Identification and Preliminary mRNA Validation of Key Genes Associated with Tolerogenic Dendritic Cells in Keloid Based on Transcriptome Data

Abstract

Purpose

Keloid is a cutaneous fibrotic disorder characterized by excessive fibrous tissue proliferation during wound healing. Tolerogenic dendritic cells (tolDCs) may participate in keloid immunopathology, but the related molecular signatures remain unclear. This study aimed to identify candidate tolDC-associated genes in keloid by integrating tolDC-related transcriptomic profiles with keloid transcriptomic datasets.

Patients and Methods

Public keloid and tolDC transcriptome datasets were analyzed. Candidate genes were screened by intersecting keloid differentially expressed genes (DEGs) with tolDC-related DEGs. Key genes were selected using protein-protein interaction analysis, machine learning algorithms, expression comparison, and ROC analysis. An ANN model based on the selected genes was constructed and validated. Enrichment analysis, subcellular localization, immune infiltration analysis, regulatory network construction, drug prediction, and molecular docking were performed to explore their potential biological relevance. The expression of key genes was preliminarily validated at the mRNA level in clinical samples by RT-qPCR.

Results

A total of 2,182 keloid DEGs and 3,541 tolDC-related genes were intersected, and CCL20 and NLRC4 were identified as candidate key genes. The ANN model showed potential diagnostic value for distinguishing keloid from control samples. CCL20 and NLRC4 were mainly localized in the cytoplasm and were associated with pathways including Reactome antimicrobial peptide signaling. Immune infiltration analysis showed associations with six immune cell types, including eosinophils and memory B cells. Regulatory network analysis identified 31 miRNAs potentially targeting these genes, and drug prediction suggested associations with 79 chemical compounds. RT-qPCR analysis in clinical samples supported the bioinformatics findings at the mRNA level.

Conclusion

This study identified CCL20 and NLRC4 as candidate tolDC-associated genes in keloid. These genes may provide candidate diagnostic biomarkers and hypotheses for future immunomodulatory therapeutic targets in keloid.

Introduction

Keloid disorder is a fibroproliferative skin disease characterized by excessive collagen deposition and dermal fibrosis after injury. Unlike hypertrophic scars, keloids extend beyond the wound and persist as raised, thickened lesions.¹ Common types include classic keloids, hypertrophic scars, acne keloidalis, and keloidalis nuchae.² Their incidence shows marked ethnic variation, ranging from 4.5–16% in Black and Hispanic populations but less than 1% in Caucasians.³ Clinically, keloids impair cosmetic appearance, restrict mobility, and cause pain, pruritus, and psychological distress.^4,5 Current treatments—such as intralesional corticosteroids, silicone sheeting, cryotherapy, and surgical excision—can reduce scar size or symptoms, yet recurrence is frequent and no single therapy is universally effective.⁶ Currently, the etiology of keloid formation remains unclear, and treatment responses vary considerably, with current treatments benefiting only a subset of patients.^7,8 Moreover, therapeutic agents and adjuvant modalities are associated with significant adverse effects, including skin atrophy,⁹ pigmentation changes,¹⁰ severe pain, blistering, and radiation dermatitis,¹¹ which restrict safe use and compromise long-term outcomes. Consequently, existing regimens fail to resolve the health burden keloids impose.¹² Therefore, it is essential to deepen our understanding of keloid diagnosis, etiology, and pathogenesis,¹³ and to identify key genes and molecular pathways that may serve as potential targets—which is helpful for improving the diagnosis, treatment, and disease prevention of keloids.¹⁴

Tolerogenic dendritic cells (tolDCs) are specialized antigen-presenting cells that promote immune tolerance by suppressing effector T-cell responses and inducing regulatory T cells.¹⁵ They can be generated both in vitro and in vivo and are recognized as crucial modulators in maintaining immune homeostasis.¹⁶ Increasing evidence highlights their involvement in skin-related disorders.¹⁷ For example, in systemic sclerosis, a fibrotic autoimmune disease sharing pathological similarities with keloid disorder, tolDC dysfunction contributes to chronic inflammation, fibroblast activation, and excessive extracellular matrix deposition.¹⁸ Similarly, ultraviolet radiation—induced tolDCs have been implicated in cutaneous immune tolerance,¹⁹ while redox regulation has been shown to influence their stability and functional persistence.²⁰ These findings suggest that tolDCs may also participate in the immune dysregulation underlying keloid formation.²¹ However, despite their recognized role in skin fibrosis and immune-mediated diseases, direct gene-level evidence linking tolDCs to keloid disorder remains limited.²² Current studies focus largely on autoimmune and fibrotic contexts,^19,23,24 leaving a gap in understanding of how tolDC-associated genes regulate keloid pathogenesis.²⁵ Therefore, exploring the molecular mechanisms by which tolDCs exert protective or pathogenic functions in keloid disorder, particularly at the genetic level, is essential to advancing diagnostic, preventive, and therapeutic strategies for this challenging condition.

