The Genomic and Epigenomic Landscapes of AML

Luca Mazzarella,a Laura Riva,b Lucilla Luzi,a,c Chiara Ronchini,b and Pier Giuseppe Peliccia,d

A progressively better understanding of the genetic and epigenetic abnormalities underlying acute myeloid leukemia has changed clinical practice and affected the outcome of thousands of patients. Over the past decades, approaches focused on cloning, sequencing, and functional characterization of one or a few genes were the preferred (and the only possible) modality of investigation. The advent of disruptive new sequencing technologies brought about an unprecedented acceleration in our learning curve. Our view of the abnormalities required to generate and sustain leukemia is evolving from a piecemeal account based on individual lines of research into a comprehensive view of how all the important components (eg, transcriptional program, modifications, DNA sequence, alterations in noncoding genome) interact, in each patient and each leukemic cell. In this article, we provide an overall look at this complicated landscape and highlight outstanding issues for future research. Semin Hematol 51:259–272. C 2014 Elsevier Inc. All rights reserved.

he idea that underlying genetic abnormalities A complex picture has emerged, with some AML- might dictate clinical decisions, now a common distinctive features. concept in oncology, was pioneered for acute The average number of mutations in coding sequences T 4 1–3 myeloid leukemia (AML) 25 years ago. Treatment of is very low in AML, compared with the majority of solid AML was indeed revolutionized by the recognition of tumors (13 mutations per patient in AML vs 52 in microscopically visible chromosomal abnormalities, as breast , 290 in bladder cancers, and 500 in exemplified by the case of acute promyelocytic leukemia smoking-associated lung adenocarcinomas). A functional – and the associated t(15;17) PML-RARa translocation.4 7 logic governs the type of mutations, which recurrently About one half of all AML cases are cytogenetically affect specific pathways controlling lineage identity, cell normal,8 and it took a long time to appreciate the role of survival, and proliferation, as well as genomic and smaller-scale genomic abnormalities (single nucleotide epigenomic stability. This logic is parsimonious in that variations or small insertions/deletions), such as those each pathway suffers the minimum number of hits that – involving NPM1, FLT3, or RAS.9 11 The advent of are sufficient to affect its outcome: mutations of genes high-throughput sequencing has radically changed the within each pathway are usually mutually exclusive. By field and finally allowed an assessment of the panoply of estimating the mutant allele fraction within leukemia cell genetic aberrations in AML, providing a preliminary view populations, it has become evident that each leukemia of the numbers and types of mutations, their distribution exhibits an oligoclonal anatomy, which evolves over time within leukemic and preleukemic populations, and their under the effect of a selective environment, as in the case – evolution during the natural history of the disease.12 17 of systemic treatments. Furthermore, mutations arise in a clearly ordered temporal fashion, with some being invariably present in the founding clones (eg, mutations aDepartment of Experimental Oncology, European Institute of Oncol- of NPM1) and others occurring at later stages, often in ogy, Milan, Italy. FLT3 b subclones (eg, mutations of ). All known major Center for Genomic Science of IIT@SEMM, Fondazione Istituto players appeared, together with new ones, such as Italiano di Tecnologia (IIT), Milan, Italy. fi cIFOM, FIRC Institute of Molecular Oncology Foundation, Milan, epigenetic modi ers (DNA [DNMT] 12 Italy. 3, ten-eleven translocation [TET]), metabolic enzymes dDipartimento di Scienze della Salute, Università degli Studi di (isocitrate dehydrogenase 1 and 2 [IDH1/2]), and Milano, Milan, Italy. components of the cohesin complex.18,19 Finally, dereg- Conflicts of interest: none. Address correspondence to: Luca Mazzarella MD PhD or Pier ulation of epigenetic factors, caused directly by gene Giuseppe Pelicci MD PhD, Department of Experimental Oncology, mutations or indirectly by alterations of transcriptional European Institute of Oncology, Via Adamello 16, 20139 Milan, networks, has emerged as major driving forces of the Italy. E-mail: [email protected], [email protected] 0037-1963/$ - see front matter disease. The details of this picture are evolving at & 2014 Elsevier Inc. All rights reserved. stunning pace. We provide here an overview of the http://dx.doi.org/10.1053/j.seminhematol.2014.08.007 current state-of-the-art.

Seminars in Hematology, Vol 51, No 4, October 2014, pp 259–272 259 260 L. Mazzarella et al

31–33 MECHANISMS OF GENOMIC ALTERATIONS IN not require either oxidative damage or DNA replication. BLOOD CELLS Because CpG density is 10 times higher in coding than in noncoding DNA, the exome is more prone to Genomic alterations are the net result of DNA damage replication-independent mutations, suggesting that spon- and subsequent DNA repair, coupled with the cell's taneous deamination of methylated cytosines could be inability to eliminate itself (apoptosis). Accumulation of particularly relevant to development.32,34 Notably, DNA damage is inevitable because DNA, unlike all other overrepresentation of C4T substitutions is usually con- macromolecules, cannot be substituted in its entirety served in leukemia.35 Overrepresentation of A4T trans- during the life span of a cell. Established sources of versions in the TpA dinucleotide, through unknown exogenous DNA damage for the development of AML underlying molecular mechanisms, have also been include ionizing radiation, smoke, benzene, and chemo- reported,29 which may be in line with hematopoietic- therapeutic treatment with alkylating drugs.20 However, specific mutational patterns associated with aging identi- clear exposure to any of these agents can only rarely be fied in experimental animals.33 identified. Thus, a significant amount of DNA-damaging A further mechanism leading to genomic instability in agents are thought to derive from inside the cells, as stem cells is their reliance on replication-independent byproducts of intracellular metabolic pathways. Central to strand-break repair mechanisms, such as nonhomologous the damaging effect of both exogenous and endogenous end-joining.23 This mechanism is inherently more error agents is the formation of reactive oxygen species, which prone than homologous recombination because it does not cause alterations to both bases and the sugar backbone of rely on strand complementarity to repair the DNA breaks. DNA, resulting in nucleotide mismatching or strand HSCs have evolved a specific molecular circuitry of breaks. An additional- and thus far underappreciated- resilience to DNA damage characterized by lack of p53- potential source of DNA damage comes from endogenously triggered apoptosis response,36 which is fundamental for formed aldehydes, as recently reported.21 The number of the maintenance of the HSC pool. damaging insults suffered by each DNA molecule every day Thus, HSCs accumulate mutations over time, at a is estimated on the order of thousands.22 consistent rate, with aging being the most relevant risk DNA damage is continuously generated and needs to factor for AML.37 be continuously corrected by a wide array of repair mechanisms. The bone marrow mainly relies on compo- METHODS TO MEASURE MUTATION nents of the homologous recombination and interstrand 22,23 ACCUMULATION IN BLOOD CELLS crosslink repair systems. Hereditary deficiencies in such components lead to syndromes characterized by bone Measuring the actual rate of mutation accumulation in marrow failure and leukemia, such as Fanconi anemia, the hematopoietic tissue, and in the HSCs in particular, is Bloom syndrome, and ataxia-telangiectasia.24 Treatment critical to understanding the effects of metabolism and with topoisomerase inhibitors or alkylating agents also external environments on leukemogenesis, and might entail a major risk of accumulating genetic alterations for provide clues to prevent or delay leukemia initiation. Is AML.25 the rate of spontaneous mutation in HSCs sufficient, in To be “fixed” into the genome, most DNA alterations the long run, to initiate leukemogenesis, or, alternatively, “require” a round of DNA replication; therefore, the are exogenous stimuli needed to reach a critical threshold accumulation of mutations increases with each cell's repli- of mutation accumulation? In the latter case, which type of cative history. The hierarchical organization of cell prolifer- environment favors mutation accumulation? Because ation in any given tissue, however, reduces its mutation mutations are rare events and HSCs are rare cells, burden. Most tissue cells divide and amplify rapidly but are estimating their true mutation rate has proven challenging periodically eliminated and replaced, thus impeding, in the and has relied mostly on the use of indirect methods. The long run, the accumulation of mutations. In the bone classical strategy has consisted of quantifying cells that marrow, only a small subset of stem cells persists through- spontaneously lose, through a single genetic hit, a meas- out a life span, and these are maintained in a quiescent urable biological property that is assumed to be irrelevant replicative state, thus favoring genomic integrity of the for the fitness of the bearing cell clone. The X-linked genes tissue.26 Furthermore, reactive oxygen species is maintained HPRT (measurable by using an in vitro clonogenic assay) at low levels in the hematopoietic stem cells (HSCs), and its and PIG-A (measurable by flow cytometry) have been the increase causes cell cycle entry and HSC exhaustion.27 most used for this measurement. Using lymphoblastoid However, this is not sufficient to prevent the