Identification biomarkers and therapeutic targets of disulfidptosis-related in rheumatoid arthritis via bioinformatics, molecular dynamics simulation, and experimental validation

Abstract

The relationship between disulfidptosis and rheumatoid arthritis (RA) remains unclear. We aimed to identified biomarkers disulfidptosis-related in RA and revealed potential targeted drugs. Two microarray datasets (GSE93272, GSE45291) related to RA were downloaded from the Gene Expression Omnibus (GEO) database. Disulfidptosis-related genes(DRGs) were extracted from FerrDb database. GSE93272 was used to identify DRGs, and GSE45291 was used to verify results. Multivariate Cox regression analysis was used to identify candidate disulfidptosis-associated hub genes. The differentiated values of DRGs were determined by receiver operator characteristic (ROC) monofactor analysis to judge their potential quality as biomarkers. RT-qPCR were used to validate the expression of hub genes. Additionally, we analyzed the connection between the hub genes and the filtration of immune cells in RA. We made predictions about the miRNAs, TFs and possible drugs that regulate the hub genes. Subsequently, molecular docking was carried out to predict the combination of drugs with hub targets. Finally, molecular dynamics simulation was conducted to further verify the findings. Oxoacyl-ACP Synthase Mitochondrial(OXSM) was identified as a biomarker with high diagnostic value, and an RA diagnostic model based on OXSM for a single gene was constructed. The model showed high accuracy in distinguishing RA and healthy controls (AUC = 0.802) and was validated by external datasets, showing excellent diagnostic power (AUC = 0.982). Twelve potential drugs against RA were recognized by comparative toxicogenomics database (CTD). Molecular docking results showed that ICG 001 had the highest binding affinity to OXSM, and molecular dynamics simulations confirmed the stability of this complexes. Furthermore, CIBERSORT analysis showed a significant correlation between immune cell infiltration and OXSM, and a regulatory network of TFs-gene-miRNAs comprising 8 miRNAs and 34 TFs was identified. Finally, the RT-qPCR results showed that OXSM was significantly increased in the peripheral blood of RA patients compared with healthy controls, consistent with the bioinformatics analysis. These studies suggest that OXSM may be a potential biomarker and therapeutic target for diagnosing RA, and ICG 001 may be a potential drug for RA. These findings may provide new avenues for the effective diagnosis and treatment of RA.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-93656-4.

Introduction

Rheumatoid arthritis (RA) is a pathologically complexes autoimmune disease featured by inflammation and joint degeneration, often accompanied by persistent joint pain and gradual destruction of articular cartilage and bone, ultimately leading to severe dysfunction^1–4. However, the pathogenesis of RA still needs to be fully understood. Disulfidptosis is a recently discovered new mode of cell death by the excessive accumulation of disulfides. It has been studied in cancer and other neurological diseases but has not been well described in rheumatoid arthritis. It has been shown that thiol-induced deterioration of the antioxidant defense system may lead to the pro-oxidant/antioxidant imbalance in RA⁵, and chronic inflammation and oxidative stress are the main mechanisms in the pathophysiology of RA⁶. Some studies have shown that oxidative stress is linked to the progression of rheumatoid arthritis (RA) and its associated pathologies. When the function of the antioxidant defense system is impaired, or free radicals are overproduced, the body will enter an oxidative stress mode, thus inducing the apoptosis of chondrocytes^7–10, Excessive apoptosis of chondrocytes and altered cartilage degeneration are some of the leading causes of RA¹¹. Excessive reactive oxygen species (ROS) play a vital role in the pathogenesis of RA¹². In the physiological state, a moderate amount of ROS can promote immunity and repair and resist microbial infection. When ROS is excessive, it will induce severe oxidative stress and directly attack biological macromolecules such as proteins, nucleic acids, and lipids on the cell membrane and eventually lead to cell apoptosis, necrosis, and so on⁹. A large body of evidence shows that ROS leads to pathological changes in RA pathogenesis by affecting the damage of articular cartilage and chondrocytes¹², while CD4 + T cells, B cells, dendritic cells, mast cells and macrophages are also involved in RA pathogenesis¹³, where neutrophils are considered to be the primary effector cells¹⁴. Neutrophils aggravate disease severity by excessive activation and release of reactive oxygen species ROS¹⁵. Moreover, increased cytokines such as TNF-α, IL-1, and IL-6 in patients with RA stimulate substantial production of ROS, leading to joint tissue destruction, which subsequently aggravates the disease^6,16. In the inflamed RA joint cavity, the infiltration of monocytes and macrophages will continuously release high levels of ROS. The persistence of oxidative stress will induce apoptosis and activate immune cells, triggering the release of proinflammatory cytokines such as Th-17, exacerbating joint inflammation¹⁷. When insufficient or excessive depletion of NADPH is produced, intracellular redox homeostasis is disrupted¹⁸, leading to excessive intracellular accumulation of cystine, inducing disulfide bond stress in the actin cytoskeleton, and subsequent actin network collapse¹⁹. The actin cytoskeleton partly controls the phenotypic stability of chondrocytes, and chondrocytes maintain tissue homeostasis through a balance between synthesis and catabolism²⁰. The collapse of the actin network and dysregulation of cartilage phenotypic stability lead to the destruction of chondrocytes and cartilage matrix, which then leads to irreversible chondrocyte damage and eventually induces RA^11,21. Some studies have found that including decreased cytoskeletal proteins, decreased thiols, decreased glucose concentration, disulfide accumulation, osteoclast differentiation, and inflammatory cell infiltration in cartilage, synovium, and other tissues of RA patients^22–28. Staron et al. noted that the concentration of thiol groups in the red blood cell membranes of RA patients was significantly reduced, possibly related to the aggregation of membrane proteins and an increase in disulfides^29,30. Analysis of some studies showed a significant increase in cystine in the plasma of RA patients and a positive correlation between the severity of RA and increased cystine levels in the blood^6,25,26. Excessive accumulation of intracellular cystine leads to disulfide bond stress and disulfidptosis¹⁹. It is, therefore, reasonable to assume that disulfidptosis may significantly affect the progression of RA.

