Mouse Models of Frequently Mutated Genes in Acute Myeloid Leukemia

Simple Summary Acute myeloid leukemia is a genetically heterogeneous disease and shows variable treatment outcomes. Genetic profiling has revealed different driver mutations in AML patients. Therefore, it is important to understand the biological impact of these mutations in leukemia transformation. In this review, we discuss the individual and synergistic effects of these mutations in the pathogenesis of leukemia based on the available evidence from mouse models. Abstract Acute myeloid leukemia is a clinically and biologically heterogeneous blood cancer with variable prognosis and response to conventional therapies. Comprehensive sequencing enabled the discovery of recurrent mutations and chromosomal aberrations in AML. Mouse models are essential to study the biological function of these genes and to identify relevant drug targets. This comprehensive review describes the evidence currently available from mouse models for the leukemogenic function of mutations in seven functional gene groups: cell signaling genes, epigenetic modifier genes, nucleophosmin 1 (NPM1), transcription factors, tumor suppressors, spliceosome genes, and cohesin complex genes. Additionally, we provide a synergy map of frequently cooperating mutations in AML development and correlate prognosis of these mutations with leukemogenicity in mouse models to better understand the co-dependence of mutations in AML.


Introduction
Acute myeloid leukemia (AML) is characterized by the uncontrolled proliferation of leukemic stem cells and results in cytopenia in peripheral blood [1,2]. The 5-year overall survival rate is 30-35% for AML patients up to the age of 60 years [3], underscoring the need for a better understanding of the heterogeneity of AML to develop novel treatment strategies for patients who respond poorly to currently available therapies. Deep sequencing enabled the discovery of driver mutations in primary and relapsed AML patients [4]. Papaemmanuil et al. identified more than 5000 driver mutations across 76 genes in 1540 AML patients [2]. These major driver genes can be clustered into 7 functional groups: cell signaling genes, epigenetic modifier genes, nucleophosmin 1 (NPM1), transcription factors, tumor suppressors, spliceosome genes, and cohesin genes [5]. Identification of these key molecular abnormalities improved prognosis, prediction of treatment outcomes, and measurable residual disease (MRD) monitoring [6,7]. Understanding the pathophysiology of these mutations in disease development will be useful to develop more potent targeted therapeutic options for AML patients.
Genetically engineered mouse models have been developed to study AML progression in vivo. These include induced AML models (by chemicals, viral infection, or irradiation), transgenic mouse models, and patient-derived xenograft (PDX) models [8]. Here, we summarize the biological effects of different AML mutations in overexpression, knockin, or knockout mouse models.
In this review, we discuss the in vivo oncogenic potential of the most frequently mutated genes in AML patients. Further, this review highlights the biological relevance of clinically mutually exclusive and co-occurring mutations.

Mouse Models of Genes Involved in Cell Signaling Pathways in Myeloid Malignancies
The signal transduction gene set is the most often mutated gene set in AML. These genes are involved in transducing signals from the cell membrane to the nucleus. This results in the target gene expression of the downstream pathway. In this way, it controls important cellular events such as proliferation, apoptosis, and differentiation.

KIT
KIT (CD117) is a transmembrane tyrosine kinase receptor that is activated b its ligand SCF [137]. Most common KIT mutations occur in exons 8 and 17 [138

KIT
KIT (CD117) is a transmembrane tyrosine kinase receptor that is activated by binding its ligand SCF [137]. Most common KIT mutations occur in exons 8 and 17 [138,139]. Mutations in both exons lead to the constitutive activation of KIT without the ligand. c-Kit −/− mice show postnatal death with impairment in hematopoiesis (  Figure S1) [35,36]. It has been shown that the KIT mutation at exon 17 is a negative prognostic factor in RUNX1-RUNX1T1/AML1-ETO/RUNX1-ETO AML patients [140]. In a cooperation study, c-Kit (D814V) or c-Kit (T417I∆418-419) induces AML in mice when coexpressed with the RUNX1-RUNX1T1 fusion [141]. Further, KIT D816 (at exon 17) is a hot spot mutation in inv(16) AML, which is characterized by the CBFbeta-MYH11 fusion gene [142]. The cooperation of these two alterations (KIT D816 and CBFbeta-MYH11) also shows synergistic leukemia development in mice [143].

KRAS
RAS family member genes (NRAS, KRAS, and HRAS) are frequently mutated genes in cancers [144]. Proteins encoded by these genes transmit the transcriptional signal from the cell surface to the nucleus [145]. Oncogenic mutations in these genes activate constitutive RAS signaling in the cell. NRAS and KRAS mutations are frequently detected in AML patients but HRAS mutations rarely appear in AML patients [146]. Approximately five percent of AML patients display mutations in the KRAS gene [144,146]. Homozygous deletion of Kras results in embryonic lethality (Table 1) [41,42]. Reconstituted mice with bone marrow (BM) progenitor cells that express KRASG12D do not induce any disease (  Figure S2) [39,40]. These findings support the idea that KRAS mutations are not sufficient for AML development. Therefore, many synergistic mouse models of KRAS mutations with other oncogenes have been established. Co-expression of KRAS mutations with other mutations such as DNMT3A [123], TP53 [124], NF1 [125], and BCOR [126] induces AML in mice ( Table 2; Supplementary Figure S3). The KRAS mutation also induces AML in mice when combined with fusion oncogenes such as AML1-ETO [38].

