Biomarkers in Acute Myeloid Leukemia: Leveraging Next Generation Sequencing Data for Optimal Therapeutic Strategies

Next generation sequencing (NGS) is routinely used for mutation profiling of acute myeloid leukemia. The extensive application of NGS in hematologic malignancies, and its significant association with the outcomes in multiple large cohorts constituted a proof of concept that AML phenotype is driven by underlying mutational signature and is amenable for targeted therapies. These findings urged incorporation of molecular results into the latest World Health Organization (WHO) sub-classification and integration into risk-stratification and treatment guidelines by the European Leukemia Net. NGS mutation profiling provides a large amount of information that guides diagnosis and management, dependent on the type and number of gene mutations, variant allele frequency and amenability to targeted therapeutics. Hence, molecular mutational profiling is an integral component for work-up of AML and multiple leukemic entities. In addition, there is a vast amount of informative data that can be obtained from routine clinical NGS sequencing beyond diagnosis, prognostication and therapeutic targeting. These include identification of evidence regarding the ontogeny of the disease, underlying germline predisposition and clonal hematopoiesis, serial monitoring to assess the effectiveness of therapy and resistance mutations, which have broader implications for management. In this review, using a few prototypic genes in AML, we will summarize the clinical applications of NGS generated data for optimal AML management, with emphasis on the recently described entities and Food and Drug Administration approved target therapies.


INTRODUCTION
Acute myeloid leukemia (1) is a clonal malignant expansion of immature myeloid precursors due to block in differentiation. Mutation profiling is standard for routine baseline clinical evaluation of AML. Prior to 2008, there were two functional genetic groups for leukemic pathogenesis: class I (activated signaling genes such as FLT3, KIT and RAS mutations) that conferred the proliferative potential, and class II genes involved in transcription and differentiation such as CEBPA and RUNX1 (2). The development of high-throughput NGS sequencing platforms uncovered many somatic mutations in AML. In 2013, the Cancer Genome Atlas (TCGA) project expanded the functional genetic classes to nine families involved in the pathogenesis of the myeloid neoplasms. An average of 14 mutations were identified in the AML genome, ranging between five to 23 genetic mutations in each case (3). In 2016, a large cohort that enrolled 1540 AML patients described 11 subgroups of genomic alterations with different clinical outcomes using NGS. The additional separate categories included AML with mutations in genes encoding chromatin and RNA-splicing regulators, AML with TP53 mutations and/or chromosomal aneuploidies, and AML with IDH2 R172 mutations (4). Taken together, these findings supported the fact that AML phenotype is driven by underlying mutational signature ( Figure 1).
Accordingly, both the 2016 revisions to the WHO and 2017 European LeukemiaNet (5) incorporated gene mutations into the sub-classification and risk-stratification of AML (6,7). Molecular testing plays a major role in the current World Health Organization (WHO) classification of myeloid neoplasms. Specific genetic abnormalities are regarded as disease-defining mutations and others have important prognostic and therapeutic implications. The WHO recognizes 3 distinct sub-categories of AML based on somatic mutations: AML with NPM1 mutation, AML with bi-allelic mutations of CEBPA, and AML with mutated RUNX1 (provisional entity). Genes such as TP53, RUNX1, IDH1/2 and FMS-related tyrosine kinase 3 (8), among others, are found to be altered in the different subcategories of AML with prognostic and/or therapeutic implications of utmost importance. The latest 2017 European Leukemia Net (ENL) guidelines for AML recommends molecular profiling for mutations in NPM1, CEBPA, FLT3-ITD, TP53, RUNX1, ASXL1, and BCR-ABL1. Therefore, the complete cytogenetic and molecular work-up is essential at the time of initial AML evaluation. The main goal of genome sequencing is the identification of actionable gene mutations and riskstratification. Hence, multi-gene panel next-generation sequencing (NGS) based mutation analysis is the primary mode for assessment for multiple AML associated genes (9).
In this review, we will provide an overview of the diagnostic, prognostic and therapeutic data that can be obtained from routine clinical NGS data in AML. In addition, we will describe the implications of the vast amount of additional data that can be obtained from routine clinical NGS beyond prognostication and management, such as the ontogeny of AML, clues regarding underlying germline predisposition and clonal hematopoiesis, serial monitoring to assess the effectiveness of therapy and resistance mutations. We will use a few prototypic genes for each section, with emphasis on the recently described entities and actionable mutations, specifically pertaining to the Food and Drug Administration (10) approved target therapies and the ongoing clinical trials. AML-defining translocations are beyond the scope of this review.

