Abstract
BACKGROUND
Diffuse gliomas are the most common primary malignant brain tumors, whose overall prognosis is quite dismal. Tumor-cell-secreted extracellular vesicles (EVs) participate in physiological and pathological processes and have potential applications to diagnostics of malignant tumors including diffuse gliomas. Because urine is less invasive to collect, development of early diagnosis based on urine EVs is eagerly awaited. In this study, we captured urine EVs of patients with gliomas efficiently with the nanowire device and compared expression profile of microRNAs (miRNAs) within urine EVs with that of healthy donors to identify diagnostic accuracy by a machine learning algorithm.
METHODS
62 patients with diffuse gliomas, including 27 glioblastoma and 35 lower grade gliomas, and 100 healthy donors were analyzed, along with orthotopic transplant mouse model. Urinary EVs were obtained with the nanowire device which could collect EVs more efficiently than the conventional ultracentrifugation method (Yasui et.al., Science Adv.2017). Machine learning methods were performed to select the miRNAs which could distinguish patients with gliomas from healthy control.
RESULTS
More than 2400 miRNAs were obtained from all urine samples. We identified miRNA panels that provided high diagnostic accuracy of diffuse gliomas (92.5%). There were 440 miRNAs whose expression increased by more than 1.5 fold (p< 0.05) as compared to healthy donor samples (glioma-upregulated miRNAs), whereas the expression of 87 miRNAs decreased to less than 2/3-fold (p< 0.05) (glioma-downregulated miRNAs). Mouse miRNAs which were homologous to glioma-upregulated and -downregulated miRNAs showed significantly high and low level expressions, respectively, in glioma mouse models as compared to normal control mice, confirming the reliability of urine miRNA-based diagnosis. Furthermore, some of these glioma-upregulated miRNAs has been reported to be involved in tumor progression.
CONCLUSIONS
miRNAs obtained from urine could be biomarkers for detection of gliomas by machine learning and some of these could be associated with tumor progression.