Disulfidptosis is a newly discovered mode different from conventional programmed cell death. In the absence of glucose, a novel form of cell death resulting from disulfide stress caused by excessive cystine accumulation is called disulfidptosis¹⁹. Excessive accumulation of disulfide induces elevated levels of disulfide bonds between F-actin, leading to the disintegration of the actin network and, ultimately, cell death^19,31,32. Induction or inhibition of specific cell death pathways has been used to treat various diseases, including autoimmune diseases^33,34. The discovery of the disulfidptosis mechanism provides a new strategy for treating these diseases^19,31,32.

To better understand the regulatory mechanism of disulfide ptosis genes in RA, this study integrated bioinformatics, molecular docking, molecular dynamics simulation, and experimental verification to explore the role of disulfide ptosis genes in RA and provide new targets and potential drugs for its treatment. The workflow chart is shown in Fig. 1.

Fig. 1.

Workflow diagram.

Materials and methods

Downloading and processing of the microarray data

Our study used the following screening criteria³⁵: (1) all samples were collected from RA patients and healthy controls; (2) each group had a large sample size of more than 20 participants. In GSE93272, 275 whole blood samples were collected, comprising 232 from RA patients and 43 from healthy controls. GSE45291 included 493 RA samples and 20 healthy controls; (3) Test samples were sourced from human subjects; (4) the tissue used for sequencing was either whole blood (WB) or peripheral blood (PB). The samples in datasets GSE93272 and GSE45291 were obtained from blood samples of RA patients, and the consistent specimen source ensured the robustness of the subsequent analysis results. The expression profiles (GSE93272, GSE45291) related to RA were downloaded from the Gene Expression Omnibus (GEO)(https://www.ncbi.nlm.nih.gov/geo/)³⁶ and summarized in Table 1Dataset GSE93272³⁷was used to screen hub genes, and dataset GSE45291³⁸ was used to verify the diagnostic relevance of hub genes.

Table 1.

Descriptive statistics.

Data number	Platform information	RA group	Control group	Species
GSE93272	GPL570	232	43	Homo sapiens
GSE45291	GPL13158	493	20	Homo sapiens

Differential expression analysis

The R package limma³⁹(version 3.40.6)was used to identify differentially expressed genes (DEGs) between RA and healthy controls. The analysis included log2 transformation of the data, multiple linear regression using the lmFit function, and calculation of regulatory t-statistics, regulatory f-statistics, and log probability of differential expression using an empirical Bayes conditioning approach. Genes significantly different between RA and control groups were set with settings | logFC |> 1.2 and p< 0.05^40,41.

Differential gene enrichment analysis

The GO and KEGG⁴² annotations of genes were extracted using the R package org.Hs.eg.db (version 3.1.0) and the KEGG REST API (https://www.kegg.jp/kegg/rest/keggapi.html), respectively. These annotations were used as the background, and the genes were mapped to the background set. GO and KEGG enrichment analyses were then performed using the R package clusterProfiler (version 3.14.3) to obtain the results of gene set enrichment^39,43. The significance threshold was an adjusted P-value < 0.05⁴⁴.

Weighted gene co-expression network analysis

A scale-free co-expression network was constructed using the R package Weighted gene co-expression network analysis (WGCNA)⁴⁵. The sensitivity was set to 3, and modules with a distance of less than 0.25 were merged to yield 16 modules with different colors. The correlation between each module and RA was then analyzed, and the most relevant module was selected for further analysis. Gene significance (GS) and module membership (MM) were calculated for each gene in the hub module, the threshold was set to MM > 0.8, and GS > 0.2, and potential RA-related genes were selected.

Disulfidptosis-related genes extraction

DRGs were retrieved from the FerrDb database’s extended data⁴⁶, and duplicates were removed to extract effective targets.

Feature gene identification and GSEA analysis

The DRGs, DEGs, and WGCNA module genes were fed into Venny 2.1.0(http://bioinformatics.psb.ugent.be/webtools/Venn/) to identify overlapping targets. Two feature genes, OXSM and Actinin Alpha 4 (ACTN4), were obtained, and their expression levels were analyzed. Finally, the eigengenes’ function and potential biological roles were further analyzed by Gene Set Enrichment Analysis(GSEA).

Hub gene identification, diagnostic model development and validation

To screen hub genes, we performed multivariate cox regression analysis⁴⁷of the two characteristic genes. Further, the“pROC” R package was used to perform a receiver operating characteristic (ROC) monofactor analysis⁴⁸on the training set GSE93272 and the validation set GSE45291, respectively, to judge the diagnostic value of the disulfidptosis-related biomarkers in RA patients. Normally, the closer the ROC curve is to the top left corner, the better the diagnostic performance⁴⁹, while the AUC (Area Under Curve) values range from 0.5 to 1, with values greater than 0.8 indicating good predictive accuracy and reliability in practice⁵⁰. The filtering criteria of the hub genes were set at AUC > 0.800⁵¹.

Immune infiltration analysis

CIBERSORT analyzed the relative proportions of the 22 immune cells in each sample^52,53, and bar plots showed the proportions of each immune cell type in the different samples. Furthermore, the correlation between the 22 infiltrating immune cells was analyzed using the Corrplot R package⁵⁴. Boxplots compare the proportion of immune cells between RA and healthy controls and further analyze the correlation between hub genes and 22 immune cell types.

Construction of TFs-Gene-miRNAs network

Interactions between the TFs, miRNAs, and the hub genes were analyzed using the NetworkAnalyst3.0⁵⁵ and visualized with Cytoscape. Finally, we analyzed the correlation between TFs and OXSM in the GSE93272 dataset.

Potential therapeutic drug prediction

CTD is a publicly available database including 17,100 chemicals, 54,300 genes, 6100 phenotypes, 7270 diseases, and 202,000 exposure statements⁵⁶. Potential drugs for OXSM prediction using CTD and demonstrating drugs - gene- diseases interactions via Sankey plots.

Molecular docking

The SDF format for all drugs were downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). PDB files for core proteins were obtained from the RCSB Protein Database (http://www.rcsb.org/), molecular docking by the online docking tool CB-Dock2⁵⁷, and visualized using PyMOL.

Molecular dynamics simulations

The Gromacs2023.2 program⁵⁸performed molecular dynamics simulations^59,60of 100ns^61–63for small molecule-protein complexes. Calculating the binding free energy between ligand and receptor by MM/GBSA^64–67 provides detailed insights into the binding mode between small molecules and target proteins.