NRAS
NRAS mutations account for approximately 10% of AML patients [147]. NRAS mutations often appear in codons 12, 13, or 61 [147,148]. Homozygous Nras knockout mice show normal growth and no hematological defects (Table 1) [48]. So, unlike Kras, Nras is dispensable for embryonic development. The retroviral overexpression mouse model of NRASG12D does not show any hematological disease development (Table 1; Supplementary Figure S1) [38,43]. However, another study has demonstrated that NRASG12D causes CMML and AML-like disease in mice [44]. Although they have used the MSCV promoter to express the NRAS oncogene, the major discrepancy in the results may be due to the use of a different strain of mice and a high titer of the virus. Moreover, the expression of NRASG12D under the myeloid-specific hMRP8 promoter causes abnormal hyperkeratotic skin lesions in mice [45]. Transgenic Mice expressing oncogenic Nras under the control of the Moloney murine leukemia virus (Mo-MuLV) LTR develop myeloproliferative disorders [46]. This study also observed Nras induced apoptosis in the bone marrow of some mice. Heterozygous NrasG12D knockin mice develop MPNs (Table 1; Supplementary Figure S2) [149]. Secondary transplant recipients from Nras G12D/+ bone marrow cells develop a CMML-like disease [150]. Homozygous NRASG12D mice die from severe myeloproliferative disease [47]. It suggests that biallelic NRAS mutations provide a stronger oncogenic signal. Inactivation of P53 collaborates with the NRASG12D mutation to induce a highly penetrant AML in vivo ( Table 2; Supplementary Figure S3) [127]. Previous studies have shown that overexpression of the Ras oncogene causes p53 accumulated cell senescence, which can be overcome by P53 inactivation [151]. However, the endogenous expression of the Ras oncogene does not cause cell senescence [152], indicating involvement of an alternate mechanism for this oncogenic synergy. In contrast to Nras G12D/+ alone or p53 −/− alone, cooperation of these mutations induces quiescence in megakaryocyte-erythroid progenitors (MEPs) that is sufficient to drive AML development in mice (Table 2; Supplementary Figure S3) [127]. Similarly, cooperation of the NrasG12D mutation with Dnmt3aR878H/DNMT3AR882H activates the Myc pathway and induces AML in mice (Table 2; Supplementary Figure S3) [128,132]. NRAS mutations also show synergistic AML development with mutations in other genes such as NPM1 [129], EZH2 [130], and IDH2 [25] ( Table 2; Supplementary Figure S3). Cooperation AML models of oncogenic NRAS has been also established with fusion genes such as NUP98 fusions (NUP98-NSD1, NUP98-JARID1A, and NUP98-DDX10) [43,153], KMT2A-AF9 [154], KMT2A-ENL [155], and RUNX1-RUNX1T1 [155].

NF1
The neurofibromin 1 gene (NF1) encodes a RAS GTPase activating protein that modifies active RAS-GTP into its inactive RAS-GDP and suppresses the RAS pathway [156]. Therefore, deletion of this gene or inactivating mutations in this gene triggers the RAS-MAPK signaling pathway. Almost five percent of adult de novo AML patients carry mutations in the NF1 gene [157]. Homozygous deletion of Nf1 in embryonic stem cells results in embryonic lethality (Table 1) [49]. On the other hand, Nf1 heterozygous deletion in embryonic stem cells predisposes mice to develop various tumors including leukemia [50]. Somatic deletion of Nf1 induces a myeloproliferative disorder in mice that models juvenile myelomonocytic leukemia (JMML) ( Table 1) [158]. Another study explains that GM-CSF signaling is indispensable for Nf1 −/− induced MPN development, indicating therapeutic benefit by targeting GM-CSF signaling in NF1 mutated myeloid disorders [159]. Collaboration of Nf1 inactivation and the KRASG12D mutation induces AML in mice (Table 2; Supplementary Figure S3) [125]. However, it is unclear whether the synergy is due to hyperactivated RAS signaling or activation of a non-RAS pathway by Nf1 deficiency. Further, concurrent haploinsufficiency of both Nf1 and Asxl1 induces AML in mice [131]. The above-mentioned mouse models of Nf1 suggest that the Nf1 mutation is a loss of function mutation and can induce AML in collaboration with other mutations.

PTPN11
The PTPN11 gene encodes Src homology region 2 (SH2)-containing protein tyrosine phosphatase-2 (SHP2) that regulates the RAS pathway [160]. Germline mutations in the PTPN11 cause Noonan syndrome and somatic mutations occur in leukemia patients [161]. Mutations in Q510, A72, E76, and G503 are hotspot PTPN11 mutations across different cancers [162]. PTPN11 mutations frequently occur in JMML patients (around 35%) but are less common in AML patients [163]. Homozygous Ptpn11 null mice die at the embryonic stage (Table 1) [54]. Knockout of Ptpn11 in murine hematopoietic cells causes the death of mice due to bone marrow aplasia, indicating its significance for the survival of HSCs [164]. This study also shows that constitutive expression of Kras can rescue Ptpn11 ∆/∆ HSCs and myeloid progenitor cells that indicate the presence of Kras downstream of Ptpn11 in HSCs [164]. Retrovirally overexpressed models of Ptpn11 mutants induce a JMMLlike disease in mice (Table 1; Supplementary Figure S1) [51]. Knockin mice expressing Ptpn11D61Y show fatal MPNs [52]. Similarly, mice carrying the Ptpn11E76K mutation developed MPNs (Table 1; Supplementary Figure S2) [53,165] and later few mice develop acute leukemia. Further, knock-in of the Ptpn11E76K mutation in myeloid and lymphoid progenitors causes AML and acute lymphocytic leukemia (ALL), respectively [166]. In vivo mouse models also show cooperation of PTPN11 mutations with KMT2A fusions such as KMT2A-MLLT3 [167] or KMT2A-MLLT10 [168] in leukemia development. In conclusion, most of the PTPN11 mutations are gain of function mutations and show oncogenic activity in vivo.