GENETIC ALTERATIONS WITH FDA-APPROVED TARGETED THERAPIES AML With Mutated FLT3
Category: activated signaling genes.
FLT3 is a receptor with tyrosine kinase (TK) activity, involved in proliferation and differentiation of hematopoietic progenitors. The associated mutations generally affect the juxta-membrane and the TK domain of the receptor. Broadly, mutations in FLT3 are of two types: FLT3-ITD which is the most frequent alteration, and FLT3-TKD which consists of a point mutation in the TK domain (25).

FLT3-ITD
FLT3-ITD are in-frame mutations which consist of duplication of small sequences, ranging from 3 to larger than 400 base pairs resulting in a receptor with an elongated juxta-membrane domain. This leads to constitutive activation of the receptor and activation of intracellular pathways resulting in cellular proliferation (25,32). The frequency of FLT3-ITD mutations in AML ranges from 20-50% (30). FLT3-ITD is associated with proliferative AML with a high WBC count (25). The prognostic implication of the mutation is dependent on the allelic burden (33). The allele burden of FLT3-ITD can be measured using one of the two parameters (10): allele ratio (AR), defined as the ratio of the area under the curve of mutant to wild-type and (2) allele frequency (AF) which is defined as the ratio of the area under the curve of mutant to total (mutant + wild-type). The 2017 ELN adopts a cut-off of 0.5 to differentiate between low versus high AR (34). AML patients with high AR FLT3-ITD had significantly low complete remission (CR) rates, with poor survival and relapse; only high FLT3-ITD AR patients (≥0.5) benefited from allogeneic stem cell transplantation (32). Per ELN, NPM1 mutated AML with FLT3-ITD AR <0.5 is considered as a favorable prognostic subgroup, similar to AML with absent FLT3-ITD mutation, and stem cell transplant is not recommended (34).

FLT3-TKD
FLT3-TKD mutations are less common than the FLT3-ITD; they consist mainly of missense point mutations, deletions or insertions within the TK domain. The most frequent alteration is a point mutation involving nucleotide substitution on codon 835. NGS can identify numerous mutations outside of D835, including deletions. The definitive implications of these various FLT3-TKD in prognostic stratification are still under review (25).
FLT3 mutations can be targeted using tyrosine kinase inhibitor drugs in combination with standard chemotherapy. The addition of tyrosine kinase inhibitor (TKI), Midostaurin, to standard chemotherapy protocol for AML led to significantly longer overall survival (OS) and event-free survival (EFS) in three FLT3 subgroups: FLT3-TKD, FLT3-ITD low and FLT3-ITD high AR (RATIFY-NCT00651261) (12). Midostaurin, the first FDA approved targeted therapy in AML, is a non-selective first generation TKI that targets multiple other pathways including c-KIT, PKC, PDGFR, VEGFR, resulting in higher toxicity (13). In 2018, a more selective "second-generation" FLT3 inhibitor, Gilteritinib, with fewer side-effects received FDA approval for relapsed or refractory FLT3-ITD or FLT3-TKD mutations-positive AML (14) based on the results of ADMIRAL (NCT02421939) and CHRYSALIS (NCT02014558) clinical trials, both of which demonstrated significantly improved outcomes in the Gilteritinib group (15,16). Other first and second generation of inhibitors for FLT3 such as Sorafenib (NCT01398501) (31) and Quizartinib (NCT02668653) (in newly diagnosed AML) (8), have currently reached late stages of clinical testing.
While NGS is ideal for identification of mutations across the entire coding region, specifically FLT3 TKD mutations and small FLT3 ITDs, amplicon-based targeted NGS is unable to pick up majority of the medium to larger ITDs. Hence, ITD mutations are always tested concurrently by PCR followed by fragment analysis using capillary electrophoresis (CE). Alternate computational algorithms, such as Pindel to analyze NGS data have shown promising results with 100% sensitivity and specificity in detecting medium and large insertions at 1% VAF (35). Pindel is a pattern growth algorithm to detect breakpoints of medium and large alterations from paired-end short reads (36).
