The microarray dataset for RA was retrieved from the GEO database of the National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov/geo/) using the search term “rheumatoid arthritis.” The selection criteria were as follows: (1) the study type involved expression profiling via microarrays, (2) the tissue type was synovial tissue, and (3) the organism was Homo sapiens. To minimize the influence of biological disease-modifying anti-rheumatic drugs (bDMARDs), only pre-treatment samples were included (27, 28). In total, four gene expression datasets (GSE206848, GSE55235, GSE172188, and GSE12021) were identified as suitable. The GSE206848, GSE55235, and GSE172188 datasets were used as the training set, consisting of 20 RA patients and 17 healthy controls, while the GSE12021 dataset served as the validation set, comprising 12 RA patients, 9 healthy controls, and 10 osteoarthritis (OA) patients. Information on patient age and sex is provided in Supplementary Table S1 .

The “GEOquery” package was employed to convert the probe matrix into a gene expression matrix, using probe annotation files. If multiple probes corresponded to the same gene, the expression values of the first probe were retained. Given that these three datasets were derived from different platforms and exhibited batch effects, the “sva” package was utilized to correct for platform-related batch effects.

The “limma” package was employed to analyze DEGs between RA patients and healthy controls. The selection criteria for DEGs included a p < 0.05 and an absolute fold change > 2. Functional enrichment analysis of Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways for the upregulated DEGs was performed using the “clusterProfiler” package in R. Pathways with p < 0.05 was considered significantly enriched.

We used the R package “WGCNA” to construct a weighted gene co-expression network for the DEGs. The soft threshold was selected using the “pickSoftThreshold” algorithm to build the gene co-expression modules. Each module contained at least 5 genes, with any remaining ungrouped genes assigned to the grey module. We calculated the correlation coefficients between modules and phenotypes to identify modules closely associated with RA. Additionally, gene significance and module membership were assessed to evaluate the relationship between gene modules and RA patients.

We employed machine learning algorithms for hub gene selection, including the random forest (RF) model (29) and least absolute shrinkage and selection operator (LASSO) regression (14).

The RF model is a decision tree-based machine learning approach. In this study, we utilized R package “randomForest” to build the random forest model. To determine the optimal number of variables, we computed the average error rate of candidate module genes. We then evaluated the error rate across a range of tree numbers, from 1 to 1000, and selected the number of trees that yielded the lowest error rate. Finally, we determined the feature importance scores for each candidate gene in the significant modules and selected genes with importance values greater than 0.

LASSO regression is a widely used machine learning algorithm for fitting generalized linear models. By applying an L1 penalty (λ), coefficients of less important variables are set to zero, thereby selecting the most important variables and building optimal classification models. In our study, we performed LASSO regression analysis on the candidate genes using R package “glmnet” to identify hub genes. The optimal value of λ was determined through tenfold cross-validation, selecting the value that resulted in the smallest standard error.

To evaluate the diagnostic accuracy of hub genes selected through machine learning algorithms, we employed the R package “pROC” to plot receiver operating characteristic (ROC) curves comparing RA patients with healthy controls in the training set. The area under the curve (AUC) was used to assess the diagnostic accuracy of these genes as potential hub genes for RA, with a larger AUC indicating higher accuracy. This analysis was subsequently validated in the independent validation set. Furthermore, to assess the specificity of the hub genes for RA diagnosis, we plotted ROC curves comparing RA patients to OA controls in the validation set. Expression levels of the hub genes in RA patients and controls were visualized using box plots generated by R package “ggplot2.”

We conducted gene set variation analysis using the R package “GSVA” to explore the correlation between hub genes and hallmark signaling pathways. Subsequently, we applied the CIBERSORT algorithm to determine the infiltration of different immune cells in RA synovial tissues and healthy controls. Spearman correlation coefficients were then used to analyze the relationships between hub genes and immune cell types.

Twelve participants were recruited for this study from the First Affiliated Hospital of Shantou University Medical College, comprising 6 RA patients and 6 OA patients. The age and sex information of the participants can be found in Supplementary Table S1 . Synovial tissue samples were collected from each participant during joint surgery. Our study protocol was approved by the Ethics Committee of the First Affiliated Hospital of Shantou University Medical College (Approval No. B-2024-060). Informed consent was obtained from all enrolled patients.

