What Is Abnormal in Normal Karyotype Acute Myeloid Leukemia in Children? Analysis of the Mutational Landscape and Prognosis of the TARGET-AML Cohort

Normal karyotype acute myeloid leukemia (NK-AML) constitutes 20–25% of pediatric AML and detailed molecular analysis is essential to unravel the genetic background of this group. Using publicly available sequencing data from the TARGET-AML initiative, we investigated the mutational landscape of NK-AML in comparison with abnormal karyotype AML (AK-AML). In 164 (97.6%) of 168 independent NK-AML samples, at least one somatic protein-coding mutation was identified using whole-genome or targeted capture sequencing. We identified a unique mutational landscape of NK-AML characterized by a higher prevalence of mutated CEBPA, FLT3, GATA2, NPM1, PTPN11, TET2, and WT1 and a lower prevalence of mutated KIT, KRAS, and NRAS compared with AK-AML. Mutated CEBPA often co-occurred with mutated GATA2, whereas mutated FLT3 co-occurred with mutated WT1 and NPM1. In multivariate regression analysis, we identified younger age, WBC count ≥50 × 109/L, FLT3-internal tandem duplications, and mutated WT1 as independent predictors of adverse prognosis and mutated NPM1 and GATA2 as independent predictors of favorable prognosis in NK-AML. In conclusion, NK-AML in children is characterized by a unique mutational landscape which impacts the disease outcome.


Introduction
Acute myeloid leukemia (AML) is a heterogenous malignant disease of the bone marrow resulting from the accumulation of acquired somatic genetic lesions in myeloid progenitor cells [1]. Overall survival of childhood AML has reached~70% based on intensive chemotherapy regimens with relapse still occurring in~30% of all patients [2], and therefore novel targeted therapeutics are warranted to increase cure rate [3]. Genetic characterization of AML is a hallmark both in research unravelling the disease biology and identifying new therapeutic targets, as well as in the clinic supporting disease classification, risk stratification, disease monitoring and therapy guidance [4]. Disease classification is, however, still mainly based on the identification of recurrent chromosomal lesions [5] from routine cytogenetic examinations. These cytogenetic examinations involve conventional chromosomal banding analysis and fluorescence in situ hybridization (FISH) targeting recurrent cytogenetic lesions including gene fusions, as well as reverse transcriptase-polymerase chain reaction (RT-PCR) targeting recurrent gene fusion transcripts and common gene mutations (e.g., FLT3-internal tandem duplication [ITD]) [6]. Importantly,~20-25% of pediatric AML show no cytogenetic abnormalities following routine cytogenetic examination, and this group is referred to as normal karyotype or cytogenetically normal AML (NK-AML) [7][8][9][10]. Molecular analysis by RT-PCR targeting selected genes such as FLT3, NPM1, and CEBPA is useful in these cases to identify prognostically important mutations [11][12][13][14]. These aberrations do, however, only represent a limited picture of the genetics in NK-AML. Fortunately, the advent of Next-Generation sequencing (NGS) technologies has allowed more detailed genetic investigations of AML through targeted gene panel sequencing, whole exome sequencing, whole-genome sequencing (WGS), and transcriptomic RNA sequencing. Using these techniques, major efforts have been made to unravel the molecular landscape of pediatric AML and disease progression [10,15,16] including a study of focusing on normal karyotype FLT3-ITD positive AML [17].
From the publicly available data from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) initiative AML study (TARGET-AML) [10], we here aimed to characterize the molecular landscape and identify determinants of outcome in childhood NK-AML.

Data Source
The present study was based on the publicly available data from the TARGET-AML initiative [10]. The original clinical and genetic data can be accessed at https://target-data. nci.nih.gov/Public/AML/ (accessed on 24 January 2021).

Study Population
The AML patient samples in the TARGET-AML study were obtained from the Children's Oncology Group protocols AAML0531 [18], AAML03P1 [19], and CCG-2961 [20]. The TARGET-AML study profiled a total of 815 children and young adults for somatic mutations at diagnosis (based on paired leukemia-remission samples) including WGS of 197 subjects as well as targeted capture sequencing (TCS) averaging 500× coverage in 800 subjects used for validation [10]. For our analysis, we identified the study cohort combining the clinical data files of the discovery and validation cohorts (date version: 13 December 2018).

