Next-generation sequencing with a myeloid gene panel in core-binding factor AML showed KIT activation loop and TET2 mutations predictive of outcome

Clinical outcome and mutations of 96 core-binding factor acute myeloid leukemia (AML) patients 18–60 years old were examined. Complete remission (CR) after induction was 94.6%. There was no significant difference in CR, leukemia-free-survival (LFS) and overall survival (OS) between t(8;21) (N=67) and inv(16) patients (N=29). Univariate analysis showed hematopoietic stem cell transplantation at CR1 as the only clinical parameter associated with superior LFS. Next-generation sequencing based on a myeloid gene panel was performed in 72 patients. Mutations in genes involved in cell signaling were associated with inferior LFS and OS, whereas those in genes involved in DNA methylation were associated with inferior LFS. KIT activation loop (AL) mutations occurred in 25 patients, and were associated with inferior LFS (P=0.003) and OS (P=0.001). TET2 mutations occurred in 8 patients, and were associated with significantly shorter LFS (P=0.015) but not OS. Patients negative for KIT-AL and TET2 mutations (N=41) had significantly better LFS (P<0.001) and OS (P=0.012) than those positive for both or either mutation. Multivariate analysis showed that KIT-AL and TET2 mutations were associated with inferior LFS, whereas age ⩾40 years and marrow blast ⩾70% were associated with inferior OS. These observations provide new insights that may guide better treatment for this AML subtype.


INTRODUCTION
Acute myeloid leukemia (AML) is a group of heterogeneous diseases with distinct clinicopathologic, cytogenetic and genetic characteristics. Conventional therapeutic approaches include induction and consolidation chemotherapy. Patients at high risk of relapse receive allogeneic hematopoietic stem cell transplantation (HSCT) as post-remission therapy. AMLs with translocation involving core-binding factors (CBF) including t(8;21)(q22;q22) and inv(16)(p13;q22) or t(16;16)(p13;q22) constitute a distinct clinicopathologic subtype as defined by the World Health Organization (WHO), and occur in 15-20% of adult patients. 1,2 In general, CBF-AML has a superior outcome compared with other cytogenetic subtypes, with a long-term overall survival (OS) of 40-60%. [3][4][5][6][7] However, their clinical outcome is highly heterogeneous. Mutations of type III receptor tyrosine kinase KIT in the activation loop at exon 17 (D816, N822) and the extracellular domain at exon 8, which are rare in other AMLs, occurred in 12-46% cases of t(8;21) and 9-53% of cases of inv (16). [8][9][10][11][12][13][14][15] These mutations may induce ligand-independent KIT activation and are generally associated with a higher risk of relapse and poorer prognosis. The impacts of other gene mutations frequently identified in AML, including RAS and FLT3, on outcome in these AML are less well defined. With the advent of next-generation sequencing (NGS), the genomic landscape of AML can be examined in detail. In particular, The Cancer Genome Atlas (TCGA) examined the genomic profile of 200 de novo AML and provided important genomic information of this disease. 16 However, the number of CBF-AML in TCGA was limited.
In this study, we examined the clinical outcome of a consecutive cohort of patients with CBF-AML who were treated with a uniform approach and examined their mutation spectrum using NGS. The aim was to define the clinicopathologic characteristics and identify novel gene mutations of prognostic values for the design of better therapeutic strategies in this AML subtype.

Treatments
Specific chemotherapy regimens are shown in Supplementary Table S1. Induction chemotherapy comprised daunorubicin and cytarabine. Reassessment bone marrow was performed between days 21 and 28 after induction. Consolidation comprised four courses of high-dose cytarabine. Before 2012, two courses of daunorubicin and etoposide were given as consolidation before high-dose cytarabine. Remission and relapse were defined by standard criteria. Salvage chemotherapy included ICE (idarubicin, cytarabine, etoposide), MAC (mitoxantrone, cytarabine), FLAG (fludarabine, cytarabine, granulocyte-colony stimulating factor) and CLARA (clofarabine, cytarabine). Patients who achieved second complete remission (CR2) received the same salvage chemotherapy as consolidation until HSCT or leukemic progression (Supplementary Table S2).