Total RNA was extracted from tissue samples using TRIzol (TIANGEN, China) according to the manufacturer’s instructions. RNA was then reverse transcribed into complementary DNA (cDNA) using cDNA Synthesis SuperMix for qPCR (YEASEN, China). qRT-PCR reactions were conducted using Advanced qPCR SYBR Green Master Mix (YEASEN, China). β-Actin was used as an internal control, and relative mRNA quantification was calculated using the 2^−ΔΔCt method. Primer sequences used for qRT-PCR are listed in Supplementary Table S2 .

Total protein was extracted from synovial tissue using RIPA buffer (YEASEN, China) supplemented with protease and phosphatase inhibitors (P002, NCM Biotech, China). Briefly, protein samples were separated by 10% SDS-polyacrylamide gel electrophoresis and transferred onto PVDF membranes. After blocking with 5% non-fat milk, the membranes were incubated overnight at 4°C with primary antibodies: GAPDH (1:10000, 10494-1-AP, Proteintech, USA) and CKAP2 (1:2000, 25486-1-AP, Proteintech, USA), followed by incubation with secondary antibodies at room temperature for 2 hour. Target proteins were detected using an ECL kit (4AW011-200, 4A Biotech, China).

Synovial tissue was fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned into 4 μm-thick slices. The tissue sections were deparaffinized and dehydrated in xylene. For H&E staining, sections were stained with hematoxylin (G1100, Solarbio) and eosin (G1140, Solarbio). IHC was performed using an IHC Kit (KIT-9710, MXB Biotechnologies, China). The primary antibody used was CKAP2 (1:800, PS13605M, Abmart, China). Sections were stained using a DAB Detection Kit (DAB-0031, MXB, China) and counterstained with hematoxylin. Neutral resin was applied to seal the sections, and slides were imaged in the Department of Pathology of Shantou Central Hospital using a digital slide scanner (Pannoramic SCAN, 3DHISTECH Ltd, Hungary).

The MH7A (human rheumatoid arthritis synovial cell line) and HFLS-RA cells (human fibroblast-like synoviocytes: rheumatoid arthritis) were purchased from Shanghai Guan&Dao Biological Engineering Co., Ltd. (Shanghai, China). The passage number of all cell lines used for the experiments was no greater than 10. All cell lines were maintained in DMEM (Gibco, USA) supplemented with 10% fetal bovine serum (Gibco, USA), at 37°C under 5% CO₂. For CKAP2 knockdown, shRNA sequences targeting CKAP2 or a scramble sequence were cloned into an pLKO.1-puro lentiviral cloning vector (Guangzhou IGE Biotechnology Co., Ltd., China). The shRNA and scramble sequences are shown in Supplementary Table S3 . The virus was packaged in HEK293T cells after transfection with lentiviral packaging vectors using EZ Cell Transfection Reagent II (Shanghai Life-iLab Biotech, China), and then used to infect MH7A and HFLS-RA cells. Transfected cells were selected with puromycin (2 µg/mL) for one week, and expression levels of the target RNA were confirmed by qRT-PCR.

Cell proliferation was assessed using the Cell Counting Kit-8 (CCK-8) assay (YEASEN, China) according to the manufacturer’s protocol. RA cells were seeded in 96-well plates at a density of 3,000 cells per well. Once the cells had adhered, CCK-8 solution (10 µL) was added to 90 µL of serum-free medium at 0, 24, 48, 72, and 96 hours, followed by a 2-hour incubation at 37°C. Absorbance at 450 nm was measured to determine the optical density (OD) values. Cell proliferation was calculated as follows: cell proliferation = absorbance value at each time point/absorbance value at 0 hours.

Cells were digested with trypsin, counted, and seeded at a density of 1 × 10⁷ cells per well in a 6-well plate, allowing them to grow until confluence. A scratch wound was made using a 200 µL pipette tip. Closure of the wound was monitored at 0, 24, and 48 or 72 hours to assess cell migration.