Resistance mutations to FLT3 inhibitors can emerge over the course of therapy via activation of alternative alternate intracellular pathways. Certain type II (second generation) FLT3 inhibitors do not have activity against TKD mutations, therefore the emergence of a FLT3-TKD mutation during treatment, particularly FLT3 D835 mutation would confer resistance to TKI (16). Crenolanib is a second generation inhibitor with activity against ITD and TKD mutations, hence, able to overcome the treatment resistance resulting from FLT3-TKD alteration (25). Other mechanisms of resistance to FLT3 inhibitors include the emergence of leukemia clones harboring mutations that activate RAS/MAPK pathway signaling, or BCR-ABL1 fusions (37). Hence, sequential mutation analysis is important for early detection of these resistant clone to modulate the therapy accordingly, prior to overt morphologic relapse (Tables 1, 2 and Figure 2).

AML With Mutated IDH1 and IDH2
Category: DNA methylation related.
IDH1 and IDH2 are DNA methylation genes; the mutations in IDH induce dysregulation of epigenetic methylation particularly the function of the TET family of methylators. These aberrations will ultimately result in muting the pathways involved in differentiation of hematopoietic progenitors leading to maturation arrest (5). Moreover, IDH mutations diminish the DNA repair mechanism and result in the accumulation of secondary mutations (39). IDH1 and IDH2 mutations are associated with AML in 4-9% and 8-19% of the cases respectively; they are generally a founding clone sufficient to cause overt leukemia without additional genetic alterations (23). The hotspot mutations involve amino-acid substitutions at codon 132 in exon 4 of the IDH1 gene and codons 140 or 172 in exon 4 of IDH2. IDH2-R172 is mutually exclusive with NPM1 mutation and is regarded as an independent sub-category by genomic analysis, but not currently recognized by the WHO (40).
Similar to FLT3, mutations in IDH1/2 are a prototypic example of targeted therapy in AML. Enasidenib is an oral selective inhibitor of mutant IDH2 enzyme variants R140Q, R172S, and R172K (11); it was FDA approved in 2017 for the treatment of relapsed/refractory AML patients (11). Ivosidenib targets mutant IDH1enzyme leading to normal differentiation and maturation of malignant cells, and was FDA approved in 2019 for treatment of relapsed/refractory AML cases (10).
Despite the presence of hotspot mutations, NGS is ideally suited for baseline identification, serial monitoring of response to therapy and relapse emergence. Droplet digital PCR for specific mutations during serial monitoring can provide a higher sensitivity than standard NGS, however, can miss detection new emerging mutations (Tables 1, 2 and Figure 2).

AML With Mutated KIT
Category: Kinase signaling pathway.
Clinicopathologic and morphologic features: Seen mostly in core-binding factor (CBF) leukemias that encompass AML with inv(16) and t (8;21). In the absence of either translocations, detection of KIT mutation is a helpful clue to search for underlying mastocytosis.
Activating mutations of KIT (encoding a transmembrane glycoprotein) leading to constitutional activation of receptor tyrosine kinase pathway, similar to systemic mastocytosis, GIST and germ cell tumors, are most commonly observed in "core binding factor" (CBF) leukemias which encompass AML with t(8;21) and AML with inv (16). Gain of function mutations in KIT have been found in 2% of AML overall and in a 33% of the CBF leukemias. These mutations tend to occur within exon 17,  (42). Although CBF-AML has a favorable overall prognosis, most reports indicate that co-occurrence of KIT mutations confer an adverse prognosis in AML with both inv(16) and t (8;21), and a higher incidence of relapse in AML with inv(16) (41)(42)(43)(44).
Therefore, CBF-AML with KIT mutations has been reclassified into intermediate-risk group in the National Comprehensive Cancer Network recommendations (43). Nevertheless, data is still unclear since some authors attributed the negative prognostic effect of KIT alterations to only those present at an allelic burden higher than 25% or 35% (43,45,46). Moreover, the prognostic effect of KIT mutation in pediatric CBF AML patients is still uncertain (47).
Given the prognostic implication of the gain-of-function KIT mutations, and over-expression of KIT observed in most CBF-AML including those with KIT mutations (48), studies have explored the addition of KIT inhibitors such as dasatinib and avapritinib to frontline therapy (49) to improve the outcome. Results are promising in terms of reducing relapse rates of KIT mutated CBF-AML to levels comparable to non-KIT mutated CBF AML (49) (Tables 1, 2 and Figure 2).

WHO CLASSIFICATION DEFINING GENETIC MUTATIONS AML With Mutated NPM1
Category: DNA replication and cell cycle.
