SYNOVIAL FLUID PROTEOMICS-BASED PREDICTION MODEL INTEGRATING PAIN AND STRUCTURAL BIOMARKERS FOR OSTEOARTHRITIS DIAGNOSIS

초록

Purpose (the aim of the study): Osteoarthritis (OA) poses a significant challenge in clinical diagnostics due to its gradual progression, the absence of predictive markers, and the discordance between radiographic findings and symptomatic presentations. Conventional diagnostic methods, such as imaging and pain assessment, are often subjective, reactive, and inadequate in capturing the complex nature of OA. To address these limitations, we developed a novel prediction model integrating biomarkers indicative of both pain and morphological changes. Synovial fluid (SF) samples from OA patients were analyzed using high-performance liquid chromatography-mass spectrometry (HPLC-MS) to identify protein biomarkers, stratified by Kellgren-Lawrence (KL) grade (indicating structural progression) and Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain scores. Methods: SF samples were obtained from 43 OA patients and pretreated to deplete high-abundance proteins, enhancing detection sensitivity. Proteomic analysis using mass spectrometry identified 401 proteins, with differentially expressed proteins (DEPs) via Welch’s t-test (fold change ≥ |1.2|, p-value ≤ 0.05) across groups (healthy vs early vs late) stratified by KL (low vs high) and WOMAC score (low vs high). Gradient boosting machine (GBM) analysis employed to evaluate the importance of protein expression and demographic variables (sex, age, and BMI) in predicting KL grade and WOMAC scores. The dataset was randomly divided into a training (70%) and a test (30%) set for variable selection. The selected features were normalized using the z-score and further analyzed with elastic net regression, where optimal λ and α values were determined through 10-fold cross-validation. The predictive performance of various biomarker combinations was assessed using multivariate logistic regression models, with coefficients validated against external datasets. Results: From 401 identified proteins, DEPs included 66 proteins for high KL grades, 20 proteins for low KL grades, and 40 proteins linked to WOMAC scores. The GBM analysis identified 132 significant predictors for KL grade and 149 for WOMAC scores. This analysis included known OA biomarkers such as COMP (relative importance: 3.45), MMP3 (relative importance: 0.99), both ranked among the top 20 variables, along with CRTAC1 (relative importance: 0.69). The GBM models demonstrated strong predictive performance, with an R2 of 0.86 and a root mean square error (RMSE) of 0.63 for KL grades, and an R2 of 0.73 and RMSE of 17.9 for WOMAC scores. The proteins selected by the GBM model were then used to train an elastic net regression model, which further identified 22 proteins significantly correlated with KL grades and 14 proteins with WOMAC scores. Overlapping DEPs, including LUM, and CFI, were highlighted as key components of a feature selection set associated with both KL grades and WOMAC scores, as indicated by the overlap between DEPs and coefficient variables within the elastic net regression model. Multivariate logistic regression was conducted to assess the probability of OA using multiple variables, including: 1) CRTAC1 only, 2) MMP3 only, 3) the feature selection set, and 4) the essential biomarker set proposed by Zhou et al. (Sci. Adv, 2023). The feature selection set demonstrated superior predictive performance in multivariate logistic regression, achieving an area under the curve (AUC) of 0.76 for total OA and 0.84 for early OA. These values exceeded those of established biomarkers such as CRTAC1 (AUC = 0.68) and MMP3 (AUC = 0.73), as well as the essential biomarker set (AUC = 0.66). For early OA, the feature selection set (AUC = 0.84) also outperformed CRTAC1 (AUC = 0.63), MMP3 (AUC = 0.7), and the essential biomarker set (AUC = 0.55). Conclusions: Our findings highlight the inadequacy of current OA diagnostic paradigms and emphasize the potential of integrating biomarkers that reflect pain and structural changes.