Previous transcriptomic and machine-learning-based studies have identified several molecular biomarkers in keloids and provided valuable insights into keloid diagnosis and pathogenesis. For example, Sun et al identified ferroptosis-related molecular signatures in keloids based on multiple transcriptomic datasets, while Nurzati et al reported keloid-related miRNA–mRNA regulatory networks and potential diagnostic biomarkers using GEO datasets.^26,27 However, most existing biomarker studies have focused on broad immune-, fibrosis-, ferroptosis-, or miRNA-related signatures, whereas the potential contribution of tolDC-related molecular programs to keloid immunopathology remains insufficiently explored.

In this study, we utilized transcriptomic datasets of keloid from public databases and integrated them with tolDC-related transcriptomic profiles to identify candidate molecular associations between tolDCs and keloid pathogenesis. Unlike previous keloid biomarker studies that mainly focused on broad immune- or fibrosis-related signatures, this study specifically focused on tolDC-associated genes that may be involved in keloid immunopathology. Through bioinformatics approaches, we screened candidate tolDC-associated key genes in keloid. Based on these findings, we constructed an artificial neural network (ANN) diagnostic model for keloid and evaluated its predictive accuracy using ROC analysis. Furthermore, we explored the correlation and subcellular localization of the key genes, performed functional enrichment analysis, assessed immune cell infiltration, and built regulatory and drug–gene interaction networks, followed by molecular docking to predict potential therapeutic compounds. These analyses provide exploratory insights into tolDC-related molecular signatures in keloid and may help establish a basis for developing candidate diagnostic biomarkers and future immunomodulatory therapeutic strategies.

Materials and Methods

Data Sources

The research team obtained two keloid-related transcriptome datasets from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo).²⁸ Specifically, dataset GSE185309 (platform: GPL24676) included 9 skin tissue samples from keloid patients (keloid group) and 8 normal control skin tissue samples (normal group). Dataset GSE158395 (platform: GPL24676) consisted of 7 skin tissue samples from keloid patients (keloid group) and 6 normal control skin tissue samples (normal group). TolDC-associated genes were retrieved via differential expression analysis, drawing on four tolDC-related datasets: GSE182528 (platform: GPL570), GSE23371 (platform: GPL570), GSE52894 (platform: GPL10558), and GSE56017 (platform: GPL570).

Differential Expression Analysis and Identification of Candidate Genes

To identify DEGs associated with keloid pathogenesis, we performed differential expression analysis between keloid specimens and control specimens in dataset GSE185309 by employing the DESeq2 package.²⁹ The significance thresholds were set as |log2 fold change (FC)| > 0.5 and p-value < 0.05. A volcano plot was generated using the ggplot2 package (v 3.5.1)³⁰ to visualize DEGs, with labels for the top 10 most significantly upregulated and downregulated DEGs. Additionally, a heatmap illustrating the expression levels of the top 25 upregulated and top 25 downregulated DEGs was constructed using the Heatmap package (v 2.21.1).³¹

Similarly, to get tolDC-related genes, we performed differential expression analysis on the four tolDC-related datasets with the limma package (v 3.56.2)³² utilized, adhering to the same significance thresholds of |log2 FC| > 0.5 and p-value < 0.05. Through the use of the ggplot2 package (v 3.5.1) and Heatmap package (v 2.21.1), volcano plots and heatmaps were generated, respectively. Subsequently, duplicate entries were removed from the upregulated and downregulated DEGs derived from the four datasets to determine the final set of tolDC-associated DEGs.

Finally, to identify tolDC-related genes involved in keloid pathogenesis, the intersections between upregulated DEGs of keloids and upregulated tolDC-associated DEGs, as well as between downregulated DEGs of keloids and downregulated tolDC-associated DEGs, were determined using the ggvenn package (v 0.1.10).³³ The intersecting genes were designated as candidate genes.

Functional Exploration of Candidate Genes and Identification of Hub Genes

Later, to ascertain the functions and signaling pathways linked to candidate genes in disease pathogenesis, we performed Gene Ontology (GO) enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis by employing the clusterProfiler package (v 4.1.5),³⁴ with a significance threshold set at p-value < 0.05. GO analysis typically describes gene functions at three levels: biological process (BP), cellular component (CC), and molecular function (MF). We selected 15 leading significantly enriched GO terms and 10 leading KEGG pathways for visualization.

Moreover, we employed the STRING online database (https://string-db.org)³⁵ to build a protein-protein interaction (PPI) network of candidate genes at the protein level (interaction score > 0.4). Isolated nodes were first removed, after which the network was visualized using Cytoscape software (v 3.8.2). Finally, we identified the top 50 genes in the PPI network by means of three algorithms in the cytoHubba plugin: degree, maximum neighborhood component (MNC), and maximum clique centrality (MCC), each used individually. The intersections among the top 50 genes from the three algorithms were pinpointed via the ggvenn package (v 0.1.10), with these overlapping genes designated as hub genes.

