Molecular Profiling Defines Distinct Prognostic Subgroups in Childhood AML: A Report From the French ELAM02 Study Group

Supplemental Digital Content is available in the text


Introduction
Approximately 20% of childhood acute leukemia is of myeloid origin. Acute myeloid leukemia (AML) is defined as a clonal disorder caused by stepwise accumulation of successive genetic defects. In recent years, the use of genetic data to inform disease classification and clinical practice has been an active field of research. Improvements in identifying such molecular and cytogenetic aberrations have revealed the heterogeneity of this group of diseases. Recurrent mutations and gene fusions have been shown to affect a wide range of genes that have been classified into 8 functional categories: kinase signaling, transcription factors, tumor suppressors, DNA methylation, chromatin modifiers, cohesin, spliceosome, and the NPM1 gene. 1 Consequently, some genetic alterations with major prognostic significance -such as inv(16)(p13.1q22)/CBFB-MYH11, t(8;21)(q22;q22)/RUNX1-RUNX1T1, single NPM1 mutations and CEBPA double mutations (CEBPAdm) -have been implemented into the World Health Organization (WHO) classification of AML. 2 Nevertheless, most of investigations are based on the study of large cohorts of adult AML patients 3,4 while genetic profiles are known to be quite different between adults and children with AML. 5 Moreover, despite major treatment improvements over the past decades, pediatric AML is still associated with relapse rates up to 30% and survival rates below 75%. 6 In this context, a better description of the pattern of molecular aberrations in childhood AML remains a great challenge to refine prognostication and improve outcome in such patients.
We report here the comprehensive molecular landscape of a large and well-annotated cohort of de novo pediatric AML enrolled in the prospective ELAM02 trial and propose a new prognostic molecular classifier in this particular group of patients.

Patients
The present study focuses on 385 patients of the 438 children treated in the ELAM02 trial (Treating Patients with Childhood Acute Myeloid Leukemia with Interleukin-2; ClinicalTrials.gov NCT00149162). Patient selection was based on the availability of genomic DNA at AML diagnosis. Children aged 0 to 18 years with newly diagnosed AML were enrolled between March 2005 and December 2011. Acute promyelocytic leukemia, therapyrelated AML and Down syndromes were excluded from the ELAM02 trial. The study was approved by the Ethics Committee of Saint-Antoine Paris University Hospital (Assistance Publique-Hôpitaux de Paris) and by the Institutional Review Board of the French Regulatory Agency and was conducted in accordance with the Declaration of Helsinki.

Mutational analysis
Genomic DNA from bone marrow aspirates at diagnostic was studied by high-throughput sequencing (HTS) of 36 genes recurrently mutated in myeloid malignancies. The studied panel included genes encoding proteins involved in kinase signaling (Screening for Non-Acceptable Polymorphisms). 8 The presence of the FLT3-internal tandem duplication (ITD) was performed for all patients by fragment analysis as previously described. 9 Statistical methods Event-free survival (EFS) and overall survival (OS) were estimated by the Kaplan-Meier method and compared by cause-specific hazard Cox models. EFS was measured from the date of diagnosis to the date of the first event (induction failure, relapse, or death) or to the date of last follow-up. Patients who failed to achieve complete remission (CR) were considered as failures at day 60. OS was measured from the date of diagnosis to the date of death from any cause or last follow-up. Data were analyzed and compared without censor at transplant for patients who received allogeneic stem cell transplantation in first CR. Comparisons between patient subgroups were performed by the Mann-Whitney test for continuous variables and by Chi-square or Fisher exact test for categorical variables. Hazard ratios (HRs) are given with 95% confidence interval (CI). Multivariate analyses assessing the independent effect of the covariates were performed using Cox proportional hazard model. Variables associated with the outcome and a P-value < 0.10 in univariate analysis or known as validated factors were included in the multivariable models. Then a backward and forward stepwise selection was performed. All P-values were 2-sided and values <0.05 were considered statistically significant. All statistical tests were performed with the SPSS 22.0 (IBM Corp., Armonk, NY) and R3.2.3 software packages (R Development Core Team, Vienna, Austria).

Impact of molecular abnormalities on complete remission rate and clinical outcome
Among the 385 patients included in this study, 350 (91%) achieved CR after 2 courses of intensive induction chemotherapy. In univariate analysis, FLT3-ITD, WT1 mutations, WBC count higher than 30 Â 10 9 /L, "other" cytogenetics and NUP98 fusions were associated with more induction failures (Supplemental Table S3, Supplemental Digital Content, http://links.lww.com/ HS/A1). Despite the small number of cases, only the presence of a NUP98 fusion remained associated with induction failure in multivariate analysis (P = 0.038) ( Table 1). Characteristics of NUP98-rearranged cases are detailed below.