DNMT3A
DNA methylation is one of the important epigenetic control mechanisms in both normal development and cancer [169]. DNA methylation usually refers to the conversion of cytosine to 5' methyl cytosine in CpG islands, which is orchestrated by DNA methyltransferases (DNMTs) [170]. DNMT3A mutations are reported in approximately 20% of AML patients and most commonly affect amino acid R882 [171]. Dnmt3a −/− mice are not embryonic lethal but die at 4 weeks of age (Table 1) [57]. Hematopoietic-specific conditional Dnmt3a knockout mice develop a myelodysplastic syndrome (MDS) or MPN phenotype in mice [58]. In this study, loss of Dnmt3a shows unaltered homing to BM but a selective increase in liver homing compared to Dnmt3a wild-type bone marrow cells [58]. It suggests that alterations in liver homing play a role in the development of hematopoietic neoplasms. Mice transplanted with Dnmt3a knockout HSCs in a competitive microenvironment do not develop any disease [172] but develop a broad spectrum of hematopoietic malignancies in a non-competitive environment (lethally irradiated mice) [173]. Retroviral overexpression of the hotspot DNMT3A-Arg882His (R882H) mutation induces CMML in mice (Table 1; Supplementary Figure S1) [55]. Further, knockin of the Dnmt3a R878H mutation drives AML development in mice with activation of the mTOR pathway (Table 1; Supplementary Figure S2) [56]. Mutations in Dnamt3a or loss of Dnamt3a synergize with other alterations such as Kras (G12D/+) [123], FLT3-ITD [22,23], Npm1 cA/+ [134], Bcor −/− [135] and IDH2 neomorphic mutations [133] to develop AML in mice ( Table 2; Supplementary Figure S3). In an Flt3-ITD/DNMT3A cooperation AML model, haploinsufficiency of Dnmt3a hypomethylates genes such as Gata3 that causes the transformation of FLT3-ITD-induced MPNs to AML. In addition, genes involved in Wnt signaling such as Cxxc5 and Emilin2 are also dysregulated by loss of Dnmt3a, but the correlation with its downstream target c-Myc is not established [22]. Another study has shown that the dosage of Dnmt3a determines myeloid or lymphoid transformation. FLT3-ITD overexpression in homozygous Dnamt3a knockout or heterozygous Dnmt3a knockout cells causes T-ALL or AML in mice, respectively [174]. This suggests that the heterozygous Dnmt3a knockout predisposes mice to myeloid malignancy, while the homozygous Dnamt3a knockout is more prone to lymphoid malignancies. Celik et al. found that secondary transplantation of primary MDS BM cells from Dnmt3a knockout mice progress to AML by acquiring C-kit mutations [175]. Another study has shown that the Ptpn11D61Y mutation cooperates with Dnmt3a −/+ to induce rapid myeloproliferative neoplasms in mice [176].

TET2
Ten eleven translocation 2 (TET2) belongs to the TET family protein that converts 5-methylcytosine (5-mC) to 5-hydroxymethylcytosine (5-hmC), an epigenetic modification important for the regulation of transcription [177][178][179]. Loss of function genomic alterations of TET2 is implicated in 10-20% of AML patients [180,181]. Inactivation of TET2 in both murine bone marrow (BM) hematopoietic stem/progenitor cells (HSPCs) and human cord blood CD34+ cells show unbalanced myeloid directed differentiation [178,182]. Further, heterozygous loss of Tet2 initiates aberrant hematopoiesis in vivo [61] and the homozygous loss leads to a wide spectrum of myeloid malignancies including MDS, CMML, and sarcoma (Table 1) [59,60]. Several studies have explored the cooperative effect of TET2 loss and mutation/ deletion in other sets of genes. Loss of TET2 cooperates with the FLT3-ITD mutation ( Table 2; Supplementary Figure S3) [28] or the TET3 deletion [183,184] to induce AML in mice. Hypermethylation of Gata2 is observed as a synergistic effect of loss of Tet2 and Flt3-ITD mutation that promotes AML, and this can be reversed by restoration of Gata2 [28]. The double knockout of Tet2 and Tet3 causes DNA damage and impaired DNA repair compared to the single loss of Tet genes [183]. Apart from FLT3-ITD, loss of TET2 synergizes with other mutations in signal transduction genes such as NRASG12D to induce a fully penetrant CMML phenotype by combined suppression of negative regulators of the RAS pathway [185]. Loss of TET2 with loss of EZH2 induces an aggressive MDS/MPN phenotype [186]. TET2 also cooperates with loss of the transcription factor BCOR [94] or the Sf3b1K700E mutation in developing a myeloid disease in mice [114]. Tara et al. reported distinct DNA hypermethylation patterns as a synergistic result of combined loss of Tet2 and Bcor [94]. Collectively, loss of TET2 cooperates with genes of most functional classes to induce aberrant hematopoiesis in vivo.