Quantitative realtime PCR

In this study, the peripheral blood of three healthy controls and five RA patients was used(our study is conducted in strict accordance with the Declaration of Helsinki.), and the total RNA was extracted using the column whole blood total RNA extraction and purification kit (Sangon Biotech, Shanghai, China). Reverse mRNA transcription was performed using the PrimeScript RT kit (TaKaRa, Dalian, China). RT-qPCR experiments were performed according to the instructions of SYBR Premix Ex Taq Kit (TaKaRa, Dalian, China) and performed on LightCycler 480 (Roche). GAPDH was used as the reference gene. Relative expression of genes was calculated via the 2^{−ΔΔ Ct} method. Primers for RT-qPCR are listed in Table 2:

Table 2.

PCR primer sequence.

Number	Gene name	Primer sequence (5′ to 3′)	Species
1	OXSM-F	TGATGCTGGTCACATAACTGC	Homo sapiens
2	OXSM-R	TCCCAATGGTGTGGAAGTAGC	Homo sapiens
3	GAPDH-F	GGAGCGAGATCCCTCCAAAAT	Homo sapiens
4	GAPDH-R	GGCTGTTGTCATACTTCTCATGG	Homo sapiens

Statistical analysis

Statistical analysis was conducted in SPSS Statistics 20. Our data met a normal distribution. Therefore, they were analyzed using an unpaired t-test. P-value < 0.05 was considered significant.

Results

Differential genes identification

Limma is a differential expression screening method based on generalized linear models, and we used the R package limma (version 3.40.6) from the GEO data set (GSE93272), setting |logFC |>1.2 and p < 0.05, identifying a total of 1608 DEGs including 1329 up-regulated and 279 down-regulated genes (Supplementary Table 1). Volcano plots of DEGs are shown in Fig. 2A. Figure 2B shows the heat map of the first 20 DEGs.

Fig. 2.

(A) Volcano plot of DEGs (|log2FC| > 1.2 and p < 0.05 ). (B) Heatmap of top 20 DEGs.

Differential genes functional enrichment analysis

GO and KEGG enrichment analysis of the 1,608 DEGs were performed using the R software package clusterProfiler (version 3.14.3). Set P-value < 0.05. The bubble plot was then used to show the top 10 DEGs GO categories (Fig. 3A-C). The top 10 enrichment terms in KEGG(Fig. 3D). BP enrichment in the cell activation, immune system process, immune system development, regulation of body fluid levels, immune system process, cellular response to stress, positive regulation of immune system process, actin cytoskeleton organization, lymphocyte activation, regulation of T cell receptor signaling pathway(Fig. 3A). CC was mainly enriched in the cell-substrate adherens junction, nucleoplasm part, nuclear part, nuclear lumen, actin filament, cytoskeleton, immunological synapse, actin cytoskeleton, actomyosin, interleukin-6 receptor complex (Fig. 3B). MF was enriched in the cytokine binding, cytokine receptor activity, signaling receptor binding, G protein-coupled chemoattractant receptor activity, C-C chemokine receptor activity, ATPase activity coupled, cytoskeletal protein binding, phosphatase binding, interleukin-17 receptor activity, CD8 receptor binding(Fig. 3C). Moreover, KEGG analysis revealed that DEGs are mainly involved in the Regulation of actin cytoskeleton, Adherens junction, Viral protein interaction with cytokine and cytokine receptor, Transcriptional misregulation in cancer, Th17 cell differentiation, Th1 and Th2 cell differentiation, Pathways in cancer, Cytokine-cytokine receptor interaction, Leukocyte transendothelial migration, ECM-receptor interaction(Fig. 3D).

Fig. 3.

GO and KEGG enrichment analysis of DEGs. (A) BP term analysis. (B) CC term analysis. (C) MF term analysis. (D) KEGG term analysis.

Weighted gene co-expression network analysis

Scale-free co-expression networks were constructed using the sample gene method to remove outliers and samples in the R package WGCNA. Through the Sangerbox platform, the power of β = 12 (scale-free R²= 0.85) was automatically set as the soft threshold to produce a scale-free network, ensuring the robustness of the network construction^68–70(Fig. 4A, B). A module merge threshold of 0.25, a sensitivity of 3, and a minimum module size of 30 yielded 16 different co-expression modules (Fig. 4E). Then, the correlation with clinical characteristics (Fig. 4C, D), module membership (MM) and gene significance (GS), and greenyellow module (r = 0.51,p = 2.8e⁻⁵⁹; Fig. 4F) and green module (r = 0.35,p = 1.2e⁻⁸¹; Fig. 4G) were selected for further analysis.

Fig. 4.

Results for WGCNA. (A) Scale-wise independence of the expression matrix. (B) Average connectivity values of the expression matrix. (C) Cluster dendrogram of the genes. (D) Correlation between the different modules and the clinical characteristics. Red indicates a positive correlation, and green indicates a negative correlation. (E) Module vector clustering heatmap. (F) Correlation between the module membership degree and the gene significance in the green-yellow module. (G) Correlation between the module membership degree and gene significance in the green.

Feature genes identification and GSEA analysis

1608 DEGs, 934 module genes(Supplementary Table 2), and 84 DRGs were crossed to obtain two feature genes, OXSM and ACTN4 (Fig. 5A). We next validated the expression levels of the two feature genes in GSE93272. Violin plot results show that the OXSM gene is highly expressed in RA (Fig. 5B). In contrast, ACTN4 is lowly expressed in RA (Fig. 5C). To understand further the role of feature genes in the pathogenesis of RA, we performed single-gene GSEA analysis. The results showed that OXSM was in the Spliceosome, Peroxisome, Oxidative phosphorylation, and Proteasome, Regulation of the actin cytoskeleton, and Medium enrichment (Fig. 5D). ACTN4 was mainly enriched in the Proteasome, Peroxisome, Regulation of the actin cytoskeleton, Oxidative phosphorylation, and Chemokine signaling pathway (Fig. 5E).

Fig. 5.

Feature genes identification and GSEA analysis. (A) Venn diagram of LIMMA, WGCNA, and DRGs. (B) OXSM gene expression levels in the GSE93272 experiment set. (C) Expression levels of the ACTN 4 genes in the GSE93272 experiment set. (D) GSEA analysis of the OXSM gene. (E) A GSEA analysis of the ACTN4 gene.

