Integrating Single-cell Analysis, Spatial Transcriptomics, Mendelian Randomisation And Machine Learning Frameworks To Identify TNFAIP6 and CHI3L2 As Diagnostic Biomarkers For Osteoarthritis

Background Osteoarthritis (OA) is a highly disabling degenerative joint disease, characterized pathologically by a progressive disruption of joint homeostasis. This manifests primarily as degradation of the cartilage extracellular matrix, dysregulation of chondrocyte phenotypes, and osteophyte formation. Clinically, patients commonly present with joint pain and swelling, stiffness, restricted mobility, and functional impairment, predominantly affecting the knees, hands, and hips. Severe cases often necessitate joint replacement surgery. Early diagnosis is paramount for improving patient prognosis. Recent investigations have focused on the role of oxidative stress-related genes (OSGs) in the early prediction of OA. This study aims to leverage single-cell analysis-derived OSG expression profiles, integrated with machine learning (ML) approaches, to identify prognostic biomarkers and enhance the efficacy of early OA diagnosis. Methods Single-cell RNA sequencing (scRNA-seq) datasets and spatial transcriptomics datasets related to osteoarthritis (OA) were collected from the GEO database. scRNA-seq analytical methods were applied to investigate the composition and interrelationships of distinct chondrocyte subpopulations in OA, followed by visualization of the transcriptional landscape. Gene sets associated with oxidative stress were retrieved from GeneCards. The expression levels of oxidative stress were evaluated and scored across different chondrocyte subpopulations using multiple algorithms. The CellChat R package was employed to explore potential cell-cell communication mechanisms. Pseudotime analysis revealed distinct states of chondrocyte subpopulations under oxidative stress. Core oxidative stress-related genes (OSGs) were identified and screened using a gene module analysis method based on co-expression network analysis. Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Set Variation Analysis (GSVA) enrichment analyses were performed on the core OSGs. Furthermore, we integrated three machine learning algorithms—XGBoost), random forest, CatBoost, GBDT and Dung Beetle Optimization (DBO)—to identify the best feature genes and construct a diagnostic risk model. Model performance was evaluated using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. The roles of core OSGs in cell-cell communication, immune cell infiltration, and interactions with immune-related molecules at the single-cell level were analyzed. Mendelian randomization (MR) and colocalization analyses were further employed for validation. Spatial transcriptomics was applied to reveal the spatial expression landscape of core genes in ProC within OA. Results We identified TNFAIP6 and CHI3L2 as key OSGs associated with OA, demonstrating significant differential expression at the single-cell level. The diagnostic model constructed using these OSGs exhibited significant accuracy, with consistently high AUC values. Their potential as diagnostic biomarkers was confirmed through the integration of multiple algorithms and single-cell data analysis. The findings were further corroborated by spatial transcriptomics, Mendelian randomization, and colocalization analyses. Conclusions TNFAIP6 and CHI3L2 are highlighted as primary biomarkers, underscoring their promise as therapeutic targets for osteoarthritis. The identification of these core biomarkers may facilitate early diagnosis, potentially altering disease trajectory and improving patient outcomes.