Genetic Data
Cytogenetic data including International System for Human Cytogenetic Nomenclature (ISCN) karyotypes used for disease classification were obtained from the clinical data files.
For mutational status in each sample, we used the spreadsheets reporting proteincoding somatic variants identified using WGS (n = 197) and TCS (n = 684) (Supplementary Tables S5 and S6 in Bolouri et al. [10]). For survival analyses, we also included mutational status from routine testing of FLT3, NPM1, CEBPA, and WT1 provided in the clinical data. For mutational burden of single nucleotide variants (SNVs) and insertions and deletions (indels) in each sample, we used the spreadsheet reporting variant calls from Strelka and SAMtools mpileup variant callers (Supplementary Table S4a in Bolouri et al. [10]). Our SNV/indel analysis depended on the accuracy of these processed variant data and we did not apply any additional variant filtering.

Genetic and Statistical Analysis
Based on the cytogenetic information provided in the clinical data, samples were categorized as core-binding factor AML (CBF-AML) including t(8;21)(q22;q22) and inv16(p13.1q22), KMT2A (MLL) rearranged AML, another abnormal karyotype AML and NK-AML. The first three groups are collectively referred to as abnormal karyotype AML (AK-AML) in this study. The assigned categorization based on ISCN karyotypes remained throughout all analyses in this study.
Recurrently mutated genes (mutated in ≥2 samples) were identified in NK-AML. The SNV/indel mutational burden in NK-AML vs AK-AML was plotted and tested for equal distributions using the Wilcoxon rank sum test. The SNV/indel mutational landscapes of NK-AML vs AK-AML were visualized using two-level doughnut charts [21], considering all genes mutated in ≥2% of the samples within both groups with all genes categorized according to The Cancer Genome Atlas (TCGA) functional groups [22]. Mutually exclusive and co-occurring mutation events between genes in NK-AML vs. AK-AML were identified using Maftools (v2.2.10) [23], which uses pairwise Fisher's exact test on a 2 × 2 contingency table containing frequencies of mutated and non-mutated samples. Computed odds ratios indicated either co-occurrence or mutual exclusivity.
Baseline patient characteristics and mutation frequencies were summarized using descriptive statistics. For comparing NK-AML vs AK-AML characteristics Wilcoxon rank sum test was used for continuous variables and χ 2 test for categorical variables including mutation status. Event-free survival (EFS; events defined as induction failure, induction death, death in first complete remission, or relapse) and overall survival (OS) with 95% confidence intervals (CIs) were computed using the Kaplan-Meier estimator. Cumulative incidence of relapse (CIR) was estimated using the pseudo values method with death as competing risk [24]. Log rank tests were used for comparison of different EFS and OS distributions by cytogenetic group, patient characteristics, and mutational status. Cox proportional hazards regression was used to compute hazard ratios as a measure of the relative risk of any event or death during follow-up in univariate analysis as well as multivariate analysis adjusted for baseline risk factors, FLT3-internal tandem duplication (ITD), and mutational status of other co-occurring genes as relevant. Stratified analyses included cytogenetic group for the entire study cohort and for the NK-AML cohort, baseline characteristics and mutational status for genes mutated in ≥10 samples with wildtype NK-AML patients as reference. Stem cell transplantation was excluded from the analysis due to no time-points provided in the clinical data necessary for its inclusion in the model as a time-dependent variable. The assumption of proportional hazards was verified using Schoenfeld residuals test and we checked for multicollinearity computing variance inflation factor and tolerance values. Significance level was set at p < 0.05.
Data management and analysis of the genetic and clinical data were performed using Stata Statistical Software Package (v14.2; StataCorp. LP, College Station, TX, USA). Analysis of co-occurrence and mutual exclusivity was performed in R (v3.6.2) using the Maftools library (v.2.2.10).

Study Cohort and Clinical Characteristics
From the TARGET-AML study cohort of 966 patients, we excluded patients with no clinical data available (n = 31), aged 20 years or above (n = 16), French-American-British classification (FAB) M3 morphology (n = 2), and no cytogenetic information (n = 44), to obtain the final study cohort 873 patients in this analysis. We identified 208 (24%) NK-AML and 665 (76%) AK-AML patients (CBF-AML, n = 248; KMT2A rearranged, n = 175; other AK-AML, n = 242). Table 1 summarizes the patient characteristics stratified by cytogenetic group. Patients with NK-AML were characterized by older age at diagnosis (median: 12 years, p < 0.001) which has previously been reported [25], a high prevalence of FAB-M1 morphology (p < 0.001), and a high frequency of stem cell transplantation (SCT) in first complete remission (CR1) (p < 0.001) compared to AK-AML. No significant differences were seen in 5-year EFS and OS. However, 5-year CIR in NK-AML was lower compared with KMT2A rearranged and other AK-AML (Table 1).

