Osteoarthritis (OA) is a major cause of human disability. Despite receiving treatment, patients with the middle and late stage of OA have poor survival outcomes. Therefore, within the framework of predictive, preventive, and personalized medicine (PPPM/3PM), early personalized diagnosis of OA is particularly prominent. PPPM aims to accurately identify disease by integrating multiple omic techniques; however, the efficiency of currently available methods and biomarkers in predicting and diagnosing OA should be improved. Disulfidptosis, a novel programmed cell death mechanism and appeared in particular metabolic status, plays a mysterious characteristic in the occurrence and development of OA, which warrants further investigation.
In this study, we integrated three public datasets from the Gene Expression Omnibus (GEO) database, including 26 OA samples and 20 normal samples. Via a series of bioinformatic analysis and machine learning, we identified the diagnostic biomarkers and several subtypes of OA. Moreover, the expression of these biomarkers were verified in our in-house cohort and the single cell dataset.
Three significant regulators of disulfidptosis (NCKAP1, OXSM, and SLC3A2) were identified through differential expression analysis and machine learning. And a nomogram constructed based on these three regulators exhibited ideal efficiency in predicting early- and late-stage OA. Furthermore, based on the expression of three regulators, we identified two disulfidptosis-related subtypes of OA with different infiltration of immune cells and personalized expression level of immune checkpoints. Notably, the expression of the three regulators was demonstrated in a single-cell RNA profile and verified in the synovial tissue in our in-house cohort including 6 OA patients and 6 normal people. Finally, an efficient disulfidptosis-mediated diagnostic model was constructed for OA, with the AUC value of 97.6923% in the training set and 93.3333% and 100% in two validation sets.