Next-generation sequencing reveals gene mutations landscape and clonal evolution in patients with acute myeloid leukemia

ABSTRACT Objectives The study aims to understand geneome diversiﬁcation and complexity that developed in Acute myeloid leukemia (AML). Methods Next-generation sequencing (NGS) was used to identify the genetic proﬁles of 22 genes relevant to hematological malignancy in 204 patients with de novo non-M3 AML. Results At time of initial diagnosis, at least one mutation was identified in 80.9% of patients (165/204). The most commonly mutated gene was NPM1 (22.1%), followed by ASXL1 (18.1%), TET2 (18.1%), IDH2 (15.7%), CEBPA (14.7%), FLT3-ITD (13.2%) and DNMT3A (11.8%). Mutations landscape analysis indicated several patterns of co-occurring and mutual exclusive gene mutations. Some correlation was observed between gene mutations and clinicohematological features. Multivariate analysis showed that age >60 years, karyotypes, IDH2 and KIT mutations were the independent unfavorable prognostic factors for OS; NPM1-mut/ FLT3-ITD-wt was independently correlated with prolonged OS; whereas the independent poor risk factors for RFS were karyotypes, high WBC and RUNX1 mutation. According to different genotype demonstrated by multivariate analysis, 163 patients with intermediate-risk cytogenetics were classified into three subgroups: patients with NPM1-mut/ FLT3-ITD-wt or biallelic CEBPA mutation as favorable risk, patients with KIT, IDH2, TP53 or NRAS mutations as unfavorable risk, and the remaining was the intermediate risk. We also obtain information of clonal evolution during leukemia progression by observing five patients who underwent repeat NGS at relapse in our cohort. Conclusion NGS techniques is a useful tool for discovering related gene mutations and clonal evolution in AML genomes, leading to novel targeted therapeutic approaches that could improve patients outcomes.

subclass, it is very important to establish refined stratification through a combination of molecular analysis of gene alterations. The role of genetic changes in AML has been emphasized by the European LeukemiaNet (ELN) risk classification system, for example, FLT3-ITD, NPM1 and CEBPA mutations have been shown to have a significant impact on prognostic evaluation and therapeutical strategies for patients with normal karyotypes [4]. Recent studies have demonstrated that patients with NPM1 mutation lacking FLT3internal tandem duplication (ITD) [6][7][8] or biallelic CEBPA mutation [9,10] are associated with a favorable prognosis. In addition, a variety of additional molecular markers have drawn our attention and may be available for greater comprehensive molecular risk stratification. The mutation profiling of these novel genes including TET2, RUNX1, DNMT3A, TP53, KIT, NRAS, IDH1, IDH2 and ASXL1 may also be correlated with treatment and prognosis outcomes [11][12][13][14]. It has been described that RUNX1, ASXL1 and TP53 mutations have an adverse effect in patients with AML, which have also been added to the recommendations of 2017 ELN genetic risk stratification [15].
The development of next-generation sequencing (NGS) technologies help us identify a greater number of novel and clinically impactful gene mutations and understand the pathophysiological process of AML; which highlights the possibility of translating these insights into improved individual risk assessment and novel therapeutic approaches [16][17][18]. According to whole-genome and exome sequencing in 260 genes in 200 AML patients, Ley et al. suggested that genetic abnormalities in AML could be organized into nine functional categories. Moreover, a complex interplay of genes was identified among the nine functional categories, indicating the important role of genetic events in AML pathogenesis [19]. Researchers have long recognized the importance of genetic changes for AML, but only recently, with NGS technologies has there been a new recognition that discovering AML-related gene mutations profiles are the key for understanding AML oncogenesis and progression. Previous studies have focused on a class of genes modifying DNA methylation, such as DNMT3A, IDH1, IDH2 and TET2 [20][21][22][23]. It is believed that these gene alterations may have prognostic implications and also involved in leukemia pathogenesis. However, the systematic blueprint of recurrent genetic mutations has not been comprehensively established.
In addition, AML are tumors with a fundamental property of genetic and epigenetic heterogeneity. Reflecting the clonal evolution within the disease as a dynamic process driven by constant loss and acquisition of somatic mutations during the course of tumor. More recently research of clonal evolution is emerging as an important new area, contributing to understand geneome diversification and complexity develop in leukemia [24,25].
In this study, we identified the frequently and clinical signicance in 22 genes relevant to hematological malignancy by using NGS in a group of de novo AML. We also report information on clonal evolution by repeating NGS at relapse, therefore, may benefit from personalized therapy.