Hub genes identification and diagnostic model construction

We identified hub genes by multivariate Cox regression analysis and receiver operating characteristic (ROC) curves. Using the R package survival, the prognostic value of survival time and two feature genes was evaluated by integrating the survival time data and survival status in 275 samples. As shown in Fig. 6A, OXSM had a significant prognosis compared to ACTN4 (p = 0.02). The feature genes were also evaluated for diagnostic value, and the accuracy of the prediction power of the two feature genes was assessed by calculating the Area Under Curve (AUC)value. As shown in Fig. 6B, C, the AUC values of the two feature genes were 0.802 and 0.778, and the hub genes were screened using AUC > 0.800. In summary, OXSM was selected as the hub gene for further validation in this study.

Fig. 6.

Multivariate Cox regression analysis and diagnostic value of the feature genes in training set GSE93272. (A) Multivariate Cox results for two feature genes. (B) ROC curve of the diagnostic model of OXSM in the GSE93272 training set. (C) ROC curve of the diagnostic model of ACTN4 in the GSE93272 training set.

Diagnostic model validation

To verify the expression levels of hub genes and the precision of the diagnostic model, we used GSE45291 external independent datasets for validation. The boxplot results showed that the OXSM genes were highly expressed in RA patients (Fig. 7A), consistent with the results in the training set. The risk score was calculated using coefficient values obtained from multivariate Cox regression analysis through the Sangerbox online platform, and the OXSM gene diagnostic model was validated in the validation set GSE45291. As shown in Fig. 7B, the ROC curve showed an AUC value of 0.982 (95%CI 0.970–0.993), which was higher than the training set, indicating that the model has excellent diagnostic efficiency (Fig. 6B). Notably, GSE45291 is a blood sample dataset⁷¹, showing excellent test results, highlighting the advantage of OXSM as a novel potential biomarker for the diagnosis of RA.

Fig. 7.

The expression patterns and diagnostic performance validation of OXSM in the GSE45291 dataset. (A) The expression pattern of OXSM in the GSE45291 dataset. (B) Diagnostic efficacy of OXSM in the GSE45291 dataset.

Immune infiltration analysis

The immune microenvironment is essential for clinical treatment sensitivity and disease diagnosis⁷². This study used CIBERSORT to calculate the proportion of 22 immune-infiltrating cells in RA and healthy controls (Fig. 8A, B). The results showed that eight immune cells differ between RA and healthy controls (Fig. 8C). Specifically, Gammadelta T cells and M2 macrophages were highly infiltrated in the RA group, while CD8 + T cells, Regulatory T Cells (Tregs), and Naive CD4 + T cells were significantly reduced. Further, we explored whether OXSM is associated with immune cell infiltration by using the Spearman analysis (Fig. 8D). Correlation analysis showed that OXSM showed a significant positive correlation with Gammadelta T cells, macrophage M2 and a significant negative correlation with Regulatory T Cells (Tregs) and Neutrophils. These lines of evidence suggest that changes in the immune microenvironment in RA patients may be related to OXSM. They also highlight the critical role of OXSM in the immune microenvironment of RA.

Fig. 8.

Analysis of immune cell infiltration between RA and healthy control. (A) Relative abundance of the 22 infiltrating immune cells between the RA and healthy control groups. (B) Heatmap of the correlation between 22 immune cells. (C) Boxplots showing the difference in immune-infiltrating cells between the RA and healthy control. *p < 0.05, **p < 0.01, ***p < 0.001. (D) Correlation between the OXSM and immune-infiltrating cells.

Construction TFs-Gene-miRNAs network

The Transcription factors(TFs) and miRNAs for OXSM were predicted using the NetworkAnalyst online tool. Finally, 34 TFs and 8 miRNAs for OXSM were obtained (Supplementary Tables 3, 4). A TFs-gene-miRNAs network(Fig. 9A)containing 43 nodes and 42 edges were constructed using cytoscape, where hsa-mir-7-5p has been linked to the pathological progression of RA⁷³. Further, we analyzed the correlation of OXSM with TFs in dataset GSE93272 using spearman. As shown in Fig. 9B, OXSM positively correlated with embryonic ectoderm development(EED)⁷⁴, TATA-box binding protein associated factor 7(TAF7) and zinc finger protein 639 (ZNF639) and negatively with non-centrosomal microtubule array protein 1(NOCA1) and cut like homeobox 1(CUX1), suggesting that the above TFs may play a role in the regulation of OXSM expression. We analyzed the TFs-gene-miRNAs interaction network and predicted five significant TFs and eight miRNAs as regulators of OXSM expression.

Fig. 9.

The TFs-gene-miRNAs network construction. (A) The TFs-gene-miRNAs network based on the hub gene.The purple balls represent hub gene, green triangles represent miRNAs and blue diamonds represent TFs. (B) Heatmap of the correlation between OXSM and TFs in GSE93272.

Potential therapeutic drug prediction

Twelve predicted drugs against OXSM were obtained through CTD (Supplementary Table 5). Through molecular docking, we selected the top 10 drugs with the lowest binding energy, and the relationship between targets, drugs, and diseases was revealed through the sankey diagram (Fig. 10A). Among the predicted drugs, doxorubicin⁷⁵, sunitinib⁷⁶, and schizandra B⁷⁷ have been documented or studied as targeted drugs for RA, demonstrating the high confidence in drug candidates for RA.

Fig. 10.

Prediction of hub gene-targeted drugs. (A) The Sankey plot demonstrates the interaction relationship between targets-drugs-disease.

Molecular docking validation

Molecular docking is used to assess the binding capacity of the ligand and the receptor, and the smaller the binding energy, the tighter the ligand binds to the protein. A binding energy of less than − 4.25 kcal/mol indicates a specific binding capacity, less than − 5.0 kcal/mol indicates a good binding activity, and less than − 7.0 kcal/mol indicates a solid binding activity⁷⁸. OXSM core proteins were docked to the corresponding 12 predicted drugs using the CB-Dock2 online tool, and all docking binding energy was − 6.0 kcal/mol (Supplementary Table 6). The top three were ICG 001 (-10.1 kcal/mol), sunitinib (-8.7 kcal/mol), and doxorubicin (-8.4 kcal/mol), indicating binding solid activity between drug and target protein, especially OXSM and ICG 001. As shown in Fig. 11A, ICG 001 can form hydrogen bonds with ARG252 (length = 2.2 Å) THR315 (length = 3.1Å), which greatly increases the stability of the complex. To compare the stability of ICG 001 and known rheumatoid drug binding to the target, six FDA-approved drugs of Auranofin, Ibuprofen, Leflunomide, MethylprednisoloneTablets, Nimesulide, and Methotrexate for rheumatoid RA were selected with OXSM for molecular docking analysis. The binding energy of these drugs with OXSM is -6.1 kcal/mol, -6.7 kcal/mol, -8.0 kcal/mol, -8.5 kcal/mol, -8.5 kcal/mol, -9.3 kcal/mol, respectively (Supplementary 7). Finally, the docking results were visualized using pymol software (Fig. 11A-C).