EZH2
Enhancer of zeste homolog 2 (EZH2) incorporates the H3K27me3 mark on its target and causes transcriptional repression [194]. A higher expression level of Ezh2 induces myeloproliferative disease in mice [68]. Ezh2 null mice show developmental defects and embryonic lethality (Table 1) [69]. Complete ablation of Ezh2 in the hematopoietic system made the mice susceptible to myelodysplastic disorders [70]. Additional loss of Tet2 accelerates the disease development in mice [186]. Similarly, Ezh2 loss cooperates with the RUNX1S291fs mutant to accelerate MDS onset in mice [195]. RUNX1 S291fs/Ezh2 promotes MDS development by activating inflammatory cytokine responses but attenuates leukemia development via PRC1 mediated repression of Hoxa9 [195]. In contrast to the other studies, the loss of Ezh2 combined with constitutive expression of the NRasG12D mutation-induced leukemic transformation in mice (Table 2; Supplementary Figure S3) [130]. This cooperation amplifies branched-chain amino acid (BCAA) metabolism and enhances mTOR signaling, which is crucial to induce AML in this cooperation model [130]. Deletion of Ezh2 attenuates the leukemogenicity of KMT2A-MLLT3 expressing cells [196,197]. These data indicate that EZH2 can function either as a tumor suppressor or oncogene, while the determinants of these distinctive roles are not well understood.

ASXL1
The addition of sex combs-like 1 (ASXL1) is an epigenetic modifier that binds to polycomb repressive complex 2 (PRC2) and regulates target gene expression through the H3K27me3 repressive histone modification [198]. Nearly 10% of de novo AML patients show mutations in ASXL1 [199]. ASXL1 mutations also occur in clonal hematopoiesis [200]. 80% of Asxl1 −/− mice showed embryonic lethality (Table 1) and the remaining mice dis-played features of MDS. Further, heterozygous loss of Asxl1 also induced an MDS-like disease [75]. The retroviral overexpression of C-terminal truncated mutant ASXL1 induced an MDS-like disease in mice (Table 1; Supplementary Figure S1) [71], but knockin Asxl1 mutant mice did not develop any blood disease in mice (Table 1; Supplementary Figure S2) [73]. However, another study showed the development of MDS phenotypes in Asxl1 G643fs/+ knockin mice (Table 1; Supplementary Figure S2) [74]. Additionally, Asxl1 Y588X transgenic mice showed a wide spectrum of myeloid malignancies, including AML, MPNs, and MDS [72]. The discrepancy among ASXL1 mouse models may be due to differences in the expression level of the mutants, different promoters to drive the expression of ASXL1 mutants, and the length/type of ASXL1 mutants. Loss of Asxl1 and NRasG12D collaborates to promote leukemia in mice [201]. Heterozygous loss of both Asxl1 and Nf1 cooperates to accelerate myeloid leukemia in mice ( Table 2; Supplementary Figure S3) [131]. The combined loss of both genes induces an MYC-driven transcription signature through H3K4me3 enrichment that prominently contributes to the acceleration of disease development in mice. These reports suggest that the chromatin modifier ASXL1 cooperates with the RAS signaling pathway by NRAS or NF1 mutations to develop leukemia in mice. A recent study also shows that the Asxl1G643W mutant accelerates mutant CEBPA driven AML development in mice ( Table 2; Supplementary Figure S3) [136].

Mouse Models of Nucleophosmin 1 (NPM1) in Myeloid Malignancies
Nucleophosmin 1 (NPM1) is one of the most frequently mutated genes in AML [2,205] and its mutation is exclusively restricted to myeloid malignancies [206]. However, it is also overexpressed in different solid cancers [206]. NPM1 is a multifunctional protein that primarily resides in the nucleus and plays an active role in different basic biological processes such as ribosome biosynthesis, DNA repair, cellular growth, and stress response [207]. Falini et al. first reported NPM1 mutations in its exon 12 in more than 35% AML cases, characterized by a translocation of the NPM1 protein to the cytoplasm [208]. Insertions or duplications of 4 base pairs in exon 12 disrupt the C terminal nucleolar localization signal (NoLS) that leads to cytoplasmic localization of NPM1, which was later validated in NPM1 transfected cells [209]. NPM1 mutations are always heterozygous [210], which may indicate that the interaction of the wild type and mutant proteins is required for the survival of leukemia cells. While homozygous deletion of Npm1 shows embryonic lethality in mice (Table 1) [83], heterozygous loss of Npm1 mostly predisposes to MDS in mice [83,84]. This implies that wild-type NPM1 is required for cell survival and supports the biological importance of the heterozygous mutation status in AML patients. Different models of NPM1 mutations show that it induces myeloproliferative disorders. For example, expression of the cytoplasmic NPM1 mutant (NPMc+) under the human myeloid-specific MRP8 promotor or in a conventional knockin mouse model induced myeloproliferative disease [79,80] Figure S3) [81,82]. Two different cooperation models of mutant NPM1 with two different signal transduction genes, FLT3-ITD [82,121,122] and NRASG12D [129], showed rapid leukemia onset in mice (Table 2; Supplementary Figure S3). In the double mutant mouse models, Dovey et al. have shown that the Npmc+/Flt3-ITD combination leads to AML with a shorter latency in mice compared to the Npmc+/NrasG12D combination [129]. In these cooperation models, a dependence was noted on the Hoxa network for the maintenance of leukemic cells [129], and these results coincide with upregulation of the HOXA genes in NPM1 mutant AML patients [211]. The other common FLT3 mutation, FLT3-TKD, also synergizes with mutated NPM1 and induces a short-latency AML in mice [21]. Clinically, NPM1 mutations also co-occur with both FLT3-ITD and FLT3-TKD mutations. Another study has discovered cooperation of NPM1 and IDH2 mutations for leukemia development in mice through activation of the Hoxa9/Meis1 pathway, where NPMc and IDH2/R140Q increase the expression of Hoxa9 and Meis1, respectively [193]. Chou et al. have observed that the NPM1 mutation downregulates CXCR4/CXCL12 pathway genes to induce myeloproliferation in mice and a similar observation has been made in human NPM1 positive AML patients [80]. Another study has shown that the NPM1 mutation drives Dnmt3a mutant clonal hematopoiesis to AML in mice [134]. In this line, a recent interesting study has demonstrated that disruption of the KMT2A-Menin chromatin complex using a small molecule inhibitor (VTP-50469) significantly increases the leukemic latency of Npm1c+/Dnmt3a double mutant mice [212]. Additionally, combined menin and FLT3 inhibition show potent antileukemic effect in NPM1 and FLT3 double mutant primary AML patient cells [213].