Molecular classifier in childhood AML
Considering results from multivariate analysis and strong molecular markers validated among studies 6,18-20 (i.e., NPM1 mutations and CEBPAdm), we defined a molecular classifier, refining the prognosis in childhood AML.  Finally, the molecular classifier was compared to the 2017 European LeukemiaNet (ELN) classification 21 which is currently used to stratify adult patients with AML. A total of 139 patients were classified in the favorable subgroup with both classifications. Only 3 NPM1-mutated-AML were classified as favorable according to the molecular classifier and as intermediate or adverse according to the ELN classification because of high FLT3-ITD ratio (n = 2) or complex karyotype (n = 1). Interestingly, the ELN classification fails to separate intermediate and adverse subgroups in our pediatric cohort (Fig. 6B, Supplemental Table S5, Supplemental Digital Content, http://links.lww.com/ HS/A1). Together, these data show that the ELN classification lacks of prognostic significance in childhood AML, especially in nonfavorable AML and the use of the present molecular classification could improve risk stratification in pediatric patients.

Discussion
The better knowledge of molecular aberrations in AML has greatly improved the management of AML patients over the past decades. However, most of reported studies have focused on adult cohorts. The ELAM02 trial gave us the opportunity to investigate incidences and prognostic significances of molecular aberrations in childhood AML which currently remains a lifethreatening malignancy with poor outcome compared to acute lymphoblastic leukemia.
The most common mutations involved genes controlling kinase signaling (especially NRAS/KRAS, FLT3-ITD, KIT mutations). These mutations concerned 61% of the whole cohort and were   found in all cytogenetic subgroups. All but FLT3-ITD had no independent impact on outcome. To date, the prognostic significance of FLT3-ITD in pediatric AML remains controversial. 22 In the present study, FLT3-ITD was associated with reduced EFS but did not influence OS in multivariate analysis in the whole cohort. Importantly, FLT3-ITD were found in heterogeneous diseases including NPM1-mutated or CBF AML which have shown to have a highly favorable outcome but also in NUP98-rearranged and WT1-mutated AML which are associated with poor prognosis. Among NPM1-mutated childhood AML, FLT3-ITD did not impact outcome in line with a previous study. 23 Transcription factors were the second most common class of mutations (16% of the whole cohort). CEBPAdm and RUNX1 mutations defined independent molecular subgroups of patients (4.2% and 6.2% respectively) associated with highly favorable and poor outcome respectively. Both mutations were mutual exclusive with NPM1 mutations and occurred almost exclusively in normal karyotype-AML. By contrast, all other classes of mutations were found in less than 10% of patients. Importantly, while mutations within DNA-methylation-related genes (DNMT3A, TET2, IDH1/2) are highly prevalent in adult AML (together higher than 50%), 24   Moreover, the systematic use of LD-RT-PCR allowed the detection of recurrent transcript fusions in about a half of pediatric patients. Fusions involving 1 of the 2 CBF subunits or the KMT2A gene were found in 24% and 21% of patients respectively. Among KMT2A-rearranged cases, the KMT2A-MLLT3 fusion was by far the most common, representing nearly the half of KMT2A fusions. While CBF rearrangements were associated with a favorable prognosis, KMT2A rearrangements were associated with an intermediate outcome in the present study. KMT2A-MLLT3 fusion did not show a better prognosis than other KMT2A rearrangements in line with a recent large retrospective study of KMT2A-rearranged pediatric AML. 26 In accordance with previous reports, 6,[18][19][20]27 CBF rearrangements, NPM1 mutations and CEBPAdm defined a particular subgroup with good prognosis. Together, these aberrations were found in more than one third of childhood AML. By contrast, NUP98 fusions were associated with the worse prognosis, mostly due to induction failures. Other aberrations associated with poor outcome included RUNX1, PHF6, and WT1 mutations. In a previous report by the Children's Oncology Group, WT1 mutations were shown to be an independent factor of poor prognosis both on EFS and OS. 28 Interestingly, AML with RUNX1 mutations has been added to the last WHO classification as a provisional entity, 2 considering they represent a biologically distinct group with a possibly worse prognosis in adults AML. 29 Our results show that RUNX1 mutations also defined a distinct subgroup with poor outcome in childhood AML. Finally, PHF6 mutations are a rare event in childhood AML and to our knowledge, their prognosis impact has not been reported in a large series. 30 Importantly, by contrast to the adult-based-ELN classification, the present molecular classification identified a group of pediatric patients with particular poor prognosis. Moreover, the cooccurrence of FLT3-ITD in this subgroup identified patients with the worst outcome. This result remains of great interest in the context of FLT3 inhibitors use.
In conclusion, we reported the comprehensive genomic landscape of a large cohort of pediatric de novo AML enrolled in the ELAM02 trial and proposed a prognostic classification based on gene mutations and fusions in this particular group of patients. Despite some overlaps between childhood and adult AML, pediatric patients harbored a different pattern of molecular aberrations, especially with fewer mutations within epigeneticrelated genes. We confirmed the favorable-risk group including CBF fusions, NPM1 mutations, and CEBPA biallelic mutations and refined the poor-risk group including RUNX1, WT1, and PHF6 mutations as well as NUP98 fusions. KMT2A-rearranged AML were included in the intermediate-risk group with no difference between KMT2A-MLLT3 and other KMT2A fusions in this study. Overall, these results have important implications to contribute in refining risk stratification of pediatric AML and show the need for further validations in independent pediatric cohorts.