Fig. 11.

Diagram of the top three drug binding modes with the lowest molecular docking binding energy to the OXSM protein (PDB id: 2IWZ). (A) ICG 001 with OXSM ((− 10.1 kcal/mol). (B) Sunitinib with OXSM (− 8.7 kcal/mol). (C) Doxorubicin with OXSM (− 8.4 kcal/mol).

Molecular dynamics simulations validation

Affinity values indicate the stability of the bond between the receptor and the ligand, with lower affinity indicating that the bond is more stable⁷⁹. Considering that OXSM has the lowest binding energy to ICG 001, we chose the OXSM-ICG 001 complexes for molecular dynamics simulations. These simulations help to analyze the dynamic interactions within the protein-ligand complexes and to validate the molecular docking results⁸⁰. Root mean square deviation (RMSD) is used to observe the overall protein conformational changes of the system relative to the initial structure during the simulation. The larger values indicate a more significant structural change of the protein in the system. As shown in Fig. 12A, the system protein and small molecule RMSD have some fluctuations in the early kinetic period (0-10ns), and at 10-100ns, the RMSD remains around 0.18 nm and 0.40 nm, respectively, indicating the stability of the interaction between ICG 001 and OXSM. Root Mean Square Fluctuation (RMSF) reflects the fluctuation of the site structure of amino acid residues during the simulation, and the larger the value is, the more flexible the residue at this position is. As shown in Fig. 12B, the RMSF of all residues is less than 0.20 nm, with slight overall fluctuation and relatively stable.

Fig. 12.

(A) RMSD of the OXSM-ICG 001 complexes under a 100 ns MD simulation. (B) RMSF of the OXSM-ICG 001 complexes under a 100 ns MD simulation.

The Radius of gyration (Rg) is used to evaluate the tightness of the architecture. The smaller values indicate a tighter protein. It can be found from Fig. 13A that the protein system is within 0-100ns, with a little tight degree, and stable at about 2.60 nm, which is relatively balanced overall. Solvent-accessible surface area (SASA) is the area of the protein surface accessible by the solvent, which measures the area of the protein surface in contact with the solvent. According to the atomic charge range from − 0.20 to 0.20 as the hydrophobic region and the other area as the hydrophilic region, the area of the two parts is counted. As shown in Fig. 13B, the solvent-accessible surface area of the whole system is relatively stable.

Fig. 13.

The radius of gyration (A) and solvent-accessible surface area (B) of the OXSM-ICG 001 complexes during the 100ns time frame.

Further, we also analyzed the hydrogen bonds formed between proteins and ligands, as shown in Fig. 14A, a 100ns molecular dynamics simulation of the protein-ligand hydrogen bond interaction, and the number of hydrogen bonds formed between 1 and 3 that contributed to the binding. As shown in Fig. 14B, after 100ns of simulation, the ligand remained tightly bound in the protein cavity.

Fig. 14.

(A) The number of hydrogen bonds in complexes between ICG 001 and OXSM. (B) Details of the 3D interaction between ICG 001 and OXSM, yellow stick tick for ICG 001 in the figure and cyan stick tick for amino acids acting with ICG 001. Blue solid lines indicate the hydrogen-bonding interactions.

The binding free energy of protein and small molecules changes with time, and the lower the binding free energy, the stronger the binding force⁸¹. As shown in Fig. 15A, the binding free energy contribution of small molecules and protein mainly comes from ARG-252, LEU-354, ASP-272, and ALA-253 of protein A chain. The binding free energy − 41.23 kcal/mol (Fig. 15B) was calculated by MM/GBSA, and the lower the binding energy, the stronger the binding capacity. Free energy landscape (FEL) characterizes the change in the free energy of matter experienced during the simulation. Calculating the free energy topography of the protein and ligand complexes can guide the characteristic conformations of the extracted complexes. Meanwhile, the protein’s free energy topography shows the protein conformation’s stability. Figure 15C shows the free energy topography of the protein, which dominates the complexes conformation ranging from 0.10 to 0.14 nm and RMSD from 2.525 to 2.540 nm.

Fig. 15.

(A) Residues and energetic analysis that contribute substantially to the ICG 001 binding energy. (B) The contribution of various interactions to the MM/GBSA score of ligands. (C) The free-energy landscape map of the ICG 001-OXSM complexes.

Ramachandran plot is a visualization method developed by G. N. Ramachandran et al. in 1963 in the protein structure⁸². It can also reflect whether the conformation of the protein is reasonable. Ramachandran Plot displays the structural features of the protein backbone by drawing the dihedral angles (φ and ψ angles). The scatter concentration region contains reasonable regions of dihedral angles. These regions represent stable conformations in the protein structure. Generally, reasonable dihedral angle regions include the α helical region, β folding region disordered region, etc. Regions where scatter does not exist are the abnormal or forbidden dihedral angles. These regions represent conformations with little or no appearance in the standard protein structure.

Figure 16A, B is the standard Ramachandran Plot, Fig. 16C is the overall diagram of complexes proteins, Fig. 16D is the glycine, Fig. 16E is the dihedral angle distribution of the previous residue of proline, and Fig. 16F is the dihedral angle distribution of proline. The unique structural properties of glycine and proline affect the distribution of their dihedral angles (φ and ψ), thus affecting the folding and function of the entire protein. Compared with the standard Ramachandran Plot (Fig. 16A, B), most scattered points are distributed in reasonable areas, and only a few parts are scattered in other regions. Overall, the OXSM proteins bound to ICG 001 showed a spatially acceptable molecular conformation after the MD simulation, indicating the stability of the complexes.

Fig. 16.

