Distinct Metabolite Profiles Are Linked to Gender and Obesity in Plasma of Patients with Knee Osteoarthritis
J. S. Rockel,W. Zhang,S. Likhodi,G. Sun,A. Furey,E. Randell,K. Sundararajan,R. Ghandi,G. Zhai,M. Kapoor
DOI: https://doi.org/10.1016/j.joca.2017.02.157
IF: 7.507
2017-01-01
Osteoarthritis and Cartilage
Abstract:Purpose: To identify if distinct phospholipid metabolite profiles matched to gender and BMI of patients with osteoarthritis (OA) exist in plasma compared to normal volunteers. Methods: Plasma from patients undergoing total knee replacement surgery due to primary OA or normal adult volunteers with no self-reported diagnosis of OA, both from Newfoundland and Labrador (Canada), were collected with informed consent. Patient plasma was isolated and concentrations of 87 phosphatidylcholine (PC) and lysophosphatidylcholine (lysoPC) analogues were determined by LC-MS/MS. Individuals ≥ age 50 were segregated based on gender and/or obesity (BMI≥30) for further analysis using the web based tool suite Metaboanalyst v3.0 (metaboanalyst.ca). Metabolites were median normalized and log-transformed prior to analysis. PC and lysoPC analogue concentrations were analyzed by Student’s t-test in segregated cohorts to determine metabolites significantly changed in OA vs normal volunteer plasma. Plasma levels were also analyzed by PLS-DA and variable in importance (VIP) scores were determined. Metabolites with VIP scores≥1 and significantly different (P<0.05) were selected for further analysis. Metabolites common to multiple subgroups were combined into a master list applicable to the entire cohort. Potential biomarkers were selected for further analysis by entering common metabolites into a logistic regression model of OA, selecting those with odds ratios significantly different from 1, and determining the area under the ROC (AUROC). As a sensitivity analysis, ElasticNet logistic regression with 10-fold cross-validation to select tuning parameters was also used for metabolite selection. Results: From a cohort of 379 individuals, 262 were age≥50. Of the 262 individuals whose plasma was further analyzed, 114 were normal volunteers and 148 were patients receiving TKR surgery. This cohort was further segregated into males (124 individuals) with BMI < 30 (32 normal, 30 OA) or BMI≥30 (20 normal, 42 OA), and females (138 individuals) with BMI < 30 (33 normal, 29 OA) or BMI≥30 (26 normal, 47 OA). There were no overlapping PCs or lysoPCs meeting cutoff criteria between males and females with BMI < 30 while a single metabolite was common in lists generated between males and females with BMI≥30. Overall, this would suggest a substantial difference in metabolite generation between males and females with OA when segregated based on BMI. Furthermore, there was greater similarity between lists generated between males compared to females (6 vs 0 common metabolites, respectively), suggesting a more homogeneous change in metabolite production in males compared to females. Lists generated from all males compared to all females, regardless of BMI had 4 overlapping metabolites that were changed with OA, suggesting some degree of similarity in metabolite level changes in males and females. Using Venn analysis, we selected common metabolites present within individual lists generated and determined that three metabolites, namely PC ae 38:0, lysoPC a C16:0 and PC ae C38:2 were potential biomarkers for OA, with a logistic regression AUROC of 0.75 (CI:0.674–0.814) and odds ratios of 0.27, 2.69 and 0.09, respectively (all p≤0.001). ElasticNet analysis confirmed these metabolites were strongly associated with OA, both overall and in the gender/obesity subgroups. Conclusions: Plasma metabolites associated with knee OA differ substantially with gender and obesity while metabolite predictors of OA vary more with obesity in women compared to men. Candidate PC and lysoPC metabolites show promise as a panel of plasma biomarkers for knee OA that span gender and BMI. We are currently generating a validation cohort to further evaluate biomarker lists generated from our segregated cohorts and confirm the usefulness of the panel of 3 metabolites identified for identification of knee OA.