accumulation cell lines derived from healthy donors, Araten et al of mutations. Certain types of DNA damage accumulate in estimated a PIG-A cellular mutation rate (ie, frequency a replication- and/or oxidation-independent manner. In of PIG-A mutations per cell per division) of 10.6 – particular, spontaneous deamination of methylated cytosine 10 7.38 By considering the PIG-A gene structure, Lynch32 at CpG dinucleotides, which results in C4T substitutions, translated this cellular mutation rate into a base substitu- is the most common single nucleotide variation in the tion rate of 1.47 10 9 (base substitutions per division per germline28 and somatic (in cancer)29,30 DNA, and it does cell). The PIG-A cellular mutation rate increases in AML’s genomic and epigenomic landscape 261 hereditary instability syndromes: 4-fold in ataxia- Lastly, only a minor fraction of the transcribed genome telangiectasia and 40-fold in Fanconi anemia.38 By apply- is truly able to influence cellular fitness and promote ing the same approach to primary cells from healthy tumor progression. Identifying this fraction is of crucial donors, it is evident that a certain degree of genetic importance and is probably the single most important mosaicism is physiologic in the hematopoietic compart- challenge for cancer genomic research at this stage. ment, with a median prevalence of mutated cells on the – – order of 10 6 (4.9 10 6 in Rondelli et al39 and 3.8 – 10 6 in Dobrovolsky et al40). Not All Mutations Are Created Equal: Drivers However, these methods rely on proliferating cells Versus Passengers and the Driver Mutation (primary or immortalized in vitro) with a medium to Rate fi short life span and a history of divisions signi cantly The Cancer Genome Atlas (TCGA) collection of 200 longer than that of HSCs. This makes extrapolation of AML cases (50 analyzed by using whole-genome sequenc- these findings to quiescent stem cells difficult. To 41 ing [WGS] and 150 by using whole-exome sequencing) is circumvent this problem, Dollé et al generated trans- the basis for much of our current understanding of AML genic animals with multiple copies of a reporter bacterial genetics. Ley et al identified 2315 somatic single nucleo- gene (lacZ), which allows enumeration of mutants by tide variants and 270 small insertion/deletions in protein- bacterial transformation of recovered genomic DNA. coding regions, with an average of 13 mutations per Although this method was not applied to analyses of patient.17 Only a minority of these mutations can confer a the bone marrow, it showed that mutation rates are selective growth advantage and promote clonal expansion fi strongly tissue speci c and that mutation prevalence (“driver mutations”) because most of them are “passenger increases with age and tissue proliferation, suggesting mutations” acquired during the lifetime of the leukemic that it is modulated by both systemic and cellular 33 cell (including its preleukemic phase), with no role in metabolism. leukemogenesis. Which are the true drivers, and how do Importantly, all analyses that are based on reporter we identify them? An initial attempt estimated a total of genes actually measure frequency of inactivating muta- 377 putative driver genes, based on mutational patterns in tions, remaining blind to neutral or gain-of-function the COSMIC database (considering data up to 2010). The fi mutations. This nding is important because, on average, 377 driver genes can theoretically be affected by non- only a minority of cancer-associated mutations are synonymous mutations in a total of 34,000 nucleotide truncating and are known or predicted to lead to loss positions, thus predicting a "driver mutation rate" of of protein function and/or expression (14% across 5 43 42 3.4 10 /cell per division. Because driver genes cancers, 17% in AML). In addition, the reference of were identified across cancer types, tissue-specific expres- measurement is usually the gene in its entirety. Extrap- sion patterns were not taken into account. There are, olation of frequencies of single-base mutations from indeed, several examples of disease-specificdrivergenesthat frequencies of gene mutations is solely based on theoret- play no or little role in other cancer types. Notable examples ical derivation and lacks hard evidence. Next-generation in AML are NPM1 and FLT3. Therefore, mutated drivers fi sequencing has nally provided the most direct measure- are probably tissue specific and may dictate the average ment of the physiologic mutation rate in the hematopoi- 26 number of total mutations required for cancer generation. etic compartment to date. By sequencing DNA from Another issue to be considered is that a gene, to clonal colonies derived from HSCs of healthy donors of function as driver, does not necessarily need to be different age groups, the authors estimated accumulation mutated; it may drive tumorigenesis due to its ectopic or of an average of 0.13 exonic mutation per year of life, otherwise-inappropriate expression. In AML, a paradig- resultinginanaverageof 10 total mutations by 70 matic example is HOXA9, which is very rarely mutated years of age. (only as consequence of the NUP98-HOXA9 transloca- fi These gures, however, are too preliminary to obtain a tion); however, it is very often upregulated (through fi fi de nitive understanding of the role and signi cance of epigenetic mechanisms), and its overexpression (in concert mutation rates during leukemogenesis. It is becoming with MEIS1) is sufficient to cause leukemia in mouse clear, in fact, that the overall background mutation rate models.44 This class of deregulated driver genes will be probably represents a suboptimal measurement of the risk harder to identify and requires dedicated approaches (eg, of cancer development. First, mutation rate is now known functional genomics), but in principle, these genes are as to be unevenly distributed across the genome because it is important as the mutated genes for clinical purposes. influenced by several functional states, such as replication timing and chromatin state.29 Second, although there are 23,000 protein-coding genes in the human genome, Approaches to Identifying Driver Genes only a variable fraction of them is expressed in each cell. Therefore, “silent” genes are unlikely to play a causative Sifting through DNA-sequencing data to identify role, consistent with the finding that silent genes accumu- drivers requires precise operative definitions. A gold stand- late more mutations in cancers. ard would be the demonstration, in experimental settings, 262 L. Mazzarella et al of a functional role in leukemogenesis (as described by identification of all driver genes. Bioinformatics Perry and Attar in this issue of the Journal). This has been approaches have become more specific, but the sensitivity accomplished for some genes; it is clear that mutations that we have reached with the available AML genomes is hitting TP53, RAS, or RUNX1 contribute to leukemo- harder to discern. During our analyses of AML genomes,49 genesis. For incompletely characterized genes, which are for example, we identified several mutated genes that have the majority, the task often relies on bioinformatics been causally implicated in the pathogenesis of AML (eg, analyses. Several computational approaches have been ETV6, JAK2, NOTCH1, PRDM16, CBL, CBFB, developed that variably combine elements of 2 fundamen- SETBP1, NSD1) or other cancers (eg, PTEN, MYC, tal approaches: frequentist and pathway oriented. The ARID1A, SF3B1, EGFR, NF1, DAXX, SETD2) but had former evaluates frequency of mutations in a gene across not been identified as drivers by using any statistical many cancer genomes, compared with a background approach. Lawrence et al46 explored this issue and found mutation rate.29,45,46 The latter, by examining large gene that although we are already approaching saturation for interaction networks, identifies the mutated genes that mutations present in 420% of patients, the number of belong to the most significantly mutated sub-networks and candidate cancer genes at lower frequencies grows rapidly produces patterns of co-occurrence or mutual exclusivity with sample size, and near-saturation may be achieved between patients.47,48 Progressive refinements of these with a number of samples oscillating between 600 and approaches have allowed elimination of several genes with 5000 for any given tumor type, depending on their dubious roles in tumorigenesis, mainly genes of large size background mutation rate. In particular, the same authors and high background mutation rate.46 estimated a need of 650 samples for tumors with Analyses of AML genomes have been reported in 0.5 mutation/Mb (a frequency that is in the range of several publications. To compare results, we restricted AML mutation frequency) to identify a catalogue of cancer our survey to 4 publications. These described analyses of a genes mutated at 2%. Indeed, there is still plenty of room similar dataset of AMLs (Z200 samples) using different for more AML sequencing. bioinformatics strategies, all based on a frequentist approach (predicted driver genes for each pipeline are reported in Supplemental Table 1).17,29,46,49 The Muta- FUNCTIONAL ANATOMY OF AML tional Significance in Cancer (MuSiC) algorithm was used MUTATIONS AND EPIMUTATIONS in the initial TCGA analysis and, with slight modifica- Cancer genome sequencing offers the possibility of tions, in a subsequent pan-cancer analysis17,35; Lawrence investigating the clonal anatomy of AML and evolution et al29 used the Mutation Significance package of tools over the natural history of the disease. The relative (MutSig). Our group used a frequentist strategy but abundance of mutant alleles within the DNA molecules analyzed separately the different AML cytogenetic sub- sequenced (variant allele fraction) has revealed the exis- groups.49 The 4 studies led to the identification of a total tence of intrapatient