Expression-based decision tree model reveals distinct microRNA expression pattern in pediatric neuronal and mixed neuronal-glial tumors

Abstract

Abstract Background The understanding of the molecular biology of pediatric neuronal and mixed neuronal-glial brain tumors is still insufficient due to low frequency and heterogeneity of those lesions which comprise several subtypes presenting neuronal and/or neuronal-glial differentiation. Important is that the most frequent ganglioglioma (GG) and dysembryoplastic neuroepithelial tumor (DNET) showed limited number of detectable molecular alterations. In such cases analyses of additional genomic mechanisms seem to be the most promising. The aim of the study was to evaluate microRNA (miRNA) profiles in GGs, DNETs and pilocytic asytrocytomas (PA) and test the hypothesis of plausible miRNA connection with histopathological subtypes of particular pediatric glial and mixed glioneronal tumors. Methods The study was designed as the two-stage analysis. Microarray testing was performed with the use of the miRCURY LNA microRNA Array technology in 51 cases. Validation set comprised 107 samples used during confirmation of the profiling results by qPCR bioinformatic analysis. Results Microarray data was compared between the groups using an analysis of variance with the Benjamini-Hochberg procedure used to estimate false discovery rates. After filtration 782 miRNAs were eligible for further analysis. Based on the results of 10 × 10-fold cross-validation J48 algorithm was identified as the most resilient to overfitting. Pairwise comparison showed the DNETs to be the most divergent with the largest number of miRNAs differing from either of the two comparative groups. Validation of array analysis was performed for miRNAs used in the classification model: miR-155-5p, miR-4754, miR-4530, miR-628-3p, let-7b-3p, miR-4758-3p, miRPlus-A1086 and miR-891a-5p. Model developed on their expression measured by qPCR showed weighted AUC of 0.97 (95% CI for all classes ranging from 0.91 to 1.00). A computational analysis was used to identify mRNA targets for final set of selected miRNAs using miRWalk database. Among genomic targets of selected molecules ZBTB20, LCOR, PFKFB2, SYNJ2BP and TPD52 genes were noted. Conclusions Our data showed the existence of miRNAs which expression is specific for different histological types of tumors. miRNA expression analysis may be useful in in-depth molecular diagnostic process of the tumors and could elucidate their origins and molecular background.

Keywords

• Brain tumor
• Differentiation model
• Dysembryoplastic neuroepithelial tumor
• Expression
• Ganglioglioma
• microRNA