Identification and validation of KLRB1-related biomarkers in rheumatoid arthritis

Abstract

Background

Rheumatoid arthritis (RA) is a systemic autoimmune inflammatory disorder. KLRB1 (killer cell lectin like receptor B1), which is intricately linked to immune modulation and inflammatory responses, represents a promising biomarker for the identification of RA. This study mainly explores the relationship between KLRB1 and RA, and identifies biomarkers related to KLRB1 in RA, providing theoretical support for the diagnosis and treatment of RA.

Methods

The transcriptome data of RA were sourced from the public database. Differential expression analysis was used to identify differentially expressed genes (DEGs) and KLRB1-related DEGs. Additionally, key module genes associated with RA were determined using weighted gene co-expression network analysis (WGCNA). Subsequently, the DEGs, KLRB1-related DEGs, and key module genes were subjected to an intersection analysis to identify candidate genes. Afterwards, machine learning, expression validation, and diagnostic evaluation of the aforementioned genes were conducted to identify biomarkers, and a nomogram was constructed to evaluate the diagnostic value of the biomarkers. Furthermore, enrichment analysis and immune microenvironment analysis were carried out for further evaluation of the role of biomarkers in the regulatory mechanisms in RA. Ultimately, the expression of biomarkers in clinical samples was validated through the utilization of reverse transcription quantitative polymerase chain reaction (RT-qPCR).

Results

The study identified 1,264 DEGs, 293 KLRB1-related DEGs, and 1,379 key module genes, which resulted in the selection of 36 candidate genes. Thereafter, 2 biomarkers (ADAMDEC1 and CXCL13) associated with KLRB1 in RA were identified through machine learning, expression validation, and diagnostic evaluation. The nomogram model indicated that these biomarkers possess considerable diagnostic value for patients with RA. Besides, these biomarkers were notably enriched in the “cytoskeleton in muscle cells” and “motor proteins” pathways. Moreover, ADAMDEC1 and CXCL13 demonstrated positive correlation with plasma cells, CD8 + T cells, and activated CD4 + T memory cells, and an inverse association with activated mast cells and activated NK cells. The RT-qPCR analysis demonstrated a significant increase in ADAMDEC1 and CXCL13 expression levels in the RA group (P < 0.05).

Conclusions

This study identified 2 effective biomarkers (ADAMDEC1 and CXCL13) for RA, thereby providing potential therapeutic targets for RA patients.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-42924-y.

Introduction

Rheumatoid arthritis (RA) is a persistent, systemic autoimmune condition characterized by inflammation primarily in the joints and periarticular soft tissues¹. The global prevalence of RA from 1990 to 2020 was approximately 0.21%, representing a 14.1% increase since 1990². The etiology of RA remains incompletely understood, but a combination of genetic predispositions and environmental factors such as tobacco smoking, obesity, and occupational exposures are believed to contribute to its pathogenesis^3,4. Current therapeutic strategies for RA encompass a multifaceted approach, including pharmacological interventions, physical therapy, surgical interventions, and lifestyle modifications⁵. Although the age standardized mortality rate of RA decreased by 23.8% from 1990 to 2020², all of these treatment methods cannot completely prevent joint damage and may result in adverse reactions⁶. The standardized management of RA, which requires a personalized approach, regular monitoring, and adjustment of treatment plans as needed, can greatly improve the comprehensive treatment effect of RA. Therefore, exploring new effective biomarkers and investigating the molecular mechanisms of RA can help predict RA and develop personalized treatment plans, ultimately achieving the goal of improving the quality of life of RA patients.

KLRB1, also known as CD161, is a C-type lectin-like receptor predominantly expressed on natural killer (NK) cells and specific subsets of T cells⁷. It plays a pivotal role in modulating the immune responses of NK and T cells, particularly in terms of cytotoxicity and cytokine production⁸. Research has shown that the proportion of CD161 + T cells is positively correlated with RA disease activity (DAS28, CRP, ESR levels)⁹, and CD161 has been shown to be a biomarker for human Th17¹⁰. CD161 + Th17 plays an important pathogenic role in RA, which may be related to its secretion of IL-17A and IL-22¹¹. An increasing number of researches on KLRB1 emphasize its potential as a biomarker for various diseases, particularly in the fields of neoplastic disorders and autoimmune diseases^12–14. Nevertheless, the precise mechanisms by which KLRB1 operates in RA remain inadequately defined, necessitating further exploration to elucidate its involvement in the onset and progression of RA, which could underpin the development of novel biomarkers or therapeutic strategies.

This study is based on transcriptomic data of RA provided by public databases, and evaluates biomarkers related to KLRB1 in RA through bioinformatics analysis, further exploring the potential mechanisms of biomarkers in RA and opening up new paths for clinical precision diagnosis and personalized treatment of RA patients.

Methods

Data source

In this study, the GSE55235 (platform: GPL96) and GSE55457 (platform: GPL96) datasets were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). The GSE55235 dataset was used as the training dataset, comprising synovial tissue samples from 10 RA patients and 10 non-rheumatoid arthritis (NRA) controls. The GSE55457 dataset served as the validation set, containing synovial tissue samples from 13 RA patients and 10 NRA controls. Furthermore, 200 inflammation-related genes (IRGs) were retrieved from the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/). A total of 6 RA-related inflammatory factors were identified from the available literature: IL17A¹⁵, IL6, IL1B, IL13, IL4 and IL10^16–18.