clonal heterogeneity. This hetero- of 53 drivers, with 21 common to all (CEBPA, DNMT3A, geneity varies over time in response to environmental cues, EZH2, FLT3, IDH1, IDH2, KIT, KRAS, NPM1, NRAS, such as systemic treatment,15 and mutations are acquired PHF6, PTPN11, RAD21, RUNX1, SMC1A, SMC3, in a specific temporal order. The knowledge of preferred STAG2, TET2, TP53, U2AF1, and WT1). For the combinations of drivers and their presence in the founding majority of these genes, there is overwhelming evidence clones versus the subclones, and the possibility to better for their involvement in AML, suggesting a good specific- define the pathways involved in AML onset and progres- ity of the bioinformatics analyses. Thirty-two genes were sion, will allow the design of better curative strategies. identified as drivers by using one method alone. Many of Integration of genomics with epigenomics and transcrip- these genes have been directly or indirectly implicated in tomics might shed further light on the multilayered basis AML (BCOR, ASXL1, GATA2, SUZ12, KDM6A, and of leukemogenesis. DDX41) or solid tumors (CTCF, PLCE1, and CHD4). Interestingly, one gene identified was CALR, shortly Initiator and Cooperating Mutations thereafter found to be frequently mutated in myeloproli- ferative diseases.50,51 A mutation can be defined as an initiator when its We have recently proposed a new bioinformatics presence leads to spontaneous development of leukemia. In approach (DOTS-Finder),42 which integrates the fre- transgenic mice, NPM1 mutants or translocation products quentist strategy with a method that exploits the known such as PML-RARa behave as typical initiator mutations structural properties of mutations in tumor suppressor because they give rise to a disease that recapitulates clinical genes and oncogenes (as proposed by Vogelstein et al52). and morphologic features of human AML. However, This approach led to the identification of 2 additional disease arises after long latency periods and with incom- drivers: CBFB and CBX7 (both discussed in the plete penetrance, suggesting the requirement of additional following text). cooperating mutations that are stochastically accumulated These reports raise the question of the number of fully over time. If these are experimentally introduced in the sequenced AML genomes that is sufficient for the genetic context of an existing initiator, leukemia occurs AML’s genomic and epigenomic landscape 263

– with dramatically shorter latency.53 57 Thus, an initiator with strong patterns of mutual exclusivity. The statistically by itself is not sufficient to cause disease and is almost most significant group includes the transcription factor never the sole mutation identified; once the initiator fusion genes NPM1, RUNX1, TP53, and CEBPA. In the mutation is present, however, the endogenous mechanisms second group, there are mutations in FLT3 and genes of mutation accumulation (physiologic or accelerated by encoding tyrosine or serine/threonine kinases, tyrosine the initiator itself) is sufficient to give rise to the selected phosphatases, and RAS family proteins. Lastly, ASXL1, cooperating mutations. In a branched evolution model cohesins, other myeloid transcription factors, and epige- of cancer, initiator mutations mark the totality of netic modifiers are also mutated in a mutually exclusive leukemic clones (the founding clone as well as possible fashion. To identify cooperativity of mutations, Ley et al subclones) and are therefore theoretically useful calculated the likelihood that mutations or groups of as biomarkers for monitoring minimal residual disease mutations occurred together by chance.17 They found and as therapeutic targets to tackle the totality of the that NPM1, FLT3, and DNMT3A mutations show the disease mass. It must be noted, however, that there is most significant pattern of co-occurrence. currently no definitive evidence that the product of an initiator mutation must persist for the entire life span of all Grouping Mutations Into Functional Pathways leukemic cells. It cannot be excluded that initiators sometimes become dispensable during disease evolution, By assembling genomic alterations into molecular sub- implying that targeting the initiator may be biologically networks and examining patterns of mutual exclusivity ineffective. and co-occurrence between these subsets of genes, the In sequencing studies, clonal anatomy reconstructed majority of driver genes in the TCGA AML dataset was through variant allele fraction analysis can help distinguish assigned to 9 "sub-networks": transcription factors (18% initiators from cooperators. In an WGS analysis of 50 of cases), nucleophosmin (NPM1) (27%), tumor suppres- AML cases,17 the authors found that 450% contained a sors (16%), DNA-methylation–related (44%), signaling founding clone and at least 1 subclone, with a maximum (59%), chromatin modifiers (30%), myeloid transcription of 3 subclones per tumor sample. The analysis of exome factors (22%), cohesin complex (13%), and splicing sequencing, which due to higher coverage is more suitable factors (14%).17 Of 200 samples, 199 contained at least for the detection of variants at lower abundance, produced 1 nonsynonymous alteration in 1 of the aforementioned comparable results. Mutations in some genes always categories. This functional compartmentalization might appear (DNMT3A) or almost always appear (eg, RUNX1, prove useful for identifying common mechanisms of NPM1, U2AF1, DNMT3A, IDH2, IDH1, KIT)in leukemogenesis. founding clones. In contrast, mutations in NRAS, TET2, KRAS, CEBPA, WT1, PTPN11, and FLT3 are often Mutations in Signal Transduction present in subclones.35 A recent study by Shlush et al58 Components attempted a reconstruction of the hierarchy of events leading to leukemia: mutations in DNMT3A usually arise Abnormal intracellular signaling was among the first earlier than NPM1 mutations, and do not necessarily hallmarks of cancer to be discovered, and genes that provoke disease, because they are also found in normal regulate signaling are the most frequently mutated in hematopoietic cells and confer a multilineage repopulating AML (60% of cases in the TCGA cohort). Mutations of advantage in xenografts. These types of mutations might the tyrosine kinases FLT3 and KIT, and of the GTPases of be classified as predisposing mutations. NPM1 and FLT3 the RAS family (NRAS and KRAS), are found in the vast mutations occur only in leukemia, with FLT3 usually majority of cases. In addition, protein phosphatases found in subclones. (among which PTPN11 is the most common) were found Analyses of AML molecular subgroups (based on the to be mutated in 6% of cases, frequently associated with presence of known recurring translocations or mutations other kinases. The functional consequences of these in NPM1) suggest that the number of cooperating mutations (eg, the extent of activation of the correspond- mutations required for the development/maintenance of ing signaling pathways) vary significantly, depending on AML depends on the primary initiator mutation. Indeed, the type of mutation and possibly their variant allelic MLL and PML-RARa AML cases harbor the lowest fraction, thus exerting different effects on tumor growth number of mutations per patient in coding genes, whereas and disease outcome. For instance, FLT3 can be mutated RUNX1-RUNX1T1 AML and cases with TP53 mutations either by internal tandem duplications (ITDs), which have the highest. disrupt the autoinhibitory juxtamembrane domain, or by Mutual exclusivity of mutations, within and among ITDs and point mutations that render the tyrosine kinase samples, was investigated in the TCGA AML dataset using domain hyperactive. ITDs are associated with poorer the Dendrixþþ (De novo Driver Exclusivity) algorithm.47 prognosis. Interestingly, this correlates with their ability Dendrixþþ, which is able to identify the genes that are to activate the STAT5 pathway, which only occurs with mutated in large subsets of patients and are exclusively juxtamembrane ITDs.59 Although controversial, some stud- mutated, allowed the identification of 3 groups of genes ies have found a correlation between the mutated/wild-type 264 L. Mazzarella et al

FLT3 allelic ratio and disease outcome (as reviewed by the expression of leukemia drivers; conversely, several Kindler et al60). leukemia drivers are themselves epigenetic modifiers or Mutations of signaling proteins have attracted partic- affect their own expression. In addition, the leukemogenic ular interest in the last decade, due to the availability of potential of many transcription factors depends on their specific chemical inhibitors with good in vitro efficacy. In physical interactions with epigenetic modifiers (see AML, however, kinase inhibitors have experienced little Supplemental Table 2). success so far when tested in clinical trials. FLT3 inhib- Epimutations might affect chromatin globally or at itors, in particular, despite promising results in preclinical specific sites. Recent findings suggest that global changes studies, did not pass the scrutiny of Phase II clinical trials of chromatin are unlikely to be relevant per se for the (reviewed elsewhere60,61), raising doubts concerning transformed phenotype. For instance, the mixed-lineage FLT3's role as a bona fide driver gene. However, 60% leukemia (MLL) and CALM-AF10 translocation products of patients developing secondary resistance to FLT3 induce hypermethylation of H3K79 at the HOXA9 locus, inhibitors acquire specific mutations in the FLT3 drug- a critical event in leukemogenesis. The same translocation binding pocket,62 suggesting that FLT3-mutated leuke- products, however, induce, respectively, increased or mias depend on continuous FLT3 activity for their decreased global methylation of H3K79.65 survival. The hypothesis that the FLT3 mutation is a Two main groups of epigenetic alterations and the corres- cooperator event during