Current findings for recurring mutations in acute myeloid leukemia

The development of acute myeloid leukemia (AML) is a multistep process that requires at least two genetic abnormalities for the development of the disease. The identification of genetic mutations in AML has greatly advanced our understanding of leukemogenesis. Recently, the use of novel technologies, such as massively parallel DNA sequencing or high-resolution single-nucleotide polymorphism arrays, has allowed the identification of several novel recurrent gene mutations in AML. The aim of this review is to summarize the current findings for the identification of these gene mutations (Dnmt, TET2, IDH1/2, NPM1, ASXL1, etc.), most of which are frequently found in cytogenetically normal AML. The cooperative interactions of these molecular aberrations and their interactions with class I/II mutations are presented. The prognostic and predictive significances of these aberrations are also reviewed.


Introduction
The identification of mutations in certain genes, such as the fms-related tyrosine kinase 3 (FLT3), CCAAT/ enhancer binding protein alfa (C/EBPα), runt-related transcription factor 1 (RUNX1), myeloid-lymphoid or mixed lineage leukemia (MLL), Wilms tumor (WT1) and nucleophosmin (NPM) 1 genes, in acute myeloid leukemia (AML) has significantly improved our understanding of leukemogenesis [1][2][3][4]. This is particularly the case for patients with normal cytogenetics, who comprise the largest subgroup (approximately 45%) of AML patients [5][6][7]. In fact, assessments of the presence of internal tandem duplications in the FLT3 receptor gene (FLT3-ITD) [8] and mutations in the NPM1 gene [4] are currently routine practices in guiding therapeutic decisions in AML patients with a normal karyotype [9]. Recent studies have revealed prevalent mutations, such as DNA methyltransferase (Dnmt) 3a mutations [10], ten-eleven-translocation oncogene family member 2 (TET2) [11] mutations and isocitrate dehydrogenase (IDH) 1 gene mutations [3], using novel technologies like high-throughput massively parallel DNA sequencing [12]. Indeed, AML is increasingly subclassified as a unique entity in the 2008 revision of the World Health Organization classification of myeloid neoplasms and acute leukemia [13], based on specific recurring genetic abnormalities that can predict the prognosis and response to therapy [7].
AML development is considered to be a multistep process that requires the collaboration of at least two classes of mutations to obtain full-blown leukemia. Almost a decade ago, Gilliland and Griffin [14] presented a paradigm model for this process, designated the "two-hit model". This model comprises class I mutations that activate signal transduction pathways and confer a proliferation advantage on hematopoietic cells, and class II mutations that affect transcription factors and primarily serve to impair hematopoietic differentiation [15,16]. Mutations leading to activation of the receptor tyrosine kinase (RTK) FLT3, c-kit (KIT) and Ras signaling pathway are considered to be class I mutations. Recurring chromosomal aberrations such as t (8; 21), inv (16) and t (15; 17), which generate fusion transcripts of RUNX1/ETO, CBFβ/MYH11 and PML/RARα, respectively, fall into class II mutations. Not only chromosomal abnormalities but also mutations of the transcription factors RUNX1, C/EBPα and MLL are classified into this group. These "classical" class I and class II mutations are presented in Figure 1. However, most of the newly identified genetic alterations, such as those in Dnmt, TET2, IDH1, IDH2, NPM1, ASXL1, which are addressed mainly in this review, have not been classified because the consequences of these mutations have not been identified. In this review, each of these "unclassified mutations" is detailed individually. The author summarizes the current findings of these novel gene mutations, most of which are frequently found in cytogenetically normal AML (CN-AML). The cooperative interactions of the molecular aberrations are also presented.

Dnmt mutations
Dnmts are enzymes that catalyze the addition of a methyl group to the cytosine residue of CpG dinucleotides. Aberrant CpG island methylation has long been hypothesized to contribute to the pathogenesis of cancer [17]. Recently, using massively parallel DNA sequencing [12], Ley [10], there were alterations of specific gene DNA methylation patterns and/or gene expression profiles, such as HOXB genes, in samples with Dnmt3a mutations compared with those without such changes [19]. Consistent with these observations, the Dnmt3a mutations, which frequently occurred in arginine (R) 882, caused reduced enzymatic activity in vitro [19].