A transwell assay was used to measure cell invasion and migration. Cells were digested with trypsin, and 4 × 10⁴ cells were seeded into the upper chamber of a transwell insert containing serum-free culture medium (for the migration assay) or pre-coated with Matrigel (for the invasion assay). The lower chamber was filled with complete medium to stimulate cell migration. After 48 hours of incubation, the medium and non-migrated or non-invaded cells in the upper chamber were removed. The remaining cells were fixed and stained with 0.1% crystal violet. Migration or invasion was quantified by counting the cells in five random fields at 200X magnification on each membrane.

Data were analyzed using R software (version 4.3.0) and GraphPad Prism (version 9.5.1). Continuous variables were compared using t-tests or Wilcoxon tests, while categorical variables were evaluated using chi-square tests. Statistical significance was defined as p < 0.05 (two-sided).

Microarray datasets GSE206848, GSE55235, and GSE172188 were obtained from the GEO database, comprising a total of 20 RA patients and 17 healthy controls. After rigorous quality control, the batch effect was effectively eliminated, ensuring a robust foundation for subsequent analyses ( Figures 1A, B ). Differential expression gene analysis revealed 242 DEGs in RA, with 146 upregulated genes, and 96 downregulated genes. Volcano plots and heatmaps further highlighted the differences in gene expression profiles between RA patients and healthy controls ( Figures 1C, D ). Functional enrichment analysis was conducted on the 146 upregulated genes. GO enrichment analysis revealed that these genes were enriched in pathways related to white blood cell-mediated immunity, immunological synapse formation, immunoglobulin complex, and chemokine activity, indicating their potential involvement in immune responses. In KEGG enrichment analysis, the most significantly enriched pathway was associated with RA, with additional pathways related to Th1 and Th2 cell differentiation and type 1 diabetes, all of which are closely linked to immune responses or immune-related diseases ( Figures 1E, F ).

To identify hub genes associated with RA, we performed WGCNA on the DEGs between healthy and RA synovial tissues. A soft threshold of 16 was set to ensure a scale-free network distribution ( Figure 2A ). By evaluating gene correlations, we constructed a hierarchical clustering dendrogram, revealing 6 distinct gene modules with similar co-expression patterns ( Figure 2B ). The “turquoise” module exhibited a correlation coefficient of 0.60 and a p of 8 × 10–⁵ with the RA, indicating a strong association between the module’s gene expression and RA ( Figure 2C ). Furthermore, within the “turquoise” module, a significant correlation was observed between gene significance and module membership, with a correlation coefficient of 0.39 and a p of 0.00017 ( Figure 2D ). As a result, the “turquoise” module was identified as a key module associated with RA.

To further identify hub genes in RA, we applied the RF method and identified 56 core genes ( Figures 2E, F ). Subsequently, LASSO regression analysis was conducted, leading to the identification of 3 hub genes: CKAP2, POU2AF1, and HLA-DOB, all of which were selected as hub genes associated with RA ( Figures 2G, H ).

To assess the diagnostic efficacy of the identified hub genes, we performed ROC curve analysis. In the training set, the AUC values were 0.876 for CKAP2, 0.885 for POU2AF1, and 0.897 for HLA-DOB ( Figure 3A ). In the validation set, compared to healthy controls, the AUC values were 0.898 for CKAP2, 0.935 for POU2AF1, and 0.898 for HLA-DOB ( Figure 3B ). Additionally, to evaluate the specificity of the hub genes for RA, we compared RA to OA in the validation set and calculated the AUC values for the hub genes. The results showed AUC values of 0.808 for CKAP2, 0.833 for POU2AF1, and 0.750 for HLA-DOB ( Figure 3C ). These findings indicate that the 3 hub genes possess high diagnostic value for RA patients, regardless of whether they are compared with healthy controls or OA controls. Furthermore, we compared the expression profiles of the hub genes in RA patients, healthy controls, and OA controls in both the training and validation sets. The results demonstrated that these hub genes were expressed at higher levels in RA patients ( Supplementary Figures S1A–C ).