Identification of Key Genes

To identify key genes, machine learning analyses were performed on the hub genes using dataset GSE185309. First, least absolute shrinkage and selection operator (LASSO) regression analysis was conducted using the glmnet package (v 4.1–8)³⁶ with 10-fold cross-validation. The optimal lambda value was determined, and feature genes were selected at the minimum error point. Meanwhile, support vector machine-recursive feature elimination (SVM-RFE) analysis of the hub genes was performed using the caret package (v 2.0.3.1).³⁷ The combination with the highest accuracy and lowest error rate was selected to determine the required number of features for screening feature genes. Next, the ggvenn package (v 0.1.10) served to find the overlap of feature genes obtained via the two machine learning algorithms, resulting in candidate key genes. Next, we carried out Wilcoxon rank-sum tests for the candidate key genes in both datasets GSE185309 and GSE158395 to compare their expression differences and trends (p < 0.05). We plotted receiver operating characteristic curves (ROC) of the candidate key genes using the pROC package (v1.18.5).³⁸ An AUC value greater than 0.7 was used as a preliminary screening threshold because AUC values of 0.7–0.8 are commonly interpreted as indicating acceptable discriminatory performance in diagnostic model assessment.³⁹ Ultimately, genes were classified as key genes only if they met all of the following criteria: AUC > 0.7 in ROC analysis, statistically significant differential expression between keloid and control groups, and consistent expression trends across the two independent datasets. This combined screening strategy was used to improve the robustness of candidate gene selection and to avoid relying on ROC performance alone.

Construction of an Artificial Neural Network (ANN) Model Based on Key Genes and Functional Enrichment Analysis

To assess the capacity of the screened key genes to predict keloid pathogenesis, we constructed an ANN model with the neuralnet package (v 1.44.2),⁴⁰ using dataset GSE185309. Afterward, ROC curves of the ANN model were generated via the pROC package (v 1.18.5) for datasets GSE185309 and GSE158395. An AUC value greater than 0.7 was considered to indicate acceptable predictive performance of the ANN model, consistent with the criterion used for candidate gene screening.³⁹

Moreover, for exploring the biological functions of key genes involved in keloid pathogenesis, we downloaded the dataset c2.cp.all.v2024.1.Hs.symbols.gmt from the Molecular Signatures Database (MSigDB)⁴¹ to serve as the background gene set. For dataset GSE185309, we calculated Spearman’s rank correlation coefficients between each key gene and all other genes using the cor function from the stats package (v 4.3.1),⁴² a then ranked all genes in descending order according to these coefficients. We then performed gene set enrichment analysis (GSEA) using the clusterProfiler package (v 4.1.5), with significance thresholds set to |normalized enrichment score (NES)| > 1, false discovery rate (FDR) < 0.25, and p < 0.05. Finally, the top 10 significantly enriched pathways were visualized.

Correlation Analysis, Subcellular Localization Prediction, and Molecular Regulation of Key Genes

For investigating the interaction relationships between key genes during keloid pathogenesis, Spearman correlation analysis (|cor| > 0.3, p < 0.05) on the key genes was conducted with the psych package (v 2.4.6) (https://CRAN.R-project.org/package=psych,⁴³ utilizing dataset GSE185309. We predicted the subcellular localization of the key genes using the mRNALocater database.⁴⁴ Additionally, we used the miRWalk database (https://mirwalk.umm.uni-heidelberg.de/)⁴⁵ to predict miRNAs targeting the key genes, aiming to explore the molecular interactions between key genes and non-coding RNAs. We selected the top 30 miRNAs with the highest scores and then visualized the regulatory network using Cytoscape software (v 3.8.2).

Immune Microenvironment Analysis

The immune microenvironment plays a crucial role in disease pathogenesis. Through the use of the single-sample gene set enrichment analysis (ssGSEA) method incorporated in the GSVA package (v 1.53.28),⁴⁶ we determined the enrichment scores of 28 immune cell types in keloid and control samples from dataset GSE185309. To visually show these scores, a heatmap was generated via the pheatmap package (v 1.0.12).³² Moreover, Wilcoxon tests were carried out to compare immune cell infiltration score differences between the keloid and control groups in GSE185309 (p < 0.05), with the results visualized via the ggplot2 package (v 3.5.1). Subsequently, using the cor function from the psych package (v 2.4.6), Spearman correlation analysis was conducted to assess the correlations both among differentially infiltrated immune cells and between key genes and differentially infiltrated immune cells (|cor| > 0.3, p < 0.05).

Drug Prediction and Molecular Docking

To investigate potential therapeutic agents for keloids, we predicted drugs targeting each key gene using the Drug Signatures Database (DSigDB) (https://dsigdb.tanlab.org/DSigDBv1.0/),⁴⁷ and visualized the drug-key gene interaction network with Cytoscape software (v 3.8.2).

Subsequently, the interactions between key genes and their targeted drugs were investigated. First, we downloaded the three-dimensional (3D) structures of proteins encoded by key genes from the Protein Data Bank (PDB) (https://www.rcsb.org/).⁴⁸ At the same time, the 3D structures (SDF format) of selected drugs (combined.score > 10,000) were fetched from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/).⁴⁹ These structures were processed using AutoDock (v 4.2.6) and AutoGrid tools. Finally, molecular docking simulations were performed, and the docking results were visualized using PyMOL software (v 3.1.3). In the molecular docking analysis, lower predicted binding energy values were interpreted as indicating more favorable ligand–target interactions. Because docking scores are computational estimates and may vary across software, targets, and ligands, the results were interpreted cautiously. Based on previous molecular docking and virtual screening studies, a predicted binding energy lower than −5 kcal/mol was used as an inclusive preliminary screening criterion for potential ligand–target interactions rather than as definitive evidence of strong binding.^50,51 Therefore, the docking results were considered exploratory computational predictions requiring further experimental validation.

