Collagen synthesis-associated pivotal biomarkers identification and ceRNA networks construction in frozen shoulder by comprehensive bioinformatics and machine learning strategies

Abstract

Abstract Background Frozen shoulder is characterized by aberrant collagen synthesis and fibrosis. Long non-coding RNAs (lncRNAs) have been implicated in collagen production and fibrosis development. However, the specific alterations in lncRNA expression in frozen shoulder patients remain poorly understood. Therefore, this study aimed to identify collagen synthesis-related genes and provide a competitive endogenous RNA (ceRNA) networks for frozen shoulder. Methods We acquired dataset GSE140731 from the Gene Expression Omnibus (GEO) database and used the 'limma' R software package to identify differentially expressed lncRNAs (DELs) and mRNAs (DEGs). These DEGs were intersected with collagen synthesis-related genes to obtain collagen synthesis-related DEGs (CS-DEGs). Subsequently, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were conducted using the 'ClusterProfiler' package in R. Machine learning algorithms were employed to select candidate core genes based on CS-DEGs, which were then validated using an external dataset (GSE190023). This validation process led to identifying two core genes, COL11A1 and ADAMTS14. TargetScan, miRTarBase, and miRDB databases were utilized to predict target microRNAs for these core genes, while ENCORI was employed to predict target lncRNAs for these microRNAs. The intersection of predicted lncRNAs with DELs yielded core lncRNAs. Lastly, the 'ggalluvial' package in R was used to construct the lncRNA-miRNA-mRNA ceRNA networks. The ceRNA networks was further validated using quantitative real-time polymerase chain reaction (RT-qPCR). Result A total of 427 DELs and 549 DEGs were identified. Combined with the Molecular Signatures Database (Msigdb), we discovered 23 upregulated and 1 downregulated CS-DEGs. These genes were primarily associated with collagen synthesis regulation. Using three machine learning algorithms, we selected three candidate core genes, and after validation with an external dataset, two core genes (COL11A1 and ADAMTS14) were identified. By cross-predicting miRNAs and establishing lncRNA-miRNA interactions, we constructed a ceRNA networks of 2 lncRNAs, 9 miRNAs, and 2 mRNAs. Histological staining revealed increased collagen fibres and fibroblast cell numbers in frozen shoulder synovial tissues using H&E staining, and significantly higher fibrosis was observed in frozen shoulder patients compared to the control group using Masson's trichrome staining. RT-qPCR results were consistent with the sequencing data. Conclusion This study represents the first attempt to construct a ceRNA networks related to collagen synthesis in frozen shoulder using a combination of bioinformatics approaches and experimental validation. The identified ceRNA networks has the potential to regulate the development and progression of fibrosis in frozen shoulder, thereby presenting promising biomarkers and therapeutic targets for the diagnosis and treatment of frozen shoulder and associated joint stiffness.