leukemogenesis suggests that even ponding modifiers can be identified based on their molec- if 100% killing of FLT3-mutated clones is achieved, this ular targets: DNA methylation and modifications. would unlikely result in disease eradication. Perturbations of the DNA Methylation The Normal and Leukemic Epigenome Landscape in AML and Mutations in Related Genes One third of the “driver” genes recurrently mutated in AML code for proteins that regulate chromatin (as noted DNA methylation is the most studied epigenetic mark elsewhere17,63, see Supplemental Table 2). In addition, the in AMLs (recently reviewed by Schoofs et al66). DNMT1, availability of NGS-based approaches to study chromatin DNMT3A, and DNMT3B are the enzymes responsible at a global level (epigenome) has revealed the existence of for de novo methylation of cytosines and for its main- different types of epigenetic alterations in AML (epimu- tenance.67 Demethylation, once thought to only occur tations). Such degree of epigenomic disruption, visibly passively through cell division, occurs actively through the higher than in most solid tumors, probably reflects an intermediate oxidation of 5mC to 5-hydroxymethyl- elevated reliance on chromatin-mediated mechanisms cytosine catalyzed by TET proteins, a recently discovered to carefully orchestrate a continuous transition from family of dioxygenases.68 Demethylation is then completed multipotent stem cells to a multitude of differentiated by active excision of the oxidized base by the thymine- lineages with an extremely heterogeneous transcriptional DNA glycosylase.69 The oxidation reaction depends on program. Fe(II) and α-ketoglutarate cofactors. α-Ketoglutarate is The nature of epigenetic modifications (both normal normally produced in nonlimiting amounts by IDH1 and and pathologic) is extremely wide: chemical modifications IDH2. The symmetric CpG dinucleotide is the preferred to DNA bases and , remodelling of nucleosome context for 5-C methylation, and in human somatic cells occupancy and positioning, switching of histone variants, 60% to 80% of the 28 million CpGs are normally transcriptional and posttranscriptional regulation by non- methylated.67 CpG islands (CG-dense regions of o1kb coding RNAs, and long-range intra- and intermolecular on average) enclose 10% of all CpGs, are frequently DNA interactions.64 A key common aspect is their associated with promoters, and are prevalently unmethy- plasticity; epigenetic marks can be stably maintained across lated. Overall, CpGs are depleted in the genome compared generations, but they can rapidly change to adapt to with other dinucleotides because of the spontaneous developmental or environmental modifications. A wealth tendency of 5mC to mutate (as discussed earlier). Various of epigenetic “writers,”“erasers,” and “readers” work methods allow study of DNA methylation together to guarantee this equilibrium. The same writers, at the genomic level, with different levels of coverage erasers, and readers can all be altered in AML: as targets of and sensitivity. Bisulfite conversion followed by DNA mutations, as interactors with mutated transcription fac- sequencing is the most comprehensive because it reaches tors, or as readers of aberrant epigenetic information. genome-wide extent and base-pair resolution, but due to Thus, chromatin alterations might be selected to increase cost limitations, such studies in AML are not yet available. the fitness of cancer cells, such as DNA mutations, and Our knowledge remains mostly limited to CpG bring in the additional property of rapid adaptation to a islands.70 changing environment (in the absence of selection). In 2010, the Melnick and Bullinger laboratories The interaction between epimutations and DNA pioneered large-scale DNA methylation investigations of mutations is bidirectional, engaging in a sort of yin-yang selected genomic loci in AML cohorts of considerable size relationship: changes in epigenetic marks selectively alter (92 and 385 patients, respectively),71,72 subsequently AML’s genomic and epigenomic landscape 265 enhanced with higher throughput methods.73 These enhancers (eg, H3K3me1), transcript elongation (eg, H3 studies found clearly distinct DNA methylation patterns K79me2), or transcriptional repression (eg, ). in leukemias, compared with CD34þ cells from the In addition, the same modification can be associated with normal bone marrow, and specific associations with the different or even opposite activities depending on their expression of mutated proteins, including AML1-ETO, genomic context, as exemplified by methylation at H3K36, CEBPa, PML-RARa, MLL, NPM1, RUNX1-RUNX1T1, which is associated with active transcription when it and MYH11-CBFB.1,2 For example, extensive and site- accumulates on coding regions, or with repression when specific losses of DNA methylation are associated with on promoters.79 Each histone mark is actively imposed or MLL fusions and some of the often co-occurring muta- removed by separate enzymatic activities. Acetylation is tions (NPM1, DNMT3A, and FLT3).12,73 Of note, in catalyzed by histone acetyltransferases (HATs) and removed 5 of 16 well-defined epigenetic signatures, no evident by histone deacetylases (HDACs). Methylation is again correlation with genetic abnormalities could be initially more complex, and each specificmarkhasitsown identified. Only a few months later, however, 2 were methyltransferase and demethylase activity. Most histone shown to correlate with mutations in IDH1 and IDH2. bear the evolutionarily conserved SET IDH mutations had been previously identified in the first domain (hDOT1L being a significant exception). Histone WGS of one AML case and then confirmed in 8% of an demethylases are biochemically distinguished in 2 classes: additional 187 AML genomes.13 Mutant IDHs favor the flavin adenine dinucleotide–dependent amine oxidases, anomalous reduction of α-ketoglutarate to D-2-hydroxy- among which the best-characterized is the H3K4 demethy- glutarate (D-2HG), a potent competitive inhibitor of lase LSD1, and proteins bearing the Jumonji domain, TET2-dependent 5-meC hydroxylation,74,75 thus altering which demethylate in an oxidation-dependent reaction that, DNA demethylation. Consistently, leukemias harboring similarly to DNA demethylation, requires α-KG as a IDH mutations exhibited extensive promoter hyperme- cofactor. This biochemical similarity has led to studies thylation, fitting with 2 of the previously identified investigating the possible role of IDH1/2 and D-2HG in leukemia-associated epigenetic patterns73,76 and in stark H3K27 methylation maintenance (in addition to their role contrast to the pattern observed in MLL leukemias. in DNA methylation). Understanding IDH biochemistry and genetics was bene- ficial, with a clinically relevant return. First, D-2HG is a Genome-Wide Analysis of Histone biomarker: its serum levels are selectively elevated in IDH Modifications in AML mutant carriers, and it may be used in the future for posttreatment follow-up and prognostic stratification.77 Genome-wide analysis of histone modifications by Second, IDH is a target for pharmacologic interventions: using chromatin immunoprecipitation-based techniques selective IDH-mutant inhibitors have been developed and relies on antibody quality and antigen abundance, and recently shown to be efficacious in AML and myelodys- thus often requires elevated amounts of sample material. plastic syndrome (MDS) patients with advanced and Consequently, only 2 genome-wide studies of histone refractory disease.78 marks (H3K9me3 and H3 pan-acetylation) have been conducted thus far on large cohorts of primary AML samples.80,81 The authors found a clear segregation of The Histone Modification Landscape in both acetylation and methylation patterns for AML AML samples compared with control samples (normal CD34þ Among the several possible posttranslational modifica- cells and white blood cells). In particular, loss of histone tions of histones, acetylation and methylation are partic- acetylation at a subset of promoters seemed to be a distinc- ularly relevant to leukemias. tive feature of AML, whereas patterns of H3K9me3 Acetylation occurs on lysine residues and consists of the showed prognostic power. However, no correlation with addition of a single, negatively charged acetyl residue, an underlying genetic abnormality was evident, probably which decreases the strength of the interaction between due to the small sample size and cytogenetic heterogeneity the negatively charged DNA backbone and the positive of the cohort. histone tail, leading to chromatin decompaction and Much of our current knowledge on histone modifica- increased accessibility for transcription factors. Several tions in AML comes from studies in experimental models histone lysine residues can be acetylated, but the precise (cell lines or mouse models) but this is likely to change in site makes little difference, as all histone acetylation is the near future. associated with active transcription at gene promoters, with few exceptions (H4K16). Alterations of Active Transcription-Associated The picture is much more complicated for histone Histone Methylation: The MLL-hDOT1L methylation. This can occur on lysine and arginine residues, Association as a Paradigm with the attachment of 1, 2, or 3 methyl groups. Each residue and degree of methylation is associated with specific The best-studied epigenetic histone modifier in AML is functions, such as active gene transcription (eg, H3K4me3), MLL, which represents a paradigm of how genetic and 266 L. Mazzarella et al epigenetic mechanisms interact to deregulate transcription. the complex and confer target specificity. Methylated MLL bears a conserved SET domain that catalyzes H3K4- H3K27 mediates recruitment of the polycomb repressive methylation, a set of modifications associated with pro- complex 1 (PRC1), composed of RING1A/B (the catalytic moters or enhancers. MLL-dependent H3K4 