The Reverse Transcription Quantitative PCR (RT-qPCR)

In this study, we acquired skin samples from keloid patients (n = 5) and controls (n = 5) at Beijing Hospital. From each skin tissue sample, total RNA was purified using Trizol (Ambion), followed by reverse transcription into complementary DNA (cDNA) for qPCR with HP All-in-one qRT Master Mix II (RT203-Ver.1). 2 × Universal Blue SYBR Green qPCR Master Mix was utilized for RT-qPCR assays. Key gene primers were manufactured by Sangon Biotech (Shanghai, China) (Table S1). Three technical replicates were performed for each biological sample. The use of these human tissue specimens for RT-qPCR validation was reviewed and approved by the Research Ethics Committee of Beijing Hospital (Approval No. 2026BJYYEC-KY262-01), and the requirement for written informed consent was waived. The GAPDH gene served as an endogenous reference gene. The resultant data were statistically analyzed and described by the Graphpad Prism package (v 10.0).⁵²

Statistical Analysis

Statistical analyses were executed with R software (v 4.3.3). Analysis of differences between groups was executed via the Wilcoxon test. Statistical significance was defined as a p-value lower than 0.05. Additionally, the differences in RT-qPCR experimental results distinctions between groups were assessed by applying the t-test.

Results

A Total of 158 Candidate Genes Were Obtained

Within dataset GSE185309, we identified 2,182 DEGs in keloid samples relative to normal samples; this set of genes included 1,116 upregulated ones and 1,066 downregulated ones. The top 10 upregulated DEGs were NLGN4X, PRKY, HORMAD1, CCL5, SLC30A8, PRR9, FCGBP, DMGDH, PLA2G4C, and MME; the top 10 downregulated DEGs were MT1H, PNMA8A, MYCBPAP, TCEAL7, ANPEP, LOC105372981, RIBC2, CRISPLD1, ROS1, and SLC47A1 (Figure 1A). For the top 50 upregulated DEGs and top 50 downregulated DEGs, their expression levels showed distinct differences between keloid and normal samples (Figure 1B).

Figure 1.

Identification of tolDC-related candidate genes in keloid pathogenesis. (A) Volcano plot showing differentially expressed genes between keloid and normal tissues in the GSE185309 dataset. (B) Heatmap showing representative differentially expressed genes between keloid and normal tissues in the GSE185309 dataset. (C–F) Volcano plots showing differentially expressed genes between tolDCs and control groups in the GSE182528 dataset (C), GSE56017 dataset (D), GSE23371 dataset (E), and GSE52894 dataset (F). Red dots indicate upregulated genes, blue dots indicate downregulated genes, and gray dots indicate genes without significant differential expression.

Subsequently, differential expression profiles of the tolDC-related datasets were obtained separately: GSE182528 contained only 7 upregulated DEGs with no downregulated DEGs (Figure 1C and Supplementary Figure 1A); GSE56017 included 19 upregulated DEGs and 2 downregulated DEGs (Figure 1D and Supplementary Figure 1B); GSE23371 comprised 178 upregulated DEGs and 365 downregulated DEGs (Figure 1E and Supplementary Figure 2); and GSE52894 contained 1,837 upregulated DEGs and 1,449 downregulated DEGs (Figure 1F and Supplementary Figure 3A).

After removing duplicate genes, the tolDC-associated DEGs finally included 1,947 upregulated DEGs and 1,594 downregulated DEGs. Furthermore, the intersection of 1,116 upregulated DEGs from dataset GSE185309 and 1,947 upregulated DEGs from the tolDC-associated DEGs yielded 84 overlapping genes. The intersection of 1,066 downregulated DEGs from GSE185309 and 1,594 downregulated DEGs from the tolDC-related set yielded 74 overlapping genes. In total, 158 candidate genes were obtained (Supplementary Figure 3B).

Functional Analysis of Candidate Genes and Screening of 39 Hub Genes

Analysis of GO terms for the 158 candidate genes revealed 399 functional terms that met the significance criterion (p < 0.05). Among the 323 enriched BPs, these candidate genes primarily participated in pathways including mitotic cell cycle phase transition, cellular response to inorganic substances, cellular response to metal ions, carbohydrate derivative transport, and dicarboxylic acid metabolic process. For the 20 enriched CCs, the candidate genes were significantly enriched in cell leading edge, basal part of cell, early endosome, basal plasma membrane, and ruffle membrane. In 56 enriched MFs, the candidate genes were notably enriched in amide binding, active transmembrane transporter activity, GTPase activator activity, ATPase-coupled transmembrane transporter activity, and amide transmembrane transporter activity (Figure 2A and Table S2).

Figure 2.

Candidate gene enrichment analysis and PPI network construction (A) Gene Ontology (GO) enrichment analysis. (B) KEGG pathway enrichment analysis. (C) Protein–Protein Interaction (PPI) Network. (D) Venn diagram of the intersection of hub genes.