Differential expression analysis

The objective was to obtain genes that exhibited differential expression in the RA and NRA groups in the GSE55235 dataset, DEGs in both groups were identified by limma package (v 3.58.1)¹⁹, with a threshold of |log₂FoldChange (FC)|> 1 and P < 0.05. Moreover, to gain further insight into the potential mechanism of KLRB1 action in RA, differential expression levels of KLRB1 were analyzed between the RA and NRA groups within the GSE55235 dataset using the Wilcoxon test (P < 0.05). Subsequently, the RA samples were categorized into two groups based on their KLRB1 expression levels, with the high and low expression groups defined by the median value of KLRB1 expression. The limma package (v 3.58.1) was employed to ascertain the disparities in gene expression between the high and low expression groups (|log₂FC|> 0.5, P < 0.05), and these genes were designated as KLRB1-related DEGs. Furthermore, the volcano plot and heatmap of DEGs and KLRB1-related DEGs were plotted utilizing the ggplot2 package (v 3.5.1)²⁰ and ComplexHeatmap package (v 2.18.0)²¹ respectively. The top 10 genes that were most significantly upregulated and downregulated were labelled in the volcano plot, while a heatmap illustrated their expression profiles (ranked by log₂FC value).

Weighted gene co-expression network analysis (WGCNA)

To obtain the module genes most strongly association with RA, WGCNA was conducted utilizing the WGCNA package (v 1.72–5)²² on all samples in the GSE55235 dataset. The preliminary stage of the analysis entailed the clustering of the samples. The application of hierarchical clustering enabled the identification of any outlier samples, thus ensuring the accuracy and reliability of the results. To maximize the scale-free topological fit of the interactions between genes, a soft threshold (power) was chosen to construct the co-expression network based on a scale-free fit index (R²) exceeding 0.9, with a mean connectivity approaching 0. The filtered expression matrix was utilized to construct the co-expression network, adhering to the hybrid dynamic tree cutting algorithm, with a criteria of at least 100 genes per module and merge cut height set at 0.25. Subsequently, hierarchical clustering trees were constructed to further delineate co-expression modules. In addition, the gene expression data from the RA and control samples were employed as phenotypic traits. The Pearson correlation analysis was conducted utilizing the psych package (v 2.4.3)²³ to compute the correlation matrix between these traits and co-expression modules (|correlation coefficients (cor)|> 0.3, P < 0.05). The modules most pertinent to the phenotypic traits were selected, and the genes in their modules were designated as key module genes.

Identification and functional analysis of candidate genes

The intersection of DEGs, KLRB1-related DEGs, and key module genes was performed using the ggvenn package (v 0.1.10)²⁴ to identify genes associated with KLRB1 in RA, which were recorded as candidate genes. Subsequently, the biological functions of the candidate genes were elucidated using the clusterProfiler package (v 4.10.1)²⁵, which facilitated the performance of Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis on the candidate genes (P < 0.05)^26–28. GO analysis was divided into 3 categories: biological process (BP), cellular component (CC), and molecular function (MF). The GO entries and KEGG pathways were ordered in descending order of their P-values, thereby displaying the top 10 results exhibiting the most significant enrichment, respectively. To examine the extent of interrelationships among candidate genes, a Pearson correlation analysis was conducted on all samples in the GSE55235 dataset, leveraging the psych package (v 2.4.3). Following this, the correlation outcomes were represented visually through the pheatmap package (v 1.0.12)²⁹.

Identification of biomarkers

The interactions between the candidate genes were further investigated. This involved the use of the search tool that retrieves interaction genes (STRING) (https://www.string-db.org) in conjunction with the construction of a protein–protein interaction (PPI) network (confidence = 0.4). Then, the candidate genes that were found to exhibit interactive relationships within the PPI network were designated as hub genes. To identify biomarkers associated with KLRB1 in RA, a preliminary screening of hub genes was conducted using the GSE55235 dataset, employing two distinct machine learning techniques. A least absolute shrinkage and selection operator (LASSO) regression analysis was conducted using the glmnet package (v 4.1–8)³⁰ to identify LASSO feature genes. In this instance, the LASSO regression analysis was conducted with tenfold cross-validation. Concurrently, the support vector machine-recursive feature elimination (SVM-RFE) feature genes were identified utilizing the caret package (v 6.0–94)³¹. Subsequently, potential biomarkers were identified by taking the intersection of the feature genes that had been identified by the two different machine learning methods using the ggvenn package (v 0.1.10). Next, to determine the expression patterns of the potential biomarkers, we first performed the Shapiro—Wilk test on the expression data of the biomarkers (P < 0.05). Then, we used the Wilcoxon test to compare the expression differences of the potential biomarkers between the RA and NRA groups in the GSE55235 and GSE55457 datasets (P < 0.05). Subsequently, the diagnostic value of the potential biomarkers for RA was then evaluated by plotting the receiver operating characteristic curve (ROC) and calculating the area under the curve (AUC) value utilizing the pROC package (v 1.18.5)³². Eventually, potential biomarkers displaying significant expression differences and consistent trends among groups across both datasets, with the AUC values exceeding 0.7 in both, were identified.