NPM1 is a molecular chaperone involved in cell cycle progression with multiple critical functions including ribosome biogenesis and transport, apoptotic response to stress stimuli, maintenance of genomic stability, and DNA repair (52). NPM1 mutations are the most common genetic alteration in AML occurring in 25% to 41% of cases (53). These alterations affect exclusively the C-terminal region, leading to cytoplasmic mislocalization of the NPM1 and leukemogenesis induction by inhibition of p53 activity (54). The most frequent mutation is a 4 base pair insertion of TCTG at position 956-959 (55). Multiple gene variants of NPM1 have been identified, all engendering the  (55); these epigenetic alterations, unlike NPM1, are usually identified in pre-leukemic cells, and NPM1 mutations are believed to be a later occurrence (56). NPM1 mutated AML is considered a distinct biological subtype of AML in the latest (2016) WHO classification (57). In the settings of myelodysplastic syndrome (MDS) or myelodysplastic syndrome/ myeloproliferative neoplasms (MDS/MPN), independent studies have suggested that the presence of NPM1 mutation should be an AML-defining mutation, irrespective of the blast percentage, since these patients benefit from AML-type treatment with intensive chemotherapy (58,59). NPM1 mutated AML is generally considered a favorable prognostic marker with good response to induction therapy (60). The prognosis is influenced by concurrent mutations in other genes particularly FLT3 allele ratio. AML patients with NPM1 mutation with absent or low FLT3-ITD AR (<0.5) have similar OS, and are classified as favorable risk by 2017 ELN, whereas AML patients with mutated NPM1 and high FLT3 ITD AR are classified as intermediate-risk along with wild-type NPM1 with absent or low FLT3 ITD AR. NPM1 mutations in AML disappear with CR. However, the persistence of the mutation during follow-up is a sign of adverse prognosis (1). Unfortunately, residual leukemia is not always apparent by morphology or flow cytometry. Cocciardi et al. described the loss of NPM1 mutation at relapse in 9% of NPM1 mutated AML, a finding that alters the prognosis through the selection of clones that harbor FLT3 or DNMT3A mutations exclusively (61). Both of the above findings highlight the importance of serial NGS as a follow up tool for AML to identify the clonal evolution of the disease.
A significant improvement of OS was observed in NPM1 mutated AML with Venetoclax, a BCL2 inhibitor, in combination with hypomethylating agents in patients > 65 years old (17). Venetoclax was approved by the FDA in combination with azacitidine or decitabine for the treatment of newly diagnosed AML in patients older than 75 years old or who have severe comorbidities (14,18). Multiple other targeted therapeutic options for NPM1 mutated AML are ongoing; these include: 1-Deguelin, a selective silencer of the NPM1 mutation that stimulates apoptosis and induces differentiation in AML cells (62); 2-NSC34884 a molecula that disrupts the hydrophobic region that induce NPM1 oligomerization leading to apoptosis; 3-CIGB-300 a molecule that binds NPM1 to prevent the phosphorylation process resulting in induction of apoptosis (63); 4-Selinexor (KPT 330), an inhibitor of exportin 1 responsible of the cytoplasmic mis-localization of mutated NPM1 (64); 6-EAPB0503, a molecule that promotes NPM1 degradation and corrects the NPM1 mis-localization causing an inhibition of the leukemia cell growth (65) ( Table 2).
Recommended method of testing: NGS that includes the whole coding region (66).
RUNX1 is a TF located on chromosome 21q22.12 recurrently involved in leukemia due to multiple types of alterations including chromosomal translocations, mutations, and copy number changes. Somatic mutations occur in up to 15% of the AML, and the frequency is higher in secondary AML arising from MDS (67). De novo AML with mutated RUNX1 have characteristic clinicopathologic features that include male predominance, higher frequency of SRSF2 and ASXL1 mutations, normal karyotype, and absent NPM1 mutations (68).
Per 2017 ELN guidelines, mutated RUNX1 is considered as an adverse prognostic factor. You et al. reported the results of a study on 219 patients with AML, those with RUNX1 mutations had shorter relapse-free survival than patients with wild type-RUNX1 (69). In a meta-analysis of four studies, RUNX1 mutation was associated with dismal prognosis in AML (70). Stengel et al. reported a better outcome when RUNX1 mutation was associated with IDH2 mutation, but worse when associated with ASXL1, SF3B1, SRSF2 and PHF6 mutations (71). Based on the poor outcome observed in studies that evaluated AML with RUNX1 mutations, the 2016 revised WHO AML classification system regards de novo AML with mutated RUNX1 as a provisional entity (57,69,70). However, when only de novo AML cases are evaluated, the outcome is similar to AML with wild-type RUNX1 cases (68).