Patients
We retrospectively analyzed 204 patients newly diagnosed as de novo non-M3 AML who were performed with targeted NGS of marrow samples from First Affiliated Hospital of Nanjing Medical University. All patients were diagnosed and classified according to the FAB Cooperative Group Criteria [26]. Information on treatment regimen was available in 190 AML cases. A total of 135 patients received standard induction chemotherapy with cytarabine for 7 days and mitoxantrone or anthracyclines (idarubicin or doxorubicin), and 49 patients received induction chemotherapy, which consisted of homoharringtonine 3 mg, D1-7; and cytarabine 100 mg/m 2 , D1-7. The remaining patients received low dose therapy (cytarabine 10 mg/m 2 , D1-14; aclarithromycin 10 mg, D1-14 and G-CSF 200 μg/ m 2 , D1-14) and/or supportive care because of underlying comorbidities or decided by the physicians. After achieving complete remission (CR), 166 patients received consolidation chemotherapy with high-dose cytarabine based chemotherapy for two to four courses. This study was approved by the hospital ethics committee.

Cytogenetics
Chromosome analysis was performed according to standard protocols. Bone marrow cells were collected after 24-48 h of unstimulated culture. Mitotic metaphase cells were banded by trypsin-Giemsa technique and the karyotypes were described according to the International System for Human Cytogenetic Nomenclature (ISCN).

Statistical analyses
Analysis of associations between different mutations, and between clinical features and gene mutations was performed using χ 2 test and Fisher exact test for categorical variables and One-way Anova test for continuous variables. OS was measured from the date of diagnosis to death or alive at last follow-up, and relapse free survival (RFS) was measured from complete response (CR) to death, relapse or alive in CR at last follow-up. Univariable and multivariable Cox proportional hazard regression model was used to identify the prognostic value of gene mutations. The Kaplan-Meier method and log-rank test were applied to analyze OS in intermediate-risk group. All P-values are two-sided, and P < 0.05 was considered as statistically significant. Statistical analyses were performed using SPSS version 17.0 software.

Patient characteristics
A total of 204 patients were analyzed in the study, including 101 females and 103 males with median age 54.4 years (range, 20-86 years). According to 2016 World Health Organization classification criteria, 101 patients were classified as AML with recurrent genetic abnormalities, 18 patients as AML with myelodysplasia-related changes, and 85 patients as AML, not otherwise specified. CBF leukemias, whose cytogenetic subtypes are t(8;21) and inv (16)
Pervious study suggested greater mutation heterogeneity in intermediate-risk cytogenetic group than favorable risk group or adverse risk cytogenetic group [12,27]. In patients with favorable risk cytogenetic AML, KIT (43.8%) was found to be the most frequent mutation. The most commonly mutated genes in adverse risk cytogenetic group were TP53 (30.8%) and ASXL1 (23.1%), while other genes present with low frequency. With respect to 162 patients in intermediate-risk cytogenetic group, eight mutated genes (NPM1, TET2, IDH2, CEBPA, ASXL1, FLT3-ITD, DNMT3A and IDH1) present with greater than 10%, suggesting great molecular heterogeneity among this group. Among them, the most commonly mutated gene was NPM1 (27.2%), TET2 (19.8%), IDH2 (17.9%), CEBPA (16.7%) and ASXL1 (16.7%). When we analyzed mutations landscape and several patterns of co-occurring mutations, significantly cooccurrence was found between NPM1mutation and IDH2, FLT3-ITD, DNMT3A and TET2 mutation; and between ASXL1 mutation and TET2 mutation. We also noted significant pairwise associations of FLT3-ITD and DNMT3A with IDH2 mutations. In contrast, mutual exclusivity of gene mutations was observed between TP53 and FLT3-ITD, RUNX1and NPM1, RUNX1and CEBPA, and KIT and IDH2. We also found a high exclusivity between IDH1 and NRAS. The pairwise cooperativeness between gene mutations is illustrated in Figure 2.

Integrative stratification of cytogenetics and gene mutations
Among 15 patients with favorable risk cytogenetics, 6 patients had KIT mutation, but it presented an    (Table 5).

Variant allelic frequencies between different genetic groups
We examined variant allelic frequencies (VAFs) of 13 genes that mutated in ≥8 patients. The median and mean VAFs of mutated gene were 35.4% and 38.3%, respectively. These somatic mutations can be classified into several classes according to their different biological function, including tumor suppressors, chromatin modifiers, DNA methylation, NPM1, transcription factors and activated signaling. As shown in Figure 4, only PHF6 and ASXL1 had a mean VAF slightly greater than 50%. TP53 mutation and DNA methylation related mutations involving TET2, IDH2, DNMT3A and IDH1 mutation had a similar VAF about 40%, which was consistent with the overall mutations VAF in our study. Moreover, NPM1 and transcription factors (CEBPA and RUNX1) had a VAF about 30%. It is of note that, mutations in FLT3, NRAS and KIT that involved in signaling pathways were present with the lowest mean VAF, indicating that they are later events and exist only in subpopulation of leukemia stem cells during the course of AML.