Besides, the 158 candidate genes displayed significant enrichment in 12 KEGG pathways (p < 0.05), among which were mineral absorption, ABC transporters, antifolate resistance, folate transport and metabolism, and oocyte meiosis (Figure 2B and Table S3). At the protein level, no relevant protein data were identified for three genes, the PPI network consisted of protein targets encoded by 155 genes, with a total of 144 regulatory interactions. Proteins encoded by genes such as LMNB1, ZWILCH, ECT2, FBXO5, and EZH2 interacted frequently with those encoded by other genes (Figure 2C).

Later, we detected 39 overlapping genes among the top 50 genes selected through the Degree, MCC, and MNC algorithms, which were named hub genes (Figure 2D).

CCL20 and NLRC4 Were Identified as Key Genes for Keloids

First, in the LASSO analysis, the model exhibited the minimal error when log(lambda, min) = −3.75, leading to the selection of 6 feature genes: CCL20, CRABP2, CYP27A1, GDF15, LRP1, and NLRC4 (Supplementary Figure 4A and 4B).

Meanwhile, the SVM-RFE model achieved the optimal combination of error rates and the highest predictive accuracy when the number of feature genes was set to 20. The 20 feature genes selected were LRP1, ABCC3, ABCG1, CCL20, ALDH5A1, CRABP2, NLRC4, CMPK2, GDF15, CPT1A, ANPEP, PARPBP, TRIB1, PCK2, POLA2, EZH2, MT1F, PICALM, PTTG1, and BTG2 (Supplementary Figure 4C and 4D). The intersection of feature genes from both algorithms yielded 5 candidate key genes: CCL20, CRABP2, GDF15, LRP1, and NLRC4 (Figure 3A). Subsequently, among these 5 candidate key genes, only CCL20 and NLRC4 showed significant differential expression between the keloid and normal groups in both datasets GSE185309 and GSE158395, with upregulated expression in the disease group (p < 0.05) (Figure 3B and C). ROC analysis revealed that both CCL20 and NLRC4 had AUC values > 0.9 in both datasets (Figure 3D and E). Therefore, CCL20 and NLRC4 were ultimately identified as the key genes in this study.

Figure 3.

Identification and validation of diagnostic feature genes. (A) Venn diagram showing the overlapping feature genes identified by LASSO regression and SVM-RFE algorithms. (B and C) Expression levels of candidate key genes in the validation dataset GSE158395 (B) and training dataset GSE185309 (C). (D and E) ROC curves evaluating the diagnostic performance of CCL20 and NLRC4 in the validation dataset GSE158395 (D) and training dataset GSE185309 (E).

Key Genes Exhibited Good Accuracy in Disease Prediction and Were Involved in Multiple Functional Pathways

An ANN model was constructed using the key genes CCL20 and NLRC4 (Figure 4A). In datasets GSE185309 and GSE158395, the AUC values under the ROC curves of the ANN model were 0.986 and 0.928, respectively, indicating the ANN model excellent predictive accuracy (Figure 4B and C). These analyses validated the accurate predictive ability of the screened key genes for keloid pathogenesis. GSEA showed that CCL20 was enriched in 847 pathways (Figure 4D and Table S4), with the top 5 being medicus reference translation initiation, wp interactions between immune cells and microRNAs in tumor microenvironment, metabolism of xenobiotics by cytochrome p450, wp cytoplasmic ribosomal proteins, and medicus reference regulation of GF-RTK-Ras-ERK signaling pathway adaptor proteins. NLRC4 was significantly enriched in 676 signaling pathways (Figure 4E and Table S5), and the top 5 were sequentially medicus reference autophagy vesicle nucleation elongation maturation sequestosome 1-like receptor, reactome regulation of mitf m dependent genes involved in lysosome biogenesis and autophagy, medicus reference tbk1 mediated autophagosome formation, wp hypertrophy model, and reactome tbc rabgaps. Notably, among the top 10 enriched pathways, CCL20 and NLRC4 were co-enriched in medicus reference tbk1 mediated autophagosome formation.

Figure 4.

Construction and evaluation of an artificial neural network (ANN) model and functional enrichment analysis of CCL20 and NLRC4. (A) ANN model constructed based on CCL20 and NLRC4. (B and C) R OC curves evaluating the diagnostic performance of the ANN model in the training set (B) and validation set (C). (D and E) GSEA enrichment plots showing pathways associated with CCL20 and NLRC4.

Interaction Between the Two Key Genes and Their Molecular Regulatory Mechanisms

For the two key genes, a significant positive correlation was detected (cor = 0.57, p < 0.05), indicating that their correlation degree is high (Figure 5A). Through subcellular localization analysis, it was found that CCL20 and NLRC4 were most concentrated in the cytoplasm, with the endoplasmic reticulum and nucleus coming next in terms of abundance (Figure 5B).

Figure 5.

Characterization of the CCL20–NLRC4 axis. (A) Heatmap of the correlation between CCL20 and NLRC4 expression.(B) Stacked plot of mRNA subcellular localization predicted by mRNALocater.(C) miRNA–gene regulatory network predicted by miRWalk.

