Synovial exosomal type II collagen as a biomarker for osteoarthritis Progression: From molecular evaluation to AI-powered SERS-based diagnosis

Osteoarthritis (OA) is a degenerative joint disease which lacks reliable biomarkers for monitoring disease progression. Current assessment methods rely primarily on radiographic Kellgren–Lawrence (K-L) grading and symptom-based scores, which often show poor concordance with underlying molecular changes. Here, we present a label-free diagnostic strategy for classifying OA severity based on type II collagen in synovial exosomes, using surface-enhanced Raman spectroscopy (SERS). Exosomal surface type II collagen (ExoCOL2A1), quantified from synovial fluid of OA patients, exhibited a significant inverse correlation with radiographic severity and demonstrated superior diagnostic capability compared to other protein markers. To enable non-destructive detection, a plasmonic gold nanoparticle substrate was used to acquire the SERS spectra of whole synovial exosomes, which were subsequently analyzed using a deep neural network (DNN). The DNN model accurately classified OA severity based on spectral characteristics, achieving a prediction accuracy of 95.3 %, and outperforming traditional machine learning models such as LDA, SVM, and kNN. Principal component analysis further identified ExoCOL2A1-associated spectral peaks as key contributing factors for the exosomal SERS spectrum. These findings establish the clinical potential of exosomal type II collagen as OA severity–associated markers and highlight the utility of AI-integrated SERS as a non-invasive diagnostic platform.