Mouse Models of Transcription Factor Genes in Myeloid Malignancies
A transcription factor regulates the expression of its target gene by controlling its transcription. Myeloid transcription factors are usually differentially expressed between healthy and disease states. Mutations in different transcription factors have been discovered in AML patients.

CEBPA
The CCAAT enhancer-binding protein alpha (CEBPA) gene encodes two isoforms: the 42 kDa isoform (p42) and the 30 kDa isoform (p30) [214]. Approximately one-tenth of AML patients display CEBPA mutations [215,216]. The most predominantly occurring mutations involve loss of p42 [217]. Homozygous Cebpa knockout mice die shortly after birth (Table 1) [86]. Conditional knockout of Cebpa in adult mice shows a selective block of granulocytic development and accumulation of blasts in the bone marrow [218]. Interestingly, mice lacking the p42 isoform but retaining the p30 isoform develop AML. Biallelic loss of the p42 isoform rapidly induced AML, suggesting a tumor suppressor function of the p42 isoform [87]. This result explains why mutations in AML patients frequently occur in the p42 isoform. CEBPA mutations broadly occur as two types of mutations: N terminal mutations and C terminal mutations. N terminal mutations lead to loss of the p42 isoform and C terminal mutations are located in the basic region-leucine zipper DNA binding domain. AML patients with biallelic CEBPA mutations show an N terminal mutation on one allele and a C terminal mutation on the other allele. In this line, Bereshchenko Figure S2) [85]. An additional ASXL1-G643W mutation accelerated the development of AML in vivo [136]. Further, mutations in CEBPA and granulocyte colony-stimulating factor receptor (CSF3R) show synergy in the development of AML in mice, where CSF3R signaling induced both proliferation and differentiation, and the Cebpa mutation blocked differentiation through inactivation of differentiation-associated enhancers [219]. Another mouse study showed that wild-type CEBPA is required for KMT2A-MLLT1 driven leukemia through activation of Hoxa9/Meis1 [220]. In this line, Collins et al. have shown that inactivation of Cebpa impairs Hoxa9/Meis1-mediated leukemogenesis [221]. These studies prove that wildtype CEBPA is indispensable for Hoxa9/Meis1 mediated transformation and provides an explanation for the absence of CEBPA null mutations in AML patients.

RUNX1
RUNX1 (AML1) belongs to the Runt-related transcription factor (RUNX) family of proteins that recognize 5 -TGTGGT-3 and binds to this motif by forming a complex with core-binding factor beta (CBFβ) [222]. RUNX1 controls hematopoiesis by regulating the transformation of hematopoietic stem cells to differentiated cells [223]. Runx1 is required for fetal hematopoiesis, and its loss causes embryonic lethality in mice (Table 1) [89]. In contrast, Runx1 is not essential for adult hematopoiesis and a conditional Runx1 knockout in adult mice showed a myeloproliferative phenotype [224]. Genetic or chromosomal alterations of RUNX1 frequently occur in AML patients [222,225]. Retroviral overexpression of two mutants of RUNX1, D171N, and S291fsX300, induce an MDS phenotype in mice (Table 1; Supplementary Figure S1) [88,226]. Further, the RUNX1-D171N mutation collaborates with overexpression of Evi1 to induce leukemia in mice [88]. Behrens et al. demonstrated that FLT3-ITD collaborates with RUNX1 mutations to induce an aggressive AML in mice ( Table 2; Supplementary Figure S3) [20]. Cooperation of Runx1 deficiency and U2af1S34F mutation induces AML in mice [111]. RUNX1 also cooperates with ASXL1 to accelerate leukemogenesis through activating the HIF1-α pathway [227]. Runx1 has also been shown to be required for KMT2A-MLLT3 leukemogenesis in mice [228].

MYC
The MYC gene family consists of C-MYC, N-MYC, and L-MYC and encodes proteins that function as transcription factors [229]. MYC proteins are tightly regulated in the healthy state but are dysregulated in cancers [230]. Dysregulated MYC expression, rearrangements (particularly in lymphoma), and overexpression have been identified in hematological neoplasms [231]. Germline ablation of murine C-Myc causes embryonic lethality (Table 1) [92]. Overexpression mouse models of C-Myc, and N-Myc show rapid AML development (Table 1; Supplementary Figure S1) [90,91], underscoring the importance of MYC overexpression in AML patients.