Allogeneic hematopoietic stem cell transplantation
Indications of HSCT at CR1 for eligible patients included two or more inductions to achieve CR1, additional chromosomal abnormalities and presence of KIT mutations. HSCT was recommended for all eligible patients in CR2. Allogeneic HSCT was performed in Queen Mary Hospital. Bu-Cy (busulfan; cyclophosphamide) was employed as myeloablative conditioning, and Flu-Bu (fludarabine, busulfan) and Cy-TBI (total body irradiation) for non-myeloablative conditioning. Patients received standard antimicrobial and graft versus host disease prophylaxis as described previously. 17 Patients who relapsed after HSCT received one of the salvage regimens aforementioned followed by infusion of mobilized peripheral blood HSC from the original donors. 18 Next-generation sequencing  Table S3). Workflows of MiSeq sequencing library preparation, variant calling and annotation as well as detection of FLT3 internal tandem duplication by ITDseek has been previously described. 19 Complex insertions and deletions were detected by an in-house designed algorithm INDELseek on a Cray XC30 supercomputer (Cray Inc., Seattle, WA, USA).  Abbreviations: AML, acute myeloid leukemia; BM, bone marrow; CBF, core-binding factor; CR1, first complete remission; CR2, second complete remission; FLAG, fludarabine, cytarabine, granulocyte-colony stimulating factor; HSCT, hematopoietic stem cell transplantation; ICE, idarubicin, cytarabine, etoposide; MAC, mitoxantrone, cytarabine; R1, first relapse; WCC, white cell count; 7:3, cytarabine, daunorubicin. a Additional chromosomal abnormalities included trisomy-X (n = 10), trisomy-Y (n = 24) and others (n = 31). b A total of 88 patients achieved CR1.

Orthogonal validation of detected variants
Mutations with variant allele frequency (VAF) o20% were confirmed by one of the following methods according to the specific gene mutations involved. These included microfluidic PCR using Access Array 48.48 (Fluidigm, South San Francisco, CA, USA) and a different primer panel followed by MiSeq NGS, bi-directional Sanger sequencing or PCR fragment analysis by capillary electrophoresis using ABI 3130xl genetic analyzer (Applied Biosystems, Foster City, CA, USA). Only confirmed variants were analyzed in this study.

Survival and statistical analyses
Leukemia-free survival (LFS) was defined as the time between first complete remission (CR1) to first relapse. Unless otherwise specified, LFS was censored at the date of the last follow-up or death. Overall survival (OS) was defined as the time between diagnosis and death or the date of the last follow up. Treatment-related mortality was defined as death within 30 days of the last chemotherapy or HSCT. Numerical data were compared using Mann-Whitney U-test for nonparametric and Student's t-test for parametric parameters. Categorical data were compared using χ 2 test. Different thresholds for age (increment of 10 years, from 20 to 50 years), white cell counts (increments of 10 × 10 9 /l, from 10 to 100 × 10 9 /l) and marrow blasts percentage (increment of 10%, from 30 to 90%) were tested to identify the optimal cutoffs that best defined LFS and OS. Parameters that fulfilled the proportional hazard assumption with a P-valueo0.1 in univariate analyses were further evaluated by multivariate analysis with the Cox proportional hazards model. Survival curves were constructed using the Kaplan-Meier method and compared by the log-rank test. All analyses were performed using SPSS (IBM Corp., Armonk, NY, USA). A P-value of 0.05 was considered statistically significant.