TET2 mutations
In 2009, Delhommeau et al. [11] conducted high-resolution single-nucleotide polymorphism and comparative genomic hybridization arrays to identify a candidate tumor-suppressor gene common to patients with MDS, myeloproliferative disorders and AML. They identified inactivating mutations of the TET2 gene in about 15% of patients with various myeloid malignancies, such as MDS (19%), myeloproliferative disorders (12%), secondary AML (24%) and chronic myelomonocytic leukemia (CMML) (22%) [11]. TET2 can convert 5-methylcytosine (5-mC) to 5-hydroxymethylcytosine (5-hmC) [   The model of the "classical" class I and class II mutations in AML. This model comprises class I mutations that activate signal transduction pathways and confer a proliferation advantage on hematopoietic cells, and class II mutations that affect transcription factors and primarily serve to impair hematopoietic differentiation [15,16]. The development of AML is a multistep process that requires at least these two genetic abnormalities for the development of the disease. Class I mutations are shown in yellow boxes and class II mutations are in red boxes. The combination of each mutation is demonstrated as blue rings. Runx 1 mutations and MLL rearrangements may be exception in this model, as shown in orange boxes, since co-occurrence is observed between these two mutations.

IDH1 and IDH2 mutations
Mardis et al. [3] used massively parallel DNA sequencing to obtain a very high level of coverage of a primary, cytogenetically normal, de novo genome for AML with minimal maturation (AML M1) and a matched normal skin genome, and identified 12 acquired mutations within the coding sequences of genes and 52 somatic point mutations in conserved or regulatory portions of the genome. Many of these were mutations that had already been identified, such as those in NRAS and NPM1, but they also found novel mutations of the IDH1 gene [3]. They further found that IDH1 gene mutations were present in 15 of 187 AML genomes and strongly associated with a normal cytogenetic status. IDH1 and IDH2 function at a crossroads involving cellular metabolism in lipid synthesis, cellular defense against oxidative stress, oxidative respiration and oxygen-sensing signal transduction [23]. Recently, AML patients harboring IDH1 and IDH2 mutations were found to show aberrant hypermethylation [24]. In fact, these mutations led to the production of an abnormal cellular metabolite, 2-hydroxyglutarate, which can inhibit the hydroxylation of 5-mC by TET2 [24]. Consistent with Metzeler et al. [21], IDH1 and IDH2 mutations were mutually exclusive with TET2 mutations (p = 0.009) [24]. Boissel et al. [25] analyzed the prognosis of patients with IDH1 mutations and IDH2 mutations in a [ [25][26][27] 6.6-9.6% in AML

IDH2
In IDH2 mutation CN-AML patients, there is a higher risk of induction failure, a higher RR and shorter OS.
The analysis of the clinical impact in 4 groups (NPM1 and FLT3-ITD single mutants, double mutants, and wild-type for both) revealed that patients having only an NPM1 mutation had a significantly better OS and DFS and a lower cumulative incidence of relapse. [28,29] 27.5-35.2% in AML 45.7-53% in CN-AML ASXL1 Patients with ASXL1 mutations had a shorter OS than patients without, but the mutation was not an independent adverse prognostic factor in multivariate analysis.

WT1
Multivariate analysis demonstrated that the WT1 mutation was an independent poor prognostic factor for OS and RFS among total patients and the CN-AML group. [41,42] 8.3-10.7% in CN-AML 6.8% in de novo non-M3 AML cohort of 520 adults with AML homogeneously treated in the French Acute Leukemia French Association (ALFA)-9801 and -9802 trials. The prevalences of IDH1 mutations and IDH2 mutations were 9.6% and 3.0%, respectively, and the mutations were mostly associated with CN-AML. In patients with CN-AML, IDH1 mutations were associated with higher risk of relapse (RR) and shorter OS ( They also showed that IDH1 mutations had strong associations not only with NPM1 mutations (p < 0.001), but with MLL-partial tandem duplicaton (PTD) as well (p = 0.020) ( Table 1). In addition, they revealed that the prognosis was adversely affected by IDH1 mutations with trends for shorter OS (p = 0.110), shorter EFS (p < 0.003) and higher cumulative RR (p = 0.001) ( Table 1).