Notably, as RUNX1 is a gene implicated in germline predisposition disorders, attention should be given to specific findings in RUNX1 alterations. The presence of double mutations or mutations with near heterozygous or homozygous VAF, presence of mutation in familial clusters and history of thrombocytopenia should prompt investigation for an underlying familial platelet disorder with predisposition to myeloid malignancy (72).
While there are no direct targeted therapies at this time, enhancer suppression using bromodomain and extra-terminal motif (BET) inhibitor prevents aberrant RUNX1 and ERG signalinduced transcription in pre-clinical studies (73). Using CRISPR/ Cas9, the BET protein antagonist induced inhibition of RUNX1 resulting in more apoptosis of leukemic cells expressing mutated RUNX1 compared to wild-type cells, inducing an improvement of the survival of mice (74) ( Table 2).
Clinicopathologic and morphologic features: AML with or without maturation.
Recommended method of testing: NGS and PCR followed by direct sequencing (sanger-sequencing) (75).
CCAAT enhancer binding protein (CEBPA) is a TF located on chromosome 9q13.11 expressed in myeloid lineage. It plays a major role in proliferation and differentiation of the myeloid precursors to granulocytes or monocytes (76).
CEBPA mutations are found in 10-15% of patients (77). AML with biallelic CEBPA mutations in a heterozygous or homozygous pattern are significantly associated with better overall prognosis and outcomes independently of other molecular markers (77,78). AML with single CEBPA mutations are uncommon, and more studies are needed to elucidate the clinical significance. Detection of biallelic CEBPA alterations should also prompt investigation for an underlying germline mutation on constitutional DNA with skin fibroblast culture and investigation of underlying familial predisposition (57,79).
Due to the high GC content, CEBPA is a notoriously difficult gene to sequence. There are reports of successful CEBPA sequencing by NGS. This requires extensive modifications and tweaking of the PCR conditions and probe sequences (75).
Jakobsen et al. identified NT5E that encodes CD73 to be upregulated in bi-allelic CEBPA mutant leukemia (80). The efficacy of CD73 inhibitors, including synergy with immune checkpoint inhibitors such as PD-1/PD-L1 are being evaluated as targeted therapy in bi-allelic CEBPA mutant AML (NCT03454451) ( Table 2).

GENES INVOLVED IN PROGNOSTIC RISK-STRATIFICATION OF AML
Tumor Protein p53 (TP53) Category: tumor suppressor.
Clinicopathologic and morphologic features: associated with complex karyotype and shorter OS.
Recommended method of testing: NGS for detections mutations across the whole coding region and low mutant burden (81).
TP53 is mutated in 8-14% of de novo AML cases, but as high as 73% in AML in older patients and therapy-related AML (26)(27)(28). The mutations induce loss-of-function, dominant negative and gain-of-function phenotypes (82). The presence of a TP53 mutation is an independent predictor of poor survival and is associated with a high risk of recurrence and treatment resistance. The 2017 ELN classification regards the presence of TP53 mutation as an adverse risk category (7).
One important additional information obtained from NGS for prognostication, beyond just the presence of a TP53 mutation, is the variant allele frequency (VAF). In MDS, there is significant difference in prognosis between MDS with TP53 VAF >40% vs. <20% (median OS of 124 months vs. OS not reached) (83). Prochazka et al. categorized 98 de novo AML cases using similar cut-offs: VAF >40%, VAF 20% -40% and VAF <20%; sub-clonal TP53 mutation (VAF <20%) showed a negative prognostic effect in terms of CR rate, OS and EFS (81). These findings highlight the importance of a sensitive molecular assay that detects minute sub-clones of TP53 mutation, and the importance of the genetic data provided by NGS, including accurate VAF.
The development of targeted therapy in TP53 mutated cases should take into consideration multiple factors such as alterations affecting related pathways as well as other therapeutic options such as BCL2 inhibitors' in the latter, the TP53 activation may overcome resistance to BCL-2 inhibitors (84). Idasanutlin, an MDM2 inhibitor (85) and Cobimetinib a MEK inhibitor (86), both affecting TP53 expression, are being evaluated in combination with Venetoclax (BCL2 inhibitor) in phase I and phase II trials. Also, clinical trials exploring APR-246, a mutant P53 activator, are underway (NCT03931291); recently, in 2020, FDA granted breakthrough therapy designation for APR-246 in Combination with Azacitidine for the treatment of MDS with a TP53 Mutation (87). The different roles of TP53 in chemotherapy response and particularly the work of Zuber et al. on the activation of TP53 gene in mice provide supportive data of the significant role of exogenous activation of TP53 pathway in regard to treatment response (79) ( Table 2).