Subsequently, 30 miRNAs were predicted to target CCL20, including hsa-miR-1225-5p, hsa-miR-1296-3p, hsa-miR-194-3p, hsa-miR-145-5p, hsa-miR-330-5p, hsa-miR-421 and hsa-miR-410-3p. In contrast, only one miRNA, hsa-miR-6742-5p, was predicted to target NLRC4 (Figure 5C).

The Pathogenesis of Keloids Was Regulated by Immune Cells and Key Genes

In dataset GSE185309, the infiltration levels of immune cells varied among different samples. Cells and eosinophils showed differential infiltration between the keloid and normal groups (Figure 6A). Among the 28 cell types, 6 immune cells exhibited differential expression between groups (p < 0.05) and eosinophils were significantly upregulated in keloid samples, while memory B cells, natural killer T cells, and type 2 T helper cells were significantly downregulated in keloid samples (Figure 6B). Among the 6 differentially infiltrated immune cells, memory B cells and type 2 T helper cells showed a significant positive correlation (cor = 0.877, p < 0.001), whereas effector memory CD8 T cells and memory B cells exhibited the strongest negative correlation (cor = −0.571, p < 0.01) (Figure 6C and Table S6). Besides, CCL20 and NLRC4 were negatively associated with natural killer T cells, memory B cells, and type 2 T helper cells (meeting the criterion of |cor| > 0.3), and positively associated with activated dendritic cells, effector memory CD8 T cells, and eosinophils (also satisfying |cor| > 0.3) (Figure 6D and Table S6).

Figure 6.

Analysis of immune cell infiltration and its correlation with keloid pathogenesis.(A) Heatmap of immune infiltration of 28 immune cell types in the training set.(B) Boxplots showing the proportions of immune cells in keloid versus normal tissues. (C) Correlation heatmap of differentially infiltrated immune cells. (D) Correlation heatmap between biomarkers and differentially infiltrated immune cells. Statistical significance is indicated as follows: ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001. Correlations were assessed using Spearman correlation analysis.

Binding Capacity of CCL20 and NLRC4 with Related Drugs

Based on the DSigDB database, 72 drugs were predicted to target CCL20, such as quercetin, gossypol, clenbuterol, and luteolin, while 7 drugs were predicted to target NLRC4, including iron, trimethoprim, alpha-Neu5Ac, and silver (Figure 7A). The binding energy between CCL20 and 3-nitrofluoranthene was the lowest at −6.95 kcal/mol, suggesting a favorable predicted ligand–target interaction between them, and the docking conformation of the drug and target protein is shown in Figure 7B. The binding energy between NLRC4 and the predicted drugs was greater than −5 kcal/mol, suggesting a relatively weak binding capacity.

Figure 7.

Prediction and validation of drug–gene interactions. (A) Predicted gene–drug interactions visualized by Cytoscape based on DSigDB analysis (B) Molecular docking result of CCL20 with 3-nitrofluoranthene visualized by AutoDock.”.

Expression Analysis of Key Genes

RT-qPCR data indicated that expression of CCL20 and NLRC4 was found to be upregulated in keloid skin tissues compared with normal group samples, showing significant differences (p < 0.05). The expression trends and significant differences of the two genes were consistent with the conclusions drawn from bioinformatics analysis, supporting the consistency of the bioinformatics findings at the mRNA level (Figure 8).

Figure 8.

Validation of CCL20 and NLRC4 expression in keloid tissues. (A) CCL20 mRNA expression (B) NLRC4 mRNA expression. Data are presented as mean ± SD. Comparisons between two groups were performed using an unpaired two-tailed Student’s t-test. Statistical significance is indicated as follows: ns, not significant; *P < 0.05; **P < 0.01.

Validation of key gene expression by RT-qPCR. The relative mRNA expression levels of CCL20 and NLRC4 were significantly upregulated in keloid tissue samples compared with normal controls (n = 5 per group, p < 0.05). The trends were consistent with bioinformatics analysis, supporting the robustness of our computational predictions.

Discussion

Keloid is a fibroproliferative skin disorder characterized by excessive extracellular matrix deposition during wound healing, resulting in raised, invasive scars that can cause pain, pruritus, functional limitations, and psychological burden.^26,53–55 TolDCs are a specialized subset of dendritic cells that promote immune tolerance by suppressing effector T-cell activity and inducing regulatory T-cell differentiation.^15,20 TolDCs have been implicated in diverse immune-mediated diseases, including autoimmunity, transplantation, cancer, and fibrotic disorders,^15,16,20,56 suggesting their potential relevance to keloid pathogenesis. TolDCs have been shown to regulate immune responses by inducing T cell tolerance and suppressing fibrotic pathways, as demonstrated in models of skin tolerance,⁵⁷ UV-induced immunoregulation,¹⁹ hapten-induced tolerance,⁵⁸ and transplantation tolerance.⁵⁹ Given the parallels between keloid fibrosis and other autoimmune or fibrotic diseases such as scleroderma, where tolDC-based therapies are being explored, it is plausible that impaired tolDC function may contribute to aberrant immune activation and excessive extracellular matrix deposition in keloids.²³ Building on this concept, our study integrated public transcriptomic datasets with tolDC-related gene sets and identified two candidate tolDC-associated genes, CCL20 and NLRC4, that may be related to keloid immunopathology. The artificial neural network model based on these genes demonstrated potential diagnostic performance. Through enrichment, immune infiltration, subcellular localization, correlation, molecular regulatory network, and drug prediction analyses, we found that the key genes were predominantly localized in the cytoplasm and were associated with pathways such as TBK1-mediated autophagosome formation and antimicrobial peptide signaling. We further identified six dysregulated immune cell subsets associated with CCL20 and NLRC4, as well as miRNAs and candidate drugs potentially related to these genes. Importantly, these findings should be interpreted as associative and hypothesis-generating, because the current study was based mainly on transcriptomic integration and preliminary mRNA-level validation rather than direct functional evidence in tolDC models or keloid tissues. Overall, these results provide exploratory insights into tolDC-related molecular signatures in keloid and offer candidate biomarkers for further validation, rather than establishing causal mechanisms.