methylation subunits), CBX1-7, MEL-18, RYBP, and BMI1, which is directed at several targets, most importantly homeobox- ubiquitinate H2A on lysine 119, reinforcing repression. containing transcription factors such as MEIS1 and the H3K27 demethylases include JMJD3 and KDM6A. Inter- HOX gene cluster, through which MLL regulates body estingly, H3K27 demethylase activity is coordinated with plan and hematopoietic development. In leukemias, MLL H3K4 methylation, as exemplified by the physical interaction is most commonly disrupted by chromosomal transloca- of KDM6A with the MLL analogues MLL2 and MLL3.88 tions and form a fusion protein with many different A significant fraction of H3K27-methylated genes are partners (recently reviewed by de Boer et al82 and also associated with H3K4me3, in what seems a para- Muntean and Hess83). HOXA9 and MEIS1 are directly doxical coexistence of active and repressive histone marks. upregulated by MLL fusions, and their overexpression is This "bivalent" chromatin is frequently found at the both necessary (in a mutated MLL context) and sufficient promoters of lineage-specifying transcription factors to initiate disease.44,84 Because MLL translocations invar- (including either oncogenes or tumor suppressors); it is a iably lead to loss of the SET domain, it was initially prominent feature of embryonic and adult stem cells and is – difficult to reconcile the selective upregulation of critical partially resolved upon differentiation.89 91 Importantly, targets with the loss of a transcription-associated epigenetic the repressive half of the bivalent chromatin dominates activity. It was subsequently shown that several MLL over the active mark: transcription is poised (as evidenced translocation partners share the same property to interact by the recruitment of RNApol in its poised phosphor- with hDOT1L, a methyltransferase that imposes the ylation state) and does not lead to productive transcript H3K79 mark. Thus, MLL fusion partners substitute the elongation (only small abortive mRNAs are produced). H3K4 methyltransferase activity of MLL, which is asso- Thus, strict polycomb regulation must be enforced to ciated with promoter/enhancer marking, with the H3K79 maintain appropriate transcriptional regulation, with im- methyltransferase activity, which is instead associated with mediate consequences in case of disruption. On one hand, productive transcript elongation. H3K4 methylation at loss of polycomb activity is often sufficient to trigger promoters, required to prime genes for transcription, full transcription of target genes, leading to inappropriate would be provided by the always-persisting wild-type expression of oncogenes; on the other, excessive polycomb- MLL. Consistent with this model, MLL target genes are mediated repression (through mistargeting, gain-of-function H3K79 hypermethylated in their gene body (downstream mutations, or loss of demethylase activity) can silence tumor of promoters) and transcriptionally upregulated. It must suppressors. be noted that these findings were mostly generated by PRC1 and PRC2 play a critical role in AML. Genomic using lymphoid MLL leukemias. In MLL-AML cases, alterations (point mutations or copy number alterations) however, ablation or inhibition of hDOT1L activity occur in a small but sizeable fraction of cases: in the reduces H3K79 methylation at target genes and selectively TCGA cohort, PRC2 components EZH2, EZH1, kills leukemic stem cells, suggesting similar mechanisms of SUZ12, and EED are altered in 7% of cases; PRC1 MLL fusions in AML.85,86 components CBX5 or 7 are altered in 1%. Experimental Swapping and ectopic retargeting of epigenetic enzy- evidence suggests that both the PRC2 and PRC1 com- matic domains seems a common thread of mutated plexes can also be involved in the leukemogenic process in histone modifiers in AML. MLL itself can be fused the absence of underlying mutations, due to their ectopic directly with histone acetyltransferases (CBP, p300) or recruitment by other mutated transcription factors (most arginine methyltransferases (EEN). In the t(5;11) trans- notably seen for MLL fusions and PML-RARa). There- location, the HAT-binding NUP98 fuses with the H3K36 fore, PRC2 is being investigated as a therapeutic target in – methyltransferase NSD1. HOXA9 and MEIS1 are again blood malignancies.92 95 overexpressed (due to H3K36 hypermethylation within Histone demethylases can also be mutated, in particular gene bodies) and are critical targets for leukemogenesis.87 the H3K27 demethylase KDM6A (2% in the TCGA cohort, usually through inactivating mutations, nonsense single nucleotide variants, or deletions). Notably, KDM6A Alterations of Repressive Histone-Methylation: is one of the few genes escaping X inactivation, and its Polycomb, Jumonji, and the Battle for Bivalent mutation is frequently associated with functional loss of Genes the other allele (in female subjects) or of the UTY Among histone modifications associated with gene repres- homologous gene on the Y chromosome.96 Loss of sion, regulation of H3K27 methylation has attracted signifi- KDM6A may be tumorigenic by deregulating expression cant attention in recent years. H3K27 methylation is of HOX genes, which are physiologic targets of catalyzed by the polycomb repressive complex 2 (PRC2), KDM6A.88 composed of EZH1 or EZH2 (which contain a SET Finally, ASXL1, which encodes a H2AK119 deubiqui- domain) and EED, SUZ12, and JARID2, which stabilize tylase (possibly antagonizing PRC1-dependent activity) is AML’s genomic and epigenomic landscape 267 mutated in 2% of AML cases,17 with a loss-of-function in mice initiates leukemogenesis but is not sufficient to mutational pattern. Mutations of ASXL1 are prevalent in induce full-blown leukemias. AML with intermediate-risk karyotypes, in which they Altered levels of wild-type RUNX1 can predispose to occur at a frequency of 17%,97 with frequent co- leukemia. RUNX1 maps to chromosome 21 and is occurrence with RUNX1 translocations and decreased co- amplified in Down syndrome, which is associated with occurrence with NPM1, FLT3, and DNMT3A mutations. increased incidence of AMLs and other leukemias; con- Loss or alteration of its deubiquitylase activity does not versely, RUNX1 haploinsufficiency through germline seem to play a leukemogenic role. In model systems, hemizygous deletions causes familial platelet disorder with ASXL1 mutations lead instead to loss of PRC2-mediated predisposition to AMLs (FPD/AML)102 (as described by H3K27 methylation and upregulation of the HOX locus, Godley in this issue of the Journal). probably through perturbations of the PRC2 complex, with which ASXL1 interacts.98 PML-RARa: The Godfather of the Epigenetic Hijackers INTERACTION BETWEEN EPIGENETIC The activating mechanisms of CBF fusions are remi- FACTORS AND TRANSCRIPTION FACTORS niscent of PML-RARa, the first oncogenic protein shown As mentioned earlier, aberrant recruitment of chroma- to induce epigenetic changes at target genes.103 PML- tin modifiers is critical to the leukemogenic potential of RARa recruits several repressors (HDACs, DNMTs, and several mutated transcription factors. PRC2) at RAR target genes, leading to decreased H3 acetylation, hypermethylation of CpGs, increased H3K27 – methylation, and transcriptional silencing.103 105 Func- Core-Binding Factor Aberrations tional interaction with other transcription factors, such The core-binding factor (CBF) is a heterodimeric as PU.1, dictates DNA-binding patterns.104 Loss of complex between the CBFB subunit and one of the PML-RARa genomic occupancy through pharmacologic RUNX transcription factors. CBFB is a non–DNA-bind- treatments is accompanied by extensive histone mark ing regulatory protein that allosterically enhances DNA remodeling, in both cell lines and primary acute promye- binding of the CBF complex. RUNX1 is the most locytic leukemia cells.104,105 As with RUNX1-RUNX1T1, important RUNX factor in AML (RUNX2 is not PML-RARa forms homo-oligomers that greatly enhance expressed in the hematopoietic system, and RUNX3 is the stability of the repressive complex.100 The clinical rarely mutated in leukemias). It binds DNA, through its success of PML-RARa–destabilizing therapies suggests that RUNT domain, and several chromatin modifiers, through disruption of repressive complex formation is a promising the RUNT and C-terminal domains (HATs [CBP, p300 avenue for drug targeting of other oncogenic fusions. and MOZ] and components of the HDAC- [Groucho/ TLE, mSin3A], histone methyltransferase– [MLL, CCAAT Enhancer Binding Protein Alpha: A SUV39H1, and PRMT6], PRC1- [Ring1b], and SWI- Victim of Aberrant DNA methylation SNF nucleosome remodeling complexes); this topic was recently reviewed by Koh et al.99 These interactions are The CCAAT enhancer binding protein alpha (CEBPA) cell- and context-specific and critical for the function is a transcription factor that acts as a central hub for of RUNX1. granulocytic differentiation.106 It binds DNA through its CBF is frequently altered in AML. RUNX1 is mutated basic domain, possesses 2 N-terminal transactivation in 10% of AML cases and is translocated with domains, and dimerizes with itself and other CEBP family RUNX1T1 (also named ETO)in40% of the AML members through its leucine-zipper domain. It is M2 subtype.99 CBFB is translocated with MYH11 in expressed as a full protein (p42) or as a short isoform 10% of cases, one half of which are the AML-M4Eo with weak transactivation activity (p30). subtype. Only the RUNT domain is maintained in the CEBPA is mutated in 5% to 14% of AML cases RUNT fusion proteins, which thus retain DNA-binding (6.5% in the TCGA). These are loss-of-function activity and acquire aberrant patterns of protein–protein mutations in which p42 is truncated, leaving either intact interactions. RUNX1-RUNX1T1 and CBFB-MYH11 expression of p30 or loss of its dimerization potential. recruit repressive factors at target promoters (SIN3A, Both types of mutations can be accompanied by mutations HDAC8, and DNMT1) and induce their silencing of the other alleles (usually an N-terminal mutation on (reviewed by Di Croce et al100). A specific feature of one allele and a C-terminal on the other).17,106 Hetero- RUNX1-RUNX1T1 AML is a higher number of accom- zygous mutations are instead associated with familial forms panying mutations, compared with other AML subtypes of AML.106 (8 per patient vs a general average of 5),17,101 CEBPA is frequently altered in AML through epige- suggesting that RUNX1-RUNX1T1 is a relatively "weak" netic mechanisms. In particular, a region located 1kb initiating mutation and/or that it induces genetic instability. upstream of the transcription start site is hypermethylated Notably, transgenic expression of RUNX1-RUNX1T1 in 50% of AML cases.107 DNA methylation at the 268 L. Mazzarella et al