GENE MUTATION SIGNATURES FOR IDENTIFICATION OF AML ONTOGENY
Lindsley et al. evaluated the differences in gene mutation patterns between secondary AML, therapy-related AML, and de novo AML to decode the ontogeny of different subsets of AML. The authors identified a set of gene mutations that can provide an objective evidence of AML ontogeny irrespective of clinical information. The presence of a mutations in any of these genes: SRSF2, SF3B1, U2AF1, ZRSR2, ASXL1, EZH2, BCOR, or STAG2 was >95% specific for the diagnosis of secondary AML (88). Majority of these genes, SRSF2, SF3B1, U2AF1, ZRSR2, are genes encoding proteins that belong to the spliceosomal complex that encompasses a large number of members including small nuclear ribonucleoproteins and protein factors responsible for removing introns from a transcribed pre-mRNA (89). EZH2 and BCOR mutations are associated with worse OS in AML (55,90,91) while STAG2 is a part of the cohesion complex.
The findings have major clinical and diagnostic implications. When detected in "de novo" setting, the presence of secondary AML mutations can identify a subset of patients with worse outcome (88). The "secondary-type" mutations, as expected, are frequently present in MDS (92) and chronic myelomonocytic leukemia (CMML) (93). These findings imply these mutations are unlikely to promote the development of acute leukemia without a co-operating event. It is also to be noted that these mutations occur early in the disease course and persist during follow-up despite morphologic CR, hence should not be used for MRD follow up purposes ( Figure 2 and Table 2).

UNCOVERING UNDERLYING GERMLINE MUTATIONS IN AML AND FAMILIAL PREDISPOSITION
Somatic NGS sequencing in AML can potentially uncover incidental germline mutations. The latest WHO recognizes myeloid neoplasms with germline predisposition syndromes as a distinct diagnostic category (57). Recognition of these conditions has major therapeutic implications (94) and different diagnostic and monitoring strategies for the patient himself, as well as the family members. When clinically suspicious, germline mutations testing and proper family screening are important, especially when an allogeneic bone marrow transplant is under consideration.
The new WHO classification introduced four broad categories of AML with germline predisposition: 1-nMyeloid neoplasms with germline predisposition without a pre-existing disorder or organ dysfunction, it includes AML with CEBPA and DDX41 mutations 2-Myeloid neoplasms with germline predisposition and pre-existing platelet disorders, which encompass RUNX1, ANKRD26 and ETV6 mutations 3-Myeloid neoplasms with germline GATA2 mutation and 4-Myeloid neoplasms with germline predisposition associated with inherited bone failure syndromes and telomere biology disorders. Hence, it is recommended to include these genes, CEBPA, DDX41, RUNX1, ANKRD26, ETV6, TERT and GATA2, within the standard NGS panel.
Since most of the mutated genes associated with germline predisposition disorders are also recurrently mutated in sporadic leukemia cases, attention should be given when interpreting the results of sequencing to identify clues suggesting a germline origin, such as double mutations, one with a near heterozygous or homozygous VAF (95). The presence of any of these mutations in a newly diagnosed leukemia is not sufficient to diagnose an AML with germline predisposition. Germline origin by testing constitutional DNA (skin fibroblast culture in most patients with active hematological malignancies) should be performed for confirmation (72,88). It is important to highlight that targeted sequencing will not detect all the germline alterations such as intronic regions and large deletions spanning multiple exons; additional tests should be done to further investigate when clinically suspicious, such as array comparative hybridization for deletions and whole exome sequencing for novel mutations (96,97). The confirmation of germline origin requires a prior genetic counseling since the results may cause significant disturbance of the affected individuals and families if not properly interpreted and handled ( Figure 2 and Table 2).

VARIANT ALLELE FREQUENCY OF GENE MUTATIONS ESTIMATED BY NGS IS IMPORTANT FOR TREATMENT DECISIONS
NGS is a quantitative assay that provides information on mutant allele burden, which is critical for management decisions and predict outcome. VAF is defined as the ratio of reads with mutation against total (mutant + wild-type) reads. Hence, current NGS reports are generally not limited to presence or absence of mutations but include the VAF of each alteration. VAF is important for management decisions and outcome prediction in AML and MDS. This was previously elaborated in the context of TP53 mutations in AML and MDS (81,83).