Establishment and assessment of nomogram

A nomogram was constructed to evaluate biomarkers to predict the occurrence of disease. In the GSE55235 dataset, a nomogram for estimating the likelihood of developing RA based on biomarkers was constructed using the rms package (v 6.8–1)³³. Furthermore, a calibration curve was plotted using the regplot package (v 1.1)³⁴ to ascertain the accuracy of the nomogram. The slope approaching 1 on the calibration curve indicated greater accuracy in the nomogram model prediction. Subsequently, ROC curve was plotted using the pROC package (v 1.18.5), and the diagnostic value of the nomogram model for RA in the GSE55235 dataset was assessed by calculating AUC value. Further, fivefold cross-validation was conducted to verify the reliability of the ROC curve.

Gene set enrichment analysis (GSEA)

To gain further insight into the signalling pathways that the biomarkers were involved in, the GSEA was conducted on the GSE55235 dataset for the biomarkers. Initially, RA patients were classified into high and low expression groups based on the median value of each biomarker expression. Then, the high and low expression groups were analysed for differences using the limma package (v 3.58.1), with the results sorted in descending order based on the log₂FC value. Subsequently, we utilized “c2.cp.kegg.v7.4.entrez.gmt” from the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/) as the background gene set. The GSEA was conducted for the sorted genes (P < 0.05), utilizing the enrichplot package (v 1.22.0)³⁵, and the top 5 significantly enriched pathways for each biomarker were displayed.

Inflammatory factor correlation analysis

RA being a chronic inflammatory disease, a deeper examination of the connection between biomarkers and inflammatory factors was undertaken. Initially, the Spearman correlation analysis was conducted between the IRGs exhibiting expression in the GSE55235 dataset and KLRB1. Following this, the 5 most significantly positively correlated and the 5 most significantly negatively correlated inflammatory factors, which demonstrated a notable association with KLRB1 (|cor|> 0.5, P < 0.05), were identified as KLRB1-related inflammatory factors. The expression of 10 KLRB1-related and 6 RA-related inflammatory factors was analyzed in RA and NRA samples using the Wilcoxon test (P < 0.05). Subsequently, a Spearman correlation analysis was conducted using the corr.test function from the psych package (v 2.4.3) to examine the relationship between biomarkers and differentially expressed inflammatory factors (|cor|> 0.3, P < 0.05).

Immune microenvironment analysis

To gain further insight into the level of immune infiltration in the RA and NRA groups, the abundance of 22 immune cells³⁶ per sample from the GSE55235 dataset was calculated using the CIBERSORT algorithm in the IOBR (v 0.99.9) package³⁷. In addition, immune cells with a result of 0 in 30% of the samples were excluded from subsequent analysis. Next, the difference in immune cell infiltration between two groups was compared by Wilcoxon test (P < 0.05). To evaluate the relationship between biomarkers and various immune cell types, as well as the associations among immune cells themselves, a Spearman correlation analysis was conducted on the GSE55235 dataset (|cor|> 0.3, P < 0.05).

Construction of regulatory networks and drug prediction

Molecular regulatory networks facilitated a deeper understanding of the intricate mechanisms of gene regulation and the processes involved in disease occurrence. To investigate the regulation of biomarkers by microRNAs (miRNAs), the miRNAs present in the DIANA-miTED (https://dianalab.e-ce.uth.gr/mited/) and miRDB (https://mirdb.org/) databases were predicted using the get_multimir function. Subsequently, the intersection of the predicted miRNAs from the aforementioned databases was identified in order to ascertain the target miRNAs. Additionally, the transcription factors (TFs) interacting with biomarkers were predicted utilizing the TRRUST database (https://ngdc.cncb.ac.cn/). The resulting TF-mRNA- miRNA was visualised using the Cytoscape (v 3.9.1) software. The drugs with the potential to target biomarkers were identified through the Drug-Gene Interaction database (DGIdb) (https://www.dgidb.org). Subsequently, the drug-biomarker network diagrams were visualised using Cytoscape (v 3.9.1) software.

Reverse transcription-quantitative polymerase chain reaction (RT-qPCR)