BCOR
The BCOR gene, which encodes the BCL-6 corepressor (BCOR), is located on the X chromosome. It acts as a corepressor for BCL6 to cause BCL6 mediated transcriptional repression [232]. Approximately 4% of CN-AML patients show mutations in the BCOR gene [233]. Bcor knockout male mice die before birth (Table 1) [95]. (Bcor ∆E4/y ) mice that lack the BCL6 binding domain develop T-ALL [93]. Moreover, after the deletion of exons 9 and 10, BCOR fails to interact with polycomb repressive complex 1.1 and also causes lethal T-ALL in mice. However, Bcor ∆E9−10/y shows a proliferative advantage in the myeloid compartment and combined with loss of Tet2, these mice develop MDS [94]. Concurrent knockout of Bcor and Dnamt3a cause acute erythroid leukemia in mice ( Table 2; Supplementary Figure S3) [135]. The combined loss of Bcor and Dnamt3a shows expression changes in Gata genes and p53 family members that may contribute to their collaboration [135]. Bcor loss and KrasG12D cooperate to induce AML in mice ( Table 2; Supplementary Figure S3), and Hoxa9 is required for Bcor ∆E9−10 Kras G12D tumors [126]. MLLT3, a fusion partner of KMT2A, directly interacts with BCOR, and loss of this interaction abrogates the leukemogenic potential of KMT2A-MLLT3/MLL-AF9 in mice [234]. Loss of this interaction results in the reduction of c-Myc expression.

CUX1
The CUT-like homeobox 1 (CUX1) gene, a homeodomain-containing transcription factor, is present on chromosome 7, and mutations are frequently reported in del(7q) AML patients [235,236]. CUX1 mutations are often haploinsufficient in del(7q) AML patients, suggesting a role as a tumor suppressor. Homozygous Cux1 null mice show a high postnatal death rate (Table 1) [97]. shRNA-mediated knockdown of Cux1 induces MDS in mice. Additionally, different expression levels of Cux1 show different disease phenotypes, suggesting that the development of the disease depends on the Cux1 dose [237]. Transgenic mice expressing the p75 Cux isoform, which is overexpressed in breast cancers, develop a myeloproliferative disease-like myeloid leukemia (Table 1; Supplementary Figure S1) [96]. Cux1 haploinsufficiency combined with the Flt3-ITD mutation induces AML and CMML in mice (Table 2; Supplementary Figure S3) [30]. These mice show apoptosis defects in the hematopoietic stem and progenitor cell compartments. Cux1 inactivation increases the expression of CASP8 and FADD-like apoptosis regulator (CFLAR), which may contribute to the defect in apoptosis.

SETBP1
SETBP1 mutations are frequently found in different myeloid malignancies such as MDS, JMML, and AML [238,239]. Although SETBP1 mutations are less frequent in primary AML patients, more than 15% of secondary AML patients display mutations in SETBP1 [240]. Setbp1 overexpression induces myeloid leukemia in mice by transcriptional repression of Runx1 (Table 1; Supplementary Figure S1) [98]. However, overexpression of human SETBP1-D868N only causes splenomegaly (Table 1; Supplementary Figure S1) [99]. Transplantation of hematopoietic cells expressing both SETBP1 and ASXL1 mutants causes AML in mice ( Table 2; Supplementary Figure S3) [99]. This study showed that the addition of mutated SETBP1 further enhances Hoxa9 and Hoxa10 expression in ASXL1 mutant cells [99]. Using the Sleeping Beauty transposon system, Pacharne et al. have shown that all FLT3-ITD mice developed AML with Setbp1 being the most frequent integration site. (Table 2; Supplementary Figure S3) [29]. It was demonstrated that Setbp1 overexpression activates the HOXA gene signature and Flt3 ITD/+ /Setbp1 IM+ AML is vulnerable to Kdm1a and Brd3 inhibition [29]. These studies indicate that Setbp1 contributes to leukemogenesis predominantly through the regulation of Hox genes.

PHF6
Mutations in the Plant homeodomain finger gene 6 (PHF6) gene are commonly observed in Börjeson-Forssman-Lehmann syndrome patients [241]. PHF6 mutations are more frequent in T-ALL (about 20%) patients but less frequent in adult AML (about 3%) patients [242]. PHF6 mutations are associated with reduced overall survival in AML [243]. Germline Phf6 deletion causes perinatal death in male mice (Table 1) [101]. Conditional knockout of Phf6 in hematopoietic cells causes myelodysplasia-like disease in mice, suggesting its role as a tumor suppressor in leukemia pathogenesis [100]. PHF6 mutations frequently co-occur with RUNX1 mutations in AML [171,244], but no synergistic model of both mutations is described yet.

Mouse Models of Tumor Suppressor Genes in Myeloid Malignancies
Like in other cancers, genetic abnormalities in tumor suppressor genes were also reported in AML patients. WT1 and TP53 mutations are frequently recurring tumor suppressor mutations in AML patients.

WT1
Wilms tumor 1 (WT1) was initially discovered as a tumor suppressor gene, but later it was also identified as an oncogene in various cancers. Although WT1 is dispensable for fetal hematopoiesis [245], it plays a vital role in adult hematopoiesis [246]. Approximately 6-15% of AML patients show mutations in the WT1 gene [247][248][249][250]. Mutations of this gene result in loss of function of the corresponding protein [248,251], which contributes to oncogenic transformation. Complete loss of Wt1 causes embryonic lethality in mice [102]. Heterozygous Wt1R394W knockin mice display an MDS phenotype (Table 1; Supplementary Figure S2), but Flt3 +/ITD /Wt1 +/R394W mice show a more aggressive phenotype with few AML phenotypes [26]. This suggests that a third hit is required for fully penetrant AML development in mice. In another study using a conditional knockout model, Pronier et al. have shown that Wt1 haploinsufficient mice develop T-ALL, and Wt1 haploinsufficient mice carrying an additional FLT3-ITD mutation develop a lethal AML (Table 2; Supplementary Figure S3) [27]. WT1 acts as a cofactor for TET2 in mediating 5-hydroxymethylation of cytosines (5-hmC), therefore deletion of WT1 or a mutation in WT1 disrupts this pathway [252,253]. Mutations in either of these genes disrupt the formation of 5-hydroxymethylation of cytosines (5-hmC), which is supported by the exclusive occurrence of WT1 and TET2 mutations in AML patients [243].