SEQUENTIAL NGS ASSESSMENT FOR MRD CHEMORESISTANCE AND EARLY RELAPSE DETECTION
Serial NGS is ideal for monitoring AML patients for mutational clearance and/or clonal evolution and relapse. Clearance of somatic mutations in non-preleukemia genes at the time of CR was associated with better OS and decreased risk for relapse (100). Sequential analysis for persistent mutations is particularly helpful to evaluate residual disease in patients treated using novel targeted therapeutic agents, such as IDH inhibitors, as these can pose diagnostic challenges on morphology and/flow cytometry. Therefore, NGS is particularly useful to objectively evaluate evidence of residual disease in these circumstances.
The implications of NGS include assessment of effectiveness of therapy and detection of resistance mutations which can have implications on management and therapeutic regimen choices. Most importantly, sequencing of blast cells can detect TP53 mutations which are independent predictors of poor survival and treatment resistance (7). Moreover, mutations that trigger resistance to FLT3 inhibitors can be identified in AML cases including emergence of leukemic clones harboring mutations that activate RAS/MAPK pathway signaling (37). On the other hand, emergence of FLT3-TKD mutation during treatment can engender resistance to TKI; therefore, sequential mutational analysis is mandatory for early detection of potential treatment resistance and the choice of alternative drugs. Finally, the acquisition of IDH1, WT1, ASXL1 variants in certain AML clones, either present at diagnosis or gained at relapse confer chemotherapy resistance (101).
While NGS facilitates identification of disease evolution and treatment resistance mutations, there are several caveats. First, the sensitivity of standard NGS panels used in clinical laboratories, currently limited to 2-5% VAF (102) is less or comparable to standard flow cytometry techniques for detection of residual disease at the time of morphologic remission. Yet, NGS results are more helpful than latter in settings where blasts show monocytic differentiation or in settings with limited sample quality. The identification of the mutations is partly limited by the intrinsic error rates (0.1% to 1%), that can be potentially overcome using the error correction methodologies such as molecular barcoding (103). Alternately, ultra-deep sequencing NGS with an ultra-high depth of coverage or individual gene assays such as ddPCR assays can be helpful (104). In one study, a high-throughput deep sequencing NGS method showed a detection sensitivity of 10 -4 for SNVs and 10 -5 for insertions/deletions (105).
Second, persistence of "pre-leukemia" mutation signature cannot be used for MRD detection. NGS enables the identification of background clonal hematopoiesis. Clonal hematopoiesis of indeterminate potential (CHIP), defined as the presence of a somatic alterations (either a somatic mutation associated with myeloid malignancy present at least 2% VAF or cytogenetic abnormality) in apparently healthy individuals. These patients have an increased risk of developing a hematologic malignancy, higher risk of developing therapy related myeloid neoplasms following chemotherapy for a solid tumor and increased frequency of adverse cardiovascular events (106).
Most of the CHIP alterations belong to genes DNMT3A, TET2 and ASXL1 (DTA) (106,107). All three genes are involved in epigenetic regulation of myeloid differentiation and are considered to be within preleukemic hematopoetic stem cells (108). DTA gene mutations cannot induce leukemia without a co-operating second hits (109). The mutations in either one of the DTA genes impart negative prognostic outcomes in AML (110)(111)(112). The 2017 ELN classifies ASXL1 mutations as adverse risk category (7), and its presence was >95% specific for AML of secondary origin (88).
Mutations in all three genes can be seen within the entire coding region, and include missense, frameshift, nonsense and splice-site mutations leading to a non-functional protein. The only exception being DNMT3A, which has a hotspot point mutation in codon R882 (113,114). Loss of TET2 function can also occur via mutations in IDH1, IDH2 and WT1 (115,116), that explain the mutual exclusivity between IDH1/2-TET2-WT1 mutations in AML (116). Hence, NGS is an ideal technique for detection of all these mutations.
