Osteoarthritis (OA) is a common degenerative joint disorder marked by progressive cartilage loss, synovial inflammation, and changes in subchondral bone that lead to pain and disability. Current therapies focus on symptom control rather than disease modification, creating a need for molecular biomarkers that could enable early diagnosis, risk stratification, and new therapeutic targets. Post-translational modifications such as acetylation have been implicated in processes relevant to OA pathogenesis, including regulation of gene expression, inflammation, and cartilage matrix degradation. However, the specific molecular mechanisms linking acetylation to OA remain incompletely defined. This study used integrated bioinformatics and experimental approaches to identify acetylation-related differentially expressed genes and evaluate their diagnostic potential in OA.
The study combined computational analyses with experimental validation. Public microarray data corresponding to synovial membrane tissue were obtained and processed to identify differentially expressed genes (DEGs). Co-expression network analysis isolated OA-associated modules, and intersecting results with an externally curated acetylation gene list produced acetylation-related DEGs (ACEDEGs). Protein–protein interaction (PPI) mapping and machine learning algorithms prioritized candidate hub genes. A predictive nomogram and ROC analysis assessed diagnostic performance. Immune cell infiltration was estimated by CIBERSORT. Selected candidates were validated by qRT-PCR and Western blot.
Three GEO synovial membrane datasets (GSE55235, GSE55457, GSE12021) were used as the training cohort. After probe-to-gene conversion and background correction, datasets were normalized using the limma package and batch effects were removed using the sva package. The integrated training cohort included 29 OA samples and 29 control samples. GSE82107 served as an independent validation cohort containing 10 OA and 7 control samples. Acetylation-associated genes were retrieved from GeneCards (Version 5.25, updated Jul 16, 2025) using a relevance score threshold > 2, yielding 4,405 acetylation-related genes.
DEGs between OA and control synovial samples were identified using the limma package with filtering criteria of |log2 fold change| > 0.5 and p < 0.05. Differentially expressed genes were subjected to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses to characterize the biological processes and pathways associated with the transcriptional changes observed in OA.
WGCNA was applied to construct co-expression modules from the processed expression data. A Spearman correlation-based similarity matrix was computed and transformed into an adjacency matrix using a soft threshold power of 14, followed by conversion to a topological overlap matrix (TOM). Average-linkage hierarchical clustering with a minimum module size of 70 genes produced initial modules, which were merged based on similarity. Five final modules were identified, and the red module was selected as the core module related to OA for downstream intersection analyses.
ACEDEGs were defined as genes present in the intersection of core WGCNA module genes, the list of acetylation-related genes from GeneCards, and the set of DEGs. The protein–protein interaction network for ACEDEGs was constructed using the STRING platform with a minimum required interaction score > 0.4. The resulting PPI network was imported into Cytoscape (v3.9.1), and the top 13 genes were prioritized using the MCC algorithm in the cytoHubba plugin for further evaluation.
Three machine learning methods were applied to refine hub gene selection: Least Absolute Shrinkage and Selection Operator (LASSO) regression (glmnet package with tenfold cross-validation), Random Forests (randomForest package), and eXtreme Gradient Boosting (XGBoost package). Each algorithm contributed to narrowing candidate genes toward robust OA-associated markers.
A diagnostic prediction model (nomogram) incorporating selected hub genes was constructed and its discriminative performance assessed by receiver operating characteristic (ROC) analysis. Reported model performance included an area under the ROC curve (AUC) of 0.983 in the training cohort and 0.743 in the validation cohort.
Integrated analysis of the three training GEO datasets produced a set of DEGs meeting |log2FC| > 0.5 and p < 0.05. Functional enrichment linked these DEGs to biological processes and pathways relevant to OA pathogenesis. WGCNA identified five merged gene modules; the red module was chosen for its association with OA. Intersection of the red module genes with the GeneCards-derived acetylation gene list (4,405 genes) and the DEGs yielded a set of ACEDEGs. PPI mapping and MCC ranking highlighted a subset of core genes, which were then refined using LASSO, Random Forest, and XGBoost approaches. A multi-gene nomogram derived from these analyses demonstrated high discriminative ability in the training set and moderate performance in the independent validation set (AUCs reported above).
Immune infiltration analysis using CIBERSORT indicated significant differences in immune cell composition between OA and control synovial samples, suggesting an altered immune microenvironment in OA. Experimental validation by quantitative reverse transcription PCR (qRT-PCR) and Western blotting (WB) confirmed the down-regulation of two prioritized biomarkers, JUN and MYC, in OA synovial tissue compared with controls.
CIBERSORT-based deconvolution revealed alterations in immune cell populations associated with OA in the analyzed synovial samples. The study reports notable changes in immune infiltration patterns, supporting the view that immune microenvironment alterations accompany transcriptional changes in OA synovium.
This integrative study combined public transcriptomic datasets, curated acetylation gene annotations, co-expression network analysis, PPI mapping, machine learning selection, and experimental validation to identify acetylation-related biomarkers in OA. The workflow prioritized ACEDEGs derived from the red WGCNA module and validated JUN and MYC as down-regulated in OA synovium by qRT-PCR and Western blot. The diagnostic nomogram exhibited strong performance in training and moderate generalizability in validation. The observed changes in immune cell infiltration further link acetylation-related transcriptional alterations to the OA immune microenvironment.
Limitations include reliance on publicly available microarray cohorts with limited sample sizes and potential heterogeneity between datasets; details of some intermediate gene lists, specific enrichment results, and exact model features were not reported in the source summary. The study does, however, provide a reproducible bioinformatics framework for identifying post-translational modification–related biomarkers in OA.
Using an integrated bioinformatics pipeline and experimental confirmation, the study identifies JUN and MYC as novel acetylation-related biomarkers associated with OA synovium. The findings suggest these genes may have diagnostic utility and may inform future mechanistic or therapeutic studies focused on acetylation-related pathways in OA.
All relevant data are reported within the manuscript and supporting information files. The research was funded by the Guangxi Natural Science Foundation (2023GXNSFAA026402). The authors declared no competing interests.