The CIBERSORT algorithm revealed significant differences in immune cell infiltration between RA patients and healthy controls in the synovium. Specifically, RA patients exhibited markedly increased infiltration of plasma cells (p = 0.011) and follicular helper T (Tfh) cells (p = 0.033), while healthy controls showed higher levels of resting memory CD4+ T cells (p = 0.016), activated NK cells (p = 0.040), and activated mast cells (p = 0.004). A relatively large proportion of M2 macrophages were found in both RA patients and healthy controls, but they were not statistically different ( Figure 4A ). Further analysis of immune cell infiltration in RA revealed associations between 3 hub genes and specific immune cell types. These genes were positively correlated with the infiltration of plasma cells, Tfh cells, and activated CD4+ T cells, while showing negative correlations with the infiltration of resting memory CD4+ T cells, activated NK cells, and activated mast cells ( Figure 4B ).

GSVA demonstrated that CKAP2, POU2AF1, and HLA-DOB were associated with pathways, such as IL6/JAK/STAT3 signaling, and involved in inflammatory response, complement activation, and interferon response ( Figure 4C ). Scatter plots with Spearman correlation analysis were performed to compare the relationship between the hub genes and the IL6/JAK/STAT3 signaling pathway, as well as the inflammatory response ( Figures 4D, E ).

MR analysis integrates GWAS and expression quantitative trait locus (eQTL) data, and was used to investigate the association between RA and the eQTLs of CKAP2, POU2AF1, and HLA-DOB. Results from the IVW, WM, and weighted models consistently supported a positive correlation between CKAP2 and RA ( Figure 5A ). A scatter plot illustrating the SNP effect size for CKAP2 and RA shows the SNP effect on CKAP2 along the x-axis and the SNP effect on RA along the y-axis ( Figure 5B ). MR-Egger analysis revealed no evidence of horizontal pleiotropy (p = 0.77, Figure 5C , Supplementary Table S4 ). The funnel plot demonstrated no significant heterogeneity among the SNPs (p < 0.05, Figure 5D , Supplementary Table S5 ). Leave-one-out analysis indicated that individual SNPs did not influence the MR analysis results ( Figure 5E ). These findings support a positive causal relationship between CKAP2 and RA.

We examined CKAP2 expression levels in synovial tissues from patients with RA and OA. qRT-PCR and western blot analyses revealed an upregulation of CKAP2 mRNA ( Figure 6A ) and protein ( Figure 6B ) in RA synovial tissues. Histopathological analysis using H&E staining showed marked synovial hyperplasia and increased immune cell infiltration in RA compared to OA ( Figure 6C ). IHC confirmed CKAP2 expression in both RA and OA synovium, with higher expression in RA synovium, compared to OA synovium ( Figure 6D ). To determine whether CKAP2 expression is specific to synovial tissue, we analyzed CKAP2 expression levels in peripheral blood mononuclear cells from 232 RA patients and 43 healthy controls using the GSE93777 dataset. No significant difference was observed between the two groups ( Supplementary Figure S2 ).

We knocked down CKAP2 in MH7A and HFLS-RA cells, using shRNA, and validated the extent of silencing by qRT-PCR. The results indicated that CKAP2 expression was reduced in the shCKAP2 groups compared to the scramble group.

Meanwhile, CKAP2 sh3 showed the superior silencing efficiency and was thus selected for subsequent experiments. ( Figures 7A, B ).

To assess the effect of CKAP2 on cell proliferation, we performed a CCK-8 assay. shCKAP2 reduced the proliferation of MH7A and HFLS-RA cells compared to the scramble group ( Figures 7C, D , both p < 0.001), suggesting that CKAP2 knockdown suppressed cell proliferative ability.

To evaluate the impact of CKAP2 on migration and invasion, we conducted wound-healing and transwell migration and invasion assays. Wound-healing assays showed wound surface areas for the shCKAP2 group were larger than for the scramble group in both MH7A and HFLS-RA cells ( Figures 7E, F ). Similarly, transwell assays demonstrated that knockdown of CKAP2 inhibited both migration and invasion of MH7A and HFLS-RA cells ( Figures 7G, H ). Therefore, these results suggest that CKAP2 promotes the migration and invasion of MH7A and HFLS-RA cells.