To further validate the expression levels of biomarkers between RA and NRA groups, RT-qPCR was performed. A total of 5 clinical samples of patients with a confirmed diagnosis of RA and 5 samples of NRA controls were collected from the First Affiliated Hospital of Anhui Medical University. The study was approved by the Institutional Review Board of the First Affiliated Hospital of Anhui Medical University (LLSC20211280). The total RNA of the frozen RA and NRA tissue samples was extracted by means of the TRizol kit (Ambion, 15,596-018CN, USA). All experimental steps for total RNA extraction were performed according to the instructions. 1 μL of extracted RNA was taken for concentration detection with a NanoPhotometer N50 and the purity/concentration was recorded to calculate the amount of RNA for subsequent reverse transcription steps. Subsequently, the RNA was reverse transcribed into cDNA utilizing Hifair® III 1st Strand cDNA Synthesis SuperMix for qPCR Kit (Yeasen Biotechnology, Shanghai, China) in accordance with the instructions. Next, the cDNA was diluted 5–20 times with ddH₂O (without RNase/ARase), added 3uL cDNA, 5uL 2xUniversal Blue SYBR Green qPCR Master Mix, 1uL forward primer (10 µM) and 1uL reverse primer (10 µM). In addition, 40 cycles (exclusive of pre-denaturation) of reactions were performed utilizing the CFX Connect real-time quantitative PCR instrument (BIO-RAD, XLFZ006), and program information was provided in Table S1. Primer sequence information for biomarkers was shown in Table S2, and GAPDH served as the reference gene, and relative gene expression levels were determined by employing the 2^−ΔΔCT method. Moreover, the generation of histograms depicting the differences in biomarkers mRNA expression levels between the RA and NRA groups was conducted using GraphPad Prism 5.

All methods employed in this study adhere to relevant guidelines and regulations and comply with the requirements of the Declaration of Helsinki. All patients provided written informed consent authorizing the use of their medical records for research purposes.

Statistical analysis

R language (v 4.2.2) was utilized to process and analyze the data. The P < 0.05 was considered statistically significant. In the RT-qPCR, the Ct values were compared using unpaired, independent-sample t test, which were computed utilizing the GraphPad Prism 5.

Results

Identification of 1,264 DEGs, 293 KLRB1-related DEGs, and 1,379 key module genes

A differential expression analysis revealed the presence of 1,264 DEGs between the RA and NRA groups. Of these, 542 were up-regulated genes and 722 were down-regulated genes in the RA group (Fig. 1A–B, Table S3). Among them, the expression of KLRB1 was significantly decreased in the RA group (log2FC = -2.24, P < 0.001, adj.P.Val < 0.001) (Table S3). Further investigation into KLRB1 expression in RA and NRA groups revealed a significantly elevated level in the RA group compared to the control group (P < 0.0001) (Fig. S1A). The consistency between the results of the limma analysis and the Wilcoxon test laid the differential foundation for the investigation of the molecular mechanism of KLRB1. The median value of 6.8079 for KLRB1 expression was employed to categorise RA samples into high and low expression groups. In addition, a total of 293 KLRB1-related DEGs were identified between the high and low expression groups, with 104 up-regulated and 189 down-regulated in the high expression group (Fig. 1C–D). Subsequently, a WGCNA network was constructed based on all samples in the GSE55235 dataset. The sample clustering tree showed no outlier samples, indicating that all samples were included in the construction of the WGCNA network (Fig. S1B). The soft threshold (power) was screened at 6 and the R² value exceeded 0.9 (Fig. S1C). And the hierarchical clustering tree identified a total of 11 co-expression modules (Fig. 1E). Furthermore, correlation analysis of the modules with the RA revealed that the highest positive correlation was with the MEturquoise module (cor = 0.97, P < 0.001), which comprised 1,379 key module genes (Fig. 1F).

Fig. 1.

Different expression gene (DEGs) analysis of GSE55235 dataset. (A) Volcanic map of DEGs between RA and NRA group. (B) Heat map of DEGs between RA and NRA group. The upper part reflects the trend of all genes expression, while the lower part shows the expression of top 10 DEGs. (C) Volcanic map of DEGs between KLRB1 high expression group and low expression group in RA synovial tissue. (D) Heat map of DEGs between KLRB1 high expression group and low expression group in RA synovial tissue. The upper part reflects the trend of all genes expression, while the lower part shows the expression of the top 10 DEGs. (E) Gene dendrogram obtained by average linkage hierarchical clustering. The color row below the dendrogram shows the module assignment determined by the dynamic tree cut. (F) Model-trait relationships. Each row in the heatmap corresponds to one module (labeled by black, pink, purple, magenta, yellow, brown, turquoise, greenyellow, red, blue, and green). The blue color in the heatmap represents negative correlation, the red color represents positive correlation.

Identification of 36 candidate genes and exploration of their functions

Following this, 1,264 DEGs, 293 KLRB1-related DEGs, and 1,379 key module genes were taken for intersection, resulting in 36 candidate genes (Fig. 2A). Subsequently, an enrichment analysis was conducted to provide preliminary insights into the signalling pathways implicated by the candidate genes. The candidate genes were significantly enriched in 87 GO entries with 74 BPs, 4 CCs, and 9 MFs. The top 10 significantly enriched pathways (P < 0.05) included “leukocyte cell–cell adhesion” and “lymphocyte differentiation” (Fig. 2B), which indicated that the candidate genes might play a crucial role in regulating immune response. Additionally, a KEGG enrichment analysis of the candidate genes revealed the enrichment of 22 pathways, such as “Cytokine − cytokine receptor interaction” (Fig. 2C). Furthermore, there was a significant correlation between almost all candidate genes (Fig. 2D).

Fig. 2.