CEBPA locus is prevented by a novel species of non- suggesting that they were selected for their ability to confer polyadenylated RNA, which interacts with DNMT1 and resistance to drug-induced apoptosis.115 prevents its binding to hemimethylated CEBPA during TP53 might be functionally inactivated in AML, as the S phase, thus leading to passive loss of methylation and happens in other cancers. Interestingly, the TP53 protein maintenance of active transcription. Whether this mech- is destabilized by miR-3151, which is overexpressed in anism of expression regulation of CEBPA is altered in AML in the elderly and increases leukemogenesis upon leukemias has yet to be determined.108 ectopic expression in mice.116 Two additional tumor suppressors are mutated in AML: WT1 (10% of cases) and PHF6 (3%). Both NPM1, a Central Player Awaiting a Mechanism of these genes have been associated with significantly The NPM1 gene (encoding nucleophosmin) resides on poorer prognosis and resistance to chemotherapy.114,117 chromosome 5q and is the most frequently mutated gene in AML (30% of cases).17,108,109 Fourteen different mutant proteins have been described thus far, which result from NEW KIDS ON THE BLOCKS: COHESIN AND the insertion of short nucleotide stretches (4 or 10 bps) at SPLICING FACTORS NPM1 2 different positions in the exon 12. Despite Cohesin genetic heterogeneity, all mutations cause the same frame- shift, and, as a consequence, the resulting 14 different The cohesin complex is composed of a core tetramer of NPM1mut proteins possess a novel C-terminus that shares SMC1, SMC3, RAD21, and STAG1 or STAG2, which the same CRM1-dependent nuclear export signal.109 This forms a ring around the sister chromatids and holds them event forces the cytoplasmic localization of both mutant together during cell division and DNA repair. Loading and and wild-type proteins (which form stable complexes), unloading of the cohesin complex on chromatin is thus causing the pathognomonic staining pattern of regulated by other proteins, including PDS5B, ESCO1, NPM1-AML (referred to as NPM1c). NPM1 transloca- ESCO2, and NIPBL. Beyond its function in the regu- tions are very rare in AML (NPM1-RAR; o1%), whereas lation of chromosomal integrity, recent studies have they are very frequent in lymphomas (NPM-ALK; 30%- revealed an additional role of the cohesin complex in the NPM1 regulation of gene expression, through stabilization of 40% of anaplastic large-cell lymphomas). muta- 118 tions are usually associated with a normal karyotype long-range chromosomal interactions. Six percent to (85% of cases). In the remaining 15%, NPM1 muta- 12% of AML cases harbor mutations of genes encoding components of the cohesin complex, often with a predict- tions are mutually exclusive with the most common AML 17–19 translocations,9 whereas the other chromosomal aberra- able loss-of-function outcome. Mutations are usually fi tions usually occur in a fraction of blasts, suggesting that heterozygous and signi cantly associated with mutations NPM1 TET2 ASXL1, EZH2 they represent secondary subclonal events. NPM1c is of , , and , suggesting a fi instead frequently associated with FLT3 mutations, and possible synergism with speci c epigenetic abnormalities. their strict initiator–cooperator relationship has recently The lack of increased chromosomal aberrations in AML been shown both in experimental animals54,55 and human with cohesin mutations suggests that the contribution of primary AML.106 Despite extensive studies, the ultimate cohesin mutants to leukemogenesis is mainly transcrip- nature of the pathogenic mechanisms associated with tional deregulation. Accordingly, cohesin loading onto NPM1 mutations is not clear. It associates with rRNA chromatin is reduced in cell lines expressing cohesin 19 – chromatin, interacts with a great number of regulatory mutants. Interestingly, Down syndrome associated proteins (p53, p14ARF, pRB, the ribosomal protein S9, acute megakaryoblastic leukemias often bear elevated CTCF and the transcription factor ATF5), possesses biochemical frequency of cohesin ( 53%) or ( 20%) muta- EZH2 activity of a histone chaperone, and can interact with tions, together with mutations of or other epige- 119 G-Quadruplex DNA elements (enriched at regulatory netic regulators ( 45%). These associations suggest a – regions, most notably at Myc targets).110 Although a synergism with chromosome 21 associated genes (eg, RUNX1 transcription/epigenetic-dependent mechanism has been ). fi hypothesized, it has not been formally demonstrated. Speci c topologic interactions are perturbed when cohesin or CTCF are depleted in experimental conditions, leading to alterations of gene expression.118 Investigating TP53 and Other Tumor Suppressors the 3-dimensional architecture of the genome in AML The frequency of TP53 mutations in AML is low, may allow exploring an entirely new dimension of gene compared with most other tumors (7.5% in the TCGA expression regulation and its aberrations. and in Seifert et al111). They are most frequently associated – with complex karyotypes112 114 and treatment-related Splicing Factors AML (in the context of MDS). In the latter, the same TP53 mutations were present several years before treat- Mutations involving components of the splicing machi- ment (in low-frequency clones of the MDS bone marrow), nery were perhaps one of the most surprising findings in AML’s genomic and epigenomic landscape 269 the last few years in the field of leukemia. Three articles in only partially answered. How many different mutations 2011 first reported the occurrence of these mutations in initiate leukemogenesis? How are they generated, and are – MDS (8.7%-55%).120 122 The same mutations were then they influenced by our environment? What is the role of found in 14% of TCGA AML cases,17 most frequently mutations occurring at very low clonal burden (eg, o5%)? affecting the splicing factor U2AF1 (6.6%). Most muta- How do they influence the natural history of the disease, tions found to date exhibit a relatively well-defined pattern including response to therapy? Also, does treatment select of point substitutions, affecting residues in crucial func- rare mutations or does it accelerate mutation accumula- tional domains.123 It is probably too early to delineate tion? How does the noncoding genome contribute to general features associated with this set of mutations. leukemogenesis? How many pathways need to be targeted to eradicate the disease, and how many for each patient? No doubt, the years to come will be more exciting than Noncoding Genome: The Dark Matter those behind us. Noncoding mutations may also contribute to onco- genesis. Of note, 3 of the 200 TCGA AML cases that were processed by using whole-exome sequencing showed no APPENDIX A. SUPPORTING INFORMATION mutations in coding regions, suggesting the existence of key mutations in noncoding regions. More AML cases Supplementary material cited in this article is available should be subjected to WGS to explore the contribution online at http://dx.doi.org/10.1053/j.seminhematol.2014. of noncoding mutations in leukemogenesis. 08.007.

CONCLUSIONS Back in the era when cancer gene discovery slowly REFERENCES progressed through candidate gene analyses, a popular 1. Berger R, Bernheim A, Ochoa-Noguera ME, et al. Prog- "2-hit" mutation model for myeloid leukemogenesis was nostic significance of chromosomal abnormalities in acute proposed by Kelly and Gilliland in 2002124: one hit nonlymphocytic leukemia: a study of 343 patients. Cancer contributed by mutations in signaling proteins, to increase Genet Cytogenet. 1987;28:293-9. fi proliferation, the other by mutations in transcription 2. Samuels BL, Larson RA, Le Beau MM, et al. Speci c factors, to block differentiation. One of each type would chromosomal abnormalities in acute nonlymphocytic leu- kemia correlate with drug susceptibility in vivo. Leukemia. be required for the expression of the full acute leukemia 1988;2:79-83. phenotype; if alone, either of the 2 would cause a chronic, 3. Grimwade D, Walker H, Oliver F, et al. The importance preleukemic disease (CML or MDS, respectively). The of diagnostic cytogenetics on outcome in AML: analysis of basic idea of this model still holds in 2014, as we have 1,612 patients entered into the MRC AML 10 trial. The discovered hundreds of mutated genes in AML. However, Medical Research Council Adult and Children's Leukaemia the vast majority of AML mutations can be functionally Working Parties. Blood. 1998;92:2322-33. assigned to a handful of cellular pathways. Some aspects of 4. Huang ME, Ye YC, Chen SR, et al. Use of all-trans this model, however, require updating. retinoic acid in the treatment of acute promyelocytic First, changes in the DNA sequence are not the only leukemia. Blood. 1988;72:567-72. mechanism of gene function alterations. Epigenetic 5. Castaigne S, Chomienne C, Daniel MT, et al. All-trans changes, splicing pattern modifications, and posttransla- retinoic acid as a differentiation therapy for acute promye- locytic leukemia. I. Clinical results. Blood. 