NPM1 mutations
Falini et al. [4] described abnormal localization of NPM1 in AML patients. Cytoplasmic NPM1 was detected in 208 of 591 specimens (35.2%) from patients with primary AML, but not in 135 secondary AML specimens or in 980 hematopoietic or extrahematopoietic neoplasms other than AML [4]. Thiede et al. [28] performed a larger study. They investigated 1485 AML patients for NPM1 exon 12 mutations, and found that the C-terminus of this protein was mutated in approximately 27.5% of the patients. NPM1 mutations were more prevalent in patients with a normal karyotype (324 of 709; 45.7%) than in patients with karyotype abnormalities (58 of 686; 8.5%; p < 0.001). They suggested that NPM1 mutations are strongly associated with FLT3-ITD mutations in patients with a normal karyotype (mutated NPM1/FLT3-ITD, 43.8% vs. wild-type NPM1/FLT3-ITD, 19.9%; p < 0.001) [28]. Analyses of the clinical impacts in four groups (NPM1 single mutants, FLT3-ITD single mutants, NPM1/FLT3-ITD double mutants, and wild-type for both) revealed that patients with only NPM1 mutations had significantly better OS and DFS and a lower cumulative incidence of relapse [28] (Table  1). Schlenk et al. [29] focused their analyses on CN-AML patients. Among 872 patients examined, they found that 53% of the patients had NPM1 mutations. In addition, 31% had FLT3-ITD, 11% had FLT3-TKD, 13% had C/EBPα mutations, 7% had MLL-PTD and 13% had NRAS mutations. They further demonstrated that FLT3-ITD (p < 0.001) and FLT3-TKD mutations (p = 0.03) were significantly associated with NPM1 mutations, while NRAS mutations were not (p = 0.46). NPM1 is thought to have relevant roles in diverse cellular functions, including ribosome biogenesis, centrosome duplication, DNA repair and response to stress [30]. NPM1 is also involved in the functions of p53 and p19ARF [31,32]. Li et al. [33] demonstrated that wildtype NPM1 protects hematopoietic cells against p53induced apoptosis under conditions of cellular stress. Therefore, it is possible that failure of the mutated NPM1 to protect cells may make them more sensitive to high-level genotoxic stress induced by chemotherapy. Consequently, it is possible to speculate that patients with mutated NPM1 have a better prognosis.

ASXL1 mutations
The additional sex comb-like 1 (ASXL1) gene belongs to a family with three identified members that encode poorly characterized proteins involved in the regulation of chromatin remodeling. The ASXL proteins contain a C-terminal plant homeodomain (PHD) finger and belong to the polycomb and trithorax complexes that regulate the genetic program of stem cells. ASXL1 is involved in the regulation of histone methylation by cooperation with heterochromatin protein-1 to modulate the activity of lysine-specific demethylase (LSD) 1 [34], a histone demethylase for H3K4 and H3K9 that is also important for global DNA methylation [35]. However, the hematopoietic function of ASXL1 is still unclear, since an ASXL1 knockout mouse model shows only a mild hematopoietic phenotype [36]. In 2009, Gelsi-Boyer et al. [37] first identified mutations of ASXL1 in 40 MDS/AML samples using high-density comparative genomic hybridization arrays. They found mutations in the ASXL1 gene in 4 of 35 MDS patients (11%) and 17 of 39 CMML patients (43%). Another study identified mutations of the ASXL1 gene in 12 of 63 (19%) secondary AML patients transformed from MPN [38]. Recently, Chou et al. [39] examined ASXL1 gene mutations in exon 12 in 501 adults with de novo AML. ASXL1 mutations were detected in 54 patients (10.8%), with 8.9% among patients with a normal karyotype and 12.9% among patients with abnormal cytogenetics. The mutations were closely associated with older age, male sex, isolated trisomy 8, RUNX1 mutations, and expression of human leukocyte antigen-DR and CD34, but inversely associated with t(15; 17), complex cytogenetics, FLT3-ITD, NPM1 mutations, WT1 mutations and expression of CD33 and CD15. Patients with ASXL1 mutations had shorter OS than patients without such mutations, but the mutations were not an independent adverse prognostic factor in a multivariate analysis [39] (Table 1).

