drive transformation and activate a cell migration gene expression program normally driven by RAS, even in the absence of RAS signaling [59]. Thus, ETS family overexpression in prostate cancer can mimic RAS pathway activation [59]. We speculate that Smc3 haploinsufficiency may partially mimic RAS-activating mutations often seen in inv(16) leukemias through overactivation of Fli1 and its targets. As active RAS mutations are strong leukemic drivers, their early acquisition in inv(16) AML likely makes selection of cohesin mutations redundant. Further studies on how cohesin, the core binding factor complex, and Fli1 cooperate to regulate inv(16) leukemia will likely uncover additional mechanisms common to AML development and progression. Supplementary information Supplemental Figure 1 Supplemental Figure 2 Supplemental Figure 3 Supplemental Figure 4 Supplemental Figure 5 Supplemental Figure 6 Supplemental Figure 7 Supplemental Figure 8 Supplemental Table 1 Supplemental Table 2 Supplemental Table 3 Supplemental Table 4