1990;76: tional modifications can all determine transcriptional 1704-9. deregulation and contribute to leukemogenesis. An impor- 6. Grignani F, Fagioli M, Alcalay M, et al. Acute promyelo- tant consequence is that integrating transcriptional, cytic leukemia—from genetics to treatment. Blood. 1994; genetic, and epigenetic signatures may lead to a better 83:10-25. characterization of each patient's disease biology and more 7. Avvisati G, Lo-Coco F, Diverio D, et al. AIDA (all-trans accurate prognostic scores and treatment algorithms. retinoic acid þ idarubicin) in newly diagnosed acute Second, the number and type of cooperating mutations promyelocytic leukemia: a Gruppo Italiano Malattie Ema- differ in relation to the initiating genetic event. Some tologiche Maligne dell'Adulto (GIMEMA) pilot study. initiating mutations seem to follow a "few hit" model (eg, Blood. 1996;88:1390-8. PML-RARa), whereas others may be weaker and require 8. Walter MJ, Payton JE, Ries RE, et al. Acquired copy several additional hits (eg, RUNX1-RUNX1T1). This number alterations in adult acute myeloid leukemia genomes. Proc Natl Acad Sci U S A. 2009;106:12950-5. might be relevant for the design of pathway-specific 9. Falini B, Mecucci C, Tiacci E, et al. Cytoplasmic nucle- targeting strategies. ophosmin in acute myelogenous leukemia with a normal fi A well-de ned and nuanced picture of the AML karyotype. N Engl J Med. 2005;352:254-66. genomic landscape, however, remains a chimera, as the 10. Kiyoi H, Naoe T, Nakano Y, et al. Prognostic implication current gold standard NGS technologies still suffer from of FLT3 and N-RAS gene mutations in acute myeloid sensitivity and cost limitations, leaving several questions leukemia. Blood. 1999;93:3074-80. 270 L. Mazzarella et al

11. Neubauer A, Dodge RK, George SL, et al. Prognostic 31. Kim SH, Elango N, Warden C, Vigoda E, Yi SV. importance of mutations in the ras proto-oncogenes in de Heterogeneous genomic molecular clocks in primates. novo acute myeloid leukemia. Blood. 1994;83:1603-11. PLoS Genet. 2006;2:e163. 12. Ley TJ, Ding L, Walter MJ, et al. DNMT3A mutations in 32. Lynch M. Rate, molecular spectrum, and consequences of acute myeloid leukemia. N Engl J Med. 2010;363: human mutation. Proc Natl Acad Sci U S A. 2010;107: 2424-33. 961-8. 13. Mardis ER, Ding L, Dooling DJ, et al. Recurring 33. Dollé ME, Snyder WK, Dunson DB, Vijg J. Mutational mutations found by sequencing an acute myeloid leukemia fingerprints of aging. Nucleic Acids Res. 2002;30:545-9. genome. N Engl J Med. 2009;361:1058-66. 34. Subramanian S, Kumar S. Neutral substitutions occur at a 14. Link DC, Schuettpelz LG, Shen D, et al. Identification of faster rate in exons than in noncoding DNA in primate a novel TP53 cancer susceptibility mutation through genomes. Genome Res. 2003;13:838-44. whole-genome sequencing of a patient with therapy- 35. Kandoth C, McLellan MD, Vandin F, et al. Mutational related AML. JAMA. 2011;305:1568-76. landscape and significance across 12 major cancer types. 15. Ding L, Ley TJ, Larson DE, et al. Clonal evolution in Nature. 2013;502:333-9. relapsed acute myeloid leukaemia revealed by whole- 36. Insinga A, Cicalese A, Faretta M, et al. DNA damage in genome sequencing. Nature. 2012;481:506-10. stem cells activates p21, inhibits p53, and induces sym- 16. Walter MJ, Shen D, Ding L, et al. Clonal architecture of metric self-renewing divisions. Proc Natl Acad Sci U S A. secondary acute myeloid leukemia. N Engl J Med. 2012; 2013;110:3931-6. 366:1090-8. 37. Juliusson G, Antunovic P, Derolf A, et al. Age and acute 17. The Cancer Genome Atlas Research Network. Genomic myeloid leukemia: real world data on decision to treat and and epigenomic landscapes of adult de novo acute myeloid outcomes from the Swedish Acute Leukemia Registry. leukemia. N Engl J Med. 2013;368:2059-74. Blood. 2009;113:4179-87. 18. Thol F, Bollin R, Gehlhaar M, et al. Mutations in the 38. Araten DJ, Golde DW, Zhang RH, et al. A quantitative cohesin complex in acute myeloid leukemia: clinical and measurement of the human somatic mutation rate. Cancer prognostic implications. Blood. 2014;123:914-20. Res. 2005;65:8111-7. 19. Kon A, Shih LY, Minamino M, et al. Recurrent mutations 39. Rondelli T, Berardi M, Peruzzi B, et al. The frequency of in multiple components of the cohesin complex in myeloid granulocytes with spontaneous somatic mutations: a wide neoplasms. Nat Genet. 2013;45:1232-7. distribution in a normal human population. PLoS One. 20. Deschler BB, Lübbert MM. Acute myeloid leukemia: 2013;8:e54046. epidemiology and etiology. Cancer. 2006;107:2099-107. 40. Dobrovolsky VN, Elespuru RK, Bigger CAH, Robison 21. Garaycoechea JI, Crossan GP, Langevin F, Daly M, Arends TW, Heflich RH. Monitoring humans for somatic muta- MJ, Patel KJ. Genotoxic consequences of endogenous tion in the endogenous PIG-a gene using red blood cells. aldehydes on mouse haematopoietic function. Environ Mol Mutagen. 2011;52:784-94. Nature. 2012;489:571-5. 41. Dollé ME, Giese H, Hopkins CL, Martus HJ, Hausdorff 22. Ciccia A, Elledge SJ. The DNA damage response: making JM, Vijg J. Rapid accumulation of genome rearrangements it safe to play with knives. Mol Cell. 2009;40:179-204. in liver but not in brain of old mice. Nat Genet. 23. Mohrin M, Bourke E, Alexander D, et al. Hematopoietic 1997;17:431-4. stem cell quiescence promotes error-prone DNA repair and 42. Melloni GEM, Ogier AGE, de Pretis S, et al. DOTS- mutagenesis. Cell Stem Cell. 2010;7:174-85. Finder: a comprehensive tool for assessing driver genes in 24. West AH, Godley LA, Churpek JE. Familial myelodys- cancer genomes. Genome Med. 2014;6:44. plastic syndrome/acute leukemia syndromes: a review and 43. Bozic I, Antal T, Ohtsuki H, et al. Accumulation of driver utility for translational investigations. Ann N Y Acad Sci. and passenger mutations during tumor progression. Proc 2014;1310:111-8. Natl Acad Sci U S A. 2010;107:18545-50. 25. Cowell IG, Austin CA. Mechanism of generation of 44. Kroon E, Krosl J, Thorsteinsdottir U, Baban S, Buchberg therapy related leukemia in response to anti-topoiso- AM, Sauvageau G. Hoxa9 transforms primary bone merase II agents. Int J Environ Res Public Health. marrow cells through specific collaboration with Meis1a 2012;9:2075-91. but not Pbx1b. EMBO J. 1998;17:3714-25. 26. Wilson J, Ley TJ, Link DC, et al. The origin and evolution 45. Wood LD, Parsons DW, Jones S, et al. The genomic of mutations in acute myeloid leukemia. Cell. 2012;150: landscapes of human breast and colorectal cancers. Science. 264-78. 2007;318:1108-13. 27. Tothova Z, Kollipara R, Huntly BJ, et al. FoxOs are critical 46. Lawrence MS, Stojanov P, Mermel CH, et al. Discovery mediators of hematopoietic stem cell resistance to physio- and saturation analysis of cancer genes across 21 tumour logic oxidative stress. Cell. 2006;128:325-39. types. Nature. 2014;505:495-501. 28. Kong A, Frigge ML, Masson G, et al. Rate of de novo 47. Vandin F, Upfal E, Raphael BJ. De novo discovery of mutations and the importance of father's age to disease risk. mutated driver pathways in cancer. Genome Res. 2012;22: Nature. 2012;488:471-5. 375-85. 29. Lawrence MS, Stojanov P, Polak P, et al. Mutational 48. Ciriello GG, Cerami EE, Sander CC, Schultz NN. Mutual heterogeneity in cancer and the search for new cancer- exclusivity analysis identifies oncogenic network modules. associated genes. Nature. 2013;499:214-8. Genes Dev. 2012;22:398-406. 30. Alexandrov LB, Nik-Zainal S, Wedge DC, et al. Signatures 49. Riva L, Ronchini C, Bodini M, et al. Acute promyelocytic of mutational processes in human cancer. Nature. 2013; leukemias share cooperative mutations with other myeloid- 500:415-21. leukemia subgroups. Blood Cancer J. 2013;3:e147. AML’s genomic and epigenomic landscape 271

50. Klampfl T, Gisslinger H, Harutyunyan AS, et al. Somatic 70. Grønbæk K, Müller-Tidow C, Perini G, et al. A critical mutations of calreticulin in myeloproliferative neoplasms. appraisal of tools available for monitoring epigenetic N Engl J Med. 2013;369:2379-90. changes in clinical samples from patients with myeloid 51. Nangalia J, Massie CE, Baxter EJ, et al. Somatic CALR malignancies. Haematologica. 2012;97:1380-8. mutations in myeloproliferative neoplasms with nonmu- 71. Figueroa ME, Lugthart S, Li Y, et al. DNA methylation tated JAK2. N Engl J Med. 2013;369:2391-405. signatures identify biologically distinct subtypes in acute 52. Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, myeloid leukemia. Cancer Cell. 2010;17:13-27. Diaz LA, Kinzler KW. Cancer genome landscapes. Science. 72. Bullinger L, Ehrich M, Döhner K, et al. Quantitative DNA 2013;339:1546-58. methylation predicts survival in adult acute myeloid 53. Wartman LD, Larson DE, Xiang Z, et al. Sequencing a leukemia. Blood. 2010;115:636-42. mouse acute promyelocytic leukemia genome reveals 73. Akalin A, Garrett-Bakelman FE, Kormaksson M, et al. genetic events relevant for disease progression. J Clin Base-pair resolution DNA methylation sequencing reveals Invest. 2011;121:1445-55. profoundly divergent epigenetic landscapes in acute mye- 54. Mallardo M, Caronno A, Pruneri G, et al. NPMcþ and loid leukemia. PLoS Genet. 2012;8:e1002781. FLT3_ITD mutations cooperate in inducing acute leukae- 74. Dang L, White DW, Gross S, et al. Cancer-associated IDH1 mia in a novel mouse model. Leukemia. 2013;27:2248-51. mutations produce 2-hydroxyglutarate. Nature. 2010;465:966. 55. Mupo A, Celani L, Dovey O, et al. A powerful molecular 75. Ward PS, Patel J, Wise DR, et al. The common feature of synergy between mutant nucleophosmin and Flt3-ITD leukemia-associated IDH1 and IDH2 mutations is a neo- drives acute myeloid leukemia in mice. Leukemia. morphic enzyme activity converting alpha-ketoglutarate to 2013;27:1917-20. 2-hydroxyglutarate. Cancer Cell. 2010;17:225-34. 56. Cheng K, Sportoletti P, Ito K, et al. The cytoplasmic NPM 76. Figueroa ME, Abdel-Wahab O, Lu C, et al. Leukemic mutant induces myeloproliferation in a transgenic mouse IDH1 and IDH2 mutations result in a hypermethylation model. Blood. 