(A) and (B) is a standard Ramachandran plot. (C) complexes Protein Overall Ramachandran plot. (D) Ramachandran plot of glycine. (E) Ramachandran plot of the dihedral angle distribution of one residue preceding the proline. (F) Ramachandran plot of the dihedral angle distribution of proline.

RT-qPCR validation hub gene expression

RT-qPCR verified the expression level of the hub genes and showed significantly higher OXSM expression in peripheral of RA compared to healthy controls (Fig. 17A).

Fig. 17.

RT-qPCR results of OXSM expression in rheumatoid arthritis patients and healthy controls.** P < 0.01.

Discussion

Rheumatoid arthritis, an autoimmune disease with multiple etiology, has been extensively studied and has made some progress^83,84. However, due to the lack of sufficient specific biomarkers and the heterogeneity of RA pathogenesis, the early diagnosis and treatment of RA still face challenges^85–87. Therefore, searching for newer and more effective diagnostic markers and therapeutic targets is crucial for the early diagnosis and individualized treatment of RA⁵¹. Disulfidptosis is a newly discovered form of cell death due to the accumulation of disulfide, ultimately leading to cell death¹⁹. Studies have shown that disulfidptosis-related genes can regulate the secretion of pro-inflammatory cytokines, leading to the development of autoimmune diseases such as psoriatic arthropathy (PsA), ankylosing spondylitis (AS), juvenile idiopathic arthritis (JIA) and rheumatoid arthritis (RA)^34,88. It was shown that disulfide stress-mediated disulfidptosis is involved in the immature differentiation of upper basement cells in psoriasis⁸⁹. Although disulfidptosis has been studied in several inflammatory diseases⁹⁰, the specific mechanism, regulatory role and potential targets in RA have not been elaborated.

Our study aimed to identify and validate novel biomarkers associated with disulfidptosis in RA and to reveal the underlying mechanisms. We downloaded the GSE93272 dataset from the GEO database as a training set and identified 1329 upregulated DEG and 279 downregulated DEG by extracting gene expression profiles. According to GO functional enrichment analysis, RA-related DEG in BP was mainly enriched in the immune system process, cellular response to stress, cellular response to stress, and T cell receptor signaling pathway regulation. CC was enriched in the nuclear part, actin cytoskeleton, immunological synapse, and interleukin-6 receptor complex. MF was enriched in cytokine receptor activity, cytoskeletal protein binding, interleukin-17 receptor activity, and CD8 receptor binding. KEGG analysis showed that DEGs are enriched in the Regulation of actin cytoskeleton, Viral protein interaction with cytokine and cytokine receptor, Th17 cell differentiation, Th1 and Th2 cell differentiation. These functional enrichment analyses indicate that DEG is enriched in immune response, inflammation, and actin cytoskeleton and plays an important role in the pathogenesis of RA^19,23,24,91.

It has been found that cytoskeletal protein levels are reduced in skeletal muscle of RA patients^23,24,92, suggesting that the deterioration of RA is associated with actin destruction. The actin cytoskeleton plays an essential role in maintaining chondrocyte homeostasis and survival. The collapse of the actin cytoskeleton and the loss of chondrocyte homeostasis lead to the destruction of the cartilage matrix, leading to irreversible chondrocyte damage and ultimately inducing joint inflammation^21,23,24. Studies have shown that synovial fluid and plasma levels of thiols are reduced, and disulfide levels are significantly elevated in RA, thereby triggering oxidative stress, leading to joint damage, and ultimately inducing RA^23,25–27. However, during disulfidptosis, the intake of abundant cystine may increase the intracellular concentrations of glutathione and NADPH, thereby reducing reactive oxygen species (ROS) levels. Thus, reducing ROS levels plays a crucial role in the anti-inflammatory pathway and is critical for alleviating and inhibiting RA progression^23,24,85,92. In this study, we linked disulfidptosis to the pathogenesis of RA, integrated bioinformatics, molecular docking, molecular dynamics simulation, and experimental verification to find the potential essential genes of disulfidptosis in RA, explore the potential therapeutic targets and drugs, and provide new targets and strategies for the future treatment of RA.

We identified two feature genes associated with disulfidptosis from the GSE93272 dataset OXSM and ACTN4. GSEA enrichment showed significant enrichment of feature genes in the peroxisome, oxidative phosphorylation, proteasome, and regulation of the actin cytoskeleton. In order to find markers with higher diagnostic value in clinical diagnosis and treatment, we identified hub genes by multivariate cox regression analysis and receiver operating characteristic curve (ROC) analysis. The results of multivariate survival analysis showed that OXSM had high prognostic significance (p = 0.02), and the AUC value corresponding to ROC was 0.802(95%CI:0.738–0.866), 0.778(95%CI:0.710–0.846). The AUC > 0.800 was used as the screening criterion, and finally, OXSM was selected as the hub gene. Further, we validated the expression level of OXSM and the model’s accuracy in Dataset GSE45291. The results showed that OXSM was significantly upregulated in the peripheral blood of RA patients with an AUC of 0.982(95%CI:0.970–0.993). Therefore, OXSM performed best as a diagnostic marker.

Oxoacyl-ACP Synthase Mitochondrial (OXSM), also known as FASN2D, is a second essential synthase encoded by nuclear DNA involved in the mitochondrial fatty acid synthesis and extension pathway^93–95. OXSM, as a disulfidptosis-related gene, plays a vital role in the metabolic pathways of tumor cells and mitochondrial dysfunction diseases^96,97. The finding of a positive correlation between OXSM levels and plasma cell and T cell levels in bladder cancer samples suggests its possible involvement in regulating immune response and inflammation⁸⁴. In metastatic oral squamous cell carcinoma (OSCC), the core binding factor subunit β (CBFB) regulates OXSM expression through cis-specific enhancer-binding of OXSM, controlling the proliferation, invasion, and lipid synthesis of metastatic OSCC cells⁹⁵. In colorectal cancer, CBFB deficiency enhances cellular resistance to MEK, an inhibitor of the targeted mitogen-activated protein kinase (mAPK) pathway^98,99. Some studies have reported that OXSM can inhibit the growth of ovarian cancer cell lines, disrupt the stability of OXSM by knocking down heat shock protein D family member 1 (HSPD1), and promote the proliferation and migration of ovarian cancer cells¹⁰⁰Some studies have found that hsa-miR-338-3p participates in fatty acid biosynthesis and metabolism by regulating OXSM levels and is also implicated in the occurrence and proliferation of glioma cells¹⁰¹.