Clinicopathologic characteristics and induction chemotherapy
During the study period, 96 patients with CBF-AML were diagnosed. Of these patients, 91 received a standard 7-day regimen of cytarabine (100 mg/m 2 /day) and 3-day regimen of daunorubicin (50 mg/m 2 , N = 72; 90 mg/m 2 , N = 18; not specified, N = 1). One patient received 5:2, and one patient received MAC. Three patients did not receive chemotherapy (Figure 1 and Supplementary Table S2). All 96 patients were included in the survival analysis.  Figure S2A). The nature of these mutations is shown in Supplementary Table S5.
Functional group analysis To overview the mutation spectrum and to provide mechanistic insights into the pathogenesis of CBF-AML, we evaluated the impact of gene mutations, categorized into specific functional groups, on LFS and OS. Gene mutations involved in molecular pathways of cell signaling were associated with inferior LFS and OS (Figure 2b). Gene mutations involved in DNA methylation were associated with inferior LFS but not OS (Figure 2c). Mutations in other functional groups have no significant impact on either LFS or OS. Importantly, patients negative for gene mutations in both signaling and DNA methylation, compared with those positive for either or both mutations, had significantly superior 10-year LFS (75.0% versus 17.9%, P = 0.007) and 10-year OS (83.3% versus 36.1%, P = 0.039) (Figure 3).
KIT-AL and TET2 double mutations As KIT-AL and TET2 mutations were associated with an inferior outcome, the impact of combined KIT-AL and TET2 mutations was examined. Patients were divided into two groups. Group 1 comprised patients negative for KIT-AL mutations (including wild-type KIT and non-AL KIT mutations) and TET2 mutations (double negative, N = 41). Group 2 comprised patients with either one or both of KIT-AL and TET2 mutations (N = 31). Group 1 compared with group 2 patients had similar CR rates (97.4% versus 90.3%, P = 0.20), but significantly superior 10-year LFS (35.6% versus 11.9%, P o 0.001) and 10-year OS (52.6% versus 28.3%, P = 0.012) (Figure 5a).

Impact of allogeneic HSCT
Patients undergoing HSCT in CR1, when compared with those without, had significantly better LFS (P = 0.004) but similar OS (Figure 5b). The impact of HSCT in CR1 for each genetic subgroup was not evaluated because of the small patient number in each category. HSCT from HLA-identical sibling donors or matched unrelated donors showed comparable LFS and OS (Supplementary Figure S4). Prognostic factors Univariate analysis showed that HSCT at CR1 was significantly associated with better LFS. Mutations in cell signaling and DNA methylation as functional groups and KIT-AL and TET2 mutations as individual genes were significantly associated with inferior LFS (Table 2). Age ⩾ 40 years, presenting white blood cell count ⩾ 100 × 10 9 /l and marrow blasts ⩾ 70% were significantly associated with inferior OS. Mutations in cell signaling and KIT-AL were also associated with an inferior OS. Multivariate analysis showed that KIT-AL and TET2 mutations were significant adverse factors for LFS, whereas age ⩾ 40 years and marrow blast ⩾ 70% were adverse factors for OS. An adverse indicator for OS in univariate analysis notwithstanding, KIT-AL mutation did not fulfill the proportional hazard assumption and was excluded from multivariate analysis.

