Discovery of potential biomarkers for osteoporosis using LC-MS/MS metabolomic methods
J Wang,D Yan,A Zhao,X Hou,X Zheng,P Chen,Y Bao,W Jia,C Hu,Z-L Zhang
DOI: https://doi.org/10.1007/s00198-019-04892-0
Abstract:Our study focused on the associations of metabolites with BMD and osteoporosis, finding that several metabolites are associated with BMD, and metabolites combined with bone turnover markers tend to be more sensitive in distinguishing osteoporosis in both males and postmenopausal females, which might be meaningful for the early diagnosis of osteoporosis. Introduction: Our study aimed to evaluate the association of metabolites with bone, trying to find new metabolic markers that are distinguishing for low bone mineral density (BMD). Methods: Our study recruited 320 participants, including 138 males and 182 postmenopausal females from the Shanghai area. Bone turnover markers (BTMs), including osteocalcin, PINP and β-CTX, and other biochemical traits were tested. BMD values of the lumber spine (L1-4), femoral neck and total hip were determined using dual-energy X-ray absorptiometry and the serum metabolome profiles including 221 metabolites from five groups (acylcarnitines, amino acids, biogenic amines, glycerophospholipids, sphingolipids and hexose) were assessed by mass spectrometry. Results: No visual separation in the metabolic profiles between different BMD groups was observed in principal component analysis (PCA) or partial least squares discriminant analysis (PLS-DA) models. We compared metabolites in three groups with different BMD levels in males and postmenopausal females separately and further filtering these metabolites via random forest-based feature selection, a commonly applied machine learning algorithm which could select the features with the greatest impact on osteoporosis, then metabolites with the highest importance (≥ 5%) (5 in males and 9 in postmenopausal females) were selected to construct better models for osteoporosis classification. After adding these selected metabolites to the model, the area under the curve (AUC) of receiver operating characteristic (ROC) curves increased significantly (BTMs: AUC 0.729, 95% CI 0.647-0.802, p < 0.0001, model 1: AUC = 0.828, 95% CI 0.754-0.888, p < 0.0001; model 1 versus model of BTMs: p = 0.0158) compared to the AUC of the BTM-only model in males. Similar results were also observed in postmenopausal females (BTMs: AUC = 0.638, 95% CI 0.562-0.708, p = 0.0025; model 2: AUC = 0.741, 95% CI 0.669-0.803, p < 0.0001; model 1 versus model of BTMs: p = 0.0182). Conclusion: Metabolites combined with traditional BTMs tend to better markers for distinguishing osteoporosis in both males and postmenopausal females than BTMs alone.