A validated miRNA signature for the diagnosis of osteoporosis related fractures using SVM algorithm classification

Abstract

The aim of the present study was to develop a circulating microRNA expression signature for early prediction of osteoporotic fractures and to validate the results using Gene Expression Omnibus (GEO) datasets. The GSE70318 dataset was downloaded from GEO and used to build an osteoporotic fracture prediction model based on the receiver operating characteristic curve and support vector machine (SVM) classification index. The GSE74209 dataset was used as a validation dataset. Additionally, in vitro, alkaline phosphatase (ALP) activity was measured in the presence or absence of microRNA (miRNA/miR) treatments in human osteoblast cells. The expression of two selected genes was detected by western blotting. miR-188-3p, miR-942-3p, miR-576-3p and miR-135a-5p were differentially expressed between controls and osteoporotic patients with fractures. SVM classification using these four miRNAs provided better dichotomization. It was further confirmed that miR-576-3p and 135a-5p in the GSE74209 dataset could also significantly discriminate between the controls and fracture patients, the area under the curve of SVM2 was 0.9722 with 95% CI 0.8885-1.056. Further analysis indicated that the target genes of the two miRNAs participated in the Wingless-related integration site, Hedgehog and transforming growth factor-β signaling pathways and osteoclast differentiation. miR-576-3p and miR-135-5p transfection decreased ALP activity and ALP activity was increased in the presence of blocking antisense oligonucleotides. Western blotting indicated miR-576-3p and miR-135-5p decreased CSNK1A1L and LRP6 levels, respectively. In conclusion, two miRNA signatures were developed and validated for the prediction of osteoporotic fractures.