Mutational Landscape of NK-AML
AML samples from 661 of 873 (76%) patients were analyzed by WGS and/or TCS including samples from 168 of 208 (81%) NK-AML patients (Table 1). Of the sequenced NK-AML samples, 39 samples were examined by WGS of which 32 samples were validated by TCS. In addition, 129 samples were analyzed by TCS alone (SNV/indel detection only). From samples analyzed by both WGS and TCS, mutational events were only counted once. We compared the baseline characteristics of sequenced and non-sequenced NK-AML patients and found an older age distribution among patients with sequenced samples (median: 13 years, p = 0.004). This may reflect a higher amount of bone marrow aspirate acquired from older patients, needed for sequencing studies.
In all sequenced NK-AML samples, a total of 593 gene mutations including SNVs, indels, copy number variants (CNVs, i.e., focal deletions/duplications) and gene fusions affecting 135 different genes were identified in samples from 164/168 (97.6%) patients (see Supplemental Table S1 for an overview of the mutation status of all NK-AML samples). A total of six gene fusions not identified in the cytogenetic workup were identified including KMT2A-MLLT4, KMT2A-MLLT3, HNRNPH1-ERG, NUP98-NSD1, CBFA2T3-GLIS2, and SEPT6-KMT2A. CNVs were identified in samples from eight patients with ZEB2 as the most commonly impacted gene (n = 4) (Supplemental Table S1). In four NK-AML samples, all analyzed by WGS, no mutations were identified. No specific characteristics were found in these patients and they all varied in terms of blast percentages, white blood cell (WBC) count, and age group.
A higher SNV/indel count was identified in NK-AML compared with AK-AML (p = 0.0001) with a median of 3 called SNVs/indels in NK-AML (range 0-17, Figure 1). This is, however, overall, still a very low mutational burden compared to other cancers, as reported by Bolouri et al. [10]. only counted once. We compared the baseline characteristics of sequenced and nonsequenced NK-AML patients and found an older age distribution among patients with sequenced samples (median: 13 years, p = 0.004). This may reflect a higher amount of bone marrow aspirate acquired from older patients, needed for sequencing studies. In all sequenced NK-AML samples, a total of 593 gene mutations including SNVs, indels, copy number variants (CNVs, i.e., focal deletions/duplications) and gene fusions affecting 135 different genes were identified in samples from 164/168 (97.6%) patients (see Supplemental Table S1 for an overview of the mutation status of all NK-AML samples). A total of six gene fusions not identified in the cytogenetic workup were identified including KMT2A-MLLT4, KMT2A-MLLT3, HNRNPH1-ERG, NUP98-NSD1, CBFA2T3-GLIS2, and SEPT6-KMT2A. CNVs were identified in samples from eight patients with ZEB2 as the most commonly impacted gene (n = 4) (Supplemental Table S1). In four NK-AML samples, all analyzed by WGS, no mutations were identified. No specific characteristics were found in these patients and they all varied in terms of blast percentages, white blood cell (WBC) count, and age group.