WT1 mutations
Although mutations of WT1 were first discovered in hematological malignancies more than a decade ago, the precise roles of WT1 in normal and malignant hematopoiesis remain elusive [40]. It has been implicated in the regulation of cell survival, proliferation and differentiation, and may function as both a tumor suppressor and an oncogene [40]. Paschka et al. [41] analyzed 196 adults aged < 60 years with newly diagnosed primary CN-AML, who were treated similarly with Cancer and Leukemia Group B (CALGB) protocols 9621 and 19808, for WT1 mutations in exons 7 and 9. As a result, 21 patients (10.7%) harbored WT1 mutations. The patients with WT1 mutations had worse DFS (p < 0.001) and OS (p < 0.001) than patients with wild-type WT1. Subsequently, Hou et al. [42] examined the clinical implications of WT1 mutations in 470 de novo non-M3 AML patients aged ≥ 15 years, and their stability during the clinical course. WT1 mutations were identified in 6.8% of the total patients and 8.3% of the younger patients with CN-AML. The WT1 mutations were closely associated with younger age (p < 0.01), French-American-British M6 subtype (p = 0.006) and t(7; 11)(p15; p15) (p = 0.003). A multivariate analysis demonstrated that WT1 mutations comprised an independent poor prognostic factor for OS and relapse-free survival (RFS) among the total patients and the CN-AML patients. In addition, among 32 patients with WT1 mutations, the most frequently associated molecular event was FLT3-ITD (9 cases) [42]. Becker et al. [43] investigated 243 older (≥ 60 years) primary CN-AML patients, and found that WT1-mutated patients (7%) had more frequent FLT3-ITD (p < 0.001) and shorter OS (p = 0.08) compared with WT1-wild-type patients. The clinical features of the unclassified mutations are summarized in Table 1.

Class I mutations (FLT3, PTPN11, NRAS, KIT and CBL mutations)
FLT3 mutations FLT3 is a type III RTK and since the first description [44], numerous studies have confirmed and extended the findings that FLT3 mutations are currently one of the most frequent single mutations identified in AML. ITD mutations of the FLT3 gene occur in approximately 21-24% of adult AML patients [14,28] while activating point mutations of the FLT3-TKD, mainly at Asp 835, are found in approximately 5-7% of AML patients [1,[45][46][47][48]. The most significant impacts of an ITD are its associations with a higher leukocyte count, increased RR, decreased DFS and decreased OS, which have been reported in most studies of children and adults aged < 60 years [49]. Several groups found that an ITD is a significant predictive factor for an adverse outcome in multivariate analyses [49][50][51][52] (Table 2). Bacher et al. [46] performed a large study involving 3082 patients with newly diagnosed AML, and analyzed the mutational status and clinical significance of the FLT3-TKD. They observed FLT3-TKD mutations in 147 patients (4.8%). Unlike FLT3-ITD, the prognosis was not influenced by FLT3-TKD mutations in a total cohort of 1720 cases where follow up-data were available (97 mutated FLT3-TKD cases and 1623 wild-type FLT3 cases) ( Table 2). In addition, Ozeki et al. [53] reported that even in patients with wild-type FLT3, a clear tendency for worse OS was found in patients with high FLT3 expression (5 of 86 patients without FLT3-ITD). Another group observed a similar result for a tendency toward lower OS (12 of 24 patients, p = 0.059) and EFS (7 of 20 patients, p = 0.087) in the group with high FLT3 expression [54] ( Table 2).

PTPN11 mutations
SHP-2 is a cytoplasmic protein tyrosine phosphatase (PTP) that contains two Src homology 2 (SH2) domains. Although PTPs are generally considered to be negative regulators, SHP-2 is unusual in that it promotes the activation of the RAS-MAPK signaling pathway through receptors for various growth factors and cytokines. Mutations in the protein tyrosine phosphatase nonreceptor type 11 (PTPN11), as the human SHP-2 gene, have been shown to produce dominant active mutants in vitro [61]. Hou et al. [62] investigated the prevalence and clinical relevance of mutations of PTPN11 and their associations with other genetic changes in 272 consecutive patients with primary AML. Among 14 patients with PTPN11 mutations (5.1%), none had FLT3-ITD. On the other hand, 6 of 13 patients with PTPN11 mutations had concurrent NPM1 mutations (46.2%) [62], suggesting that PTPN11 is a class I mutation molecule similar to the case for FLT3. They further revealed that PTPN11 mutations had no prognostic significance. The CR rate (75% vs. 62%) and median OS (13 ± 8.95 vs. 25.5 ± 6.54 months) were similar between patients with and without PTPN11 mutations. However, subgroup analyses did reveal that PTPN11 mutations comprised a poor risk factor for OS of AML patients without NPM1 mutations (p = 0.001) [62] (Table 2).