TP53
TP53 is a tumor suppressor gene that is frequently dysregulated in various cancers. Approximately 8 percent of AML patients have been reported to be diagnosed with a TP53 mutation [254]. Although studies have shown that the frequency of TP53 mutations is lowest in AML compared to other cancers, AML patients carrying TP53 mutations show shorter survival compared to TP53 wildtype patients in the TCGA data set [2,254,255], underscoring the importance of TP53 in AML pathogenesis and prognosis. TP53 is highly expressed in HSCs and mice lacking Trp53, the equivalent of TP53, show an increase in the HSC population [256][257][258]. Mice carrying TP53 mutations (p53R172H, p53R172P, p53R270H) or a full deletion are often prone to hematopoietic neoplasms such as lymphoma, leukemia, and T cell or B cell malignancies (  Figure S2) [104], and its inactivation exerts chemoresistance in mice [155], indicating different oncogenic strengths among TP53 mutations. Several oncogenic cooperation mouse models show that mutations in TP53 or knockout of TP53 aggravate AML. Basova et al. have shown that deletion of P53 accelerates AML that is induced by the mutant transcription factor PU. 1 [263]. Additional models show that P53 inactivation cooperates with NRASG12D or KrasG12D to drive an aggressive AML phenotype in mice (Table 2; Supplementary Figure S3) [124,127]. These mouse model results are consistent with the clinical co-occurrence of P53 mutations with an aberrant RAS signaling pathway. It suggests that the restoration of the tumor suppressor protein function can be an effective strategy to treat AML.

Mouse Models of Spliceosome Complex Genes in Myeloid Malignancies
Splicing removes non-coding introns from precursor messenger RNA which is mediated by the spliceosome complex. Mutations in these genes cause aberrant splicing [264]. Splicing factor mutations more often appear in the founding clone rather than a subclone in the evolution of MDS [265,266] and are also detected in aging individuals lacking any hematological malignancy [267].

SRSF2
The most frequent splicing factor mutated in AML is SRSF2, which prominently occurs in the Proline 95 residue [268]. Homozygous knockout of Srsf2 causes embryonic lethality in mice [107]. Mice transplanted with murine bone marrow cells either retrovirally overexpressing the wildtype SRSF2 or the mutant SRSF2P95H variant do not develop any myeloid malignancy [107]. However, Srsf2P95H/wt knockin mice develop MDS (Table 1; Supplementary Figure S2) [108], while only a mild phenotype was observed in another knockin model [109]. A mechanistic study by Kim et al. shows that the mutant SRSF2 exhibits altered RNA binding activity that leads to the degradation of EZH2 by nonsensemediated decay [108]. In another knockin mouse model, SRSF2P95H mutant cells were found to rely on the wild-type allele for their survival [269]. The SRSF2 mutation cooperates with the IDH2 mutation and causes a lethal MDS in mice. It causes aberrant splicing of INTS3 (Integrator Complex Subunit 3) that resulted in its reduced expression through nonsense-mediated decay [270]. The collaboration of loss of Runx1 and the Srsf2P95H mutation causes MDS in mice [271].

U2AF1
U2AF1 recognizes and interacts with the AG nucleotides at the 3' splice site and the common U2AF1S34 mutation alters this interaction [272]. Biallelic deletion of U2af1 shows embryonic lethality in mice [112]. A doxycycline-inducible transgenic mouse model of mutant U2AF1 (S34F) did not develop any dysplasia, MDS, or AML, while leukopenia and progenitor cell expansion were the only phenotypes [110]. However, U2AF1(S34F) knockin mice showed MDS features such as cytopenia and dysplasia (Table 1; Supplementary Figure S2) [111]. The Runx1 deletion and the U2af1(S34F) oncogenic mutation cooperate to induce AML in mice (Table 2; Supplementary Figure S3) [111]. A recent interesting study demonstrated that mutant U2AF1 leukemia cells depend on the wildtype U2AF1 allele for their survival [112], explaining why splice factor mutations are always found in the heterozygous condition in patients.

SF3B1
While complete ablation of Sf3b1 leads to embryonic lethality in mice (Table 1) [116], haploinsufficiency caused MDS in mice [115]. This shows that the presence of the wildtype allele is important for its mutant counterpart. Moreover, hematopoietic specific SF3B1K700E expression showed some MDS-like features in two different knockin mouse models (Table 1; Supplementary Figure S2) [113,114]. Additionally, Sf3b1 +/K700E Tet2 −/− cooperation mouse model showed early onset of MDS [114].
The currently available mouse models of splicing factor genes show that the heterozygous state of the splicing factor mutations is important for the induction of myeloid malignancies. Currently, no mouse model shows leukemia induction by a splice factor mutation and there are very limited collaboration studies to show the synergy between splicing factor genes and other frequently mutated genes in AML. The dependency of spliceosome mutations on the wildtype allele suggests that the synthetic lethality properties of spliceosome mutations can be exploited in the future for drug development.

