Development of an miRNA signature for individual prognostic assessment in seminoma patients.

Abstract

645 Background: Due to the risk of long-term toxicities in young non-metastatic seminoma patients, surveillance regimens nowadays are the standard of care. However, up to 20% of patients will relapse and undergo further treatment, requiring even greater efforts in terms of adjuvant therapy. miRNAs are known to be very stable and robust biomarkers. This study aims to establish a prognostic miRNA profile utilizing bioinformatics tools to differentiate between metastatic (met) and non-metastatic (nmet) seminomas. Methods: MiRNA was extracted from formalin-fixed paraffin-embedded sections. Using microarray analysis (Agilent Technology), two cohorts of patients with seminoma (24 patients each, 12 nmet and 12 met) were examined. The following criteria were used to preselect relevant miRNAs: absolute median log2 fold change >0.5, distance correlation >0.5, accuracy in single-miRNA classifier trained with leave-one-out cross-validation (met vs. nmet) >0.7. Validation was performed in 88 patients (47 nmet and 41 met) through qPCR using the TaqMan assay, with RNU48, miR-191-5p, and miR-361-5p serving as references. Statistical and bioinformatics analyses were performed using the Mann-Whitney U test and machine learning algorithms. Results: The bioinformatics analysis of microarrays identified 16 miRNAs exhibiting an accuracy >70% in differentiating met tumors from nmet ones. The validation of 6 miRNAs by qPCR revealed significant expression differences between met and nmet tumors for miRNAs miR-371a-5p, miR-512-3p, miR−509−3−5p, miR-517a-3p (p<0.05), dependent on the reference used. All 4 miRNAs are higher expressed in met tumors. Synchronous met tumors exhibit stronger differences in the expression of the above-mentioned miRNAs compared to metachronous met tumors, except for miR-509-3-5p. miR-371a-5p was found to be the most robust miRNA to distinguish both synchronous and metachronous met from nmet seminomas. Classifiers trained on the PCR data for different combinations of the 6 investigated miRNAs using leave-one-out cross-validation show a medium performance (average accuracy of 0.59) including one model based on 4 miRNAs with high accuracy of 0.99. Conclusions: By using machine-learning algorithms on microarray data, we have established a miRNA signature that can differentiate metastatic from non-metastatic seminomas with high reliability. Validation of 4 miRNAs by qPCR confirms the high prognostic value of miRNAs in seminoma patients. Currently, we examine the suitability of this miRNA panel in liquid biopsy.