2010;115:3341-5. phenotype, disrupt TET2 function, and impair hematopoi- 57. Vassiliou GS, Cooper JL, Rad R, et al. Mutant nucleo- etic differentiation. Cancer Cell. 2010;18:553-67. phosmin and cooperating pathways drive leukemia initia- 77. DiNardo CD, Propert KJ, Loren AW, et al. Serum 2- tion and progression in mice. Nat Genet. 2011;43:470-5. hydroxyglutarate levels predict isocitrate dehydrogenase 58. Shlush LI, Zandi S, Mitchell A, et al. Identification of pre- mutations and clinical outcome in acute myeloid leukemia. leukaemic haematopoietic stem cells in acute leukaemia. Blood. 2013;121:4917-24. Nature. 2014;506:328-33. 78. Stein E, Tallman M, Pollyea DA, Flinn IW, Fathi AT, 59. Leischner H, Albers C, Grundler R, et al. SRC is a Stone RM, et al. Clinical safety and activity in a phase I signaling mediator in FLT3-ITD- but not in FLT3- trial of AG-221, a first in class, potent inhibitor of the TKD-positive AML. Blood. 2012;119:4026-33. IDH2-mutant protein, in patients with IDH2 mutant 60. Kindler T, Lipka DB, Fischer T. FLT3 as a therapeutic positive advanced hematologic malignancies [abstract]. target in AML: still challenging after all these years. Blood. Proc. Ann. Meeting AACR. 2014; CT103. 2010;116:5089-102. 79. Wagner EJ, Carpenter PB. Understanding the language of 61. Leung AY, Man CH, Kwong YL. FLT3 inhibition: a Lys36 methylation at . Nat Rev Mol Cell Biol. moving and evolving target in acute myeloid leukaemia. 2012;13:115-26. Leukemia. 2013;27:260-8. 80. Agrawal-Singh S, Isken F, Agelopoulos K, et al. Genome- 62. Smith CC, Wang Q, Chin CS, et al. Validation of ITD wide analysis of histone H3 acetylation patterns in AML mutations in FLT3 as a therapeutic target in human acute identifies PRDX2 as an epigenetically silenced tumor myeloid leukaemia. Nature. 2012;485:260-3. suppressor gene. Blood. 2012;119:2346-57. 63. Dolnik A, Engelmann JC, Scharfenberger-Schmeer M, et al. 81. Müller-Tidow C, Klein HU, Hascher A, et al. Profiling of Commonly altered genomic regions in acute myeloid leuke- histone H3 lysine 9 trimethylation levels predicts tran- mia are enriched for somatic mutations involved in chroma- scription factor activity and survival in acute myeloid tin remodeling and splicing. Blood. 2012;120:e83-92. leukemia. Blood. 2010;116:3564-71. 64. Shen H, Laird PW. Interplay between the cancer genome 82. de Boer J, Walf-Vorderwülbecke V, Williams O. In focus: and epigenome. Cell. 2013;153:38-55. MLL-rearranged leukemia. Leukemia. 2013;27:1224-8. 65. Lin YH, Kakadia PM, Chen Y, et al. Global reduction of 83. Muntean AG, Hess JL. The pathogenesis of mixed-lineage the epigenetic H3K79 methylation mark and increased leukemia. Annu Rev Pathol. 2012;7:283-301. chromosomal instability in CALM-AF10-positive leuke- 84. Faber J, Krivtsov AV, Stubbs MC, et al. HOXA9 is mias. Blood. 2009;114:651-8. required for survival in human MLL-rearranged acute 66. Schoofs T, Berdel WE, Müller-Tidow C. Origins of leukemias. Blood. 2009;113:2375-85. aberrant DNA methylation in acute myeloid leukemia. 85. Bernt KM, Zhu N, Sinha AU, et al. MLL-rearranged Leukemia. 2014;28:1-14. leukemia is dependent on aberrant H3K79 methylation by 67. Smith ZD, Meissner A. DNA methylation: roles in mam- DOT1L. Cancer Cell. 2011;20:66-78. malian development. Nat Rev Genet. 2013;14:204-20. 86. Daigle SR, Olhava EJ, Therkelsen CA, et al. Selective 68. Song CX, Szulwach KE, Dai Q, et al. Genome-wide killing of mixed lineage leukemia cells by a potent small- profiling of 5-formylcytosine reveals its roles in epigenetic molecule DOT1L inhibitor. Cancer Cell. 2011;20: priming. Cell. 2013;153:678-91. 53-65. 69. He YF, Li BZ, Li Z, et al. Tet-mediated formation of 87. Wang GG, Cai L, Pasillas MP, Kamps MP. NUP98- 5-carboxylcytosine and its excision by TDG in mammalian NSD1 links H3K36 methylation to Hox-A gene activation DNA. Science. 2011;333:1303-7. and leukaemogenesis. Nat Cell Biol. 2007;9:804-12. 272 L. Mazzarella et al

88. Agger K, Cloos PAC, Christensen J, et al. UTX and JMJD3 106. Pabst T, Mueller BU. Complexity of CEBPA dysregulation are histone H3K27 demethylases involved in HOX gene in human acute myeloid leukemia. Clin Cancer Res. regulation and development. Nature. 2007;449:731-4. 2009;15:5303-7. 89. Mazzarella L, Jørgensen HF, Soza-Ried J, et al. Embryonic 107. Hackanson B, Bennett KL, Brena RM, et al. Epigenetic stem cell-derived hemangioblasts remain epigenetically modification of CCAAT/enhancer binding protein alpha plastic and require PRC1 to prevent neural gene expres- expression in acute myeloid leukemia. Cancer Res. 2008; sion. Blood. 2011;117:83-7. 68:3142-51. 90. Bernstein BE, Mikkelsen TS, Xie X, et al. A bivalent 108. Di Ruscio A, Ebralidze AK, Benoukraf T, et al. DNMT1- chromatin structure marks key developmental genes in interacting RNAs block gene-specific DNA methylation. embryonic stem cells. Cell. 2005;125:315-26. Nature. 2013;503:371-6. 91. Stock JK, Giadrossi S, Casanova M, et al. Ring1-mediated 109. Mariano AR, Colombo E, Luzi L, et al. Cytoplasmic ubiquitination of H2A restrains poised RNA polymerase II localization of NPM in myeloid leukemias is dictated by at bivalent genes in mouse ES cells. Nat Cell Biol. gain-of-function mutations that create a functional nuclear 2007;9:1428-35. export signal. Oncogene. 2006;25:4376-80. 92. Kim W, Bird GH, Neff T, et al. Targeted disruption of the 110. Federici L, Falini B. Nucleophosmin mutations in acute EZH2-EED complex inhibits EZH2-dependent cancer. myeloid leukemia: a tale of protein unfolding and misloc- Nat Chem Biol. 2013;9:643-50. alization. Protein Sci. 2013;22:545-56. 93. Tanaka S, Miyagi S, Sashida G, et al. Ezh2 augments 111. Seifert H, Mohr B, Thiede C, et al. The prognostic impact leukemogenicity by reinforcing differentiation blockage in of 17p (p53) deletion in 2272 adults with acute myeloid acute myeloid leukemia. Blood. 2012;120:1107-17. leukemia. Leukemia. 2009;23:656-63. 94. Neff T, Sinha AU, Kluk MJ, et al. Polycomb repressive 112. Kihara R, Nagata Y, Kiyoi H, et al. Comprehensive analysis complex 2 is required for MLL-AF9 leukemia. Proc Natl of genetic alterations and their prognostic impacts in adult Acad Sci U S A. 2012;109:5028-33. acute myeloid leukemia patients. Leukemia. 2014;28: 95. Villa R, Pasini D, Gutierrez A, et al. Role of the polycomb 1586-95. repressive complex 2 in acute promyelocytic leukemia. 113. Grossmann V, Schnittger S, Kohlmann A, et al. A novel Cancer Cell. 2006;11:513-25. hierarchical prognostic model of AML solely based on 96. van Haaften G, Dalgliesh GL, Davies H, et al. Somatic molecular mutations. Blood. 2012;120:2963-72. mutations of the histone H3K27 demethylase gene UTX in 114. Patel JP, Gönen M, Figueroa ME, et al. Prognostic human cancer. Nat Genet. 2009;41:521-3. relevance of integrated genetic profiling in acute myeloid 97. Schnittger S, Eder C, Jeromin S, et al. ASXL1 exon 12 leukemia. N Engl J Med. 2012;366:1079-89. mutations are frequent in AML with intermediate risk 115. Wong TN, Ramsingh G, Young A, Shen D, Miller C, karyotype and are independently associated with an adverse Lamprecht T, et al. The role of early TP53 mutations on outcome. Leukemia. 2012;27:82-91. the evolution of therapy-related AML [abstract]. Blood. 98. Abdel-Wahab O, Adli M, LaFave LM, et al. ASXL1 mutations 2013;122(21); Meeting Abstract: 5. promote myeloid transformation through loss of PRC2- 116. Eisfeld AK, Schwind S, Patel R, et al. Intronic miR-3151 mediated gene repression. Cancer Cell. 2012;22:180-93. within BAALC drives leukemogenesis by deregulating the 99. Koh CP, Wang CQ, Ng CEL, et al. RUNX1 meets MLL: TP53 pathway. Science Signaling. 2014;7:ra36. epigenetic regulation of hematopoiesis by two leukemia 117. Owen C, Fitzgibbon J, Paschka P. The clinical relevance of genes. Leukemia. 2013;27:1793-802. Wilms tumour 1 (WT1) gene mutations in acute leukae- 100. Di Croce L, Raker VA, Corsaro M, et al. Methyltransferase mia. Hematol Oncol. 2010;28:13-9. recruitment and DNA hypermethylation of target pro- 118. Merkenschlager M, Odom DT. CTCF and cohesin: link- moters by an oncogenic transcription factor. Science. ing gene regulatory elements with their targets. Cell. 2002;295:1079-82. 2013;152:1285-97. 101. Krauth MT, Eder C, Alpermann T, et al. High number of 119. Yoshida K, Toki T, Okuno Y, et al. The landscape of additional genetic lesions in acute myeloid leukemia with somatic mutations in Down syndrome-related myeloid t(8;21)/RUNX1-RUNX1T1: frequency and impact on disorders. Nat Genet. 2013;45:1293-9. clinical outcome. Leukemia. 2014;28:1449-58. 120. Yoshida K, Sanada M, Shiraishi Y, et al. Frequent pathway 102. Ganly PP, Walker LCL, Morris CMC. Familial mutations of mutations of splicing machinery in myelodysplasia. Nature. the transcription factor RUNX1 (AML1, CBFA2) predispose 2011;478:64-9. to acute myeloid leukemia. Leuk Lymphoma. 2003;45:1-10. 121. Papaemmanuil E, Cazzola M, Boultwood J, et al. Somatic 103. Uribesalgo I, Di Croce L. Dynamics of epigenetic mod- SF3B1 mutation in myelodysplasia with ring sideroblasts. ifications in leukemia. Brief Funct Genomics. 2010;10: N Engl J Med. 2011;365:1384-95. 18-29. 122. Graubert TA, Shen D, Ding L, et al. Recurrent mutations 104. Wang K, Wang P, Shi J, et al. PML-RARa targets in the U2AF1 splicing factor in myelodysplastic syn- promoter regions containing PU.1 consensus and RARE dromes. Nat Genet. 2012;44:53-7. half sites in acute promyelocytic leukemia. Cancer Cell. 123. Scott LM, Rebel VI. Acquired mutations that affect pre- 2009;17:186-97. mRNA splicing in hematologic malignancies and solid 105. Martens JH, Brinkman AB, Simmer F, et al. PML-RARa/ tumors. J Natl Cancer Inst. 2013;105:1540-9. RXR alters the epigenetic landscape in acute promyelocytic 124. Kelly LM, Gilliland DG. Genetics of myeloid leukemias. leukemia. Cancer Cell. 2009;17:173-85. Annu Rev Genomics Hum Genet. 2002;3:179-98.