Clonal evolution of AML at relapse
To study clonal evolution patterns during leukemia progression, we observe five patients who underwent repeat NGS at relapse in our cohort. Information of clonal evolution from all five patients was shown in Figure 5. These data indicate several interesting evolution patterns during AML progression. First, in two patients, although the same gene mutations were detected both at diagnosis and relapse, some genes were present at changing VAFs from diagnosis to relapse. In patient D and E, we observed significant reduction of NRAS

Discussion
It is widely accepted that genetic diversity contributes to AML with great heterogeneity, so exploring mutational profiling is an important step to approach the truth of AML biological and clinical features. Fortunately, NGS techniques offering a powerful tool for discovering related gene mutations in AML genomes, leading to novel targeted therapeutic approaches that could improve patients outcomes. Recent advances in gene mutation studies broaden our eyes  to view AML from a more comprehensive and deeply perspective, not only the simple morphology, but also the cytogenetic and genetic aberrations [5,7,12]. Nowadays, more and more genes are found and classified in AML field [28], simultaneously examination of gene mutations in AML has been a custom program around the world. In our study, we examined 22 frequently mutated genes in 204 de novo non-M3 AML samples at diagnosis using an NGS panel currently in routinely clinical examination at our department. At least one mutation was identified in 81% of patients, indicating a clinical practicality of our panel in initial AMLs detection. Consistent with previous studies [29,30], seven genes (NPM1, ASXL1, TET2, IDH2, CEBPA, FLT3-ITD and DNMT3A) had a mutation prevalence of >10%, but NPM1 mutations were more frequent, followed by ASXL1, TET2, IDH2, CEBPA, FLT3-ITD and DNMT3A. Kihara et al. [11] reported that mutations of FLT3, NPM1, CEBPA, DNMT3A and KIT gene had high mutation frequency in AML patients in Japan. Molecular mutation of NPM1 gene is present with high  mutation rates in AMLs, ranging from24 to 29% in adults, 3-11% in children, especially in patients with a normal karyotype [16,19,31]. However, the frequency of NPM1 mutation seemed to be slightly lower in our cohort. This finding may be due to its partly younger patients selection. Clinically, patients with NPM1 mutation had higher BM blast infiltration and were common in intermediate karyotype risk group. Previous studies identified that patients in favorable or adverse risk groups were present with lower kinds of gene mutations compared to those in intermediaterisk group [11]. Some studies reported KIT mutation were the most frequently mutated genes in CBF-AML, in both adults (16-46%) and children (19-44%) [32,33]. We detected KIT mutation at 42% frequency. For patients in adverse risk cytogenetic group, the most commonly mutated genes were TP53 (30.8%) and ASXL1 (23.1%), while other genes present with low frequency. In terms of intermediate-risk cytogenetic group, eight mutated genes (NPM1, TET2, IDH2, CEBPA, ASXL1, FLT3-ITD, DNMT3A and IDH1) were present with greater than 10%, suggesting great molecular heterogeneity among this group.
Our study identified statistically significant association of mutations occurring in more than 5% of cases. In agreement with previous reports, DNMT3A and FLT3-ITD mutations were recognized as being frequently overlapped with NPM1 mutation [34]. A series of associations of co-occurring mutations were also identified, including frequent co-mutation of NPM1 with IDH2, NPM1 with TET2, TET2 with ASXL1, and IDH2 with FLT3-ITD and DNMT3A. Association of cooccurring mutations suggests functional synergism among different oncogenic pathways in AML pathogenesis. In contrast, mutual exclusivity was found for mutations in NPM1 and the transcription factor genes. Our study indicated that mutations that were common in NPM1 and CEBPA were mutually exclusive of RUNX1, thereby supporting the biological evidence that the mutations are early events during leukemogenesis [19]. Some correlation was also observed between gene mutations and clinicohematological features. Consistent with other studies, TET2 mutation was seen more often in older patients, AMLs with NPM1 mutation lack expression of both CD34 and HLA-DR, and AMLs with CEBPA mutation were associated with greater expression of CD34 [35,36]. However, we also found AMLs with FLT3-ITD mutation were associated with fewer CD34 and HLA-DR positive events, which was conflicts with another study [37]. Further studies with more patients are needed to verify such correlaton.