Studies have shown that mitochondrial dysfunction affects the development of inflammatory diseases by modulating the balance of antigen presentation and immune cells¹⁰². In RA, mitochondrial dysfunction promotes the occurrence and progression of RA through abnormal energy metabolism, overproduction of reactive oxygen species (ROS), and the activation of innate immunity⁹⁶. Furthermore, mitochondrial dysfunction in RA also affects various immune cells, including macrophages, CD4 + T cells, CD8 + T cells, and neutrophils¹⁰³. Disulfidptosis is a phenomenon caused by oxidative stress. Related disulfidptosis genes affect the recruitment and activation of immune cells by regulating the redox state of cells, the release of chemokines, and generating reactive oxygen species (ROS). At the same time, immune cells provide pro-inflammatory or anti-inflammatory signals by changing the immune microenvironment and thus influence the expression of disulfidptosis genes and their function⁹⁹.

Furthermore, diabetes and its complications are a disease of mitochondrial dysfunction; the reduction of OXSM increases the risk of nephropathy and atherosclerosis in diabetic patients, and fatty acid deficiency is a risk factor for the onset of diabetic complications⁹⁶. The latest study also found that OXSM plays an essential role in regulating OA’s inflammatory response and immune infiltration⁸⁵.

Peripheral immune cells are also involved in the pathogenesis of RA⁵¹. Compared to healthy controlsThe RA group was highly infiltrated with Gammadelta T cells and M2 macrophages, while CD8 + T cells, Regulatory T Cells, and Naive CD4 + T cells were significantly reduced. Correlation analysis showed that OXSM showed a significant positive correlation with gammadelta T cells, macrophage M2 and a significant negative correlation with regulatory T Cells and neutrophils. Studies suggest that gammadelta T cells and M2 macrophages play essential roles in RA’s pathogenesis and pathological progression^104,105. Gamma delta T cells play a unique and important role in autoimmune diseases such as juvenile idiopathic arthritis (JIA), systemic lupus erythematosus (SLE), ankylosing spondylitis (AS), systemic sclerosis (SSc), and rheumatoid arthritis(RA)^106–108. Studies have found that Gamma delta T cells participate in and activate the inflammatory process in the RA synovium¹⁰⁶. The increase of gamma delta T cells can promote the production of IL-17 inflammatory factors, thus accelerating the body’s inflammatory response¹⁰⁹. Regulatory T cells (Treg cells) are vital in preserving immune tolerance in peripheral tissues and suppressing autoimmunity and are associated with disease severity^110–112. Research shows that the imbalance between Treg cells and Th17 plays a crucial role in the occurrence and progression of RA¹¹⁰. Studies have found that the number of Treg cells in the peripheral blood of RA patients was reduced^112,113. Congenital immune cells such as Treg cells can mediate autoimmune responses when the number and/or function of these cells may lead to overproduction of pro-inflammatory cytokines such as IL-1, IL-2, IL-, IL-6, IL-8, and IL-17, accelerate the destruction of articular cartilage, promote synovitis and eventually lead to RA^110,114,115.

Transcription factors (TFs) and miRNA regulate gene expression^116,117. In GSE93272, we analyzed the correlation of OXSM with TFs and found that EED, TAF7, and ZNF639 were positively correlated with OXSM, indicating that EED, TAF7, and ZNF639 may positively regulate the transcription of OXSM. We also predicted eight miRNAs that interact with OXSM, and some studies found that the altered expression level of hsa-mir-7-5p may be involved in the pathogenesis of RA⁷³. The hsa-mir-203a-3p was associated with the occurrence of autoimmune diseases in mice¹¹⁸; the hsa-miR-200b-3p was associated with avascular necrosis of the femoral head¹¹⁹; the hsa-miR-16-5p was associated with systemic lupus erythematosus and psoriatic arthritis^120,121; the hsa-miR-1-3p was associated with systemic lupus erythematosus¹²⁰, and the hsa-mir-1-3p was associated with SARS-CoV-2 and RA in S. aureus infection¹²².

Furthermore, 12 drugs were identified by CTD, which ICG 001, carbamazepine, and sunitinib have potential drugs for the treatment of RA, and molecular docking validation predicted the strength of action between the drug and OXSM. The docking results showed that the binding energy of the 12 drug components to OXSM proteins was all − 6.0 kcal/mol, indicating a strong binding affinity. Notably, the lowest binding energy between ICG 001 and OXSM (-10.1 kcal/mol). Compared with ICG 001, the binding energies of Auranofin, Ibuprofen, Leflunom, Methylprednisolone Tablets, Nimesulide, Methotrexate, and OXSM were − 6.1 kcal/mol, -6. kcal/mol, -8.0 kcal/mol, -8.5 kcal/mol, -8.5 kcal/mol, -9.3 kcal/mol, respectively. The results showed that ICG 001 had better binding affinity with OXSM than the six drugs, highlighting its potential as a therapeutic drug. However, we have to admit that molecular docking has limitations. Specifically, our docking simulations used rigid receptor models, which may not be able to explain the dynamic nature of protein conformations. In addition, we note that the scoring functions used in the docking process may not fully capture the complexity of the ligand-receptor, which could lead to differences in the predicted binding affinities. ICG 001 is a transcriptional antagonist¹²³, and a small molecule inhibitor of the Wnt/β-catenin signaling pathway, which can effectively inhibit the proliferation and migration of a variety of cancer cells¹²⁴, reduce the chemotherapy resistance of gastric cancer stem cells¹²⁵, induce autophagy of endometrial cancer cells¹²⁶, by increasing radiation-induced DNA damage and improve the tumor immune microenvironment to enhance radiotherapy effect¹²⁷and can improve chronic lung injury and prevent progression to severe lung fibrosis¹²³. Studies have shown that there is a high level of β-catenin in FLS of RA patients, and a high level of β-catenin can maintain the stable activation of FLS, thus causing the overexpression of inflammatory factors such as TNF-α and IL, which mediate the inflammatory response of RA¹²⁸. The overactivation of the Wnt signaling pathway and abnormal expression of β-catenin play a significant role in RA bone destruction, causing severe disruption to the balance between osteoblasts and osteoclasts required for normal bone remodeling, leading to progressive bone destruction. Therefore, inhibition of the Wnt/β-catenin signaling pathway can help alleviate the occurrence and development of arthritis¹²⁹. Molecular dynamics simulations showed that ICG 001 stably binds to OXSM with a binding free energy of -41.23 kcal/mol, indicating that OXSM is the target of ICG 001. We hypothesized that under glucose starvation conditions, OXSM is aberrantly expressed in RA, triggering disulfidptosis, and ICG 001 may destroy the course of RA by inhibiting OXSM overexpression. Thus, OXSM may be an essential target for disulfidptosis-related mechanisms in RA. However, there are no studies of ICG-001 in treating rheumatoid arthritis. Since side effects of ICG-001 treatment cannot be entirely excluded, further studies are needed to determine whether the identified OXSM gene is indeed a direct target of ICG-001¹³⁰. Meanwhile, in vivo and in vitro experiments are needed to verify the targeted effect between ICG-001 and OXSM, and chemical modification or nanocarriers may be helpful for reducing the off-target effect of ICG-001 and OXSM.