Mouse Models of Cohesin Complex Genes in Myeloid Malignancies
The cohesin complex is a ring-like structure that is formed by SMC1A, SMC3, RAD21, and STAG1/2 [273]. This cohesin complex holds the sister chromatids together and thus plays an important role in chromosome segregation during cell division [274]. Therefore, cohesin mutations may lead to aneuploidy. Additionally, the cohesin complex is involved in modulating gene expression through genome organization by increasing DNA accessibility for transcription factors [275,276]. However, mutations in the cohesin complex do not cause aneuploidy in AML, suggesting that the control of the gene expression function by the cohesin complex is crucial for leukemogenesis [277]. The frequency of mutations in members of the cohesin complex is around 10% in AML patients [274,278]. Cohesin gene mutations are rarely seen as solitary mutations in AML and often co-occur with other mutations such as NPM1, DNMT3A, TET2, or RUNX1 [278][279][280]. The mechanistic role of these mutations in leukemic transformation in mouse models is largely unexplored. Biallelic deletion of cohesin genes Stag1, Stag2, Smc3, and Rad21 in mice is reported to be embryonically lethal, indicating its importance in the normal embryonic development process [117][118][119][120]. Further, abrogating the function of the cohesin complex in mice using shRNA-mediated knockdown shows features of myeloproliferative neoplasms [281]. Mutations in cohesin complex genes are usually mutually exclusive, implicating that a mutation in any of the members of the cohesin complex is sufficient to disrupt the function of the whole cohesin complex in leukemia [274,278]. Unlike Stag1 deficient hematopoietic cells, Stag2 deficient hematopoietic cells show an increase in self-renewal activity. Complete ablation of Stag2 in mice induces myeloid dysplasia and reduced differentiation to the B cell lineage through transcriptional control of the lineage-specific genes Ebf1 and Pax5 [282]. The combined loss of both Stag1 and Stag2 causes pancytopenia and bone marrow aplasia in mice [282]. Additionally, the combined knockout of Stag2 and Runx1 recapitulates an MDS phenotype in mice [283]. This study also indicates that codeletions of STAG2/RUNX1 disrupt chromatin looping, which regulates super enhancer-associated genes such as Hoxa9 and basal pausing genes involved in IFN and inflammatory responses that may contribute to leukemogenesis [283]. This model also explains the frequent co-occurrence of STAG2 and RUNX1 mutations in AML [284]. Homozygous loss of Smc3 induces bone marrow aplasia, but the haploinsufficiency of Smc3 shows the renewal of stem cell activity. Further, the combination of Smc3 −/+ and FLT3-ITD induces AML in mice with enrichment of the STAT5A gene signature (Table 2; Supplementary Figure S3) [19]. This suggests that reducing the level of Smc3a provides an accessible chromatin structure for Stat5-mediated transcription [19]. In summary, cohesin gene mutations are loss of function mutations, which are crucial for leukemogenesis.

Conclusions
AML shows a diverse genomic landscape and patients without driver mutations display low blast counts and better outcomes, indicating the importance of genomic lesions in AML pathogenesis [2]. Mouse models are considered to be ideal tools to study the in vivo pathogenesis and to understand the underlying biological interactions among driver mutations. Here we summarized the available evidence of the biological function of mutated genes available from mouse models. Frequent co-occurrence and exclusiveness of mutations in AML patients highlight the importance of gene-gene interactions in AML development. In Supplementary Figure S3, we show a map of genes that functionally cooperate in vivo. This illustrates that mutations in signaling genes are the most frequent cooperation partners of mutations in the other functional pathways. It also shows that signaling genes almost always work by overexpression/knockin (red or grey color in Supplementary Figure S3), while the cooperation partners in the other pathways are either activated (red) or inactivated (grey). This provides the basis to identify functional dependencies that may lead to novel therapeutic approaches. In addition, Supplementary Figure S3 shows the data gaps that should be closed in the near future.
Hotspot mutations in the same gene can differ in their oncogenic potential and preferential cooperation partners. We discuss how different oncogenic mutations in the same gene have a variable leukemogenic potential. ITD or TKD mutations in the FLT3 gene show the potential to induce different hematologic malignancies. Similarly, we discuss differences in oncogenic potential for hotspot mutations in TP53 and PTPN11. In a large patient cohort, it is reported that NPM1 mutations preferentially co-occur with NRASG12/13 but not with NRASQ61. Hot spot mutations in IDH2 and FLT3 also show different preferential partners [2]. This indicates that functional consequences of mutations in the same gene may have a different biologic consequence in the development of AML and therefore show distinct co-mutation patterns. AML patients are usually identified with more than one genomic abnormality. We describe how the cooperative effect of two different mutations can lead to an aggressive disease in vivo. We also highlight how a novel pathway emerges as a synergistic effect of two different mutations. TP53 lesions with complex karyotype AML constitute an adverse risk group in the AML cohort. These patients often lack RAS pathway mutations, indicating redundancy in RAS mutations and loss of RAS regulators by chromosomal alterations [2]. AML patients with ASXL1 and SRSF2 mutations show a very poor prognosis [2]. However, the additive biological effect of these two mutations has not been clearly demonstrated in mouse models and is not clearly understood yet.
In Figure 1 we show the functional synergy of mutations in different pathways. Synergy is rarely seen among genes of the same functional group. Cell signaling genes are the most promiscuous gene class that collaborates with mutations in 5 of the 6 other gene classes. Mutated genes in epigenetic modifiers and transcription factors collaborate with mutations in 3 other gene classes. Mutated NPM1 collaborates with mutations in 2 other gene classes and mutations in the cohesin complex, spliceosome complex, and tumor suppressor genes functionally synergize with only one other gene class. This suggests that mutations in signaling genes are most variable and least specific regarding their collaboration partners and may explain why they often occur as subclonal mutations in AML.
Identifying recurring mutations with modern sequencing technologies and studying the pathophysiological effect of those mutations using mouse models will provide more insights into the biology of AML that will eventually lead to the discovery of more effective treatments for AML patients.