Discussion
In this study, we exploited the publicly available data from the TARGET-AML study [10] to investigate the mutational landscape of pediatric NK-AML and its impact on prognosis. In NK-AML, the absence of any cytogenetic alterations further imposes the relevance of detailed molecular characterization to understand the genetic background of the disease, monitoring disease progression and to support informed risk stratification.
In 164 of 168 NK-AML patients not found to harbor any chromosomal aberrations from routine testing, the detailed sequencing analyses by WGS and TCS detected at least one protein-coding somatic mutation, demonstrating the utility of molecular characterization using these methods. Only four NK-AML patients were without any mutations identified.
Based on the results from WGS and TCS analysis we identified a unique mutational landscape in pediatric NK-AML characterized by a higher prevalence of mutated CEBPA, FLT3, GATA2, NPM1, PTPN11, TET2, and WT1 as well as a lower prevalence of mutated KIT, KRAS, and NRAS compared with AK-AML (Figures 2 and 3A,B). The commonly mutated genes were distributed across various TCGA functional groups compared to AK-AML. We noted that TET2 and IDH1-2 that are involved in DNA methylation, and generally are considered to be genes associated with adult AML, constituted around 7% of the mutational landscape in pediatric NK-AML. We did, however, not find mutations in these three genes to be associated with older age within pediatric NK-AML (median: 12 years, range: 0-19, p = 0.613). Of the commonly mutated genes in NK-AML, CEBPA/GATA2, FLT3/WT1, and FLT3/NPM1 were identified as frequently co-occurring events meanwhile mutated CEBPA was mutually exclusive of mutated NPM1 and FLT3. This is of importance to consider when investigating the clinical implications of mutations in these genes.
Notably, six gene fusions not reported in the cytogenetic data were detected by WGS in NK-AML. This highlights the relevance of genomic sequencing to be implemented as clinical cytogenomic analyses as recently suggested [27,28]. Certain cryptic cytogenetic events of prognostic importance are difficult to capture in routine cytogenetic examinations, including, for instance, inv(16)(p13q24) (CBFA2T3-GLIS2) [29] and rearrangements involving 11p15 (NUP98 fusions) [10]. For these cryptic events, WGS or transcriptome (RNA) sequencing may have a higher diagnostic yield detecting gene fusions or their expressed fusion transcripts, respectively [27,28,30,31].
In our multivariate analysis, we identified younger age, WBC count ≥50, FLT3-ITD, and mutated WT1 as independent predictors of a poor outcome, and mutated NPM1 and GATA2 as independent predictors of favorable outcome in NK-AML. To our knowledge, younger age has not previously been highlighted as a risk factor in NK-AML. Possible confounders should be considered for this association, especially cryptic CBFA2T3-GLIS2 fusions associated with poor prognosis, infancy, and FAB-M7 morphology (4 of 10 with known FAB class within the 0-1 year age group were M7 in this study) [32,33]. Nonetheless, this finding encourages further studies of the impact of age in independent cohorts. The clinical implications of FLT3-ITD with an inferior EFS is well established in the literature [11,12] and adverse outcomes related to mutated WT1 have been reported [10,34,35]. The favorable impact on EFS of mutated NPM1 irrespective of FLT3-ITD status has also previously been reported [36]. Mutated CEBPA has previously been reported to be an independent predictor of favorable outcome [14]; however, in our multivariate analysis, this association was obliterated following adjustment for GATA2 status. In contrast, we identified mutated GATA2 as an independent predictor of improved EFS adjusted for CEBPA status.
Some limitations must be considered interpreting these results. The complete dependency on the reported results of the study by Bolouri et al. may represent a limitation since we did not have any influence on the variant prioritization and a reanalysis of their raw data potentially could lead to the identification of previously unreported variants. Of note, we identified an older age among WGS/TCS analyzed subjects compared to non-sequenced subjects in the TARGET-AML cohort, possibly explained by the lower amount of bone marrow aspirate acquired from infant patients hindering inclusion for sequence analysis. This selection bias may have impacted the distribution of the mutational landscape of NK-AML as the prevalence of certain gene mutations, for instance, NPM1, did show some age dependency. Additionally, it should be noted that not all samples were analyzed using the same application. Hence, only 39 of 168 NK-AML were analyzed by WGS which resulted in an ascertainment bias in terms of limited detection of structural variation (e.g., gene fusions). Finally, the data used for this study did not allow us to adjust for the effect of stem cell transplantation in the multivariate analysis as stated in Section 2.4.
The TARGET-AML study [10] has proven the value of large-scale sequencing analysis to unravel the molecular background of a malignancy. With its comprehensive data, we have further described the genetics of NK-AML in comparison with AK-AML and its impact on prognosis. Such knowledge is essential in the clinic used for risk stratification and molecular minimal residual disease monitoring. Furthermore, the mutational landscape of NK-AML is also important to explore in relation to the emergence of several novel gene targeted therapies. These include, for instance, midostaurin for untreated FLT3-mutated AML currently undergoing a phase II trial [37], and inhibitors of mutated IDH1, such as ivosedinib that is Food and Drug Administration (FDA) approved for relapsed/refractory AML in children ≥12 years (NCT03245424).
Future molecular studies of pediatric AML will continue to be important to support this development. In this analysis, we only considered the genomics of NK-AML. However, to be able to obtain a more comprehensive understanding of the disease biology, multiomics approaches investigating both the epigenome, transcriptome, microRNome, and proteome are needed.

Conclusions
Normal karyotype AML in children represents a unique group within pediatric AML with its own distinct genetic background and biology compared with abnormal karyotype AML. We identified younger age, WBC count ≥50, FLT3-ITD, and mutated WT1 as independent predictors of a poor outcome and mutated NPM1 and GATA2 as independent predictors of favorable outcome.