Identification of candidate genes and exploration of their functions. (A) VN diagram among 1,264 DEGs, 293 KLRB1-related DEGs, and 1,379 MEturquoise module genes. (B) Top 10 Biological Process for GO enrichment analysis using 36 candidate genes. (C) Top 10 signaling pathways identified through KEGG enrichment analysis of 36 candidate genes. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway map was obtained from KEGG (https://www.kegg.jp/). KEGG is a publicly available resource under the terms of the academic uselicense (Kanehisa et al., 2016: Kanehisa & Goto,2000)^26–28. (D) Correlation analysis of 36 selected genes.

ADAMDEC1 and CXCL13 were identified as biomarkers for RA

The construction of PPI networks for the identified candidate genes revealed 18 interconnected genes, indicating their potential involvement in different functions, thus classifying them as hub genes (such as ADAMDEC1, CD79A) (Fig. 3A). Subsequently, the 18 hub genes were incorporated into the LASSO algorithm. By setting the lambda.min threshold to 0.00036, 4 genes (ADAMDEC1, CXCL13, GHR and LAMP3) were retained and designated as LASSO feature genes (Fig. 3B–C). Meanwhile, 5 SVM-RFE feature genes (ADAMDEC1, CD79A, CXCL13, ITGA4, SELL) were identified by SVM-RFE screening (Fig. 3D). Afterwards, the feature genes obtained from the above two algorithms were intersected to obtain the 2 candidate biomarkers (ADAMDEC1 and CXCL13) (Fig. 3E). To further analyze, since the P—value of the Shapiro—Wilk test was less than 0.05, the null hypothesis that “the data are from a normally distributed population” was rejected. This indicated that the expression data of the candidate biomarkers did not conform to a normal distribution in the GSE55235 and GSE55457 datasets (Fig. S2). Furthermore, expression verification and ROC curve evaluation were performed on the candidate biomarkers. The results showed that the expression levels of ADAMDEC1 and CXCL13 in the rheumatoid arthritis group were significantly increased (P < 0.001), and the expression patterns in the two datasets showed a high degree of consistency (Fig. 3F–G). The findings indicated stability and reliability for these two biomarkers, potentially contributing to RA diagnosis and prognostic evaluation. In addition, in the GSE55235 and GSE55457 datasets, the AUC values for ADAMDEC1 and CXCL13 were greater than 0.9 (Fig. S2). This result suggested that these two genes were effective in the diagnosis and assessment of RA patients. Therefore, ADAMDEC1 and CXCL13 were defined as biomarkers and used for further analysis.

Fig. 3.

Identification of candidate biomarkers. (A) Interaction network diagram of closely connected gene constructed based on PPI network (confidence = 0.4). (B) LASSO regression of the key genes. (C) Cross-validation in the LASSO regression model to select the tuning parameter. The abscissa shows the log (λ) value, and the ordinate shows Binomial Deviance. (D) Features screened from DEGs using the SVM-RFE algorithm. (E) VN diagram between LASSO characteristic genes and SVM-RFE feature genes. (F) ADAMDEC1 and CXCL13 expression between RA and NRA group in train (GSE55235) dataset. ADAMDEC1 and CXCL13 expression between RA and NRA group in verify (GSE55457) dataset.

Nomogram demonstrated strong performance in assessing diagnosis of RA

To further evaluate the diagnostic ability of biomarkers for RA, a nomogram model was constructed. In this model, a greater total point value corresponded to a higher rate of survival among individuals diagnosed with RA (Fig. 4A). Furthermore, the nomogram model was assessed using calibration curves and ROC curves to determine its predictive capability. The diagnostic error rate of this nomogram model was low, as shown by the calibration curve (P = 0.593) (Fig. 4B). Furthermore, the AUC value of the nomogram model in the ROC curve was 1 (Fig. 4C), indicating a reasonable level of accuracy in predicting RA by the nomogram model. After fivefold cross-validation, the biomarkers still accurately predicted the risk of RA (Fig. 4D–E, Table S4).

Fig. 4.

Diagnostic efficacy examination of nomogram model constructed by biomarkers ADAMDEC1 and CXCL13. (A) A nomogram model was constructed based on ADAMDEC1 and CXCL13 expression levels. (B) Calibration curve chart of the nomogram model. P > 0.05: the accuracy of the disease probability predicted by the nomogram model has statistical significance. (C) ROC curve of the nomogram model predicting the diagnostic value of RA patients. (D) The performance of the biomarker ADAMDEC1 in predicting the risk of RA after fivefold cross-validation. (E) The performance of the biomarker CXCL13 in predicting the risk of RA after fivefold cross-validation.

ADAMDEC1 and CXCL13-associated pathways and inflammatory profiles in RA

A comprehensive GSEA of the biological functions associated with these biomarkers revealed that 27 and 68 pathways were significantly enriched for ADAMDEC1 and CXCL13, respectively. The top 5 pathways associated with these biomarkers were found to be significantly co-enriched, including “cytoskeleton in muscle cells” and “motor proteins” (Fig. 5A). The results indicated that ADAMDEC1 and CXCL13 may influence cellular functions, including morphology, motility and signalling, by modulating these shared metabolic pathways, which were crucial in regulating physiological and pathological processes. Besides, an in-depth analysis was conducted on 6 inflammatory factors associated with RA and 10 inflammatory factors linked to KLRB1, in consideration of the inflammatory nature of RA. The RA group exhibited significantly elevated expression of APLNR, SLAMF1, and SLC1A2 (P < 0.01), and notably decreased expression levels of IL1B, OSM, and RAF1 compared to the NRA group among the studied inflammatory factors (P < 0.05) (Fig. 5B). Notably, ADAMDEC1 and CXCL13 exhibited significant positive correlations with APLNR, SLAMF1, and SLC1A2, and negative correlations with RAF1, and IL1B (Fig. 5C). Furthermore, a notable inverse correlation was observed between CXCL13 and OSM.