NRAS mutations
Ras oncogenes encode a family of guanine nucleotidebinding proteins that regulate signal transduction upon binding to a variety of membrane receptors, including KIT and FLT3, and play important roles in proliferation, differentiation and apoptosis [63]. There are three functional RAS genes: NRAS, KRAS and HRAS. The RAS genes, especially NRAS, are frequently affected by mutations in AML. Bacher et al. [64] analyzed 2502 patients with AML, and found that 257 patients (10.3%) had NRAS mutations. The subgroups with inv(16) and inv (3) showed significantly higher frequencies of NRAS mutations. In contrast, NRAS mutations were significantly underrepresented in t(15; 17) (2 of 102; 2%; p = 0.005). They did not find significant prognostic impacts of NRAS mutations for OS, EFS and DFS (Table 2).

KIT mutations
KIT is a member of the type III RTK family, and ligandindependent activation of KIT can be caused by gain-offunction mutations that have been reported in core binding factor (CBF) leukemia, and AML subgroups with inv (16)  Mutations of the c-kit gene in both exon 17 and exon 8 appeared to adversely affect OS in AML with inv (16). They also observed an adverse impact of KIT mutations on RR in t(8; 21) AML patients ( Table 2). Cairoli et al. [65] reported that among 42 patients with t(8;21), 19 patients (45.2%) had KIT mutations, whereas among 25 patients with inv(16), 12 patients (48.0%) had KIT mutations. Schnittger et al. [66] analyzed 1940 randomly selected AML patients and revealed that 33 patients (1.7%) were positive for KIT mutations in codon D816. Of these 33 patients, 8 patients (24.2%) had t(8; 21), which was significantly higher compared with the subgroup without D816 mutations. They revealed that KIT mutations had independent negative impacts on the median OS (304 vs. 1836 days; p = 0.006) and median EFS (244 vs. 744 days; p = 0.003) in patients with t(8; 21) but not in patients with a normal karyotype ( Table  2). They also revealed that other activating mutations, like FLT3 and NRAS mutations, were very rarely detected in t(8; 21) leukemia patients. On the contrary, in an analysis of 99 patients with t(8; 21), Kuchenbauer et al. [69] reported that the frequent molecular aberrations with t(8; 21) were not only KIT D816 mutations (3 of 23 patients; 13%) but also NRAS mutations (8 of 89 patients; 8.9%). Although the co-occurrence of NRAS mutations and AML1/ETO expression remains elusive, all of these reports suggest that KIT mutants play important roles in CBF leukemia, with negative impacts on the clinical course [65,66,68,69].

CBL mutations
The Casitas B-cell lymphoma (CBL) gene gives rise to the CBL protein, which has ubiquitin ligase activity and targets a variety of tyrosine kinases for degradation by ubiquitination [70]. CBL proteins also associate with the endocytic machinery and are thus important for the termination of RTK signaling [70]. CBL mutations were able to inhibit FLT3 internalization and ubiquitination [71]. In vitro experiments confirmed constitutive activation of the FLT3 pathway by the CBL mutants, and the phenotype of the altered cells resembled the phenotype of FLT3-mutated cells [72]. Reindl et al. [72] identified c-CBL gene exon 8/9 deletion mutants in 1.1% of 279 patients with AML/MDS. All the patients with CBL mutations had CBF leukemia and chromosome 11q abnormalities [72]. In a series of 37 patients with newly diagnosed inv(16) AML, Haferlach et al. [73] detected CBL splicing mutations in 6 patients (16%). The prevalence and features of these class I mutations are summarized in Table 2.

C/EBPa mutations
The transcription factor C/EBPα is crucial for the differentiation of granulocytes. C/EBPα inactivation may take place by acquired mutations, which may occur along the entire coding region. These mutations lead to increased translation of an alternative 30-kDa form with dominant negative activity on the full-length 42-kDa protein when they occur in the N-terminus, while C-terminal mutations result in deficient DNA binding and/or homodimerization activities [63,78]. As reviewed by Pabst and Mueller [79], mutations in the various portions of C/ EBPα have been reported to occur in 5-14% of AML patients. Among patients with C/EBPα mutations, 91% were in the CN-AML molecular high-risk group (FLT3-ITD-positive and/or NPM1-wild-type), although they seemed to be associated with a good prognosis in AML with an intermediate-risk karyotype [80]. Compared with C/EBPα-wild-type patients, patients with C/EBPα mutations showed a trend for a better CR rate (93% vs. 77%; p = 0.06) and significantly higher 5-year rates of EFS (55% vs. 17%; p < 0.001), DFS (53% vs. 23%; p = 0.001) and OS (58% vs. 27%; p = 0.002) [80] (Table 3). However, the reason why C/EBPα mutations confer a good prognosis remains unclear.