In this study, the key genes CCL20 and NLRC4 were identified as candidate tolDC-associated genes in keloid. CCL20 (chemokine C-C motif ligand 20), a chemokine encoded on chromosome 2q35, is normally expressed at low levels but can be strongly induced under inflammatory conditions.^60,61 It may contribute to keloid development by shaping the inflammatory microenvironment and driving aberrant fibroblast activation.⁶² It specifically recruits CCR6⁺ Th2 cells and eosinophils, which release cytokines such as IL-4, IL-13, and profibrotic mediators, thereby enhancing collagen synthesis and extracellular matrix deposition.^63–65 In addition, CCL20-induced immune responses stimulate the TGF-β/Smad pathway in fibroblasts^66–68 and promote angiogenesis,^69,70 further sustaining scar overgrowth.⁷¹ Consistent with previous reports, our mRNA-level validation showed upregulation of CCL20 in keloid tissues, supporting its potential value as a candidate marker, while the causal role of the CCL20/CCR6 axis in keloid progression requires further functional confirmation. In contrast, CARD domain-containing protein 4 (NLRC4), located on chromosome 2p22.3, encodes a highly conserved protein widely expressed in both immune (eg, dendritic cells, T cells) and non-immune cells (eg, epithelial and hepatic cells), upon stimulation by infection- or damage-associated signals, its expression is upregulated^72,73 and NLRC4 assembles into an inflammasome complex,^74–76 triggering IL-1β and IL-18⁷⁷ release⁷⁸ and driving a chronic inflammatory microenvironment.⁷⁹ These inflammatory processes may be associated with fibroblast activation, ECM deposition, and TGF-β/Smad signaling in fibrotic contexts.^79–81 Furthermore, NLRC4-mediated pyroptosis amplifies inflammatory signaling through the release of DAMPs, sustaining persistent immune activation.^82–84 However, direct evidence linking NLRC4 to tolDC function or keloid-specific fibrosis remains limited. Therefore, our findings should be interpreted as suggesting an association between NLRC4 expression and keloid-related immune–fibrotic features, rather than proving that NLRC4 is a mechanistic driver of tolDC-mediated keloid progression.

GSEA enrichment analysis conducted in this study demonstrated that both CCL20 and NLRC4 were notably enriched in five signaling pathways. TBK1-mediated autophagosome formation points to immune–autophagy crosstalk: impaired autophagy can sustain inflammasome activity (NLRC4) and profibrotic signaling in fibroblasts, while TBK1 integrates innate sensing with tolerance programs in DCs.^85,86 Reactome antimicrobial peptide signaling may extend beyond host defense to shape local inflammation, and dysregulated antimicrobial peptide-related pathways could help maintain a CCR6/CCL20-biased inflammatory milieu that favors Th2-skewing and matrix deposition.^87,88 The immune-cell–miRNA interaction pathway suggests that miRNAs may tune CCL20/NLRC4 expression and tolDC differentiation, offering a layer of post-transcriptional control over immune tolerance and fibrosis.^89,90 GF-RTK–RAS–ERK adaptor regulation aligns with the hyperproliferative, collagen-synthetic phenotype of keloid fibroblasts and may converge with CCL20-TGF-β/Smad signaling to amplify ECM production.^91,92 Finally, Ca²⁺/NMDA-driven RAS activation provides a plausible stimulus-response route in dermal cells under injury or mechanical stress, reinforcing proliferative and profibrotic loops.^93,94 These enrichment results suggest possible biological pathways associated with CCL20 and NLRC4, but they do not establish direct causal involvement of these genes in tolDC function or keloid progression.

Analysis of the immune microenvironment in keloids highlighted multiple dysregulated immune cell subsets, including activated dendritic cells, effector memory CD8⁺ T cells, eosinophils, memory B cells, natural killer T cells, and Th2 cells, all of which are known to sustain chronic inflammation and drive fibrotic remodeling. Activated dendritic cells, for example, promote T-cell activation and cytokine release,^95,96 thereby amplifying local immune responses, whereas effector memory CD8⁺ T cells rapidly exert cytotoxic functions upon antigen re-exposure.^97,98 Eosinophils^99,100 and Th2 cells^101,102 contribute to pro-fibrotic cytokine networks, while memory B cells^103,104 and NKT cells^95,105 may further modulate adaptive and innate immune balance. Notably, our analysis revealed that CCL20 expression was closely associated with activated dendritic cells, while NLRC4 correlated with effector memory CD8⁺ T cells, suggesting possible gene–immune cell associations within the keloid microenvironment. However, these correlations do not demonstrate that CCL20 or NLRC4 directly regulate these immune cell subsets. These findings underscore a coordinated gene–cell network within the keloid niche and provide hypotheses for future cell-specific and functional studies.

