Qualifiction of Proteomic Biomarkers for Knee Osteoarthritis Progression

Purpose: To qualify serum biomarkers as predictors of knee osteoarthritis (OA) progression and to evaluate the predictive performance and biological roles of selected biomarker sets in two cohorts. Methods: Participants in the Foundation for NIH (FNIH) cohort (N=600) had radiographic knee OA (Kellgren Lawrence grade I - III) at baseline. Case status (knee OA progressor vs. non-progressor) was based on change over 48-months. Radiographic progression was defined as minimum JSW loss (JSL) of ≥0.7mm and pain progression was defined as a persistent (sustained at ≥2 time points) increase of ≥9 points on the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain subscale (0-100 scale). Participants were ineligible if they were unable to progress (baseline minimal JSW <1.0 mm or WOMAC score >91), or if radiographs were of poor-quality or malposition. For these analyses, baseline sera were analyzed by highly quantitative multiple-reaction monitoring mass spectrometry. A total of 177 peptides corresponding to 101 proteins were targeted based on our prior biomarker discovery proteomics and ELISA-based studies of OA synovial fluid, urine, and serum samples. Intra/Inter-batch variation was assessed with sample pool quality control (SPQC) measurements. Peptides with dot products <90% (the transition ratios in the heavy channel compared to the transition ratios in the light channel) were removed from further analyses. Data quality was investigated to remove biological outliers. Missing at random values were imputed with the random forest method. Baseline clinical covariates were filtered by univariate logistic analyses with retention based on p-value less than 0.15. The primary analysis was the ability of biomarker (baseline) to predict case status (n=194 cases with radiographic JSL and pain progression). The secondary analyses compared JSL and pain progressors, JSL progressors (including JSL and pain progressor, and JSL only progressor), and pain progressors (including JSL and pain progressor, and pain only progressor) to the non-progressor group. Biomarkers were selected, based on their ability to predict radiographic knee OA progression, by bootstrapping elastic net and backward elimination. The performance of the set of selected biomarkers was assessed using logistic models and the area under the receiver operating characteristic curve (AUC). The selected biomarker sets were then assessed, to determine their generalizability, in the Biomarker Factory (BMF) knee OA cohort (n=124). This cohort also had radiographic knee OA (Kellgren Lawrence grade I - III) at baseline, but progression was defined as >1-unit categorical radiographic joint space narrowing (JSN) over 36 months, scored using a standardized OA atlas. Results: Among the four progressor groups (both JSW and pain progressors, JSW only progressors, pain only progressors, and OA non-progressors), baseline age (p=0.012), sex (p=0.003), race (p=0.024), KL grade (p=0.050), and WOMAC score (p=0.047) were significantly different. We observed no statistical difference in baseline body mass index (BMI), pain medication history, or JSW. With elastic net, the 30 biomarkers with highest selection frequencies for each of the four outcomes were defined as the “stable” sets. Based on backwards elimination, we defined four “essential” sets: 15 biomarkers for JSL and Pain progressors vs. other (composite comparator of the three groups); 13 biomarkers for JSL and Pain progressors vs. JSL and Pain non-progressors; 11 biomarkers for JSL progressors vs. JSL and Pain non-progressors; and 10 biomarkers for Pain progressors vs. JSL and Pain non-progressors (Figures 1, 2, 3, and 4). Among the essential biomarkers, complement C1r subcomponent (C1R_1, where the last number denotes the peptide sequence) and cartilage acidic protein 1 (CRAC1_1) appeared in all four models (the suffix numeral was added when more than one peptide was analyzed in a protein). At least one peptide from vitamin D binding protein (VTDB) appeared in the essential set for each of the four models. Other commonly selected peptides included CD44 antigen (CD44_3), dopamine beta-hydroxylase (DOPO), kininogen-1 (KNG1_2), phosphatidylinositol-glycan-specific phospholipase D (PHLD), retinol-binding protein 4 (RET4), thrombospondin-1 (TSP1), and Z-dependent protease inhibitor (ZPI). Demographic characteristics and C-terminal crosslinked telopeptide type II collagen (CTXII, urine sample), each separately, showed limited ability to predict OA progression (AUCs of 0.601 and 0.608, respectively), while models with the essential biomarker set demonstrated significantly higher AUCs (0.728 for JSL and Pain progressor vs. composite comparator, 0.740 for JSL and Pain progressor vs. non-progressor, 0.698 for JSL progressor vs. non-progressor, and 0.673 for Pain progressor vs. non-progressor). The addition of demographic characteristics and urinary CTXII only slightly but non-significantly enhanced the AUCs. Odds ratios (ORs) and 95% confidence intervals were estimated on the standardized biomarker data (Figure 1). These standardized estimates of ORs are comparable as they represent the OR for one standard deviation increase in a biomarker concentration. Peptide C1R_1 showed low ORs across the four models: 0.62 (0.44, 0.87), 0.54 (0.37, 0.78), 0.50 (0.36, 0.70), and 0.64 (0.46, 0.89) respectively. The presence of peptide CRAC1_1 favors knee OA progression: 1.24 (1.01, 1.53), 1.41 (1.10, 1.80), 1.42 (1.11, 1.82), and 1.24 (1.01, 1.54) respectively. Peptides in VTDB also showed uniformly high ORs. The other peptides selected in multiple models were CD44_3, DOPO, KNG1_2, PHLD, RET4, TSP1, and ZPI. Validation in the BMF cohort was comparable to the secondary outcome in FNIH cohort (JSL progressor vs non-progressor). Given that the BMF cohort lacked data for two (VTDB_1 and CD14_3) of the 11 essential biomarkers selected in the FNIH cohort, we evaluated the FNIH data for substitute peptides based on Spearman’s correlation. VTDB_1 was highly correlated with four other peptides from VTDB (rs > 0.89 with VTDB_2, VTDB_3, VTDB_4, and VTDB_5). We therefore chose VTDB_3, as the surrogate for VTDB_1 as it was available in the BMF data. However, there were no peptides highly correlated with CD14_3. Past literature suggested a positive correlation of serum CD14 with serum beta 2-microglobulin (B2MG) (rs = 0.63, P < 0.0001). We therefore chose B2MG as a replacement for CD14_3 as it was available in the BMF data. The model using the essential biomarker set (from the FNIH analysis of JSL and Pain progressor vs. non-progressor) showed strong predictive power (AUC=0.697) without clinical covariates in the BMF cohort. Conclusions: Our Results qualified several promising biomarkers as predictors of radiographic knee OA and knee pain progression. The predictive performance of these models demonstrates the promise of serum biomarkers in knee OA progression.View Large Image Figure ViewerDownload Hi-res image Download (PPT)View Large Image Figure ViewerDownload Hi-res image Download (PPT)View Large Image Figure ViewerDownload Hi-res image Download (PPT)