Fig. 5.

Profiles of signaling pathways and inflammatory cytokine expression associated with ADAMDEC1 and CXCL13. (A) GSEA analysis based on DEGs between high and low expression groups of ADAMDEC1 and CXCL13. (B) Differential expression of 16 inflammatory factors related to RA and KLRB1 from train dataset. (C) Inflammatory factors significantly correlated with ADAMDEC1 and CXCL13. *P < 0.05,**P < 0.01,***P < 0.001,****P < 0.0001, ns: no significance.

Immune cell distribution and correlation analysis of ADAMDEC1 and CXCL13

In order to ascertain the distinctions in the immune microenvironment between the RA and NRA groups, a stacked bar chart was employed to illustrate the infiltration abundance of 22 distinct immune cells (Fig. 6A). Following the filtration of immune cells that yielded a result of 0 in 30% of the samples, 18 immune cells were identified for further analysis. Subsequently, 8 distinct immune cell types were identified between the RA and NRA groups (Fig. 6B). Among them, neutrophils, plasma cells, activated memory CD4 T cells, CD8 T cells, and follicular helper T cells demonstrated heightened expression in the RA group, whereas activated mast cells, activated natural killer (NK) cells, and resting memory CD4 T cells exhibited reduced expression. Furthermore, a Spearman correlation analysis among immune cell types revealed a pronounced positive correlation between resting memory CD4 T cells and activated mast cells (cor = 0.72, P < 0.001). Conversely, the strongest negative correlation was observed between resting memory CD4 T cells and CD8 T cells (cor = -0.86, P < 0.001) (Fig. 6C). Moreover, ADAMDEC1 and CXCL13 exhibited the most pronounced positive correlations with plasma cells, with correlation coefficients of 0.81 and 0.82, respectively. Conversely, these biomarkers demonstrated inverse correlations with activated mast cells, with coefficients of -0.79 for ADAMDEC1 and -0.75 for CXCL13 (Fig. 6D).

Fig. 6.

Immune infiltration analysis and its correlation analysis with ADAMDEC1 and CXCL13. (A) Distribution profile of 22 immune cell infiltration in RA and NRA group. (B) Differential analysis of immune infiltration between RA and NRA group. (C) Correlation analysis of 8 differential immune infiltrating cells in RA group. (D) Correlation analysis between 8 differential immune infiltrating cells and ADAMDEC1 and CXCL13 in RA group. Red: positive correlation, blue:negative correlation.

Exploring the regulatory network and potential drug targets of ADAMDEC1 and CXCL13

To gain further insight into the regulatory factors of biomarkers, molecular regulatory network was constructed. The DIANA-miTED database predicted 169 miRNAs targeting 2 biomarkers, while miRDB forecasted 32 miRNAs. And a total of 26 target miRNAs were obtained (Fig. 7A). Furthermore, the two biomarkers predicted a total of 23 TFs, comprising 13 linked to ADAMDEC1 and 11 to CXCL13. Subsequently, a TF-mRNA-miRNA network comprising 23 TFs, 2 biomarkers (ADAMDEC1 and CXCL13), and 26 miRNAs was constructed (Fig. 7B), indicating that ADAMDEC1 and CXCL13 were regulated by multiple factors. Notably, both ADAMDEC1 and CXCL13 expression were found to be regulated by hsa-miR-186-5p and AR. The utilisation of biomarkers permitted the extraction of data on their interactions with drugs from the DGIdb, which could then be employed to elucidate the intricate interactions between drugs and biomarkers. The drug prediction results indicated that 6 and 15 potential drugs were predicted for ADAMDEC1 and CXCL13, respectively (Fig. 7C). Among them, metronidazole and tetradioxin were drugs jointly targeted by both ADAMDEC1 and CXCL13, suggesting that both may play a role in the treatment of RA.

Fig. 7.

Regulatory Networks and Potential Drug Targets of ADAMDEC1 and CXCL13. (A) MiRNAs targeting ADAMDEC1 and CXCL13 through DIANA-miTED and miRDB databases. (B) Interaction network of TF-miRNA with ADAMDEC1 and CXCL13. (C) Prediction of drugs targeting ADAMDEC1 and CXCL13.

Validation of ADAMDEC1 and CXCL13 in RT-qPCR

Following the extraction of total RNA, the results of the RNA concentration assay demonstrated that all samples were within the standard range of RNA concentration (Table S5). The mRNA expression levels of ADAMDEC1 and CXCL13 were observed to be significantly elevated in the RA group, in comparison to those in the NRA group (P < 0.05) (Fig. 8A–B). The results were found to be in accordance with the expression patterns of the 2 biomarkers predicted by the GSE55235 and GSE55457.

