**4. Materials and Methods**

SRA files of healthy and OA human knee cartilage (GEO accession number: GSE114007) and health and OA synovium (GEO accession number: GSE89408) RNA-seq data were downloaded from https://www.ncbi.nlm.nih.gov/sra. Data analyses were performed on the Galaxy platform (UseGalaxy.org, [84]). The FASTQC RNA-seq reads were aligned to the human genome (GRCh38) using HISAT2 aligner (Galaxy v. 2.1.0+galaxy 5) with default parameters [85]. Samples with an overall alignment rate >75% were used for further analysis. Raw counts of sequencing read for the feature of genes were extracted by featureCounts (Galaxy v. 1.6.4+galaxy1) [86]. Then, the limma package (Galaxy v. 3.38.3+galaxy3) was used to identify DEGs with its voom method [87,88]. Expressed genes were selected as their counts per million (CPM) value not less than 1 in at least two samples across the entire experiment, while lowly expressed genes were removed for the flowing analyses. Quasi-likelihood F-tests (ANOVA-like analysis) were achieved to identify DEGs [89]. Genes with fold change (FC) more than 2 and false discovery rate (FDR) less than 0.01 were assigned as DEGs. Heatmap, multidimensional scaling (MDS), principal component analysis (PCA), and the Venn diagram were conducted in R (v. 3.6.3) [90] with packages pheatmap (v. 1.0.12), vegan (v. 2.5-6), ggplot2 (v. 3.3.0), and VennDiagram (v. 1.6.20). Pathway enrichment of identified DEGs was firstly performed against the Reactome knowledgebase [30]. In addition, the summary of the known biofunctions for these genes was searched in the Uniprot database [31] for manually functional annotation. The STRING network [32] was also utilized for the protein-protein association and interaction assembling.