Importantly, DTA gene mutations likely persist at the time of remission of AML in pre-leukemic clones, hence they cannot be used to detect MRD (109). DNMT3A R882 mutation persists in 75% of AML patients during remission without any negative impact on outcomes (117). Interestingly, ascorbate supplementation can restore methylation patterns and minimize proliferation of blasts (118). A clinical trial evaluating the efficacy of azacitidine and high dose ascorbic acid in AML with mutated TET2 is ongoing (NCT03397173). 9-Subclonal evolution in AML using single-cell technology: All malignancies are genetically heterogeneous, composed of mutationally-defined subclonal cell populations characterized by distinct phenotypes. Precise identification of clonal and subclonal architecture is mandatory to understand the temporal evolution of tumor and the emergence of treatment resistance. Bulk sequencing cannot definitively resolve the actual complex clonal composition of neoplasms generally and AML specifically. Therefore, there is great interest in understanding the genetic alterations at a single-cell level using the newly designed sequencing platforms. Multiple reports published recently highlighted the subclonal selection during the treatment journey. Morita et al. reported the sub-clonal complexity in 37, clonal architecture and mutational histories of 123 AML patients. The authors explored single cell-level mutation cooccurrence and mutual exclusivity revealing novel clonal relationships; and emergence and selection of resistant subclones under therapies by longitudinal analysis (119). Interestingly, current emerging single-cell multi-omics technology, aid in profiling simultaneously single-cell mutations and cell surface proteins in AML cases, allowing correlation of genetic and phenotypic heterogeneity (120). Using the novel features of this technology, Petti et al., using a high-throughput platform to distinguish tumor and non-tumor cells in AML, identified tumor cells showing phenotypic aberrancies and lineage infidelity, evaluated the sub-clonal progression of tumor samples with time with a molecular signature for each sample, and cell-surface markers that could be used to isolate specific cells for downstream studies (121). Another study showed that FLT3-ITD mutation was present in the primitive cells, whereas FLT3-TKD mutation was present in the more differentiated cells within the same tumor. The study demonstrated that FLT3 variants differentially affected AML differentiation that explained the worse prognosis associated with certain alleles (122).

DISCUSSION
We believe that NGS is an exciting tool that has helped pathologists and oncologists tremendously to improve their understanding of AML pathogenesis and clonal evolution of the disease. It has played a major role in designing therapies targeting the disease and control relapse. Genomic analysis of cases at diagnosis and relapse have uncovered the alterations and clonal evolution of the genetic profile of the tumor cells during disease progression. The cited studies and clinical trial results highlight the unique genetic signature of every patient's disease, not only with respect to different combinations of mutations, but also in terms of clonal burden of different mutations, sequence of mutations, and concurrent chromosomal gain and losses. As a consequence of this, 1-the prognostic outcome of each patient can be unique and cannot be simply generalized based on the presence or absence of mutations; 2-the therapeutic molecules should be targeted to attack both the primary clone and emerging sub-clones with potential resistance mechanisms. Hence, a better understanding of each patient's AML genome by NGS is mandatory to make decisions related to appropriate personalized therapy.
Nevertheless, a lot more standardization is needed for implementing NGS in daily clinical practice; for instance, so far there is no universal consensus, yet on which target genes should be included in the sequencing panels. Moreover, the sequencing depth of coverage is still a subjective number chosen by each laboratory for the validation of their sequencer. Furthermore, there is no consistent practice for quality assessment of sequencing data; the accuracy of results is compromised at genome locations with highly repetitive sequences, or in GC-rich locations, common problems encountered during sequencing and data analysis. Finally, results interpretation can be time-consuming and requires specific expertise from bioinformatics and pathology (or equivalent degree). Further standardizations are needed when NGS is used for serial followup to assess measurable residual disease. It is important to keep in mind that bulk NGS represents the "average" findings per cell, and hence the data on genomic complexity is only inferred. On the other hand, single-cell NGS can accurately provide concrete information on sub-clonal architecture, but it has not reached the mainstream clinical work-flow yet (123).
At the same time, while NGS is an important component of AML workup, other testing needs to be performed to accurately sub-classify the disease, specify the prognosis, and determine the best targeted therapy. A comprehensive workup for AML should include a karyotype to identify the disease defining chromosomal translocations including t(8;21), t(15;17), t(16;16) and the new WHO entity AML with BCR/ABL (57); however, some fusions can be cryptic (57), and other testing is required to highlight these aberrations including Fluorescence in situ hybridization (FISH), RT-PCR, RNAseq, and other targeted fusions assays. Moreover copy-neutral loss of heterozygosity (cnLOH) has prognostic significance in patients with acute leukemia (124); Walker et al. showed that LOH mediated by uniparental disomy (UPD) is a common finding in cytogenetically normal AML. Also, UPD involving 13q and 11p are important for genetic risk stratification in these cases (125). FISH and karyotyping can be used to depict copy number aberrations; however, they are limited by low resolution or restriction to targeted assessment. The alternative method of testing is Chromosomal microarray which can characterize chromosomal copy number changes and cnLOH in myeloid malignancies (126).
Overall, NGS has enabled phenomenal advances in understanding of molecular genetics of AML and opened up new horizons for development of highly effective therapeutic molecules and protocols for individualized treatment and monitoring that are completely reshaping the management of the different subtypes of AML.