MLL rearrangement
Approximately 4-11% of patients with AML present with rearrangement of the MLL (also known as ALL1 or HRX) gene as the result of a PTD within a single MLL allele [81]. This aberration tends to be frequent in AML patients without chromosomal abnormalities [81]. MLL is a 430-kDa transcription factor with a complex structure that includes three AT-hook domains for DNA binding, a methyltransferase homology domain and a SET domain [81]. MLL is necessary for the maintenance for HOX gene expression [82]. The MLL SET domain can bind to the promoters of HOX genes [83]. Munoz et al. [84] examined 93 adult patients with de novo AML for the incidence and clinical features of MLLrearranged AMLs. As a result, they detected MLL rearrangement in 13 patients (14%). In the FAB classification, there was a significantly higher percentage of M5 subtypes in the MLL-rearranged group. The MLL-rearranged patients had lower EFS (p = 0.001) and a higher probability of relapse (p = 0.07) than MLL-wild-type patients. The clinical features of these RUNX1 and C/ EBPα mutations and MLL rearrangement are shown in Table 3.

Conclusions
The development of novel technologies has led to the identification of several important genetic mutations in AML. Along with the development of whole genome sequencing, it is probable that the major genetic aberrations have been almost completely identified. The next stage is to clarify the consequence of these molecular alterations, especially for the newly identified molecules.
As described in this review, mutations in several newly identified genes, such as TET2 and IDH1/2, lead to the aberrant hypermethylation signature in AML cells [24]. In addition, there were alterations in the DNA methylation patterns and/or gene expression profiles, such as HOXB genes, in samples with DNMT3a mutations compared with those without such mutations [19]. Collectively, these recent findings strongly suggest a link between recurrent genetic alterations and aberrant epigenetic regulation, resulting in abnormal DNA methylation statuses in myeloid malignancies.
With the identification of novel genetic aberrations, increasing numbers of cooperative interactions of these genetic alterations have been discovered. These observations indicate that these unclassified mutations may fall into several subcategories. A revised model of these combinations, including "unclassified mutations" in AML is shown in Figure 2 and 3. In general, the mutations in the same category are mutually exclusive [14,85]. However, there are some exceptions. In "classical" class II mutations, co-occurrence was reported between RUNX1 mutations and MLL rearrangement [76] (Figure 1). Likewise, in unclassified mutations, co-occurrences were observed among Dnmt3a, NPM1 and IDH1/2 mutations [10,25,26]. Dnmt3a and NPM1 mutations co-occur with both classical class I and class II mutations, therefore, these mutations may not be simply categorized into "classical" class I or class II mutations, but it would fall into new category of mutations. Further characterization of these mutations, not only from clinical studies, but also from studies of transgenic animals may be needed.
For these several years, there are admirable progress for the identification of the molecular prognostic markers for CN-AML [86]. Based on these genetic alterations, several signal transduction pathway inhibitors and DNA methyltransferase inhibitors (decitabine, azacitidine) were reported for AML treatment [87]. The newly identified combinations of genetic aberrations will lead to a refined disease classification and to the development of more rational, epigenetic or signal transduction pathway-targeted therapies.  KIT mut.
PTPN11 mut. PT P P PN11 mut. D n m t 3 a m u t . Class I Figure 2 The combination model of class I and unclassified mutations in AML. Several unclassified mutations (NPM1, WT1, Dnmt3a) co-occur with several class I mutations. These may fall into putative class II mutations (termed "class II' mutations"), shown in pink boxes. Within these mutations, co-occurrences were observed between Dnmt3a, NPM1mutations (shown in orange boxes), therefore, these mutations may be exception of this model. Ru R R nx n n 1 mut D n m t 3 a m u t .

PML/RAR
MLL re r r a e e rr r r . D n m N P M 1 m u t . Figure 3 The combination model of class II and unclassified mutations in AML. Several unclassified mutations (ASXL1, IDH1/2, TET2) co-occur with several class II mutations. These may fall into putative class I mutations (termed "class I' mutations"), shown in light yellow boxes. IDH1/2 mutations co-occurred not only with MLL rearrangement, but also Dnmt3a and NPM1 mutations, as shown in orange boxes.