DISCUSSION
Patients with CBF-AML are clinically heterogeneous despite similar cytogenetic aberrations, implying that secondary mutations might be pathogenetically important. In fact, mutations involving KIT, FLT3 and RAS were commonly reported in CBF-AML. 8,10,11,20 Knock-in mouse models of RUNX1/RUNX1T1 or CBFB/MYH11 induced a preleukemia hematopoietic state and required additional mutations for the development of AML, 21,22 supporting the proposition of a 'second-hit' leukemogenic model. With NGS, a recent study recruiting both adult and pediatric patients from two clinical trials has identified mutations in cell signaling, cohesion complex and chromatin modification, being associated with higher risk of relapse. 14 In this study, we examined the mutation spectrum in CBF-AML and the clinicopathologic features of an exclusively adult patient population who were treated with a uniform algorithm. Mutations of genes involved in cell signaling were associated with significantly inferior LFS and OS, whereas those in DNA methylation were associated with a significantly inferior LFS. Remarkably, patients without any mutations in genes involved in cell signaling and DNA methylation had long-term LFS and OS of ∼ 80%. The identity of genes in these categories was diverse, with KIT and TET2 being the most common. Other functional groups had no impact on clinical outcome. These results underscore the genetic heterogeneity in CBF-AML, and provide clinical evidence for the importance of concomitant mutations in these leukemias.
In this study, KIT mutations were identified in 450% of tested patients, with KIT-AL mutations in 35% of them. The prevalence was comparable with those reported based on targeted sequencing. 9,10 Importantly, only KIT-AL mutations but not other mutations were associated with inferior LFS and OS. Overexpression of KIT-AL mutant has been shown to cooperate with RUNX1-RUNX1T1 or CBFB-MYH11 to induce AML in mice. 21,22 However, KIT mutations are known to be unstable during disease evolution, with nearly 50% of cases losing or changing their mutations at relapse. 10 The mechanistic link between KIT-AL mutations and inferior prognosis in CBF-AML would have to be further elucidated. We also showed for the first time that TET2 mutations were an adverse prognostic factor in CBF-AML. Earlier studies examining TET2 mutations with bidirectional Sanger sequencing or 454-based NGS failed to detect any TET2 mutations in CBF-AML. 23,24 Of AML cases examined in TCGA, mutation in TET2 was shown in only 1 of 7 cases with t(8;21), and none of 11 cases with inv (16). 16 A more recent study based on RNA-sequencing has shown TET2 mutations in 3 of 20 cases with t(8;21), but none in 28 cases with inv (16). 15 In this study, we have shown a similar frequency (6/51) of TET2 mutations in t (8;21) and an apparently higher frequency (2/21) in inv (16). The scattering of mutations throughout the coding sequence and the lack of hot spot mutations supported the proposition that they were loss-offunction mutations. In all patients but one, TET2 mutations had VAF 440%, suggesting that the TET2-mutant clone was predominant and occurred in a heterozygous state. TET2 functions as a dioxygenase that converts 5 methylcytosine to 5 hydroxymethylcytosine, with a pivotal role in DNA demethylation. 25 In mice, loss of TET2 has been shown to induce genome-wide enhancer methylation that in a CBF-AML model collaborates to induce an aggressive leukemia. 26,27 Further studies are needed to validate the very poor prognostic impact and define the pathogenetic/mechanistic basis of concomitant KIT-AL and TET2 mutations.
In our CBF-AML patients not receiving allogeneic HSCT in CR1, the poor long-term LFS of only 20-30% (Supplementary Figure S5) was comparable with the results from an earlier report on t(8;21) AML from our center 28 and other reports from Asia, 29,30 but inferior to results reported from larger international studies. 31,32 It is unclear whether the inferior results of chemotherapy in our patients were because of a different gene mutation spectrum, the small size of this cohort or a difference in treatment regimens. However, the long-term OS was similar to that reported previously, 5 supporting the notion that relapsed CBF-AML remained sensitive to salvage chemotherapy and patients could still be rescued by allogeneic HSCT.
Because relapsed patients are still salvageable, it remains uncertain whether frontline allogeneic HSCT is needed for CBF-AML. Accordingly, we showed an improved LFS but not OS in our patients receiving allogeneic HSCT in CR1, consistent with results from other studies employing frontline HSCT in CBF-AML, either because of institute policy or guided by minimal residual leukemia. 29,30,33 With improvement in supportive care and advent of reduced-intensity, the indications and potential benefit of upfront HSCT for selected CBF-AML patients, particularly those with unfavorable gene mutation profiles, should be explored.
Our observations have important clinical implications. Patients without KIT-AL and TET2 mutations ('double-negative' patients) have favorable outcome, suggesting this to be a distinct genetic subgroup that may be curable with chemotherapy. In fact, patients without any mutations involved in cell signaling (which included KIT-AL mutations) and DNA methylation (which included TET2 mutations) had an even more favorable LFS and OS. With the decreasing cost of NGS, a more extensive examination of genetic mutations at diagnosis may become a plausible strategy for prognostication. Furthermore, minimal residual disease monitoring has been shown to predict treatment outcome and provide guidance to HSCT indication in CBF-AML. [33][34][35][36][37] Treatment algorithms incorporating mutational profiles and minimal residual disease monitoring should be tested to define whether better risk stratification may be achieved in this AML subtype.

AUTHOR CONTRIBUTIONS
CYC and GMKL designed research, analyzed data, performed statistical analysis and wrote the manuscript. CHA designed research, performed NGS and analyzed data. TLC and ESKM designed research and analyzed data. JPYS, HG, AKWL, RL, SFY, HKKL, HSYL, JSML, THL, CKL and SYL recruited and treated the patients. KFW, LLPS, CSPT, CCS and HWWW collected the samples and analyzed data. YLK and AYHL recruited and treated the patients, designed research, analyzed data and wrote the manuscript.