Fig. 8.

Differential expression analysis of ADAMDEC1 and CXCL13 in RT-qPCR. (A) RT-qPCR analysis of differential expression of ADAMDEC1 in RA and NRA patients’ synovium. (B) RT-qPCR analysis of differential expression of CXCL13 in RA and NRA patients’ synovium.

Discussion

RA is a chronic, systemic autoimmune disorder that predominantly impacts the joints, and early diagnosis is a crucial factor in comprehensive management⁴. Functioning as an inhibitory receptor, KLRB1 negatively regulates diverse biological functions of NK and T cells, thereby contributing to disease pathogenesis by suppressing immune effector responses^13,38. KLRB1 has been widely recognized as a prognostic biomarker for multiple cancers^12,39. Furthermore, KLRB1 is closely associated with autoimmune diseases, including primary Sjögren’s syndrome (pSS) and systemic lupus erythematosus (SLE)^40–42. In rheumatoid arthritis (RA), the expression level of KLRB1 correlates with disease activity and severity^9,43. Notably, KLRB1 has been identified as an early diagnostic biomarker for RA¹⁴, suggesting its pivotal role in RA pathogenesis through modulating immune cell functionality and orchestrating inflammatory responses.

This study identified two biomarkers, ADAMDEC1 and CXCL13, that were closely associated with KLRB1 in RA. Subsequently, we conducted GSEA analysis, inflammatory cytokine analysis, and immune microenvironment analysis on it, constructed a TF-mRNA-miRNA regulatory network, predicted targeted drugs, and revealed the functions and potential molecular mechanisms of biomarkers in RA patients. Finally, the differential expression of ADAMDEC1 and CXCL13 was confirmed in the synovial tissues of RA patients compared to a control group using RT-qPCR.

ADAMDEC1 (ADAM-like, decysin 1) belongs to the ADAM (disintegrin and metalloproteinase) family. It encodes a secreted metalloproteinase that is preferentially expressed in mature dendritic cells and macrophages⁴⁴. It plays a crucial role in various biological processes, including extracellular matrix remodeling, cell migration and infiltration, intercellular interactions, and immune regulation^45,46. The link between ADAMDEC1 and RA was initially reported in 2007⁴⁷, identifying ADAMDEC1 as the most significantly upregulated gene in synovial samples from RA patients, with a 23.8-fold increase. Subsequently, it was found that the expression of ADAMDEC1 was significantly higher in RA synovial fluid than in the control group⁴⁸. An increasing body of evidence indicates that ADAMDEC1 in synovium or synovial fluid is a reliable biomarker for RA^49,50. However, this study using the CIBERSORT algorithm revealed no significant differences in the infiltration of dendritic cells and macrophages between RA and control samples, presenting a contradiction to established conclusions^51,52. On one hand, this discrepancy may stem from the relatively small sample size of the GSE55235 dataset and the heterogeneity of RA patient conditions. This could have prevented the detection of subtle differences that may actually exist, or masked genuine variations in dendritic cell and macrophage infiltration due to individual variability. On the other hand, studies indicate that ADAMDEC1 is also expressed in both immune cell types during non-inflammatory, stable states ⁵³. This suggests it may play a role in maintaining the basal functions of immune cells and regulating immune homeostasis, with a lesser impact on the recruitment and infiltration of DCs and macrophages to inflammatory sites. This finding also provides direction for future research. Future studies could utilize single-cell sequencing technology to characterize specific subtypes of ADAMDEC1-positive cells in RA synovium, or conduct in vitro functional experiments to validate ADAMDEC1’s regulatory effects on the immune functions of dendritic cells and macrophages, thereby further elucidating its mechanism of action in RA.

CXCL13 (C-X-C Motif Chemokine Ligand 13), a regulator of B cell homing and activation, plays a pivotal role in immune modulation through its interaction with the CXCR5 receptor⁵⁴. Initially identified as a risk locus for rheumatoid arthritis (RA) in genome-wide association studies (GWAS)⁵⁵, elevated serum CXCL13 levels have been demonstrated in RA patients, with higher concentrations observed in those positive for rheumatoid factor (RF) and anti-citrullinated peptide antibodies (ACPA)⁵⁶. Plasma CXCL13 levels are significantly elevated in active RA patients compared to those in remission or healthy controls⁵⁷, establishing circulating CXCL13 as a novel biomarker for RA diagnosis and disease management⁵⁸. Notably, CXCL13 expression has been detected in the synovium and synovial fluid of RA patients^59–61, with plasma concentrations showing positive correlations with various clinical inflammatory parameters^62,63. In early RA, baseline serum CXCL13 levels predict increased joint damage progression over a 7-year follow-up period⁶⁴, positioning CXCL13 as a prognostic indicator for severe and aggressive disease⁵⁹. Therapeutic targeting of CXCL13 has shown promise in collagen-induced arthritis (CIA) models, where both prophylactic and therapeutic administration of CXCL13-neutralizing antibodies effectively suppressed disease progression, reduced joint inflammation, and attenuated cartilage damage^59,65. These were consistent with our results. In our results, ADAMDEC1 and CXCL13 were identified as biomarkers for RA, and their expression levels were significantly increased in synovial tissue of RA patients during experimental validation.