Molecular regulatory network analysis further highlights several fibrosis-related miRNAs that may contribute to keloid pathogenesis. For instance, hsa-miR-145-5p has been shown to target fibroblast-specific protein 1 and COL1A1, thereby suppressing fibroblast proliferation and ECM synthesis,¹⁰⁶ suggesting its downregulation could favor scar overgrowth.¹⁰⁷ Similarly, hsa-miR-194-3p inhibits TGF-β signaling by targeting TGFBR1, implicating its loss in enhanced fibroblast activation and collagen deposition.^108,109 In contrast, reduced expression of hsa-miR-330-5p or hsa-miR-421 may impair apoptosis and ECM degradation, leading to sustained fibroblast activation and excessive matrix accumulation.^110,111 These findings support a miRNA-mediated regulatory axis in the fibrosis-immune crosstalk of keloids. Consistently, our drug prediction analysis identified 72 compounds potentially targeting CCL20 and 7 compounds for NLRC4, with 3-nitrofluoranthene showing the most favorable predicted binding energy to CCL20 (–6.95 kcal/mol). Although the predicted binding of NLRC4-related drugs was weaker, these results highlight candidate molecular targets and chemical modulators that require further experimental validation before being considered therapeutic strategies for keloid intervention.

In summary, this study identified CCL20 and NLRC4 as candidate tolDC-associated genes in keloid, both associated with immune- and fibrosis-related pathways and correlated with immune cell infiltration. These findings suggest potential associations between tolDC-related molecular signatures and the immune–fibrosis interplay in keloids, providing candidate biomarkers and hypotheses for future mechanistic studies. Nevertheless, several limitations should be acknowledged. The conclusions rely heavily on the quality of publicly available transcriptomic datasets, the assumptions embedded in bioinformatic algorithms, and the interpretability of predictive models. Given the inherent heterogeneity of biological systems, our current analytical work may fail to fully mirror the complexity in vivo. Moreover, the present study demonstrates associations rather than causality, and the connection between tolDCs and CCL20/NLRC4 has not been directly validated in functional tolDC models or keloid tissues. Validation was limited to mRNA expression in a small clinical sample set, without protein-level confirmation by Western blotting, immunohistochemistry, or immunofluorescence. Therefore, future studies should include larger independent cohorts, cell-specific expression analysis, protein-level validation, and functional experiments to determine whether CCL20 and NLRC4 directly regulate tolDC biology or immune–fibrotic remodeling in keloids.

Conclusion

Future research should focus on validating these results through larger independent clinical cohorts and experimental approaches, such as protein-level validation, animal modeling, and gene perturbation studies, to clarify the functional roles of CCL20 and NLRC4 and their associated pathways. In particular, future studies should prioritize testing whether CCL20 and NLRC4 are directly involved in tolDC-related immune regulation in keloid, whether their expression correlates with immune–fibrotic activity or recurrence risk, and whether modulation of these pathways can influence keloid progression in experimental models. Sustained investigation into the molecular mechanisms, immune regulatory networks, and drug–target interactions will deepen our understanding of keloid biology. Clinically, this two-gene model may provide a candidate molecular signature for improving keloid diagnosis, supporting recurrence-risk stratification, and guiding future immunomodulatory therapeutic strategies for refractory keloids. However, these potential applications remain hypothetical and require further validation before clinical use. Ultimately, these efforts may pave the way toward future precision-oriented therapeutic strategies tailored to immune–fibrotic dysregulation in keloid disease.

Funding Statement

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data Sharing Statement

The original transcriptomic datasets analyzed in this study are publicly available from the GEO database under accession numbers GSE185309, GSE158395, GSE182528, GSE56017, GSE23371, and GSE52894. The individual-level RT-qPCR validation data are not publicly deposited because they were derived from anonymized human tissue samples and are subject to institutional data-governance and privacy considerations. Additional information required to reproduce the analyses may be available from the corresponding author upon reasonable request, subject to institutional and ethical requirements.

Ethics Approval

The use of human keloid and control skin tissue samples for RT-qPCR validation was reviewed and approved by the Research Ethics Committee of Beijing Hospital (Approval No. 2026BJYYEC-KY262-01). The requirement for written informed consent was waived for the use of anonymized residual clinical tissue specimens, and no identifiable private information was involved. All procedures were conducted in accordance with the Declaration of Helsinki.

Author Contributions

Zhongyang Sun and Xuejie Gao: Writing – review & editing, Writing – original draft, Visualization, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Conceptualization. Rongxin Ren: Writing – review & editing, Validation, Supervision. Hongyi Zhao: Writing – review & editing, Validation, Supervision. All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The author(s) report no conflicts of interest in this work.