To further verify our conclusions, we collected peripheral blood from three healthy controls and five RA patients and measured the relative mRNA expression levels via RT-qPCR. We found that OXSM expression was higher in the RA group than in the control group (Fig. 17A), which is consistent with the results of our bioinformatics analysis(Figs. 5B and 7A). Although, rheumatoid factor (RF) and anti-cyclic citrullinated peptide antibodies (ACPA) were used as the serum biomarkers for the early diagnosis of RA^131,132. However, both serum markers are not positive in all early RA patients and will have reduced sensitivity, especially when applied to the seronegative patient population¹³³. Our results suggest that the OXSM gene may be involved in the pathological process of RA. OXSM as a novel biomarker may be important for the early diagnosis and effective treatment of seronegative patients or RA. Finally, our results will provide new insights into the role of disulfidptosis in RA and identify OXSM potentially as a potential biomarker and target for diagnosing and treating RA.

Undeniably, there are some limitations in our study. First, we should expand the sample size and include a more diverse sample population to validate the specificity and validity of OXSM as a biomarker for diagnosing RA. Secondly, more accurate experiments should be conducted, such as surface plasmon resonance (SPR) to confirm that ICG 001 indeed treats RA through OXSM. Finally, it is necessary to combine animal models or clinical trials to demonstrate the efficacy and safety of ICG 001 in treating RA. Although we are currently limited by resources and time, and have only conducted basic experiments, these will be the focus of our next stage of research.

Conclusion

We identified OXSM as an essential gene associated with disulfidptosis in RA. We established a novel diagnostic model based on OXSM. We verified that OXSM was significantly highly expressed in the peripheral blood of RA patients by RT-qPCR, and OXSM may become a novel biomarker for diagnosing RA. Our findings may improve the accuracy and reliability of diagnostic RA, helping to provide more personalized and effective medical intervention for patients in the future. In addition to its strong association with disulfidptosis, OXSM is also associated with the immune microenvironment in RA patients. At the same time, we also predicted potential drugs targeting OXSM for RA, and molecular docking and molecular dynamics simulations suggested that OXSM and ICG 001 may be potential targets and drugs for treating RA. This is the first study to analyze the relationship between disulfidptosis and RA comprehensively. Although more mechanisms between RA and disulfidptosis have not been elaborated, disulfidptosis may become a future research hotspot for RA treatment.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ACPA

anti-cyclic citrullinated peptide antibodies

ACTN4

Actinin Alpha 4

AS

ankylosing spondylitis

AUC

Area Under Curve

BP

Biological Process

CBFB

Core binding factor subunitβ

CC

Cellular Component

CTD

The Comparative Toxicogenomics Database

CUX1

Cut like homeobox 1

DEGs

Differentially expressed genes

DRGs

Disulfidptosis-related genes

EED

Embryonic Ectoderm Development

FEL

Free energy landscape

GEO

The Gene Expression Omnibus

GO

Gene Ontology Enrichment Analysis

GS

Gene significance

GSEA

Gene Set Enrichment Analysis

HSPD1

Heat shock protein D family member 1

JIA

juvenile idiopathic arthritis

KEGG

Kyoto Encyclopedia of Genes and Genomes

mAPK

mitogen-activated protein kinase

MF

Molecular Function

MM

Module membership

NOCA1

Non-centrosomal microtubule array protein 1

OSCC

Oral squamous cell carcinoma

OXSM

Oxoacyl-ACP Synthase Mitochondrial

PB

peripheral blood

PsA

psoriatic arthropathy

RA

Rheumatoid arthritis

RF

rheumatoid factor

Rg

Radius of gyration

RMSD

Root mean square deviation

RMSF

Root Mean Square Fluctuation

ROC

Receiver Operating Characteristic

ROS

reactive oxygen species

SASA

Solvent-accessible surface area

SPR

surface plasmon resonance

TAF 7

TATA-box binding protein associated factor 7

TFs

Transcription factors

WB

whole blood

WGCNA

Weighted gene co-expression network analysis

ZNF639

Zinc finger protein 639

Author contributions

BX: Designed and conceived, data analysis, molecular biology experiments, wrote, revised, confirmed final manuscript and provided funds. QG: Collected and processed data, analyzed data, revised the manuscript and confirmed the final draft. XDL: Designed and conceived, data analysis, revised and confirmed the final draft. HLZ: Software operation. BS: Sample collection. JMW: Investigation. MTS: Statistical data analysis. All authors have reviewed and approved the submission of the manuscript.All authors reviewed the manuscript.

Funding

This work was supported by the Science and Technology Fund Project of the Guizhou Provincial Health Commission (No. gzwkj2024–523) and the Science and Technology Program of Anshun City (No. ASKC-2021–12, No. ASKC-2024–17).

Data availability

The GSE93272 and GSE45291 datasets used in this study were downloaded from the Gene Expression Omnibus (GEO)(https://www.ncbi.nlm.nih.gov/geo/ ). The software/package used in this study can be found in the Sangerbox platform (http://vip.sangerbox.com/home.html).

Declarations

Competing interests

The authors declare no competing interests.

Ethics statement

The human blood sample collection protocol was approved by the Ethics Committee of Anshun People’s Hospital, Guizhou Province. Written informed consent was obtained from all participants before enrollment.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.