The GSEA results showed that these two genes were co-enriched in "cytoskeleton in muscle cells" and “motor proteins” pathways, both of which are critically implicated in RA pathogenesis. The metazoan cytoskeleton, comprising actin filaments, microtubules, and intermediate filaments⁶⁶, governs cellular morphology, extracellular communication, and mechanotransduction. Motor proteins—myosin, kinesin, and dynein—generate mechanical forces through ATP hydrolysis, facilitating fundamental cellular processes including organelle transport, cell motility, and division⁶⁷. These structural and motile systems collectively mediate RA-related pathological mechanisms. Neutrophil extracellular traps (NETs), recognized for their pro-inflammatory roles in RA⁶⁸, require cytoskeletal regulation of membrane integrity and cellular deformation for formation⁶⁶. T cell migration in RA involves cytoskeleton-mediated perinuclear mitochondrial positioning to drive cellular polarization⁶⁹. Synovial fibroblasts (FLS) exhibit enhanced invasiveness through motor protein MYO1C-driven cytoskeletal remodeling⁷⁰. Furthermore, dynein-mediated contractility facilitates antigen clustering during B cell activation, modulating adaptive immune responses^71,72. These observations suggest synergistic regulation of cytoskeletal and motor protein pathways by ADAMDEC1 and CXCL13 in RA pathophysiology.

Immunological profiling identified eight differentially abundant immune cell populations between RA and controls: neutrophils, plasma cells, activated memory CD4 + T cells, CD8 + T cells, follicular helper T cells, activated mast cells, activated NK cells, and resting memory CD4 + T cells. Both ADAMDEC1 and CXCL13 demonstrated strongest positive correlations with plasma cells and negative associations with activated mast cells. Plasma cells, terminally differentiated antibody-secreting B cells, contribute to RA pathogenesis through autoantibody production (ACPA/RF)⁷³. Daratumumab has been found to effectively deplete plasma cells in peripheral blood mononuclear cells (PBMCs) of RA patients in a dose-dependent manner in vitro⁷⁴. In contrast, CD20 targeted therapies such as rituximab, which indirectly reduce plasma cell production by consuming precursor B cells, have already achieved sustained autoantibody reduction in clinical practice⁷⁵. Mechanistically, CXCL13 enhances BCR-mediated activation and plasma cell differentiation via CXCL13/CXCR5 signaling, a process dependent on cytoskeletal reorganization⁷⁶. Single-cell analyses identify T peripheral helper (Tph) cells as the primary CXCL13 source in RA synovium⁷⁷, with in vitro evidence demonstrating Tph-induced plasma cell differentiation from memory B cells⁶¹. Although ADAMDEC1’s role in plasma cell biology remains underexplored, its elevated expression in Gr-1⁻ monocytes correlates with lupus nephritis severity⁷⁸, suggesting potential pro-inflammatory modulation of B cell responses in autoimmunity. Paradoxically, while activated mast cells generally promote synovitis and joint destruction, emerging evidence highlights their context-dependent immunoregulatory functions⁷⁸. Accordingly, ADAMDEC1 and CXCL13 may enhance the autoimmune response of RA by synergistically promoting plasma cell differentiation and autoantibody production (positive correlation), while their negative correlation with activated mast cells suggests that they may mediate the dual contradictory function of mast cells in RA, which combines pro-inflammatory and immune regulation.

Drug prediction analysis identified metronidazole as a dual-targeting agent for ADAMDEC1 and CXCL13. This nitroimidazole antibiotic demonstrated modest clinical improvement in early RA trials, potentially via gut microbiota modulation^79,80. However, lack of laboratory parameter improvement and unacceptable toxicity halted its development⁸¹. Recent insights into gut-joint axis interactions and metabolomic advances may revive interest in repurposing this agent.

In conclusion, we identified ADAMDEC1 and CXCL13 as KLRB1-associated biomarkers in RA. While KLRB1-CXCL13 synergy in regulating Treg-mediated immunosuppression has been documented in juvenile idiopathic arthritis⁸², the interplay between KLRB1 and ADAMDEC1 remains enigmatic. Although our bioinformatics approach provides novel biomarker insights validated by RT-qPCR in clinical samples, limitations persist. Larger clinical cohorts are required for robust validation, and mechanistic studies elucidating biomarker functions in RA pathogenesis merit dedicated investigation.

Author contributions

Conceptualization: JS; formal analysis and investigation: JS and JL; Methodology: JL; data curation: HZ and FS; writing—original draft preparation: JS and JL; writing—review & editing: JS and WZ; supervision: WZ and JZ; project administration: JZ. All authors reviewed the manuscript.

Funding

This study was supported by the Natural Science Foundation of Anhui Province (Grant No. 2208085MH212) and Youth Talent Fund Project of Lianyungang First People’s Hospital (Grant No. QN2304).

Data availability

The datasets analysed during the current study are available in the Gene Expression Omnibus (GEO) database repository, [https://www.ncbi.nlm.nih.gov/geo/].

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

Not applicable.

Consent for publication

Not applicable.

Clinical trial number

Not applicable.