It is important to estimate the risk of death and relapse both from the viewpoint of physicians as well as patients. Recently, mutations in FLT3-ITD, NPM1, CEBPA, TP53, ASXL1 and RUNX1 have been revealed to have prognostic value, and they have been incorporated into the 2017 ELN risk stratification [15]. The well-established classification and prognostic stratification for AML were cytogenetics combined with FLT3-ITD, NPM1 and CEBPA. It has been described that patients with biallelic CEBPA mutations or a NPM1 mutation but no FLT3-ITD (or FLT3-ITD with a low allelic ratio) convey a favorable prognosis [9,38,39], whereas RUNX1, ASXL1 and TP53 mutations predict for a poor outcome [28,40]. Till now, for other genes, including IDH1, IDH2, TET2 or DNMT3A, etc., there was no enough evidence to warrant their implementation in the ELN stratification system. In our study, univariate analysis revealed that NPM1mut/ FLT3-ITD-wt and biallelic CEBPA mutations were associated with favorable OS, nevertheless, mutations in RUNX1, TP53 and ASXL1 were associated with a significantly worse OS. In addition, RUNX1 mutation was significantly correlated with a poorer RFS both in univariate analysis and multivariate analysis. Traditionally, karyotype is thought to be the most important prognostic factor in AML patients. A study from Tsai et al showed that older patients had a higher incidence of adverse cytogenetics, but lower incidence of the favorable karyotype as the core binding factor alternations [41]. In agreement with their finding, age and karyotype risk have significant influence both on OS and RFS in univariate analysis. Multivariate analysis indicated that age >60 years and karyotype risk were independent predictor of poor OS, while karyotype risk and high WBC were the independent poor risk factors for RFS in AML. Of note, a significantly unfavorable impact of IDH2 and KIT mutations on OS was only found in multivariable analysis, whereas IDH1 mutation did not affect disease outcome in the entire study cohort. Usually, IDH1/2 mutations are correlated with an unfavorable clinical outcome, although the prognostic impact is assessed by complex interplay of genetic aberrations [42]. According to a series of studies, the prognostic significance of IDH1/2 mutations is controversial. Patel et al. [12] reported that IDH mutations were associated with favorable prognosis in patients that were FLT3-ITD-negative and NPM1-mutation-positive. Some studies reported that IDH1 mutation was an unfavorable prognostic factor or had no effect on OS [43,44]. Several studies and reviews have reported favorable prognosis of patients with IDH2 mutation [23]. However, a study performed by Green et al indicated that IDH2 R172 mutation was significantly associated with a shorter OS than IDH2 R140 mutation [45]. There have been reports of high mutation rates of KIT mutation in patients with CBF-AML, which conveys an adverse prognosis [46]. For intermediate-risk group, we also found KIT mutation was associated with short OS. Mutation of other genes, including IDH2, NRAS, TP53, TET2 and DNMT3A had a nonsignificant trend of unfavorable OS, although some research reported an unfavorable impact of TET2 and DNMT3A on clinical outcome [21,22]. According to the above results, we have developed a risk classification system for AMLs using seven mutations and cytogenetics. AMLs in intermediate-risk cytogenetic group could be divided into three subgroups based on genotype, in which 33 patients with favorable genotype expressed favorable prognosis, whereas 42 patients with unfavorable genotype showed poor prognosis. Further work is required to determine their definite influence on prognosis.
It is well known that genetic and epigenetic evolution and clonal selection were complex events involved in the course of AML progression [24,25]. According to allelic ratio at diagnosis, DNA methylation-related mutations including TET2, IDH2, DNMT3A and IDH1 mutations had a similar VAF of about 40%, suggesting their dominant epigenetic alleles during clonal evolution. In agreement with previous reports, somatic mutations in epigenetic modifiers are commonly occur early, leading to founding clone in AML. However, mutations in FLT3, NRAS and KIT that involved in signaling pathways were present with the lowest VAF, consistent with the notion that these mutations are acquired relatively later in AML transformation [47,48]. To explore clonal patterns underlying relapse, we also analyzed five patients who underwent repeat NGS at relapse in our cohort. Interesting, although mutations in NRAS and CBL genes play established roles as drivers in leukemogenesis, they have a significant reduction of VAF or even are absent at relapse if initially present with subclone. This observation suggests that these late events may not be sufficient to drive leukemogenesis, but these mutations could potentially be the therapeutic targets considering clonal selection. The acquisition of additional clones which appeared at relapse provides the possibility of these mutations in minimal residual disease monitoring by NGS techniques. Nowadays, more and more studies support clonal selection although it is difficult to demonstrate that the emergence of new mutations results from subclonal selection or minor mutations present below the reporting threshold at initial diagnosis [49]. Some genetic mutations with role as initiators in early blood cell transformation, such as TET2, IDH1/2 and DNMT3A, may not be suitable as biomarkers to detect minimal residual disease.
In summary, we showed complex mutation profiling of several genes relevant to hematological malignancy and found that some genetic alternations may be prognostic markers in AML and invaluable in risk-stratification and guide the therapy approaches. Our results also further provided evidence of clonal architecture and highlight the genomic complexity and diversification in leukemogenesis and transformation. Whereas our study have some limitations such as heterogeneity and a relatively small number of samples. A more comprehensive research on the mutation profiling of somatic mutations involved in AML pathogenesis is critical for improved risk stratification and better therapy outcome.