Juvenile myelomonocytic leukemia (JMML) is a rare heterogeneous hematological malignancy of early childhood characterized by causative RAS pathway mutations. Classifying patients with JMML using global DNA methylation profiles is useful for risk stratification. We implemented machine learning algorithms (decision tree, support vector machine, and naïve Bayes) to produce a DNA methylation-based classification according to recent international consensus definitions using a well-characterized pooled cohort of patients with JMML (n = 128). DNA methylation was originally categorized into three subgroups: high methylation (HM), intermediate methylation (IM), and low methylation (LM), which is a trichotomized classification. We also dichotomized the subgroups as HM/IM and LM. The decision tree model showed high concordances with 450k-based methylation [82.3% (106/128) for the dichotomized and 83.6% (107/128) for the trichotomized subgroups, respectively]. With an independent cohort (n = 72), we confirmed that these models using both the dichotomized and trichotomized classifications were highly predictive of survival. Our study demonstrates that machine learning algorithms can generate clinical parameter-based models that predict the survival outcomes of patients with JMML and high accuracy. These models enabled us to rapidly and effectively identify candidates for augmented treatment following diagnosis.

Purpose

Known clinical and genetic markers have limitations in predicting disease course and outcome in juvenile myelomonocytic leukemia (JMML). DNA methylation patterns in JMML have correlated with outcome across multiple studies, suggesting it as a biomarker to improve patient stratification. However, standardized approaches to classify JMML on the basis of DNA methylation patterns are lacking. We, therefore, sought to define an international consensus for DNA methylation subgroups in JMML and develop classification methods for clinical implementation.

Experimental design

Published DNA methylation data from 255 patients with JMML were used to develop and internally validate a classifier model. Accuracy across platforms (EPIC-arrays and MethylSeq) was tested using a technical validation cohort (32 patients). The suitability of both methods for single-patient classification was demonstrated using an independent cohort (47 patients).

Results

Analysis of pooled, published data established three DNA methylation subgroups as a de facto standard. Unfavorable prognostic parameters (PTPN11 mutation, elevated fetal hemoglobin, and older age) were significantly enriched in the high methylation (HM) subgroup. A classifier was then developed that predicted subgroups with 98% accuracy across different technological platforms. Applying the classifier to an independent validation cohort confirmed an association of HM with secondary mutations, high relapse incidence, and inferior overall survival (OS), while the low methylation subgroup was associated with a favorable disease course. Multivariable analysis established DNA methylation subgroups as the only significant factor predicting OS.

Conclusions

This study provides an international consensus definition for DNA methylation subgroups in JMML. We developed and validated methods which will facilitate the design of risk-stratified clinical trials in JMML.

Esophageal squamous cell carcinoma (ESCC) is prevalent worldwide and particularly common in certain regions of Asia. Here we report the whole-exome or targeted deep sequencing of 139 paired ESCC cases, and analysis of somatic copy number variations (SCNV) of over 180 ESCCs. We identified previously uncharacterized mutated genes such as FAT1, FAT2, ZNF750 and KMT2D, in addition to those already known (TP53, PIK3CA and NOTCH1). Further SCNV evaluation, immunohistochemistry and biological analysis suggested their functional relevance in ESCC. Notably, RTK-MAPK-PI3K pathways, cell cycle and epigenetic regulation are frequently dysregulated by multiple molecular mechanisms in this cancer. Our approaches also uncovered many druggable candidates, and XPO1 was further explored as a therapeutic target because it showed both gene mutation and protein overexpression. Our integrated study unmasks a number of novel genetic lesions in ESCC and provides an important molecular foundation for understanding esophageal tumors and developing therapeutic targets.

Juvenile myelomonocytic leukemia (JMML) is a rare myelodysplastic/myeloproliferative neoplasm that develops during infancy and early childhood. The array-based international consensus definition of DNA methylation has recently classified patients with JMML into the following 3 groups: high (HM), intermediate (IM), and low methylation (LM). To develop a simple and robust methylation clinical test, 137 patients with JMML were analyzed using the Digital Restriction Enzyme Analysis of Methylation (DREAM), which is a next-generation sequencing-based methylation analysis. Unsupervised consensus clustering of the discovery cohort (n = 99) using DREAM data identified HM (HM_DREAM; n = 35) and LM subgroups (LM_DREAM; n = 64). Of the 98 cases that could be compared with the international consensus classification, 90 HM (n = 30) and LM (n = 60) cases had 100% concordance with DREAM clustering results. Of the remaining 8 cases comprising the IM group, 4 were classified as belonging to the HM_DREAM group and 4 to the LM_DREAM group. A machine-learning classifier was successfully constructed using a support vector machine (SVM), which divided the validation cohort (n = 38) into HM (HM_SVM, n = 18) and LM (LM_SVM; n = 20) groups. Patients with the HM_SVM profile had a significantly poorer 5-year overall survival rate than those with the LM_SVM profile. In conclusion, we developed a robust methylation test using DREAM for patients with JMML. This simple and straightforward test can be easily incorporated into diagnosis to generate a methylation classification for patients so they can receive risk-adapted treatment in the context of future clinical trials.

Next generation sequencing technologies have provided insights into the molecular heterogeneity of various myeloid neoplasms, revealing previously unknown somatic genetic events. In our cohort of 1444 cases analyzed by next generation sequencing, somatic mutations in the gene BRCA1-BRCA2-containing complex 3 (BRCC3) were identified in 28 cases (1.9%). BRCC3 is a member of the JAMM/MPN+ family of zinc metalloproteases capable of cleaving Lys-63 linked polyubiquitin chains, and is implicated in DNA repair. The mutations were located throughout its coding region. The average variant allelic frequency of BRCC3 mutations was 30.1%, and by a serial sample analysis at two different time points a BRCC3 mutation was already identified in the initial stage of a myelodysplastic syndrome. BRCC3 mutations commonly occurred in nonsense (n=12), frameshift (n=4), and splice site (n=5) configurations. Due to the marginal male dominance (odds ratio; 2.00, 0.84-4.73) of BRCC3 mutations, the majority of mutations (n=23; 82%) were hemizygous. Phenotypically, BRCC3 mutations were frequently observed in myelodysplastic syndromes and myelodysplastic/myeloproliferative neoplasms and associated with -Y abnormality (odds ratio; 3.70, 1.25-11.0). Clinically, BRCC3 mutations were also related to higher age (P=0.01), although prognosis was not affected. Knockdown of Brcc3 gene expression in murine bone marrow lineage negative, Sca1 positive, c-kit positive cells resulted in 2-fold more colony formation and modest differentiation defect. Thus, BRCC3 likely plays a role as tumor-associated gene in myelodysplastic syndromes and myelodysplastic/myeloproliferative neoplasms.